1efb80947Sahrens /* 2efb80947Sahrens * CDDL HEADER START 3efb80947Sahrens * 4efb80947Sahrens * The contents of this file are subject to the terms of the 5efb80947Sahrens * Common Development and Distribution License (the "License"). 6efb80947Sahrens * You may not use this file except in compliance with the License. 7efb80947Sahrens * 8efb80947Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9efb80947Sahrens * or http://www.opensolaris.org/os/licensing. 10efb80947Sahrens * See the License for the specific language governing permissions 11efb80947Sahrens * and limitations under the License. 12efb80947Sahrens * 13efb80947Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14efb80947Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15efb80947Sahrens * If applicable, add the following below this CDDL HEADER, with the 16efb80947Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17efb80947Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18efb80947Sahrens * 19efb80947Sahrens * CDDL HEADER END 20efb80947Sahrens */ 21efb80947Sahrens /* 22dc7cd546SMark Shellenbaum * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23ec5cf9d5SAlexander Stetsenko * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 24ca0cc391SMatthew Ahrens * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 25a2afb611SJerry Jelinek * Copyright (c) 2014, Joyent, Inc. All rights reserved. 26e77d42eaSMatthew Ahrens * Copyright 2014 HybridCluster. All rights reserved. 27880094b6SAndrew Stormont * Copyright 2016 RackTop Systems. 28c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 29ec5cf9d5SAlexander Stetsenko */ 30efb80947Sahrens 31efb80947Sahrens #include <sys/dmu.h> 32efb80947Sahrens #include <sys/dmu_impl.h> 33efb80947Sahrens #include <sys/dmu_tx.h> 34efb80947Sahrens #include <sys/dbuf.h> 35efb80947Sahrens #include <sys/dnode.h> 36efb80947Sahrens #include <sys/zfs_context.h> 37efb80947Sahrens #include <sys/dmu_objset.h> 38efb80947Sahrens #include <sys/dmu_traverse.h> 39efb80947Sahrens #include <sys/dsl_dataset.h> 40efb80947Sahrens #include <sys/dsl_dir.h> 4192241e0bSTom Erickson #include <sys/dsl_prop.h> 42efb80947Sahrens #include <sys/dsl_pool.h> 43efb80947Sahrens #include <sys/dsl_synctask.h> 44efb80947Sahrens #include <sys/zfs_ioctl.h> 45efb80947Sahrens #include <sys/zap.h> 46efb80947Sahrens #include <sys/zio_checksum.h> 47dc7cd546SMark Shellenbaum #include <sys/zfs_znode.h> 48cde58dbcSMatthew Ahrens #include <zfs_fletcher.h> 499e69d7d0SLori Alt #include <sys/avl.h> 508e714474SLori Alt #include <sys/ddt.h> 51c99e4bdcSChris Kirby #include <sys/zfs_onexit.h> 523b2aab18SMatthew Ahrens #include <sys/dmu_send.h> 533b2aab18SMatthew Ahrens #include <sys/dsl_destroy.h> 545d7b4d43SMatthew Ahrens #include <sys/blkptr.h> 5578f17100SMatthew Ahrens #include <sys/dsl_bookmark.h> 565d7b4d43SMatthew Ahrens #include <sys/zfeature.h> 57a2cdcdd2SPaul Dagnelie #include <sys/bqueue.h> 58efb80947Sahrens 5919b94df9SMatthew Ahrens /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */ 6019b94df9SMatthew Ahrens int zfs_send_corrupt_data = B_FALSE; 61a2cdcdd2SPaul Dagnelie int zfs_send_queue_length = 16 * 1024 * 1024; 62a2cdcdd2SPaul Dagnelie int zfs_recv_queue_length = 16 * 1024 * 1024; 63880094b6SAndrew Stormont /* Set this tunable to FALSE to disable setting of DRR_FLAG_FREERECORDS */ 64880094b6SAndrew Stormont int zfs_send_set_freerecords_bit = B_TRUE; 6519b94df9SMatthew Ahrens 663cb34c60Sahrens static char *dmu_recv_tag = "dmu_recv_tag"; 679c3fd121SMatthew Ahrens const char *recv_clone_name = "%recv"; 683cb34c60Sahrens 69a2cdcdd2SPaul Dagnelie #define BP_SPAN(datablkszsec, indblkshift, level) \ 70a2cdcdd2SPaul Dagnelie (((uint64_t)datablkszsec) << (SPA_MINBLOCKSHIFT + \ 71a2cdcdd2SPaul Dagnelie (level) * (indblkshift - SPA_BLKPTRSHIFT))) 72a2cdcdd2SPaul Dagnelie 739c3fd121SMatthew Ahrens static void byteswap_record(dmu_replay_record_t *drr); 749c3fd121SMatthew Ahrens 75a2cdcdd2SPaul Dagnelie struct send_thread_arg { 76a2cdcdd2SPaul Dagnelie bqueue_t q; 77a2cdcdd2SPaul Dagnelie dsl_dataset_t *ds; /* Dataset to traverse */ 78a2cdcdd2SPaul Dagnelie uint64_t fromtxg; /* Traverse from this txg */ 79a2cdcdd2SPaul Dagnelie int flags; /* flags to pass to traverse_dataset */ 80a2cdcdd2SPaul Dagnelie int error_code; 81a2cdcdd2SPaul Dagnelie boolean_t cancel; 829c3fd121SMatthew Ahrens zbookmark_phys_t resume; 83a2cdcdd2SPaul Dagnelie }; 84a2cdcdd2SPaul Dagnelie 85a2cdcdd2SPaul Dagnelie struct send_block_record { 86a2cdcdd2SPaul Dagnelie boolean_t eos_marker; /* Marks the end of the stream */ 87a2cdcdd2SPaul Dagnelie blkptr_t bp; 88a2cdcdd2SPaul Dagnelie zbookmark_phys_t zb; 89a2cdcdd2SPaul Dagnelie uint8_t indblkshift; 90a2cdcdd2SPaul Dagnelie uint16_t datablkszsec; 91a2cdcdd2SPaul Dagnelie bqueue_node_t ln; 92a2cdcdd2SPaul Dagnelie }; 93a2cdcdd2SPaul Dagnelie 94efb80947Sahrens static int 954e3c9f44SBill Pijewski dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) 96efb80947Sahrens { 979c3fd121SMatthew Ahrens dsl_dataset_t *ds = dmu_objset_ds(dsp->dsa_os); 98efb80947Sahrens ssize_t resid; /* have to get resid to get detailed errno */ 99c20404ffSEli Rosenthal 100c20404ffSEli Rosenthal /* 101c20404ffSEli Rosenthal * The code does not rely on this (len being a multiple of 8). We keep 102c20404ffSEli Rosenthal * this assertion because of the corresponding assertion in 103c20404ffSEli Rosenthal * receive_read(). Keeping this assertion ensures that we do not 104c20404ffSEli Rosenthal * inadvertently break backwards compatibility (causing the assertion 105c20404ffSEli Rosenthal * in receive_read() to trigger on old software). 106c20404ffSEli Rosenthal * 107c20404ffSEli Rosenthal * Removing the assertions could be rolled into a new feature that uses 108c20404ffSEli Rosenthal * data that isn't 8-byte aligned; if the assertions were removed, a 109c20404ffSEli Rosenthal * feature flag would have to be added. 110c20404ffSEli Rosenthal */ 111c20404ffSEli Rosenthal 112fb09f5aaSMadhav Suresh ASSERT0(len % 8); 113efb80947Sahrens 1144e3c9f44SBill Pijewski dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp, 115efb80947Sahrens (caddr_t)buf, len, 116efb80947Sahrens 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid); 1174e3c9f44SBill Pijewski 1184e3c9f44SBill Pijewski mutex_enter(&ds->ds_sendstream_lock); 1194e3c9f44SBill Pijewski *dsp->dsa_off += len; 1204e3c9f44SBill Pijewski mutex_exit(&ds->ds_sendstream_lock); 1214e3c9f44SBill Pijewski 1224e3c9f44SBill Pijewski return (dsp->dsa_err); 123efb80947Sahrens } 124efb80947Sahrens 12598110f08SMatthew Ahrens /* 12698110f08SMatthew Ahrens * For all record types except BEGIN, fill in the checksum (overlaid in 12798110f08SMatthew Ahrens * drr_u.drr_checksum.drr_checksum). The checksum verifies everything 12898110f08SMatthew Ahrens * up to the start of the checksum itself. 12998110f08SMatthew Ahrens */ 13098110f08SMatthew Ahrens static int 13198110f08SMatthew Ahrens dump_record(dmu_sendarg_t *dsp, void *payload, int payload_len) 13298110f08SMatthew Ahrens { 13398110f08SMatthew Ahrens ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 13498110f08SMatthew Ahrens ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 13598110f08SMatthew Ahrens fletcher_4_incremental_native(dsp->dsa_drr, 13698110f08SMatthew Ahrens offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 13798110f08SMatthew Ahrens &dsp->dsa_zc); 138*12b90ee2SMatt Krantz if (dsp->dsa_drr->drr_type == DRR_BEGIN) { 139*12b90ee2SMatt Krantz dsp->dsa_sent_begin = B_TRUE; 140*12b90ee2SMatt Krantz } else { 14198110f08SMatthew Ahrens ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp->dsa_drr->drr_u. 14298110f08SMatthew Ahrens drr_checksum.drr_checksum)); 14398110f08SMatthew Ahrens dsp->dsa_drr->drr_u.drr_checksum.drr_checksum = dsp->dsa_zc; 14498110f08SMatthew Ahrens } 145*12b90ee2SMatt Krantz if (dsp->dsa_drr->drr_type == DRR_END) { 146*12b90ee2SMatt Krantz dsp->dsa_sent_end = B_TRUE; 147*12b90ee2SMatt Krantz } 14898110f08SMatthew Ahrens fletcher_4_incremental_native(&dsp->dsa_drr-> 14998110f08SMatthew Ahrens drr_u.drr_checksum.drr_checksum, 15098110f08SMatthew Ahrens sizeof (zio_cksum_t), &dsp->dsa_zc); 15198110f08SMatthew Ahrens if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) 15298110f08SMatthew Ahrens return (SET_ERROR(EINTR)); 15398110f08SMatthew Ahrens if (payload_len != 0) { 15498110f08SMatthew Ahrens fletcher_4_incremental_native(payload, payload_len, 15598110f08SMatthew Ahrens &dsp->dsa_zc); 15698110f08SMatthew Ahrens if (dump_bytes(dsp, payload, payload_len) != 0) 15798110f08SMatthew Ahrens return (SET_ERROR(EINTR)); 15898110f08SMatthew Ahrens } 15998110f08SMatthew Ahrens return (0); 16098110f08SMatthew Ahrens } 16198110f08SMatthew Ahrens 16268ecb2ecSPaul Dagnelie /* 16368ecb2ecSPaul Dagnelie * Fill in the drr_free struct, or perform aggregation if the previous record is 16468ecb2ecSPaul Dagnelie * also a free record, and the two are adjacent. 16568ecb2ecSPaul Dagnelie * 16668ecb2ecSPaul Dagnelie * Note that we send free records even for a full send, because we want to be 16768ecb2ecSPaul Dagnelie * able to receive a full send as a clone, which requires a list of all the free 16868ecb2ecSPaul Dagnelie * and freeobject records that were generated on the source. 16968ecb2ecSPaul Dagnelie */ 170efb80947Sahrens static int 1714e3c9f44SBill Pijewski dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 172efb80947Sahrens uint64_t length) 173efb80947Sahrens { 1744e3c9f44SBill Pijewski struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); 1759e69d7d0SLori Alt 1762f3d8780SMatthew Ahrens /* 1772f3d8780SMatthew Ahrens * When we receive a free record, dbuf_free_range() assumes 1782f3d8780SMatthew Ahrens * that the receiving system doesn't have any dbufs in the range 1792f3d8780SMatthew Ahrens * being freed. This is always true because there is a one-record 1802f3d8780SMatthew Ahrens * constraint: we only send one WRITE record for any given 1819c3fd121SMatthew Ahrens * object,offset. We know that the one-record constraint is 1822f3d8780SMatthew Ahrens * true because we always send data in increasing order by 1832f3d8780SMatthew Ahrens * object,offset. 1842f3d8780SMatthew Ahrens * 1852f3d8780SMatthew Ahrens * If the increasing-order constraint ever changes, we should find 1862f3d8780SMatthew Ahrens * another way to assert that the one-record constraint is still 1872f3d8780SMatthew Ahrens * satisfied. 1882f3d8780SMatthew Ahrens */ 1892f3d8780SMatthew Ahrens ASSERT(object > dsp->dsa_last_data_object || 1902f3d8780SMatthew Ahrens (object == dsp->dsa_last_data_object && 1912f3d8780SMatthew Ahrens offset > dsp->dsa_last_data_offset)); 1922f3d8780SMatthew Ahrens 193534029e5SSimon Klinkert if (length != -1ULL && offset + length < offset) 194534029e5SSimon Klinkert length = -1ULL; 195534029e5SSimon Klinkert 1969e69d7d0SLori Alt /* 1979e69d7d0SLori Alt * If there is a pending op, but it's not PENDING_FREE, push it out, 1989e69d7d0SLori Alt * since free block aggregation can only be done for blocks of the 1999e69d7d0SLori Alt * same type (i.e., DRR_FREE records can only be aggregated with 2009e69d7d0SLori Alt * other DRR_FREE records. DRR_FREEOBJECTS records can only be 2019e69d7d0SLori Alt * aggregated with other DRR_FREEOBJECTS records. 2029e69d7d0SLori Alt */ 2034e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE && 2044e3c9f44SBill Pijewski dsp->dsa_pending_op != PENDING_FREE) { 20598110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 206be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2074e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2089e69d7d0SLori Alt } 2099e69d7d0SLori Alt 2104e3c9f44SBill Pijewski if (dsp->dsa_pending_op == PENDING_FREE) { 2119e69d7d0SLori Alt /* 2129e69d7d0SLori Alt * There should never be a PENDING_FREE if length is -1 2139e69d7d0SLori Alt * (because dump_dnode is the only place where this 2149e69d7d0SLori Alt * function is called with a -1, and only after flushing 2159e69d7d0SLori Alt * any pending record). 2169e69d7d0SLori Alt */ 2179e69d7d0SLori Alt ASSERT(length != -1ULL); 2189e69d7d0SLori Alt /* 2199e69d7d0SLori Alt * Check to see whether this free block can be aggregated 2209e69d7d0SLori Alt * with pending one. 2219e69d7d0SLori Alt */ 2229e69d7d0SLori Alt if (drrf->drr_object == object && drrf->drr_offset + 2239e69d7d0SLori Alt drrf->drr_length == offset) { 2249e69d7d0SLori Alt drrf->drr_length += length; 2259e69d7d0SLori Alt return (0); 2269e69d7d0SLori Alt } else { 2279e69d7d0SLori Alt /* not a continuation. Push out pending record */ 22898110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 229be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2304e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2319e69d7d0SLori Alt } 2329e69d7d0SLori Alt } 2339e69d7d0SLori Alt /* create a FREE record and make it pending */ 2344e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 2354e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_FREE; 2369e69d7d0SLori Alt drrf->drr_object = object; 2379e69d7d0SLori Alt drrf->drr_offset = offset; 2389e69d7d0SLori Alt drrf->drr_length = length; 2394e3c9f44SBill Pijewski drrf->drr_toguid = dsp->dsa_toguid; 2409e69d7d0SLori Alt if (length == -1ULL) { 24198110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 242be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2439e69d7d0SLori Alt } else { 2444e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_FREE; 2459e69d7d0SLori Alt } 246efb80947Sahrens 247efb80947Sahrens return (0); 248efb80947Sahrens } 249efb80947Sahrens 250efb80947Sahrens static int 2515d7b4d43SMatthew Ahrens dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, 2528e714474SLori Alt uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data) 253efb80947Sahrens { 2544e3c9f44SBill Pijewski struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); 2559e69d7d0SLori Alt 2562f3d8780SMatthew Ahrens /* 2572f3d8780SMatthew Ahrens * We send data in increasing object, offset order. 2582f3d8780SMatthew Ahrens * See comment in dump_free() for details. 2592f3d8780SMatthew Ahrens */ 2602f3d8780SMatthew Ahrens ASSERT(object > dsp->dsa_last_data_object || 2612f3d8780SMatthew Ahrens (object == dsp->dsa_last_data_object && 2622f3d8780SMatthew Ahrens offset > dsp->dsa_last_data_offset)); 2632f3d8780SMatthew Ahrens dsp->dsa_last_data_object = object; 2642f3d8780SMatthew Ahrens dsp->dsa_last_data_offset = offset + blksz - 1; 2658e714474SLori Alt 2669e69d7d0SLori Alt /* 2679e69d7d0SLori Alt * If there is any kind of pending aggregation (currently either 2689e69d7d0SLori Alt * a grouping of free objects or free blocks), push it out to 2699e69d7d0SLori Alt * the stream, since aggregation can't be done across operations 2709e69d7d0SLori Alt * of different types. 2719e69d7d0SLori Alt */ 2724e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 27398110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 274be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 2754e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 2769e69d7d0SLori Alt } 27798110f08SMatthew Ahrens /* write a WRITE record */ 2784e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 2794e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_WRITE; 2809e69d7d0SLori Alt drrw->drr_object = object; 2819e69d7d0SLori Alt drrw->drr_type = type; 2829e69d7d0SLori Alt drrw->drr_offset = offset; 2839e69d7d0SLori Alt drrw->drr_length = blksz; 2844e3c9f44SBill Pijewski drrw->drr_toguid = dsp->dsa_toguid; 285b5152584SMatthew Ahrens if (bp == NULL || BP_IS_EMBEDDED(bp)) { 2865d7b4d43SMatthew Ahrens /* 287b5152584SMatthew Ahrens * There's no pre-computed checksum for partial-block 288b5152584SMatthew Ahrens * writes or embedded BP's, so (like 289b5152584SMatthew Ahrens * fletcher4-checkummed blocks) userland will have to 290b5152584SMatthew Ahrens * compute a dedup-capable checksum itself. 2915d7b4d43SMatthew Ahrens */ 2925d7b4d43SMatthew Ahrens drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; 2935d7b4d43SMatthew Ahrens } else { 2945d7b4d43SMatthew Ahrens drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); 29545818ee1SMatthew Ahrens if (zio_checksum_table[drrw->drr_checksumtype].ci_flags & 29645818ee1SMatthew Ahrens ZCHECKSUM_FLAG_DEDUP) 2975d7b4d43SMatthew Ahrens drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP; 2985d7b4d43SMatthew Ahrens DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); 2995d7b4d43SMatthew Ahrens DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); 3005d7b4d43SMatthew Ahrens DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); 3015d7b4d43SMatthew Ahrens drrw->drr_key.ddk_cksum = bp->blk_cksum; 3025d7b4d43SMatthew Ahrens } 303efb80947Sahrens 30498110f08SMatthew Ahrens if (dump_record(dsp, data, blksz) != 0) 305be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 306efb80947Sahrens return (0); 307efb80947Sahrens } 308efb80947Sahrens 3095d7b4d43SMatthew Ahrens static int 3105d7b4d43SMatthew Ahrens dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, 3115d7b4d43SMatthew Ahrens int blksz, const blkptr_t *bp) 3125d7b4d43SMatthew Ahrens { 3135d7b4d43SMatthew Ahrens char buf[BPE_PAYLOAD_SIZE]; 3145d7b4d43SMatthew Ahrens struct drr_write_embedded *drrw = 3155d7b4d43SMatthew Ahrens &(dsp->dsa_drr->drr_u.drr_write_embedded); 3165d7b4d43SMatthew Ahrens 3175d7b4d43SMatthew Ahrens if (dsp->dsa_pending_op != PENDING_NONE) { 31898110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 3195d7b4d43SMatthew Ahrens return (EINTR); 3205d7b4d43SMatthew Ahrens dsp->dsa_pending_op = PENDING_NONE; 3215d7b4d43SMatthew Ahrens } 3225d7b4d43SMatthew Ahrens 3235d7b4d43SMatthew Ahrens ASSERT(BP_IS_EMBEDDED(bp)); 3245d7b4d43SMatthew Ahrens 3255d7b4d43SMatthew Ahrens bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 3265d7b4d43SMatthew Ahrens dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; 3275d7b4d43SMatthew Ahrens drrw->drr_object = object; 3285d7b4d43SMatthew Ahrens drrw->drr_offset = offset; 3295d7b4d43SMatthew Ahrens drrw->drr_length = blksz; 3305d7b4d43SMatthew Ahrens drrw->drr_toguid = dsp->dsa_toguid; 3315d7b4d43SMatthew Ahrens drrw->drr_compression = BP_GET_COMPRESS(bp); 3325d7b4d43SMatthew Ahrens drrw->drr_etype = BPE_GET_ETYPE(bp); 3335d7b4d43SMatthew Ahrens drrw->drr_lsize = BPE_GET_LSIZE(bp); 3345d7b4d43SMatthew Ahrens drrw->drr_psize = BPE_GET_PSIZE(bp); 3355d7b4d43SMatthew Ahrens 3365d7b4d43SMatthew Ahrens decode_embedded_bp_compressed(bp, buf); 3375d7b4d43SMatthew Ahrens 33898110f08SMatthew Ahrens if (dump_record(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) 3395d7b4d43SMatthew Ahrens return (EINTR); 3405d7b4d43SMatthew Ahrens return (0); 3415d7b4d43SMatthew Ahrens } 3425d7b4d43SMatthew Ahrens 3430a586ceaSMark Shellenbaum static int 3444e3c9f44SBill Pijewski dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data) 3450a586ceaSMark Shellenbaum { 3464e3c9f44SBill Pijewski struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); 3470a586ceaSMark Shellenbaum 3484e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 34998110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 350be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3514e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 3520a586ceaSMark Shellenbaum } 3530a586ceaSMark Shellenbaum 3540a586ceaSMark Shellenbaum /* write a SPILL record */ 3554e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 3564e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_SPILL; 3570a586ceaSMark Shellenbaum drrs->drr_object = object; 3580a586ceaSMark Shellenbaum drrs->drr_length = blksz; 3594e3c9f44SBill Pijewski drrs->drr_toguid = dsp->dsa_toguid; 3600a586ceaSMark Shellenbaum 36198110f08SMatthew Ahrens if (dump_record(dsp, data, blksz) != 0) 362be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3630a586ceaSMark Shellenbaum return (0); 3640a586ceaSMark Shellenbaum } 3650a586ceaSMark Shellenbaum 366efb80947Sahrens static int 3674e3c9f44SBill Pijewski dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) 368efb80947Sahrens { 3694e3c9f44SBill Pijewski struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); 3709e69d7d0SLori Alt 3719e69d7d0SLori Alt /* 3729e69d7d0SLori Alt * If there is a pending op, but it's not PENDING_FREEOBJECTS, 3739e69d7d0SLori Alt * push it out, since free block aggregation can only be done for 3749e69d7d0SLori Alt * blocks of the same type (i.e., DRR_FREE records can only be 3759e69d7d0SLori Alt * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records 3769e69d7d0SLori Alt * can only be aggregated with other DRR_FREEOBJECTS records. 3779e69d7d0SLori Alt */ 3784e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE && 3794e3c9f44SBill Pijewski dsp->dsa_pending_op != PENDING_FREEOBJECTS) { 38098110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 381be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3824e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 3839e69d7d0SLori Alt } 3844e3c9f44SBill Pijewski if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { 3859e69d7d0SLori Alt /* 3869e69d7d0SLori Alt * See whether this free object array can be aggregated 3879e69d7d0SLori Alt * with pending one 3889e69d7d0SLori Alt */ 3899e69d7d0SLori Alt if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { 3909e69d7d0SLori Alt drrfo->drr_numobjs += numobjs; 3919e69d7d0SLori Alt return (0); 3929e69d7d0SLori Alt } else { 3939e69d7d0SLori Alt /* can't be aggregated. Push out pending record */ 39498110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 395be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 3964e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 3979e69d7d0SLori Alt } 3989e69d7d0SLori Alt } 3999e69d7d0SLori Alt 400efb80947Sahrens /* write a FREEOBJECTS record */ 4014e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 4024e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; 4039e69d7d0SLori Alt drrfo->drr_firstobj = firstobj; 4049e69d7d0SLori Alt drrfo->drr_numobjs = numobjs; 4054e3c9f44SBill Pijewski drrfo->drr_toguid = dsp->dsa_toguid; 4069e69d7d0SLori Alt 4074e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_FREEOBJECTS; 408efb80947Sahrens 409efb80947Sahrens return (0); 410efb80947Sahrens } 411efb80947Sahrens 412efb80947Sahrens static int 4134e3c9f44SBill Pijewski dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) 414efb80947Sahrens { 4154e3c9f44SBill Pijewski struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); 4169e69d7d0SLori Alt 4179c3fd121SMatthew Ahrens if (object < dsp->dsa_resume_object) { 4189c3fd121SMatthew Ahrens /* 4199c3fd121SMatthew Ahrens * Note: when resuming, we will visit all the dnodes in 4209c3fd121SMatthew Ahrens * the block of dnodes that we are resuming from. In 4219c3fd121SMatthew Ahrens * this case it's unnecessary to send the dnodes prior to 4229c3fd121SMatthew Ahrens * the one we are resuming from. We should be at most one 4239c3fd121SMatthew Ahrens * block's worth of dnodes behind the resume point. 4249c3fd121SMatthew Ahrens */ 4259c3fd121SMatthew Ahrens ASSERT3U(dsp->dsa_resume_object - object, <, 4269c3fd121SMatthew Ahrens 1 << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)); 4279c3fd121SMatthew Ahrens return (0); 4289c3fd121SMatthew Ahrens } 4299c3fd121SMatthew Ahrens 430efb80947Sahrens if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) 4314e3c9f44SBill Pijewski return (dump_freeobjects(dsp, object, 1)); 432efb80947Sahrens 4334e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) { 43498110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 435be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4364e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 4379e69d7d0SLori Alt } 4389e69d7d0SLori Alt 439efb80947Sahrens /* write an OBJECT record */ 4404e3c9f44SBill Pijewski bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); 4414e3c9f44SBill Pijewski dsp->dsa_drr->drr_type = DRR_OBJECT; 4429e69d7d0SLori Alt drro->drr_object = object; 4439e69d7d0SLori Alt drro->drr_type = dnp->dn_type; 4449e69d7d0SLori Alt drro->drr_bonustype = dnp->dn_bonustype; 4459e69d7d0SLori Alt drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 4469e69d7d0SLori Alt drro->drr_bonuslen = dnp->dn_bonuslen; 4479e69d7d0SLori Alt drro->drr_checksumtype = dnp->dn_checksum; 4489e69d7d0SLori Alt drro->drr_compress = dnp->dn_compress; 4494e3c9f44SBill Pijewski drro->drr_toguid = dsp->dsa_toguid; 4509e69d7d0SLori Alt 451b5152584SMatthew Ahrens if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 452b5152584SMatthew Ahrens drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) 453b5152584SMatthew Ahrens drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; 454b5152584SMatthew Ahrens 45598110f08SMatthew Ahrens if (dump_record(dsp, DN_BONUS(dnp), 45698110f08SMatthew Ahrens P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0) { 457be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 45898110f08SMatthew Ahrens } 459efb80947Sahrens 4602f3d8780SMatthew Ahrens /* Free anything past the end of the file. */ 4614e3c9f44SBill Pijewski if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * 4622f3d8780SMatthew Ahrens (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0) 463be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 4643b2aab18SMatthew Ahrens if (dsp->dsa_err != 0) 465be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 466efb80947Sahrens return (0); 467efb80947Sahrens } 468efb80947Sahrens 4695d7b4d43SMatthew Ahrens static boolean_t 4705d7b4d43SMatthew Ahrens backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) 4715d7b4d43SMatthew Ahrens { 4725d7b4d43SMatthew Ahrens if (!BP_IS_EMBEDDED(bp)) 4735d7b4d43SMatthew Ahrens return (B_FALSE); 4745d7b4d43SMatthew Ahrens 4755d7b4d43SMatthew Ahrens /* 4765d7b4d43SMatthew Ahrens * Compression function must be legacy, or explicitly enabled. 4775d7b4d43SMatthew Ahrens */ 4785d7b4d43SMatthew Ahrens if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && 4795d7b4d43SMatthew Ahrens !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4))) 4805d7b4d43SMatthew Ahrens return (B_FALSE); 4815d7b4d43SMatthew Ahrens 4825d7b4d43SMatthew Ahrens /* 4835d7b4d43SMatthew Ahrens * Embed type must be explicitly enabled. 4845d7b4d43SMatthew Ahrens */ 4855d7b4d43SMatthew Ahrens switch (BPE_GET_ETYPE(bp)) { 4865d7b4d43SMatthew Ahrens case BP_EMBEDDED_TYPE_DATA: 4875d7b4d43SMatthew Ahrens if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) 4885d7b4d43SMatthew Ahrens return (B_TRUE); 4895d7b4d43SMatthew Ahrens break; 4905d7b4d43SMatthew Ahrens default: 4915d7b4d43SMatthew Ahrens return (B_FALSE); 4925d7b4d43SMatthew Ahrens } 4935d7b4d43SMatthew Ahrens return (B_FALSE); 4945d7b4d43SMatthew Ahrens } 4955d7b4d43SMatthew Ahrens 496a2cdcdd2SPaul Dagnelie /* 497a2cdcdd2SPaul Dagnelie * This is the callback function to traverse_dataset that acts as the worker 498a2cdcdd2SPaul Dagnelie * thread for dmu_send_impl. 499a2cdcdd2SPaul Dagnelie */ 500a2cdcdd2SPaul Dagnelie /*ARGSUSED*/ 501a2cdcdd2SPaul Dagnelie static int 502a2cdcdd2SPaul Dagnelie send_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 503a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb, const struct dnode_phys *dnp, void *arg) 504a2cdcdd2SPaul Dagnelie { 505a2cdcdd2SPaul Dagnelie struct send_thread_arg *sta = arg; 506a2cdcdd2SPaul Dagnelie struct send_block_record *record; 507a2cdcdd2SPaul Dagnelie uint64_t record_size; 508a2cdcdd2SPaul Dagnelie int err = 0; 509efb80947Sahrens 5109c3fd121SMatthew Ahrens ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || 5119c3fd121SMatthew Ahrens zb->zb_object >= sta->resume.zb_object); 5129c3fd121SMatthew Ahrens 513a2cdcdd2SPaul Dagnelie if (sta->cancel) 514a2cdcdd2SPaul Dagnelie return (SET_ERROR(EINTR)); 515a2cdcdd2SPaul Dagnelie 516a2cdcdd2SPaul Dagnelie if (bp == NULL) { 517a2cdcdd2SPaul Dagnelie ASSERT3U(zb->zb_level, ==, ZB_DNODE_LEVEL); 518a2cdcdd2SPaul Dagnelie return (0); 519a2cdcdd2SPaul Dagnelie } else if (zb->zb_level < 0) { 520a2cdcdd2SPaul Dagnelie return (0); 521a2cdcdd2SPaul Dagnelie } 522a2cdcdd2SPaul Dagnelie 523a2cdcdd2SPaul Dagnelie record = kmem_zalloc(sizeof (struct send_block_record), KM_SLEEP); 524a2cdcdd2SPaul Dagnelie record->eos_marker = B_FALSE; 525a2cdcdd2SPaul Dagnelie record->bp = *bp; 526a2cdcdd2SPaul Dagnelie record->zb = *zb; 527a2cdcdd2SPaul Dagnelie record->indblkshift = dnp->dn_indblkshift; 528a2cdcdd2SPaul Dagnelie record->datablkszsec = dnp->dn_datablkszsec; 529a2cdcdd2SPaul Dagnelie record_size = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; 530a2cdcdd2SPaul Dagnelie bqueue_enqueue(&sta->q, record, record_size); 531a2cdcdd2SPaul Dagnelie 532a2cdcdd2SPaul Dagnelie return (err); 533a2cdcdd2SPaul Dagnelie } 534a2cdcdd2SPaul Dagnelie 535a2cdcdd2SPaul Dagnelie /* 536a2cdcdd2SPaul Dagnelie * This function kicks off the traverse_dataset. It also handles setting the 537a2cdcdd2SPaul Dagnelie * error code of the thread in case something goes wrong, and pushes the End of 538a2cdcdd2SPaul Dagnelie * Stream record when the traverse_dataset call has finished. If there is no 539a2cdcdd2SPaul Dagnelie * dataset to traverse, the thread immediately pushes End of Stream marker. 540a2cdcdd2SPaul Dagnelie */ 541a2cdcdd2SPaul Dagnelie static void 542a2cdcdd2SPaul Dagnelie send_traverse_thread(void *arg) 543a2cdcdd2SPaul Dagnelie { 544a2cdcdd2SPaul Dagnelie struct send_thread_arg *st_arg = arg; 545a2cdcdd2SPaul Dagnelie int err; 546a2cdcdd2SPaul Dagnelie struct send_block_record *data; 547a2cdcdd2SPaul Dagnelie 548a2cdcdd2SPaul Dagnelie if (st_arg->ds != NULL) { 5499c3fd121SMatthew Ahrens err = traverse_dataset_resume(st_arg->ds, 5509c3fd121SMatthew Ahrens st_arg->fromtxg, &st_arg->resume, 5519c3fd121SMatthew Ahrens st_arg->flags, send_cb, st_arg); 5529c3fd121SMatthew Ahrens 553a2cdcdd2SPaul Dagnelie if (err != EINTR) 554a2cdcdd2SPaul Dagnelie st_arg->error_code = err; 555a2cdcdd2SPaul Dagnelie } 556a2cdcdd2SPaul Dagnelie data = kmem_zalloc(sizeof (*data), KM_SLEEP); 557a2cdcdd2SPaul Dagnelie data->eos_marker = B_TRUE; 558a2cdcdd2SPaul Dagnelie bqueue_enqueue(&st_arg->q, data, 1); 559a2cdcdd2SPaul Dagnelie } 560a2cdcdd2SPaul Dagnelie 561a2cdcdd2SPaul Dagnelie /* 562a2cdcdd2SPaul Dagnelie * This function actually handles figuring out what kind of record needs to be 563a2cdcdd2SPaul Dagnelie * dumped, reading the data (which has hopefully been prefetched), and calling 564a2cdcdd2SPaul Dagnelie * the appropriate helper function. 565a2cdcdd2SPaul Dagnelie */ 566efb80947Sahrens static int 567a2cdcdd2SPaul Dagnelie do_dump(dmu_sendarg_t *dsa, struct send_block_record *data) 568efb80947Sahrens { 569a2cdcdd2SPaul Dagnelie dsl_dataset_t *ds = dmu_objset_ds(dsa->dsa_os); 570a2cdcdd2SPaul Dagnelie const blkptr_t *bp = &data->bp; 571a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb = &data->zb; 572a2cdcdd2SPaul Dagnelie uint8_t indblkshift = data->indblkshift; 573a2cdcdd2SPaul Dagnelie uint16_t dblkszsec = data->datablkszsec; 574a2cdcdd2SPaul Dagnelie spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 575efb80947Sahrens dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; 576efb80947Sahrens int err = 0; 577efb80947Sahrens 578a2cdcdd2SPaul Dagnelie ASSERT3U(zb->zb_level, >=, 0); 579efb80947Sahrens 5809c3fd121SMatthew Ahrens ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || 5819c3fd121SMatthew Ahrens zb->zb_object >= dsa->dsa_resume_object); 5829c3fd121SMatthew Ahrens 583b24ab676SJeff Bonwick if (zb->zb_object != DMU_META_DNODE_OBJECT && 584b24ab676SJeff Bonwick DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { 58514843421SMatthew Ahrens return (0); 58643466aaeSMax Grossman } else if (BP_IS_HOLE(bp) && 58743466aaeSMax Grossman zb->zb_object == DMU_META_DNODE_OBJECT) { 588a2cdcdd2SPaul Dagnelie uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); 58988b7b0f2SMatthew Ahrens uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; 590a2cdcdd2SPaul Dagnelie err = dump_freeobjects(dsa, dnobj, span >> DNODE_SHIFT); 59143466aaeSMax Grossman } else if (BP_IS_HOLE(bp)) { 592a2cdcdd2SPaul Dagnelie uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); 593a2cdcdd2SPaul Dagnelie uint64_t offset = zb->zb_blkid * span; 594a2cdcdd2SPaul Dagnelie err = dump_free(dsa, zb->zb_object, offset, span); 59588b7b0f2SMatthew Ahrens } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { 59688b7b0f2SMatthew Ahrens return (0); 59788b7b0f2SMatthew Ahrens } else if (type == DMU_OT_DNODE) { 598efb80947Sahrens int blksz = BP_GET_LSIZE(bp); 5997adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 60088b7b0f2SMatthew Ahrens arc_buf_t *abuf; 601efb80947Sahrens 602a2cdcdd2SPaul Dagnelie ASSERT0(zb->zb_level); 603a2cdcdd2SPaul Dagnelie 6041b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 6051b912ec7SGeorge Wilson ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 6061b912ec7SGeorge Wilson &aflags, zb) != 0) 607be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 60888b7b0f2SMatthew Ahrens 609a2cdcdd2SPaul Dagnelie dnode_phys_t *blk = abuf->b_data; 610a2cdcdd2SPaul Dagnelie uint64_t dnobj = zb->zb_blkid * (blksz >> DNODE_SHIFT); 611a2cdcdd2SPaul Dagnelie for (int i = 0; i < blksz >> DNODE_SHIFT; i++) { 612a2cdcdd2SPaul Dagnelie err = dump_dnode(dsa, dnobj + i, blk + i); 6133b2aab18SMatthew Ahrens if (err != 0) 614efb80947Sahrens break; 615efb80947Sahrens } 616dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 6170a586ceaSMark Shellenbaum } else if (type == DMU_OT_SA) { 6187adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 6190a586ceaSMark Shellenbaum arc_buf_t *abuf; 6200a586ceaSMark Shellenbaum int blksz = BP_GET_LSIZE(bp); 6210a586ceaSMark Shellenbaum 6221b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 6231b912ec7SGeorge Wilson ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 6241b912ec7SGeorge Wilson &aflags, zb) != 0) 625be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 6260a586ceaSMark Shellenbaum 627a2cdcdd2SPaul Dagnelie err = dump_spill(dsa, zb->zb_object, blksz, abuf->b_data); 628dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 629a2cdcdd2SPaul Dagnelie } else if (backup_do_embed(dsa, bp)) { 6305d7b4d43SMatthew Ahrens /* it's an embedded level-0 block of a regular object */ 631a2cdcdd2SPaul Dagnelie int blksz = dblkszsec << SPA_MINBLOCKSHIFT; 632a2cdcdd2SPaul Dagnelie ASSERT0(zb->zb_level); 633a2cdcdd2SPaul Dagnelie err = dump_write_embedded(dsa, zb->zb_object, 6345d7b4d43SMatthew Ahrens zb->zb_blkid * blksz, blksz, bp); 635a2cdcdd2SPaul Dagnelie } else { 636a2cdcdd2SPaul Dagnelie /* it's a level-0 block of a regular object */ 6377adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 63888b7b0f2SMatthew Ahrens arc_buf_t *abuf; 639a2cdcdd2SPaul Dagnelie int blksz = dblkszsec << SPA_MINBLOCKSHIFT; 640b5152584SMatthew Ahrens uint64_t offset; 64188b7b0f2SMatthew Ahrens 64278f17100SMatthew Ahrens ASSERT0(zb->zb_level); 6439c3fd121SMatthew Ahrens ASSERT(zb->zb_object > dsa->dsa_resume_object || 6449c3fd121SMatthew Ahrens (zb->zb_object == dsa->dsa_resume_object && 6459c3fd121SMatthew Ahrens zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); 6469c3fd121SMatthew Ahrens 6471b912ec7SGeorge Wilson if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, 6481b912ec7SGeorge Wilson ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, 6491b912ec7SGeorge Wilson &aflags, zb) != 0) { 65019b94df9SMatthew Ahrens if (zfs_send_corrupt_data) { 65119b94df9SMatthew Ahrens /* Send a block filled with 0x"zfs badd bloc" */ 652dcbf3bd6SGeorge Wilson abuf = arc_alloc_buf(spa, blksz, &abuf, 65319b94df9SMatthew Ahrens ARC_BUFC_DATA); 65419b94df9SMatthew Ahrens uint64_t *ptr; 65519b94df9SMatthew Ahrens for (ptr = abuf->b_data; 65619b94df9SMatthew Ahrens (char *)ptr < (char *)abuf->b_data + blksz; 65719b94df9SMatthew Ahrens ptr++) 6588c76e076SBrian Behlendorf *ptr = 0x2f5baddb10cULL; 65919b94df9SMatthew Ahrens } else { 660be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 66119b94df9SMatthew Ahrens } 66219b94df9SMatthew Ahrens } 66388b7b0f2SMatthew Ahrens 664b5152584SMatthew Ahrens offset = zb->zb_blkid * blksz; 665b5152584SMatthew Ahrens 666a2cdcdd2SPaul Dagnelie if (!(dsa->dsa_featureflags & 667b5152584SMatthew Ahrens DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 668b5152584SMatthew Ahrens blksz > SPA_OLD_MAXBLOCKSIZE) { 669b5152584SMatthew Ahrens char *buf = abuf->b_data; 670b5152584SMatthew Ahrens while (blksz > 0 && err == 0) { 671b5152584SMatthew Ahrens int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); 672a2cdcdd2SPaul Dagnelie err = dump_write(dsa, type, zb->zb_object, 673b5152584SMatthew Ahrens offset, n, NULL, buf); 674b5152584SMatthew Ahrens offset += n; 675b5152584SMatthew Ahrens buf += n; 676b5152584SMatthew Ahrens blksz -= n; 677b5152584SMatthew Ahrens } 678b5152584SMatthew Ahrens } else { 679a2cdcdd2SPaul Dagnelie err = dump_write(dsa, type, zb->zb_object, 680b5152584SMatthew Ahrens offset, blksz, bp, abuf->b_data); 681b5152584SMatthew Ahrens } 682dcbf3bd6SGeorge Wilson arc_buf_destroy(abuf, &abuf); 683efb80947Sahrens } 684efb80947Sahrens 685efb80947Sahrens ASSERT(err == 0 || err == EINTR); 686efb80947Sahrens return (err); 687efb80947Sahrens } 688efb80947Sahrens 6894445fffbSMatthew Ahrens /* 690a2cdcdd2SPaul Dagnelie * Pop the new data off the queue, and free the old data. 691a2cdcdd2SPaul Dagnelie */ 692a2cdcdd2SPaul Dagnelie static struct send_block_record * 693a2cdcdd2SPaul Dagnelie get_next_record(bqueue_t *bq, struct send_block_record *data) 694a2cdcdd2SPaul Dagnelie { 695a2cdcdd2SPaul Dagnelie struct send_block_record *tmp = bqueue_dequeue(bq); 696a2cdcdd2SPaul Dagnelie kmem_free(data, sizeof (*data)); 697a2cdcdd2SPaul Dagnelie return (tmp); 698a2cdcdd2SPaul Dagnelie } 699a2cdcdd2SPaul Dagnelie 700a2cdcdd2SPaul Dagnelie /* 701a2cdcdd2SPaul Dagnelie * Actually do the bulk of the work in a zfs send. 702a2cdcdd2SPaul Dagnelie * 703a2cdcdd2SPaul Dagnelie * Note: Releases dp using the specified tag. 7044445fffbSMatthew Ahrens */ 7053b2aab18SMatthew Ahrens static int 706a2cdcdd2SPaul Dagnelie dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *to_ds, 7079c3fd121SMatthew Ahrens zfs_bookmark_phys_t *ancestor_zb, 7089c3fd121SMatthew Ahrens boolean_t is_clone, boolean_t embedok, boolean_t large_block_ok, int outfd, 7099c3fd121SMatthew Ahrens uint64_t resumeobj, uint64_t resumeoff, 7109c3fd121SMatthew Ahrens vnode_t *vp, offset_t *off) 711efb80947Sahrens { 7123b2aab18SMatthew Ahrens objset_t *os; 713efb80947Sahrens dmu_replay_record_t *drr; 7144e3c9f44SBill Pijewski dmu_sendarg_t *dsp; 715efb80947Sahrens int err; 7163cb34c60Sahrens uint64_t fromtxg = 0; 7175d7b4d43SMatthew Ahrens uint64_t featureflags = 0; 7189c3fd121SMatthew Ahrens struct send_thread_arg to_arg = { 0 }; 719efb80947Sahrens 720a2cdcdd2SPaul Dagnelie err = dmu_objset_from_ds(to_ds, &os); 7213b2aab18SMatthew Ahrens if (err != 0) { 7223b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 7233b2aab18SMatthew Ahrens return (err); 7243b2aab18SMatthew Ahrens } 725efb80947Sahrens 726efb80947Sahrens drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); 727efb80947Sahrens drr->drr_type = DRR_BEGIN; 728efb80947Sahrens drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; 7299e69d7d0SLori Alt DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, 7309e69d7d0SLori Alt DMU_SUBSTREAM); 731dc7cd546SMark Shellenbaum 732dc7cd546SMark Shellenbaum #ifdef _KERNEL 7333b2aab18SMatthew Ahrens if (dmu_objset_type(os) == DMU_OST_ZFS) { 734dc7cd546SMark Shellenbaum uint64_t version; 7353b2aab18SMatthew Ahrens if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { 7364e3c9f44SBill Pijewski kmem_free(drr, sizeof (dmu_replay_record_t)); 7373b2aab18SMatthew Ahrens dsl_pool_rele(dp, tag); 738be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 7394e3c9f44SBill Pijewski } 7403b2aab18SMatthew Ahrens if (version >= ZPL_VERSION_SA) { 7415d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; 742dc7cd546SMark Shellenbaum } 743dc7cd546SMark Shellenbaum } 744dc7cd546SMark Shellenbaum #endif 745dc7cd546SMark Shellenbaum 746ca0cc391SMatthew Ahrens if (large_block_ok && to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS]) 747b5152584SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; 7485d7b4d43SMatthew Ahrens if (embedok && 7495d7b4d43SMatthew Ahrens spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { 7505d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; 7515d7b4d43SMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 7525d7b4d43SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4; 7535d7b4d43SMatthew Ahrens } 7545d7b4d43SMatthew Ahrens 7559c3fd121SMatthew Ahrens if (resumeobj != 0 || resumeoff != 0) { 7569c3fd121SMatthew Ahrens featureflags |= DMU_BACKUP_FEATURE_RESUMING; 7579c3fd121SMatthew Ahrens } 7589c3fd121SMatthew Ahrens 7595d7b4d43SMatthew Ahrens DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, 7605d7b4d43SMatthew Ahrens featureflags); 7615d7b4d43SMatthew Ahrens 762efb80947Sahrens drr->drr_u.drr_begin.drr_creation_time = 763a2cdcdd2SPaul Dagnelie dsl_dataset_phys(to_ds)->ds_creation_time; 7643b2aab18SMatthew Ahrens drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); 76578f17100SMatthew Ahrens if (is_clone) 7663cb34c60Sahrens drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; 767a2cdcdd2SPaul Dagnelie drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(to_ds)->ds_guid; 768a2cdcdd2SPaul Dagnelie if (dsl_dataset_phys(to_ds)->ds_flags & DS_FLAG_CI_DATASET) 769ab04eb8eStimh drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; 770880094b6SAndrew Stormont if (zfs_send_set_freerecords_bit) 771880094b6SAndrew Stormont drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_FREERECORDS; 772ab04eb8eStimh 773a2cdcdd2SPaul Dagnelie if (ancestor_zb != NULL) { 774a2cdcdd2SPaul Dagnelie drr->drr_u.drr_begin.drr_fromguid = 775a2cdcdd2SPaul Dagnelie ancestor_zb->zbm_guid; 776a2cdcdd2SPaul Dagnelie fromtxg = ancestor_zb->zbm_creation_txg; 77778f17100SMatthew Ahrens } 778a2cdcdd2SPaul Dagnelie dsl_dataset_name(to_ds, drr->drr_u.drr_begin.drr_toname); 779a2cdcdd2SPaul Dagnelie if (!to_ds->ds_is_snapshot) { 78078f17100SMatthew Ahrens (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", 78178f17100SMatthew Ahrens sizeof (drr->drr_u.drr_begin.drr_toname)); 7823b2aab18SMatthew Ahrens } 7833cb34c60Sahrens 7844e3c9f44SBill Pijewski dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); 7854e3c9f44SBill Pijewski 7864e3c9f44SBill Pijewski dsp->dsa_drr = drr; 7874e3c9f44SBill Pijewski dsp->dsa_vp = vp; 7884e3c9f44SBill Pijewski dsp->dsa_outfd = outfd; 7894e3c9f44SBill Pijewski dsp->dsa_proc = curproc; 7903b2aab18SMatthew Ahrens dsp->dsa_os = os; 7914e3c9f44SBill Pijewski dsp->dsa_off = off; 792a2cdcdd2SPaul Dagnelie dsp->dsa_toguid = dsl_dataset_phys(to_ds)->ds_guid; 7934e3c9f44SBill Pijewski dsp->dsa_pending_op = PENDING_NONE; 7945d7b4d43SMatthew Ahrens dsp->dsa_featureflags = featureflags; 7959c3fd121SMatthew Ahrens dsp->dsa_resume_object = resumeobj; 7969c3fd121SMatthew Ahrens dsp->dsa_resume_offset = resumeoff; 7974e3c9f44SBill Pijewski 798a2cdcdd2SPaul Dagnelie mutex_enter(&to_ds->ds_sendstream_lock); 799a2cdcdd2SPaul Dagnelie list_insert_head(&to_ds->ds_sendstreams, dsp); 800a2cdcdd2SPaul Dagnelie mutex_exit(&to_ds->ds_sendstream_lock); 8014e3c9f44SBill Pijewski 802a2cdcdd2SPaul Dagnelie dsl_dataset_long_hold(to_ds, FTAG); 803de8d9cffSMatthew Ahrens dsl_pool_rele(dp, tag); 804de8d9cffSMatthew Ahrens 8059c3fd121SMatthew Ahrens void *payload = NULL; 8069c3fd121SMatthew Ahrens size_t payload_len = 0; 8079c3fd121SMatthew Ahrens if (resumeobj != 0 || resumeoff != 0) { 8089c3fd121SMatthew Ahrens dmu_object_info_t to_doi; 8099c3fd121SMatthew Ahrens err = dmu_object_info(os, resumeobj, &to_doi); 8109c3fd121SMatthew Ahrens if (err != 0) 8119c3fd121SMatthew Ahrens goto out; 8129c3fd121SMatthew Ahrens SET_BOOKMARK(&to_arg.resume, to_ds->ds_object, resumeobj, 0, 8139c3fd121SMatthew Ahrens resumeoff / to_doi.doi_data_block_size); 8149c3fd121SMatthew Ahrens 8159c3fd121SMatthew Ahrens nvlist_t *nvl = fnvlist_alloc(); 8169c3fd121SMatthew Ahrens fnvlist_add_uint64(nvl, "resume_object", resumeobj); 8179c3fd121SMatthew Ahrens fnvlist_add_uint64(nvl, "resume_offset", resumeoff); 8189c3fd121SMatthew Ahrens payload = fnvlist_pack(nvl, &payload_len); 8199c3fd121SMatthew Ahrens drr->drr_payloadlen = payload_len; 8209c3fd121SMatthew Ahrens fnvlist_free(nvl); 8219c3fd121SMatthew Ahrens } 8229c3fd121SMatthew Ahrens 8239c3fd121SMatthew Ahrens err = dump_record(dsp, payload, payload_len); 8249c3fd121SMatthew Ahrens fnvlist_pack_free(payload, payload_len); 8259c3fd121SMatthew Ahrens if (err != 0) { 8264e3c9f44SBill Pijewski err = dsp->dsa_err; 8274e3c9f44SBill Pijewski goto out; 828efb80947Sahrens } 829efb80947Sahrens 830a2cdcdd2SPaul Dagnelie err = bqueue_init(&to_arg.q, zfs_send_queue_length, 831a2cdcdd2SPaul Dagnelie offsetof(struct send_block_record, ln)); 832a2cdcdd2SPaul Dagnelie to_arg.error_code = 0; 833a2cdcdd2SPaul Dagnelie to_arg.cancel = B_FALSE; 834a2cdcdd2SPaul Dagnelie to_arg.ds = to_ds; 835a2cdcdd2SPaul Dagnelie to_arg.fromtxg = fromtxg; 836a2cdcdd2SPaul Dagnelie to_arg.flags = TRAVERSE_PRE | TRAVERSE_PREFETCH; 837a2cdcdd2SPaul Dagnelie (void) thread_create(NULL, 0, send_traverse_thread, &to_arg, 0, curproc, 838a2cdcdd2SPaul Dagnelie TS_RUN, minclsyspri); 839a2cdcdd2SPaul Dagnelie 840a2cdcdd2SPaul Dagnelie struct send_block_record *to_data; 841a2cdcdd2SPaul Dagnelie to_data = bqueue_dequeue(&to_arg.q); 842a2cdcdd2SPaul Dagnelie 843a2cdcdd2SPaul Dagnelie while (!to_data->eos_marker && err == 0) { 844a2cdcdd2SPaul Dagnelie err = do_dump(dsp, to_data); 845a2cdcdd2SPaul Dagnelie to_data = get_next_record(&to_arg.q, to_data); 846a2cdcdd2SPaul Dagnelie if (issig(JUSTLOOKING) && issig(FORREAL)) 847a2cdcdd2SPaul Dagnelie err = EINTR; 848a2cdcdd2SPaul Dagnelie } 849a2cdcdd2SPaul Dagnelie 850a2cdcdd2SPaul Dagnelie if (err != 0) { 851a2cdcdd2SPaul Dagnelie to_arg.cancel = B_TRUE; 852a2cdcdd2SPaul Dagnelie while (!to_data->eos_marker) { 853a2cdcdd2SPaul Dagnelie to_data = get_next_record(&to_arg.q, to_data); 854a2cdcdd2SPaul Dagnelie } 855a2cdcdd2SPaul Dagnelie } 856a2cdcdd2SPaul Dagnelie kmem_free(to_data, sizeof (*to_data)); 857a2cdcdd2SPaul Dagnelie 858a2cdcdd2SPaul Dagnelie bqueue_destroy(&to_arg.q); 859a2cdcdd2SPaul Dagnelie 860a2cdcdd2SPaul Dagnelie if (err == 0 && to_arg.error_code != 0) 861a2cdcdd2SPaul Dagnelie err = to_arg.error_code; 862a2cdcdd2SPaul Dagnelie 863a2cdcdd2SPaul Dagnelie if (err != 0) 864a2cdcdd2SPaul Dagnelie goto out; 865efb80947Sahrens 8664e3c9f44SBill Pijewski if (dsp->dsa_pending_op != PENDING_NONE) 86798110f08SMatthew Ahrens if (dump_record(dsp, NULL, 0) != 0) 868be6fd75aSMatthew Ahrens err = SET_ERROR(EINTR); 8699e69d7d0SLori Alt 8703b2aab18SMatthew Ahrens if (err != 0) { 8713b2aab18SMatthew Ahrens if (err == EINTR && dsp->dsa_err != 0) 8724e3c9f44SBill Pijewski err = dsp->dsa_err; 8734e3c9f44SBill Pijewski goto out; 874efb80947Sahrens } 875efb80947Sahrens 876efb80947Sahrens bzero(drr, sizeof (dmu_replay_record_t)); 877efb80947Sahrens drr->drr_type = DRR_END; 8784e3c9f44SBill Pijewski drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; 8794e3c9f44SBill Pijewski drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; 880efb80947Sahrens 881a2cdcdd2SPaul Dagnelie if (dump_record(dsp, NULL, 0) != 0) 8824e3c9f44SBill Pijewski err = dsp->dsa_err; 883efb80947Sahrens 8844e3c9f44SBill Pijewski out: 885a2cdcdd2SPaul Dagnelie mutex_enter(&to_ds->ds_sendstream_lock); 886a2cdcdd2SPaul Dagnelie list_remove(&to_ds->ds_sendstreams, dsp); 887a2cdcdd2SPaul Dagnelie mutex_exit(&to_ds->ds_sendstream_lock); 8884e3c9f44SBill Pijewski 889*12b90ee2SMatt Krantz VERIFY(err != 0 || (dsp->dsa_sent_begin && dsp->dsa_sent_end)); 890*12b90ee2SMatt Krantz 891efb80947Sahrens kmem_free(drr, sizeof (dmu_replay_record_t)); 8924e3c9f44SBill Pijewski kmem_free(dsp, sizeof (dmu_sendarg_t)); 893efb80947Sahrens 894a2cdcdd2SPaul Dagnelie dsl_dataset_long_rele(to_ds, FTAG); 8953b2aab18SMatthew Ahrens 8964e3c9f44SBill Pijewski return (err); 897efb80947Sahrens } 898efb80947Sahrens 89919b94df9SMatthew Ahrens int 9003b2aab18SMatthew Ahrens dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, 901b5152584SMatthew Ahrens boolean_t embedok, boolean_t large_block_ok, 902b5152584SMatthew Ahrens int outfd, vnode_t *vp, offset_t *off) 9033b2aab18SMatthew Ahrens { 9043b2aab18SMatthew Ahrens dsl_pool_t *dp; 9053b2aab18SMatthew Ahrens dsl_dataset_t *ds; 9063b2aab18SMatthew Ahrens dsl_dataset_t *fromds = NULL; 9073b2aab18SMatthew Ahrens int err; 9083b2aab18SMatthew Ahrens 9093b2aab18SMatthew Ahrens err = dsl_pool_hold(pool, FTAG, &dp); 9103b2aab18SMatthew Ahrens if (err != 0) 9113b2aab18SMatthew Ahrens return (err); 9123b2aab18SMatthew Ahrens 9133b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); 9143b2aab18SMatthew Ahrens if (err != 0) { 9153b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 9163b2aab18SMatthew Ahrens return (err); 9173b2aab18SMatthew Ahrens } 9183b2aab18SMatthew Ahrens 9193b2aab18SMatthew Ahrens if (fromsnap != 0) { 92078f17100SMatthew Ahrens zfs_bookmark_phys_t zb; 92178f17100SMatthew Ahrens boolean_t is_clone; 92278f17100SMatthew Ahrens 9233b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); 9243b2aab18SMatthew Ahrens if (err != 0) { 9253b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 9263b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 9273b2aab18SMatthew Ahrens return (err); 9283b2aab18SMatthew Ahrens } 92978f17100SMatthew Ahrens if (!dsl_dataset_is_before(ds, fromds, 0)) 93078f17100SMatthew Ahrens err = SET_ERROR(EXDEV); 931c1379625SJustin T. Gibbs zb.zbm_creation_time = 932c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_time; 933c1379625SJustin T. Gibbs zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; 934c1379625SJustin T. Gibbs zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 93578f17100SMatthew Ahrens is_clone = (fromds->ds_dir != ds->ds_dir); 93678f17100SMatthew Ahrens dsl_dataset_rele(fromds, FTAG); 937b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 9389c3fd121SMatthew Ahrens embedok, large_block_ok, outfd, 0, 0, vp, off); 93978f17100SMatthew Ahrens } else { 940b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 9419c3fd121SMatthew Ahrens embedok, large_block_ok, outfd, 0, 0, vp, off); 9423b2aab18SMatthew Ahrens } 94378f17100SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 94478f17100SMatthew Ahrens return (err); 9453b2aab18SMatthew Ahrens } 9463b2aab18SMatthew Ahrens 9473b2aab18SMatthew Ahrens int 9489c3fd121SMatthew Ahrens dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, 9499c3fd121SMatthew Ahrens boolean_t large_block_ok, int outfd, uint64_t resumeobj, uint64_t resumeoff, 9509c3fd121SMatthew Ahrens vnode_t *vp, offset_t *off) 9513b2aab18SMatthew Ahrens { 9523b2aab18SMatthew Ahrens dsl_pool_t *dp; 9533b2aab18SMatthew Ahrens dsl_dataset_t *ds; 9543b2aab18SMatthew Ahrens int err; 95578f17100SMatthew Ahrens boolean_t owned = B_FALSE; 9563b2aab18SMatthew Ahrens 95778f17100SMatthew Ahrens if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) 958be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 9593b2aab18SMatthew Ahrens 9603b2aab18SMatthew Ahrens err = dsl_pool_hold(tosnap, FTAG, &dp); 9613b2aab18SMatthew Ahrens if (err != 0) 9623b2aab18SMatthew Ahrens return (err); 9633b2aab18SMatthew Ahrens 96478f17100SMatthew Ahrens if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { 96578f17100SMatthew Ahrens /* 96678f17100SMatthew Ahrens * We are sending a filesystem or volume. Ensure 96778f17100SMatthew Ahrens * that it doesn't change by owning the dataset. 96878f17100SMatthew Ahrens */ 96978f17100SMatthew Ahrens err = dsl_dataset_own(dp, tosnap, FTAG, &ds); 97078f17100SMatthew Ahrens owned = B_TRUE; 97178f17100SMatthew Ahrens } else { 97278f17100SMatthew Ahrens err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); 97378f17100SMatthew Ahrens } 9743b2aab18SMatthew Ahrens if (err != 0) { 9753b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 9763b2aab18SMatthew Ahrens return (err); 9773b2aab18SMatthew Ahrens } 9783b2aab18SMatthew Ahrens 9793b2aab18SMatthew Ahrens if (fromsnap != NULL) { 98078f17100SMatthew Ahrens zfs_bookmark_phys_t zb; 98178f17100SMatthew Ahrens boolean_t is_clone = B_FALSE; 98278f17100SMatthew Ahrens int fsnamelen = strchr(tosnap, '@') - tosnap; 98378f17100SMatthew Ahrens 98478f17100SMatthew Ahrens /* 98578f17100SMatthew Ahrens * If the fromsnap is in a different filesystem, then 98678f17100SMatthew Ahrens * mark the send stream as a clone. 98778f17100SMatthew Ahrens */ 98878f17100SMatthew Ahrens if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || 98978f17100SMatthew Ahrens (fromsnap[fsnamelen] != '@' && 99078f17100SMatthew Ahrens fromsnap[fsnamelen] != '#')) { 99178f17100SMatthew Ahrens is_clone = B_TRUE; 99278f17100SMatthew Ahrens } 99378f17100SMatthew Ahrens 99478f17100SMatthew Ahrens if (strchr(fromsnap, '@')) { 99578f17100SMatthew Ahrens dsl_dataset_t *fromds; 99678f17100SMatthew Ahrens err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); 99778f17100SMatthew Ahrens if (err == 0) { 99878f17100SMatthew Ahrens if (!dsl_dataset_is_before(ds, fromds, 0)) 99978f17100SMatthew Ahrens err = SET_ERROR(EXDEV); 100078f17100SMatthew Ahrens zb.zbm_creation_time = 1001c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_time; 100278f17100SMatthew Ahrens zb.zbm_creation_txg = 1003c1379625SJustin T. Gibbs dsl_dataset_phys(fromds)->ds_creation_txg; 1004c1379625SJustin T. Gibbs zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; 100578f17100SMatthew Ahrens is_clone = (ds->ds_dir != fromds->ds_dir); 100678f17100SMatthew Ahrens dsl_dataset_rele(fromds, FTAG); 100778f17100SMatthew Ahrens } 100878f17100SMatthew Ahrens } else { 100978f17100SMatthew Ahrens err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); 101078f17100SMatthew Ahrens } 10113b2aab18SMatthew Ahrens if (err != 0) { 10123b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 10133b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 10143b2aab18SMatthew Ahrens return (err); 10153b2aab18SMatthew Ahrens } 1016b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, 10179c3fd121SMatthew Ahrens embedok, large_block_ok, 10189c3fd121SMatthew Ahrens outfd, resumeobj, resumeoff, vp, off); 101978f17100SMatthew Ahrens } else { 1020b5152584SMatthew Ahrens err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, 10219c3fd121SMatthew Ahrens embedok, large_block_ok, 10229c3fd121SMatthew Ahrens outfd, resumeobj, resumeoff, vp, off); 10233b2aab18SMatthew Ahrens } 102478f17100SMatthew Ahrens if (owned) 102578f17100SMatthew Ahrens dsl_dataset_disown(ds, FTAG); 102678f17100SMatthew Ahrens else 102778f17100SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 102878f17100SMatthew Ahrens return (err); 10293b2aab18SMatthew Ahrens } 10303b2aab18SMatthew Ahrens 1031643da460SMax Grossman static int 1032643da460SMax Grossman dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t size, 1033643da460SMax Grossman uint64_t *sizep) 1034643da460SMax Grossman { 1035643da460SMax Grossman int err; 1036643da460SMax Grossman /* 1037643da460SMax Grossman * Assume that space (both on-disk and in-stream) is dominated by 1038643da460SMax Grossman * data. We will adjust for indirect blocks and the copies property, 1039643da460SMax Grossman * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). 1040643da460SMax Grossman */ 1041643da460SMax Grossman 1042643da460SMax Grossman /* 1043643da460SMax Grossman * Subtract out approximate space used by indirect blocks. 1044643da460SMax Grossman * Assume most space is used by data blocks (non-indirect, non-dnode). 1045643da460SMax Grossman * Assume all blocks are recordsize. Assume ditto blocks and 1046643da460SMax Grossman * internal fragmentation counter out compression. 1047643da460SMax Grossman * 1048643da460SMax Grossman * Therefore, space used by indirect blocks is sizeof(blkptr_t) per 1049643da460SMax Grossman * block, which we observe in practice. 1050643da460SMax Grossman */ 1051643da460SMax Grossman uint64_t recordsize; 1052643da460SMax Grossman err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize); 1053643da460SMax Grossman if (err != 0) 1054643da460SMax Grossman return (err); 1055643da460SMax Grossman size -= size / recordsize * sizeof (blkptr_t); 1056643da460SMax Grossman 1057643da460SMax Grossman /* Add in the space for the record associated with each block. */ 1058643da460SMax Grossman size += size / recordsize * sizeof (dmu_replay_record_t); 1059643da460SMax Grossman 1060643da460SMax Grossman *sizep = size; 1061643da460SMax Grossman 1062643da460SMax Grossman return (0); 1063643da460SMax Grossman } 1064643da460SMax Grossman 10653b2aab18SMatthew Ahrens int 10663b2aab18SMatthew Ahrens dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep) 106719b94df9SMatthew Ahrens { 106819b94df9SMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 106919b94df9SMatthew Ahrens int err; 107019b94df9SMatthew Ahrens uint64_t size; 107119b94df9SMatthew Ahrens 10723b2aab18SMatthew Ahrens ASSERT(dsl_pool_config_held(dp)); 10733b2aab18SMatthew Ahrens 107419b94df9SMatthew Ahrens /* tosnap must be a snapshot */ 1075bc9014e6SJustin Gibbs if (!ds->ds_is_snapshot) 1076be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 107719b94df9SMatthew Ahrens 107824218bebSAndriy Gapon /* fromsnap, if provided, must be a snapshot */ 107924218bebSAndriy Gapon if (fromds != NULL && !fromds->ds_is_snapshot) 108024218bebSAndriy Gapon return (SET_ERROR(EINVAL)); 108124218bebSAndriy Gapon 10824445fffbSMatthew Ahrens /* 10834445fffbSMatthew Ahrens * fromsnap must be an earlier snapshot from the same fs as tosnap, 10844445fffbSMatthew Ahrens * or the origin's fs. 10854445fffbSMatthew Ahrens */ 108678f17100SMatthew Ahrens if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) 1087be6fd75aSMatthew Ahrens return (SET_ERROR(EXDEV)); 108819b94df9SMatthew Ahrens 108919b94df9SMatthew Ahrens /* Get uncompressed size estimate of changed data. */ 109019b94df9SMatthew Ahrens if (fromds == NULL) { 1091c1379625SJustin T. Gibbs size = dsl_dataset_phys(ds)->ds_uncompressed_bytes; 109219b94df9SMatthew Ahrens } else { 109319b94df9SMatthew Ahrens uint64_t used, comp; 109419b94df9SMatthew Ahrens err = dsl_dataset_space_written(fromds, ds, 109519b94df9SMatthew Ahrens &used, &comp, &size); 10963b2aab18SMatthew Ahrens if (err != 0) 109719b94df9SMatthew Ahrens return (err); 109819b94df9SMatthew Ahrens } 109919b94df9SMatthew Ahrens 1100643da460SMax Grossman err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep); 1101643da460SMax Grossman return (err); 1102643da460SMax Grossman } 1103643da460SMax Grossman 1104643da460SMax Grossman /* 1105643da460SMax Grossman * Simple callback used to traverse the blocks of a snapshot and sum their 1106643da460SMax Grossman * uncompressed size 1107643da460SMax Grossman */ 1108643da460SMax Grossman /* ARGSUSED */ 1109643da460SMax Grossman static int 1110643da460SMax Grossman dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1111643da460SMax Grossman const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1112643da460SMax Grossman { 1113643da460SMax Grossman uint64_t *spaceptr = arg; 1114643da460SMax Grossman if (bp != NULL && !BP_IS_HOLE(bp)) { 1115643da460SMax Grossman *spaceptr += BP_GET_UCSIZE(bp); 1116643da460SMax Grossman } 1117643da460SMax Grossman return (0); 1118643da460SMax Grossman } 1119643da460SMax Grossman 1120643da460SMax Grossman /* 1121643da460SMax Grossman * Given a desination snapshot and a TXG, calculate the approximate size of a 1122643da460SMax Grossman * send stream sent from that TXG. from_txg may be zero, indicating that the 1123643da460SMax Grossman * whole snapshot will be sent. 1124643da460SMax Grossman */ 1125643da460SMax Grossman int 1126643da460SMax Grossman dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg, 1127643da460SMax Grossman uint64_t *sizep) 1128643da460SMax Grossman { 1129643da460SMax Grossman dsl_pool_t *dp = ds->ds_dir->dd_pool; 1130643da460SMax Grossman int err; 1131643da460SMax Grossman uint64_t size = 0; 1132643da460SMax Grossman 1133643da460SMax Grossman ASSERT(dsl_pool_config_held(dp)); 1134643da460SMax Grossman 1135643da460SMax Grossman /* tosnap must be a snapshot */ 1136643da460SMax Grossman if (!dsl_dataset_is_snapshot(ds)) 1137643da460SMax Grossman return (SET_ERROR(EINVAL)); 1138643da460SMax Grossman 1139643da460SMax Grossman /* verify that from_txg is before the provided snapshot was taken */ 1140643da460SMax Grossman if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) { 1141643da460SMax Grossman return (SET_ERROR(EXDEV)); 1142643da460SMax Grossman } 114319b94df9SMatthew Ahrens 114419b94df9SMatthew Ahrens /* 1145643da460SMax Grossman * traverse the blocks of the snapshot with birth times after 1146643da460SMax Grossman * from_txg, summing their uncompressed size 114719b94df9SMatthew Ahrens */ 1148643da460SMax Grossman err = traverse_dataset(ds, from_txg, TRAVERSE_POST, 1149643da460SMax Grossman dmu_calculate_send_traversal, &size); 1150643da460SMax Grossman if (err) 115119b94df9SMatthew Ahrens return (err); 115219b94df9SMatthew Ahrens 1153643da460SMax Grossman err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep); 1154643da460SMax Grossman return (err); 115519b94df9SMatthew Ahrens } 115619b94df9SMatthew Ahrens 11573b2aab18SMatthew Ahrens typedef struct dmu_recv_begin_arg { 11583b2aab18SMatthew Ahrens const char *drba_origin; 11593b2aab18SMatthew Ahrens dmu_recv_cookie_t *drba_cookie; 11603b2aab18SMatthew Ahrens cred_t *drba_cred; 116134f2f8cfSMatthew Ahrens uint64_t drba_snapobj; 11623b2aab18SMatthew Ahrens } dmu_recv_begin_arg_t; 1163f18faf3fSek 1164f18faf3fSek static int 11653b2aab18SMatthew Ahrens recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, 11663b2aab18SMatthew Ahrens uint64_t fromguid) 1167f18faf3fSek { 11683cb34c60Sahrens uint64_t val; 11693b2aab18SMatthew Ahrens int error; 11703b2aab18SMatthew Ahrens dsl_pool_t *dp = ds->ds_dir->dd_pool; 1171f18faf3fSek 11723b2aab18SMatthew Ahrens /* temporary clone name must not exist */ 11733b2aab18SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, 1174c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, 11753b2aab18SMatthew Ahrens 8, 1, &val); 11763b2aab18SMatthew Ahrens if (error != ENOENT) 11773b2aab18SMatthew Ahrens return (error == 0 ? EBUSY : error); 11783b2aab18SMatthew Ahrens 1179feaa74e4SMark Maybee /* new snapshot name must not exist */ 11803b2aab18SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, 1181c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1182c1379625SJustin T. Gibbs drba->drba_cookie->drc_tosnap, 8, 1, &val); 11833b2aab18SMatthew Ahrens if (error != ENOENT) 11843b2aab18SMatthew Ahrens return (error == 0 ? EEXIST : error); 1185feaa74e4SMark Maybee 1186a2afb611SJerry Jelinek /* 1187a2afb611SJerry Jelinek * Check snapshot limit before receiving. We'll recheck again at the 1188a2afb611SJerry Jelinek * end, but might as well abort before receiving if we're already over 1189a2afb611SJerry Jelinek * the limit. 1190a2afb611SJerry Jelinek * 1191a2afb611SJerry Jelinek * Note that we do not check the file system limit with 1192a2afb611SJerry Jelinek * dsl_dir_fscount_check because the temporary %clones don't count 1193a2afb611SJerry Jelinek * against that limit. 1194a2afb611SJerry Jelinek */ 1195a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, 1196a2afb611SJerry Jelinek NULL, drba->drba_cred); 1197a2afb611SJerry Jelinek if (error != 0) 1198a2afb611SJerry Jelinek return (error); 1199a2afb611SJerry Jelinek 12003b2aab18SMatthew Ahrens if (fromguid != 0) { 120134f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 1202c1379625SJustin T. Gibbs uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 120334f2f8cfSMatthew Ahrens 120434f2f8cfSMatthew Ahrens /* Find snapshot in this dir that matches fromguid. */ 120534f2f8cfSMatthew Ahrens while (obj != 0) { 120634f2f8cfSMatthew Ahrens error = dsl_dataset_hold_obj(dp, obj, FTAG, 120734f2f8cfSMatthew Ahrens &snap); 120834f2f8cfSMatthew Ahrens if (error != 0) 120934f2f8cfSMatthew Ahrens return (SET_ERROR(ENODEV)); 121034f2f8cfSMatthew Ahrens if (snap->ds_dir != ds->ds_dir) { 121134f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 121234f2f8cfSMatthew Ahrens return (SET_ERROR(ENODEV)); 121334f2f8cfSMatthew Ahrens } 1214c1379625SJustin T. Gibbs if (dsl_dataset_phys(snap)->ds_guid == fromguid) 121534f2f8cfSMatthew Ahrens break; 1216c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 121734f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 121834f2f8cfSMatthew Ahrens } 121934f2f8cfSMatthew Ahrens if (obj == 0) 1220be6fd75aSMatthew Ahrens return (SET_ERROR(ENODEV)); 122192241e0bSTom Erickson 122234f2f8cfSMatthew Ahrens if (drba->drba_cookie->drc_force) { 122334f2f8cfSMatthew Ahrens drba->drba_snapobj = obj; 122434f2f8cfSMatthew Ahrens } else { 122534f2f8cfSMatthew Ahrens /* 122634f2f8cfSMatthew Ahrens * If we are not forcing, there must be no 122734f2f8cfSMatthew Ahrens * changes since fromsnap. 122834f2f8cfSMatthew Ahrens */ 122934f2f8cfSMatthew Ahrens if (dsl_dataset_modified_since_snap(ds, snap)) { 123092241e0bSTom Erickson dsl_dataset_rele(snap, FTAG); 123134f2f8cfSMatthew Ahrens return (SET_ERROR(ETXTBSY)); 123292241e0bSTom Erickson } 123334f2f8cfSMatthew Ahrens drba->drba_snapobj = ds->ds_prev->ds_object; 123492241e0bSTom Erickson } 123534f2f8cfSMatthew Ahrens 123634f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 1237ae46e4c7SMatthew Ahrens } else { 12385bae108fSAndriy Gapon /* if full, then must be forced */ 12395bae108fSAndriy Gapon if (!drba->drba_cookie->drc_force) 12405bae108fSAndriy Gapon return (SET_ERROR(EEXIST)); 12415bae108fSAndriy Gapon /* start from $ORIGIN@$ORIGIN, if supported */ 12425bae108fSAndriy Gapon drba->drba_snapobj = dp->dp_origin_snap != NULL ? 12435bae108fSAndriy Gapon dp->dp_origin_snap->ds_object : 0; 1244ae46e4c7SMatthew Ahrens } 12453cb34c60Sahrens 12463cb34c60Sahrens return (0); 12473b2aab18SMatthew Ahrens 12483b2aab18SMatthew Ahrens } 12493b2aab18SMatthew Ahrens 12503b2aab18SMatthew Ahrens static int 12513b2aab18SMatthew Ahrens dmu_recv_begin_check(void *arg, dmu_tx_t *tx) 12523b2aab18SMatthew Ahrens { 12533b2aab18SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 12543b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 12553b2aab18SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 12563b2aab18SMatthew Ahrens uint64_t fromguid = drrb->drr_fromguid; 12573b2aab18SMatthew Ahrens int flags = drrb->drr_flags; 12583b2aab18SMatthew Ahrens int error; 12595d7b4d43SMatthew Ahrens uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 12603b2aab18SMatthew Ahrens dsl_dataset_t *ds; 12613b2aab18SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 12623b2aab18SMatthew Ahrens 12633b2aab18SMatthew Ahrens /* already checked */ 12643b2aab18SMatthew Ahrens ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 12659c3fd121SMatthew Ahrens ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); 12663b2aab18SMatthew Ahrens 12673b2aab18SMatthew Ahrens if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 12683b2aab18SMatthew Ahrens DMU_COMPOUNDSTREAM || 12693b2aab18SMatthew Ahrens drrb->drr_type >= DMU_OST_NUMTYPES || 12703b2aab18SMatthew Ahrens ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) 1271be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 12723b2aab18SMatthew Ahrens 12733b2aab18SMatthew Ahrens /* Verify pool version supports SA if SA_SPILL feature set */ 12745d7b4d43SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 12755d7b4d43SMatthew Ahrens spa_version(dp->dp_spa) < SPA_VERSION_SA) 12765d7b4d43SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 12775d7b4d43SMatthew Ahrens 12789c3fd121SMatthew Ahrens if (drba->drba_cookie->drc_resumable && 12799c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) 12809c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 12819c3fd121SMatthew Ahrens 12825d7b4d43SMatthew Ahrens /* 12835d7b4d43SMatthew Ahrens * The receiving code doesn't know how to translate a WRITE_EMBEDDED 12845d7b4d43SMatthew Ahrens * record to a plan WRITE record, so the pool must have the 12855d7b4d43SMatthew Ahrens * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 12865d7b4d43SMatthew Ahrens * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 12875d7b4d43SMatthew Ahrens */ 12885d7b4d43SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 12895d7b4d43SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 12905d7b4d43SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 12915d7b4d43SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) && 12925d7b4d43SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 1293be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 12943b2aab18SMatthew Ahrens 1295b5152584SMatthew Ahrens /* 1296b5152584SMatthew Ahrens * The receiving code doesn't know how to translate large blocks 1297b5152584SMatthew Ahrens * to smaller ones, so the pool must have the LARGE_BLOCKS 1298b5152584SMatthew Ahrens * feature enabled if the stream has LARGE_BLOCKS. 1299b5152584SMatthew Ahrens */ 1300b5152584SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && 1301b5152584SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 1302b5152584SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 1303b5152584SMatthew Ahrens 13043b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 13053b2aab18SMatthew Ahrens if (error == 0) { 13063b2aab18SMatthew Ahrens /* target fs already exists; recv into temp clone */ 13073b2aab18SMatthew Ahrens 13083b2aab18SMatthew Ahrens /* Can't recv a clone into an existing fs */ 130968ecb2ecSPaul Dagnelie if (flags & DRR_FLAG_CLONE || drba->drba_origin) { 13103b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1311be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 13123b2aab18SMatthew Ahrens } 13133b2aab18SMatthew Ahrens 13143b2aab18SMatthew Ahrens error = recv_begin_check_existing_impl(drba, ds, fromguid); 13153b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 13163b2aab18SMatthew Ahrens } else if (error == ENOENT) { 13173b2aab18SMatthew Ahrens /* target fs does not exist; must be a full backup or clone */ 13189adfa60dSMatthew Ahrens char buf[ZFS_MAX_DATASET_NAME_LEN]; 13193b2aab18SMatthew Ahrens 13203b2aab18SMatthew Ahrens /* 13213b2aab18SMatthew Ahrens * If it's a non-clone incremental, we are missing the 13223b2aab18SMatthew Ahrens * target fs, so fail the recv. 13233b2aab18SMatthew Ahrens */ 1324a2cdcdd2SPaul Dagnelie if (fromguid != 0 && !(flags & DRR_FLAG_CLONE || 1325a2cdcdd2SPaul Dagnelie drba->drba_origin)) 1326be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 13273b2aab18SMatthew Ahrens 132868ecb2ecSPaul Dagnelie /* 132968ecb2ecSPaul Dagnelie * If we're receiving a full send as a clone, and it doesn't 133068ecb2ecSPaul Dagnelie * contain all the necessary free records and freeobject 133168ecb2ecSPaul Dagnelie * records, reject it. 133268ecb2ecSPaul Dagnelie */ 133368ecb2ecSPaul Dagnelie if (fromguid == 0 && drba->drba_origin && 133468ecb2ecSPaul Dagnelie !(flags & DRR_FLAG_FREERECORDS)) 133568ecb2ecSPaul Dagnelie return (SET_ERROR(EINVAL)); 133668ecb2ecSPaul Dagnelie 13373b2aab18SMatthew Ahrens /* Open the parent of tofs */ 13389adfa60dSMatthew Ahrens ASSERT3U(strlen(tofs), <, sizeof (buf)); 13393b2aab18SMatthew Ahrens (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); 13403b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, buf, FTAG, &ds); 13413b2aab18SMatthew Ahrens if (error != 0) 13423b2aab18SMatthew Ahrens return (error); 13433b2aab18SMatthew Ahrens 1344a2afb611SJerry Jelinek /* 1345a2afb611SJerry Jelinek * Check filesystem and snapshot limits before receiving. We'll 1346a2afb611SJerry Jelinek * recheck snapshot limits again at the end (we create the 1347a2afb611SJerry Jelinek * filesystems and increment those counts during begin_sync). 1348a2afb611SJerry Jelinek */ 1349a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 1350a2afb611SJerry Jelinek ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); 1351a2afb611SJerry Jelinek if (error != 0) { 1352a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 1353a2afb611SJerry Jelinek return (error); 1354a2afb611SJerry Jelinek } 1355a2afb611SJerry Jelinek 1356a2afb611SJerry Jelinek error = dsl_fs_ss_limit_check(ds->ds_dir, 1, 1357a2afb611SJerry Jelinek ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); 1358a2afb611SJerry Jelinek if (error != 0) { 1359a2afb611SJerry Jelinek dsl_dataset_rele(ds, FTAG); 1360a2afb611SJerry Jelinek return (error); 1361a2afb611SJerry Jelinek } 1362a2afb611SJerry Jelinek 13633b2aab18SMatthew Ahrens if (drba->drba_origin != NULL) { 13643b2aab18SMatthew Ahrens dsl_dataset_t *origin; 13653b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, drba->drba_origin, 13663b2aab18SMatthew Ahrens FTAG, &origin); 13673b2aab18SMatthew Ahrens if (error != 0) { 13683b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 13693b2aab18SMatthew Ahrens return (error); 13703b2aab18SMatthew Ahrens } 1371bc9014e6SJustin Gibbs if (!origin->ds_is_snapshot) { 13723b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 13733b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1374be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 13753b2aab18SMatthew Ahrens } 137668ecb2ecSPaul Dagnelie if (dsl_dataset_phys(origin)->ds_guid != fromguid && 137768ecb2ecSPaul Dagnelie fromguid != 0) { 13783b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 13793b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 1380be6fd75aSMatthew Ahrens return (SET_ERROR(ENODEV)); 13813b2aab18SMatthew Ahrens } 13823b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 13833b2aab18SMatthew Ahrens } 13843b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 13853b2aab18SMatthew Ahrens error = 0; 13863b2aab18SMatthew Ahrens } 13873b2aab18SMatthew Ahrens return (error); 1388f18faf3fSek } 1389f18faf3fSek 1390f18faf3fSek static void 13913b2aab18SMatthew Ahrens dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) 1392f18faf3fSek { 13933b2aab18SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 13943b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 13959c3fd121SMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 13963b2aab18SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 13973b2aab18SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 13983b2aab18SMatthew Ahrens dsl_dataset_t *ds, *newds; 1399f18faf3fSek uint64_t dsobj; 14003b2aab18SMatthew Ahrens int error; 14019c3fd121SMatthew Ahrens uint64_t crflags = 0; 14023b2aab18SMatthew Ahrens 14039c3fd121SMatthew Ahrens if (drrb->drr_flags & DRR_FLAG_CI_DATA) 14049c3fd121SMatthew Ahrens crflags |= DS_FLAG_CI_DATASET; 1405f18faf3fSek 14063b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 14073b2aab18SMatthew Ahrens if (error == 0) { 14083b2aab18SMatthew Ahrens /* create temporary clone */ 140934f2f8cfSMatthew Ahrens dsl_dataset_t *snap = NULL; 141034f2f8cfSMatthew Ahrens if (drba->drba_snapobj != 0) { 141134f2f8cfSMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 141234f2f8cfSMatthew Ahrens drba->drba_snapobj, FTAG, &snap)); 141334f2f8cfSMatthew Ahrens } 14143b2aab18SMatthew Ahrens dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, 141534f2f8cfSMatthew Ahrens snap, crflags, drba->drba_cred, tx); 1416f40b29ceSPaul Dagnelie if (drba->drba_snapobj != 0) 1417f40b29ceSPaul Dagnelie dsl_dataset_rele(snap, FTAG); 14183b2aab18SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 14193b2aab18SMatthew Ahrens } else { 14203b2aab18SMatthew Ahrens dsl_dir_t *dd; 14213b2aab18SMatthew Ahrens const char *tail; 14223b2aab18SMatthew Ahrens dsl_dataset_t *origin = NULL; 14233b2aab18SMatthew Ahrens 14243b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); 14253b2aab18SMatthew Ahrens 14263b2aab18SMatthew Ahrens if (drba->drba_origin != NULL) { 14273b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, 14283b2aab18SMatthew Ahrens FTAG, &origin)); 14293b2aab18SMatthew Ahrens } 14303b2aab18SMatthew Ahrens 14313b2aab18SMatthew Ahrens /* Create new dataset. */ 14323b2aab18SMatthew Ahrens dsobj = dsl_dataset_create_sync(dd, 14333b2aab18SMatthew Ahrens strrchr(tofs, '/') + 1, 14343b2aab18SMatthew Ahrens origin, crflags, drba->drba_cred, tx); 14353b2aab18SMatthew Ahrens if (origin != NULL) 14363b2aab18SMatthew Ahrens dsl_dataset_rele(origin, FTAG); 14373b2aab18SMatthew Ahrens dsl_dir_rele(dd, FTAG); 14383b2aab18SMatthew Ahrens drba->drba_cookie->drc_newfs = B_TRUE; 14393b2aab18SMatthew Ahrens } 14403b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); 14413b2aab18SMatthew Ahrens 14429c3fd121SMatthew Ahrens if (drba->drba_cookie->drc_resumable) { 14439c3fd121SMatthew Ahrens dsl_dataset_zapify(newds, tx); 14449c3fd121SMatthew Ahrens if (drrb->drr_fromguid != 0) { 14459c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, 14469c3fd121SMatthew Ahrens 8, 1, &drrb->drr_fromguid, tx)); 14479c3fd121SMatthew Ahrens } 14489c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, 14499c3fd121SMatthew Ahrens 8, 1, &drrb->drr_toguid, tx)); 14509c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, 14519c3fd121SMatthew Ahrens 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); 14529c3fd121SMatthew Ahrens uint64_t one = 1; 14539c3fd121SMatthew Ahrens uint64_t zero = 0; 14549c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, 14559c3fd121SMatthew Ahrens 8, 1, &one, tx)); 14569c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, 14579c3fd121SMatthew Ahrens 8, 1, &zero, tx)); 14589c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, 14599c3fd121SMatthew Ahrens 8, 1, &zero, tx)); 14609c3fd121SMatthew Ahrens if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & 14619c3fd121SMatthew Ahrens DMU_BACKUP_FEATURE_EMBED_DATA) { 14629c3fd121SMatthew Ahrens VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, 14639c3fd121SMatthew Ahrens 8, 1, &one, tx)); 14649c3fd121SMatthew Ahrens } 14659c3fd121SMatthew Ahrens } 14669c3fd121SMatthew Ahrens 14673b2aab18SMatthew Ahrens dmu_buf_will_dirty(newds->ds_dbuf, tx); 1468c1379625SJustin T. Gibbs dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; 14693cb34c60Sahrens 1470ae46e4c7SMatthew Ahrens /* 1471ae46e4c7SMatthew Ahrens * If we actually created a non-clone, we need to create the 1472ae46e4c7SMatthew Ahrens * objset in our new dataset. 1473ae46e4c7SMatthew Ahrens */ 14743b2aab18SMatthew Ahrens if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { 1475ae46e4c7SMatthew Ahrens (void) dmu_objset_create_impl(dp->dp_spa, 14763b2aab18SMatthew Ahrens newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); 1477ae46e4c7SMatthew Ahrens } 1478ae46e4c7SMatthew Ahrens 14793b2aab18SMatthew Ahrens drba->drba_cookie->drc_ds = newds; 1480dc7cd546SMark Shellenbaum 14813b2aab18SMatthew Ahrens spa_history_log_internal_ds(newds, "receive", tx, ""); 1482dc7cd546SMark Shellenbaum } 1483dc7cd546SMark Shellenbaum 14849c3fd121SMatthew Ahrens static int 14859c3fd121SMatthew Ahrens dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) 14869c3fd121SMatthew Ahrens { 14879c3fd121SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 14889c3fd121SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 14899c3fd121SMatthew Ahrens struct drr_begin *drrb = drba->drba_cookie->drc_drrb; 14909c3fd121SMatthew Ahrens int error; 14919c3fd121SMatthew Ahrens uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); 14929c3fd121SMatthew Ahrens dsl_dataset_t *ds; 14939c3fd121SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 14949c3fd121SMatthew Ahrens 14959c3fd121SMatthew Ahrens /* already checked */ 14969c3fd121SMatthew Ahrens ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); 14979c3fd121SMatthew Ahrens ASSERT(featureflags & DMU_BACKUP_FEATURE_RESUMING); 14989c3fd121SMatthew Ahrens 14999c3fd121SMatthew Ahrens if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == 15009c3fd121SMatthew Ahrens DMU_COMPOUNDSTREAM || 15019c3fd121SMatthew Ahrens drrb->drr_type >= DMU_OST_NUMTYPES) 15029c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 15039c3fd121SMatthew Ahrens 15049c3fd121SMatthew Ahrens /* Verify pool version supports SA if SA_SPILL feature set */ 15059c3fd121SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && 15069c3fd121SMatthew Ahrens spa_version(dp->dp_spa) < SPA_VERSION_SA) 15079c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 15089c3fd121SMatthew Ahrens 15099c3fd121SMatthew Ahrens /* 15109c3fd121SMatthew Ahrens * The receiving code doesn't know how to translate a WRITE_EMBEDDED 15119c3fd121SMatthew Ahrens * record to a plain WRITE record, so the pool must have the 15129c3fd121SMatthew Ahrens * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED 15139c3fd121SMatthew Ahrens * records. Same with WRITE_EMBEDDED records that use LZ4 compression. 15149c3fd121SMatthew Ahrens */ 15159c3fd121SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && 15169c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) 15179c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 15189c3fd121SMatthew Ahrens if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) && 15199c3fd121SMatthew Ahrens !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) 15209c3fd121SMatthew Ahrens return (SET_ERROR(ENOTSUP)); 15219c3fd121SMatthew Ahrens 15229adfa60dSMatthew Ahrens /* 6 extra bytes for /%recv */ 15239adfa60dSMatthew Ahrens char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 15249c3fd121SMatthew Ahrens 15259c3fd121SMatthew Ahrens (void) snprintf(recvname, sizeof (recvname), "%s/%s", 15269c3fd121SMatthew Ahrens tofs, recv_clone_name); 15279c3fd121SMatthew Ahrens 15289c3fd121SMatthew Ahrens if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { 15299c3fd121SMatthew Ahrens /* %recv does not exist; continue in tofs */ 15309c3fd121SMatthew Ahrens error = dsl_dataset_hold(dp, tofs, FTAG, &ds); 15319c3fd121SMatthew Ahrens if (error != 0) 15329c3fd121SMatthew Ahrens return (error); 15339c3fd121SMatthew Ahrens } 15349c3fd121SMatthew Ahrens 15359c3fd121SMatthew Ahrens /* check that ds is marked inconsistent */ 15369c3fd121SMatthew Ahrens if (!DS_IS_INCONSISTENT(ds)) { 15379c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15389c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 15399c3fd121SMatthew Ahrens } 15409c3fd121SMatthew Ahrens 15419c3fd121SMatthew Ahrens /* check that there is resuming data, and that the toguid matches */ 15429c3fd121SMatthew Ahrens if (!dsl_dataset_is_zapified(ds)) { 15439c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15449c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 15459c3fd121SMatthew Ahrens } 15469c3fd121SMatthew Ahrens uint64_t val; 15479c3fd121SMatthew Ahrens error = zap_lookup(dp->dp_meta_objset, ds->ds_object, 15489c3fd121SMatthew Ahrens DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); 15499c3fd121SMatthew Ahrens if (error != 0 || drrb->drr_toguid != val) { 15509c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15519c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 15529c3fd121SMatthew Ahrens } 15539c3fd121SMatthew Ahrens 15549c3fd121SMatthew Ahrens /* 15559c3fd121SMatthew Ahrens * Check if the receive is still running. If so, it will be owned. 15569c3fd121SMatthew Ahrens * Note that nothing else can own the dataset (e.g. after the receive 15579c3fd121SMatthew Ahrens * fails) because it will be marked inconsistent. 15589c3fd121SMatthew Ahrens */ 15599c3fd121SMatthew Ahrens if (dsl_dataset_has_owner(ds)) { 15609c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15619c3fd121SMatthew Ahrens return (SET_ERROR(EBUSY)); 15629c3fd121SMatthew Ahrens } 15639c3fd121SMatthew Ahrens 15649c3fd121SMatthew Ahrens /* There should not be any snapshots of this fs yet. */ 15659c3fd121SMatthew Ahrens if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { 15669c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15679c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 15689c3fd121SMatthew Ahrens } 15699c3fd121SMatthew Ahrens 15709c3fd121SMatthew Ahrens /* 15719c3fd121SMatthew Ahrens * Note: resume point will be checked when we process the first WRITE 15729c3fd121SMatthew Ahrens * record. 15739c3fd121SMatthew Ahrens */ 15749c3fd121SMatthew Ahrens 15759c3fd121SMatthew Ahrens /* check that the origin matches */ 15769c3fd121SMatthew Ahrens val = 0; 15779c3fd121SMatthew Ahrens (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, 15789c3fd121SMatthew Ahrens DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); 15799c3fd121SMatthew Ahrens if (drrb->drr_fromguid != val) { 15809c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15819c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 15829c3fd121SMatthew Ahrens } 15839c3fd121SMatthew Ahrens 15849c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 15859c3fd121SMatthew Ahrens return (0); 15869c3fd121SMatthew Ahrens } 15879c3fd121SMatthew Ahrens 15889c3fd121SMatthew Ahrens static void 15899c3fd121SMatthew Ahrens dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) 15909c3fd121SMatthew Ahrens { 15919c3fd121SMatthew Ahrens dmu_recv_begin_arg_t *drba = arg; 15929c3fd121SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 15939c3fd121SMatthew Ahrens const char *tofs = drba->drba_cookie->drc_tofs; 15949c3fd121SMatthew Ahrens dsl_dataset_t *ds; 15959c3fd121SMatthew Ahrens uint64_t dsobj; 15969adfa60dSMatthew Ahrens /* 6 extra bytes for /%recv */ 15979adfa60dSMatthew Ahrens char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 15989c3fd121SMatthew Ahrens 15999c3fd121SMatthew Ahrens (void) snprintf(recvname, sizeof (recvname), "%s/%s", 16009c3fd121SMatthew Ahrens tofs, recv_clone_name); 16019c3fd121SMatthew Ahrens 16029c3fd121SMatthew Ahrens if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { 16039c3fd121SMatthew Ahrens /* %recv does not exist; continue in tofs */ 16049c3fd121SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, tofs, FTAG, &ds)); 16059c3fd121SMatthew Ahrens drba->drba_cookie->drc_newfs = B_TRUE; 16069c3fd121SMatthew Ahrens } 16079c3fd121SMatthew Ahrens 16089c3fd121SMatthew Ahrens /* clear the inconsistent flag so that we can own it */ 16099c3fd121SMatthew Ahrens ASSERT(DS_IS_INCONSISTENT(ds)); 16109c3fd121SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 16119c3fd121SMatthew Ahrens dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 16129c3fd121SMatthew Ahrens dsobj = ds->ds_object; 16139c3fd121SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 16149c3fd121SMatthew Ahrens 16159c3fd121SMatthew Ahrens VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &ds)); 16169c3fd121SMatthew Ahrens 16179c3fd121SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 16189c3fd121SMatthew Ahrens dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT; 16199c3fd121SMatthew Ahrens 16209c3fd121SMatthew Ahrens ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds))); 16219c3fd121SMatthew Ahrens 16229c3fd121SMatthew Ahrens drba->drba_cookie->drc_ds = ds; 16239c3fd121SMatthew Ahrens 16249c3fd121SMatthew Ahrens spa_history_log_internal_ds(ds, "resume receive", tx, ""); 16259c3fd121SMatthew Ahrens } 16269c3fd121SMatthew Ahrens 16273cb34c60Sahrens /* 16283cb34c60Sahrens * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() 16293cb34c60Sahrens * succeeds; otherwise we will leak the holds on the datasets. 16303cb34c60Sahrens */ 16313cb34c60Sahrens int 16329c3fd121SMatthew Ahrens dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, 16339c3fd121SMatthew Ahrens boolean_t force, boolean_t resumable, char *origin, dmu_recv_cookie_t *drc) 1634efb80947Sahrens { 16353b2aab18SMatthew Ahrens dmu_recv_begin_arg_t drba = { 0 }; 1636ab04eb8eStimh 16373cb34c60Sahrens bzero(drc, sizeof (dmu_recv_cookie_t)); 16389c3fd121SMatthew Ahrens drc->drc_drr_begin = drr_begin; 16399c3fd121SMatthew Ahrens drc->drc_drrb = &drr_begin->drr_u.drr_begin; 16403cb34c60Sahrens drc->drc_tosnap = tosnap; 16413b2aab18SMatthew Ahrens drc->drc_tofs = tofs; 16423cb34c60Sahrens drc->drc_force = force; 16439c3fd121SMatthew Ahrens drc->drc_resumable = resumable; 1644a2afb611SJerry Jelinek drc->drc_cred = CRED(); 1645efb80947Sahrens 16469c3fd121SMatthew Ahrens if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { 16473b2aab18SMatthew Ahrens drc->drc_byteswap = B_TRUE; 16489c3fd121SMatthew Ahrens fletcher_4_incremental_byteswap(drr_begin, 16493b2aab18SMatthew Ahrens sizeof (dmu_replay_record_t), &drc->drc_cksum); 16509c3fd121SMatthew Ahrens byteswap_record(drr_begin); 16519c3fd121SMatthew Ahrens } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { 16529c3fd121SMatthew Ahrens fletcher_4_incremental_native(drr_begin, 16533b2aab18SMatthew Ahrens sizeof (dmu_replay_record_t), &drc->drc_cksum); 16549c3fd121SMatthew Ahrens } else { 16559c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 16563cb34c60Sahrens } 16573cb34c60Sahrens 16583b2aab18SMatthew Ahrens drba.drba_origin = origin; 16593b2aab18SMatthew Ahrens drba.drba_cookie = drc; 16603b2aab18SMatthew Ahrens drba.drba_cred = CRED(); 16613b2aab18SMatthew Ahrens 16629c3fd121SMatthew Ahrens if (DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & 16639c3fd121SMatthew Ahrens DMU_BACKUP_FEATURE_RESUMING) { 16649c3fd121SMatthew Ahrens return (dsl_sync_task(tofs, 16659c3fd121SMatthew Ahrens dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, 16669c3fd121SMatthew Ahrens &drba, 5, ZFS_SPACE_CHECK_NORMAL)); 16679c3fd121SMatthew Ahrens } else { 16689c3fd121SMatthew Ahrens return (dsl_sync_task(tofs, 16699c3fd121SMatthew Ahrens dmu_recv_begin_check, dmu_recv_begin_sync, 16709c3fd121SMatthew Ahrens &drba, 5, ZFS_SPACE_CHECK_NORMAL)); 16719c3fd121SMatthew Ahrens } 1672efb80947Sahrens } 1673efb80947Sahrens 1674a2cdcdd2SPaul Dagnelie struct receive_record_arg { 1675a2cdcdd2SPaul Dagnelie dmu_replay_record_t header; 1676a2cdcdd2SPaul Dagnelie void *payload; /* Pointer to a buffer containing the payload */ 1677a2cdcdd2SPaul Dagnelie /* 1678a2cdcdd2SPaul Dagnelie * If the record is a write, pointer to the arc_buf_t containing the 1679a2cdcdd2SPaul Dagnelie * payload. 1680a2cdcdd2SPaul Dagnelie */ 1681a2cdcdd2SPaul Dagnelie arc_buf_t *write_buf; 1682a2cdcdd2SPaul Dagnelie int payload_size; 16839c3fd121SMatthew Ahrens uint64_t bytes_read; /* bytes read from stream when record created */ 1684a2cdcdd2SPaul Dagnelie boolean_t eos_marker; /* Marks the end of the stream */ 1685a2cdcdd2SPaul Dagnelie bqueue_node_t node; 1686a2cdcdd2SPaul Dagnelie }; 1687a2cdcdd2SPaul Dagnelie 1688a2cdcdd2SPaul Dagnelie struct receive_writer_arg { 168998110f08SMatthew Ahrens objset_t *os; 16903b2aab18SMatthew Ahrens boolean_t byteswap; 1691a2cdcdd2SPaul Dagnelie bqueue_t q; 16929c3fd121SMatthew Ahrens 1693a2cdcdd2SPaul Dagnelie /* 1694a2cdcdd2SPaul Dagnelie * These three args are used to signal to the main thread that we're 1695a2cdcdd2SPaul Dagnelie * done. 1696a2cdcdd2SPaul Dagnelie */ 1697a2cdcdd2SPaul Dagnelie kmutex_t mutex; 1698a2cdcdd2SPaul Dagnelie kcondvar_t cv; 1699a2cdcdd2SPaul Dagnelie boolean_t done; 17009c3fd121SMatthew Ahrens 1701a2cdcdd2SPaul Dagnelie int err; 1702a2cdcdd2SPaul Dagnelie /* A map from guid to dataset to help handle dedup'd streams. */ 1703a2cdcdd2SPaul Dagnelie avl_tree_t *guid_to_ds_map; 17049c3fd121SMatthew Ahrens boolean_t resumable; 17059c3fd121SMatthew Ahrens uint64_t last_object, last_offset; 17069c3fd121SMatthew Ahrens uint64_t bytes_read; /* bytes read when current record created */ 1707a2cdcdd2SPaul Dagnelie }; 170898110f08SMatthew Ahrens 170968ecb2ecSPaul Dagnelie struct objlist { 171068ecb2ecSPaul Dagnelie list_t list; /* List of struct receive_objnode. */ 171168ecb2ecSPaul Dagnelie /* 171268ecb2ecSPaul Dagnelie * Last object looked up. Used to assert that objects are being looked 171368ecb2ecSPaul Dagnelie * up in ascending order. 171468ecb2ecSPaul Dagnelie */ 171568ecb2ecSPaul Dagnelie uint64_t last_lookup; 171668ecb2ecSPaul Dagnelie }; 171768ecb2ecSPaul Dagnelie 171868ecb2ecSPaul Dagnelie struct receive_objnode { 171968ecb2ecSPaul Dagnelie list_node_t node; 172068ecb2ecSPaul Dagnelie uint64_t object; 172168ecb2ecSPaul Dagnelie }; 172268ecb2ecSPaul Dagnelie 1723a2cdcdd2SPaul Dagnelie struct receive_arg { 1724a2cdcdd2SPaul Dagnelie objset_t *os; 1725a2cdcdd2SPaul Dagnelie vnode_t *vp; /* The vnode to read the stream from */ 1726a2cdcdd2SPaul Dagnelie uint64_t voff; /* The current offset in the stream */ 17279c3fd121SMatthew Ahrens uint64_t bytes_read; 1728a2cdcdd2SPaul Dagnelie /* 1729a2cdcdd2SPaul Dagnelie * A record that has had its payload read in, but hasn't yet been handed 1730a2cdcdd2SPaul Dagnelie * off to the worker thread. 1731a2cdcdd2SPaul Dagnelie */ 1732a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd; 1733a2cdcdd2SPaul Dagnelie /* A record that has had its header read in, but not its payload. */ 1734a2cdcdd2SPaul Dagnelie struct receive_record_arg *next_rrd; 17353cb34c60Sahrens zio_cksum_t cksum; 173698110f08SMatthew Ahrens zio_cksum_t prev_cksum; 1737a2cdcdd2SPaul Dagnelie int err; 1738a2cdcdd2SPaul Dagnelie boolean_t byteswap; 1739a2cdcdd2SPaul Dagnelie /* Sorted list of objects not to issue prefetches for. */ 174068ecb2ecSPaul Dagnelie struct objlist ignore_objlist; 17413cb34c60Sahrens }; 17423cb34c60Sahrens 17439e69d7d0SLori Alt typedef struct guid_map_entry { 17449e69d7d0SLori Alt uint64_t guid; 17459e69d7d0SLori Alt dsl_dataset_t *gme_ds; 17469e69d7d0SLori Alt avl_node_t avlnode; 17479e69d7d0SLori Alt } guid_map_entry_t; 17489e69d7d0SLori Alt 17499e69d7d0SLori Alt static int 17509e69d7d0SLori Alt guid_compare(const void *arg1, const void *arg2) 17519e69d7d0SLori Alt { 17529e69d7d0SLori Alt const guid_map_entry_t *gmep1 = arg1; 17539e69d7d0SLori Alt const guid_map_entry_t *gmep2 = arg2; 17549e69d7d0SLori Alt 17559e69d7d0SLori Alt if (gmep1->guid < gmep2->guid) 17569e69d7d0SLori Alt return (-1); 17579e69d7d0SLori Alt else if (gmep1->guid > gmep2->guid) 17589e69d7d0SLori Alt return (1); 17599e69d7d0SLori Alt return (0); 17609e69d7d0SLori Alt } 17619e69d7d0SLori Alt 1762c99e4bdcSChris Kirby static void 1763c99e4bdcSChris Kirby free_guid_map_onexit(void *arg) 1764c99e4bdcSChris Kirby { 1765c99e4bdcSChris Kirby avl_tree_t *ca = arg; 1766c99e4bdcSChris Kirby void *cookie = NULL; 1767c99e4bdcSChris Kirby guid_map_entry_t *gmep; 1768c99e4bdcSChris Kirby 1769c99e4bdcSChris Kirby while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { 17703b2aab18SMatthew Ahrens dsl_dataset_long_rele(gmep->gme_ds, gmep); 1771de8d9cffSMatthew Ahrens dsl_dataset_rele(gmep->gme_ds, gmep); 1772c99e4bdcSChris Kirby kmem_free(gmep, sizeof (guid_map_entry_t)); 1773c99e4bdcSChris Kirby } 1774c99e4bdcSChris Kirby avl_destroy(ca); 1775c99e4bdcSChris Kirby kmem_free(ca, sizeof (avl_tree_t)); 1776c99e4bdcSChris Kirby } 1777c99e4bdcSChris Kirby 177898110f08SMatthew Ahrens static int 1779a2cdcdd2SPaul Dagnelie receive_read(struct receive_arg *ra, int len, void *buf) 1780efb80947Sahrens { 17813cb34c60Sahrens int done = 0; 1782efb80947Sahrens 1783c20404ffSEli Rosenthal /* 1784c20404ffSEli Rosenthal * The code doesn't rely on this (lengths being multiples of 8). See 1785c20404ffSEli Rosenthal * comment in dump_bytes. 1786c20404ffSEli Rosenthal */ 1787fb09f5aaSMadhav Suresh ASSERT0(len % 8); 1788efb80947Sahrens 17893cb34c60Sahrens while (done < len) { 1790efb80947Sahrens ssize_t resid; 1791efb80947Sahrens 1792efb80947Sahrens ra->err = vn_rdwr(UIO_READ, ra->vp, 179398110f08SMatthew Ahrens (char *)buf + done, len - done, 1794efb80947Sahrens ra->voff, UIO_SYSSPACE, FAPPEND, 1795efb80947Sahrens RLIM64_INFINITY, CRED(), &resid); 1796efb80947Sahrens 17979c3fd121SMatthew Ahrens if (resid == len - done) { 17989c3fd121SMatthew Ahrens /* 17999c3fd121SMatthew Ahrens * Note: ECKSUM indicates that the receive 18009c3fd121SMatthew Ahrens * was interrupted and can potentially be resumed. 18019c3fd121SMatthew Ahrens */ 18029c3fd121SMatthew Ahrens ra->err = SET_ERROR(ECKSUM); 18039c3fd121SMatthew Ahrens } 18043cb34c60Sahrens ra->voff += len - done - resid; 18053cb34c60Sahrens done = len - resid; 18063b2aab18SMatthew Ahrens if (ra->err != 0) 180798110f08SMatthew Ahrens return (ra->err); 1808efb80947Sahrens } 1809efb80947Sahrens 18109c3fd121SMatthew Ahrens ra->bytes_read += len; 18119c3fd121SMatthew Ahrens 18123cb34c60Sahrens ASSERT3U(done, ==, len); 181398110f08SMatthew Ahrens return (0); 1814efb80947Sahrens } 1815efb80947Sahrens 1816efb80947Sahrens static void 181798110f08SMatthew Ahrens byteswap_record(dmu_replay_record_t *drr) 1818efb80947Sahrens { 1819efb80947Sahrens #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) 1820efb80947Sahrens #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) 1821efb80947Sahrens drr->drr_type = BSWAP_32(drr->drr_type); 18223cb34c60Sahrens drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); 182398110f08SMatthew Ahrens 1824efb80947Sahrens switch (drr->drr_type) { 1825efb80947Sahrens case DRR_BEGIN: 1826efb80947Sahrens DO64(drr_begin.drr_magic); 18279e69d7d0SLori Alt DO64(drr_begin.drr_versioninfo); 1828efb80947Sahrens DO64(drr_begin.drr_creation_time); 1829efb80947Sahrens DO32(drr_begin.drr_type); 18303cb34c60Sahrens DO32(drr_begin.drr_flags); 1831efb80947Sahrens DO64(drr_begin.drr_toguid); 1832efb80947Sahrens DO64(drr_begin.drr_fromguid); 1833efb80947Sahrens break; 1834efb80947Sahrens case DRR_OBJECT: 1835efb80947Sahrens DO64(drr_object.drr_object); 1836efb80947Sahrens DO32(drr_object.drr_type); 1837efb80947Sahrens DO32(drr_object.drr_bonustype); 1838efb80947Sahrens DO32(drr_object.drr_blksz); 1839efb80947Sahrens DO32(drr_object.drr_bonuslen); 18409e69d7d0SLori Alt DO64(drr_object.drr_toguid); 1841efb80947Sahrens break; 1842efb80947Sahrens case DRR_FREEOBJECTS: 1843efb80947Sahrens DO64(drr_freeobjects.drr_firstobj); 1844efb80947Sahrens DO64(drr_freeobjects.drr_numobjs); 18459e69d7d0SLori Alt DO64(drr_freeobjects.drr_toguid); 1846efb80947Sahrens break; 1847efb80947Sahrens case DRR_WRITE: 1848efb80947Sahrens DO64(drr_write.drr_object); 1849efb80947Sahrens DO32(drr_write.drr_type); 1850efb80947Sahrens DO64(drr_write.drr_offset); 1851efb80947Sahrens DO64(drr_write.drr_length); 18529e69d7d0SLori Alt DO64(drr_write.drr_toguid); 185398110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); 18548e714474SLori Alt DO64(drr_write.drr_key.ddk_prop); 18559e69d7d0SLori Alt break; 18569e69d7d0SLori Alt case DRR_WRITE_BYREF: 18579e69d7d0SLori Alt DO64(drr_write_byref.drr_object); 18589e69d7d0SLori Alt DO64(drr_write_byref.drr_offset); 18599e69d7d0SLori Alt DO64(drr_write_byref.drr_length); 18609e69d7d0SLori Alt DO64(drr_write_byref.drr_toguid); 18619e69d7d0SLori Alt DO64(drr_write_byref.drr_refguid); 18629e69d7d0SLori Alt DO64(drr_write_byref.drr_refobject); 18639e69d7d0SLori Alt DO64(drr_write_byref.drr_refoffset); 186498110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write_byref. 186598110f08SMatthew Ahrens drr_key.ddk_cksum); 18668e714474SLori Alt DO64(drr_write_byref.drr_key.ddk_prop); 1867efb80947Sahrens break; 18685d7b4d43SMatthew Ahrens case DRR_WRITE_EMBEDDED: 18695d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_object); 18705d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_offset); 18715d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_length); 18725d7b4d43SMatthew Ahrens DO64(drr_write_embedded.drr_toguid); 18735d7b4d43SMatthew Ahrens DO32(drr_write_embedded.drr_lsize); 18745d7b4d43SMatthew Ahrens DO32(drr_write_embedded.drr_psize); 18755d7b4d43SMatthew Ahrens break; 1876efb80947Sahrens case DRR_FREE: 1877efb80947Sahrens DO64(drr_free.drr_object); 1878efb80947Sahrens DO64(drr_free.drr_offset); 1879efb80947Sahrens DO64(drr_free.drr_length); 18809e69d7d0SLori Alt DO64(drr_free.drr_toguid); 1881efb80947Sahrens break; 18820a586ceaSMark Shellenbaum case DRR_SPILL: 18830a586ceaSMark Shellenbaum DO64(drr_spill.drr_object); 18840a586ceaSMark Shellenbaum DO64(drr_spill.drr_length); 18850a586ceaSMark Shellenbaum DO64(drr_spill.drr_toguid); 18860a586ceaSMark Shellenbaum break; 1887efb80947Sahrens case DRR_END: 18889e69d7d0SLori Alt DO64(drr_end.drr_toguid); 188998110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); 1890efb80947Sahrens break; 1891efb80947Sahrens } 189298110f08SMatthew Ahrens 189398110f08SMatthew Ahrens if (drr->drr_type != DRR_BEGIN) { 189498110f08SMatthew Ahrens ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); 189598110f08SMatthew Ahrens } 189698110f08SMatthew Ahrens 1897efb80947Sahrens #undef DO64 1898efb80947Sahrens #undef DO32 1899efb80947Sahrens } 1900efb80947Sahrens 1901e77d42eaSMatthew Ahrens static inline uint8_t 1902e77d42eaSMatthew Ahrens deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) 1903e77d42eaSMatthew Ahrens { 1904e77d42eaSMatthew Ahrens if (bonus_type == DMU_OT_SA) { 1905e77d42eaSMatthew Ahrens return (1); 1906e77d42eaSMatthew Ahrens } else { 1907e77d42eaSMatthew Ahrens return (1 + 1908e77d42eaSMatthew Ahrens ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT)); 1909e77d42eaSMatthew Ahrens } 1910e77d42eaSMatthew Ahrens } 1911e77d42eaSMatthew Ahrens 19129c3fd121SMatthew Ahrens static void 19139c3fd121SMatthew Ahrens save_resume_state(struct receive_writer_arg *rwa, 19149c3fd121SMatthew Ahrens uint64_t object, uint64_t offset, dmu_tx_t *tx) 19159c3fd121SMatthew Ahrens { 19169c3fd121SMatthew Ahrens int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; 19179c3fd121SMatthew Ahrens 19189c3fd121SMatthew Ahrens if (!rwa->resumable) 19199c3fd121SMatthew Ahrens return; 19209c3fd121SMatthew Ahrens 19219c3fd121SMatthew Ahrens /* 19229c3fd121SMatthew Ahrens * We use ds_resume_bytes[] != 0 to indicate that we need to 19239c3fd121SMatthew Ahrens * update this on disk, so it must not be 0. 19249c3fd121SMatthew Ahrens */ 19259c3fd121SMatthew Ahrens ASSERT(rwa->bytes_read != 0); 19269c3fd121SMatthew Ahrens 19279c3fd121SMatthew Ahrens /* 19289c3fd121SMatthew Ahrens * We only resume from write records, which have a valid 19299c3fd121SMatthew Ahrens * (non-meta-dnode) object number. 19309c3fd121SMatthew Ahrens */ 19319c3fd121SMatthew Ahrens ASSERT(object != 0); 19329c3fd121SMatthew Ahrens 19339c3fd121SMatthew Ahrens /* 19349c3fd121SMatthew Ahrens * For resuming to work correctly, we must receive records in order, 19359c3fd121SMatthew Ahrens * sorted by object,offset. This is checked by the callers, but 19369c3fd121SMatthew Ahrens * assert it here for good measure. 19379c3fd121SMatthew Ahrens */ 19389c3fd121SMatthew Ahrens ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); 19399c3fd121SMatthew Ahrens ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || 19409c3fd121SMatthew Ahrens offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); 19419c3fd121SMatthew Ahrens ASSERT3U(rwa->bytes_read, >=, 19429c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); 19439c3fd121SMatthew Ahrens 19449c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; 19459c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; 19469c3fd121SMatthew Ahrens rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; 19479c3fd121SMatthew Ahrens } 19489c3fd121SMatthew Ahrens 1949efb80947Sahrens static int 1950a2cdcdd2SPaul Dagnelie receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, 1951a2cdcdd2SPaul Dagnelie void *data) 1952efb80947Sahrens { 1953e77d42eaSMatthew Ahrens dmu_object_info_t doi; 1954efb80947Sahrens dmu_tx_t *tx; 1955e77d42eaSMatthew Ahrens uint64_t object; 1956e77d42eaSMatthew Ahrens int err; 1957efb80947Sahrens 1958efb80947Sahrens if (drro->drr_type == DMU_OT_NONE || 1959ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drro->drr_type) || 1960ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drro->drr_bonustype) || 19619e69d7d0SLori Alt drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || 1962efb80947Sahrens drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || 1963efb80947Sahrens P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || 1964efb80947Sahrens drro->drr_blksz < SPA_MINBLOCKSIZE || 1965a2cdcdd2SPaul Dagnelie drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || 1966efb80947Sahrens drro->drr_bonuslen > DN_MAX_BONUSLEN) { 1967be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1968efb80947Sahrens } 1969efb80947Sahrens 1970a2cdcdd2SPaul Dagnelie err = dmu_object_info(rwa->os, drro->drr_object, &doi); 19712bf405a2SMark Maybee 19722bf405a2SMark Maybee if (err != 0 && err != ENOENT) 1973be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1974e77d42eaSMatthew Ahrens object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT; 19752bf405a2SMark Maybee 1976e77d42eaSMatthew Ahrens /* 1977e77d42eaSMatthew Ahrens * If we are losing blkptrs or changing the block size this must 1978e77d42eaSMatthew Ahrens * be a new file instance. We must clear out the previous file 1979e77d42eaSMatthew Ahrens * contents before we can change this type of metadata in the dnode. 1980e77d42eaSMatthew Ahrens */ 1981e77d42eaSMatthew Ahrens if (err == 0) { 1982e77d42eaSMatthew Ahrens int nblkptr; 1983e77d42eaSMatthew Ahrens 1984e77d42eaSMatthew Ahrens nblkptr = deduce_nblkptr(drro->drr_bonustype, 1985e77d42eaSMatthew Ahrens drro->drr_bonuslen); 1986e77d42eaSMatthew Ahrens 1987e77d42eaSMatthew Ahrens if (drro->drr_blksz != doi.doi_data_block_size || 1988e77d42eaSMatthew Ahrens nblkptr < doi.doi_nblkptr) { 1989a2cdcdd2SPaul Dagnelie err = dmu_free_long_range(rwa->os, drro->drr_object, 1990e77d42eaSMatthew Ahrens 0, DMU_OBJECT_END); 1991e77d42eaSMatthew Ahrens if (err != 0) 1992e77d42eaSMatthew Ahrens return (SET_ERROR(EINVAL)); 1993efb80947Sahrens } 1994e77d42eaSMatthew Ahrens } 1995e77d42eaSMatthew Ahrens 1996a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 1997e77d42eaSMatthew Ahrens dmu_tx_hold_bonus(tx, object); 1998e77d42eaSMatthew Ahrens err = dmu_tx_assign(tx, TXG_WAIT); 1999e77d42eaSMatthew Ahrens if (err != 0) { 2000e77d42eaSMatthew Ahrens dmu_tx_abort(tx); 2001e77d42eaSMatthew Ahrens return (err); 2002e77d42eaSMatthew Ahrens } 2003e77d42eaSMatthew Ahrens 2004e77d42eaSMatthew Ahrens if (object == DMU_NEW_OBJECT) { 2005e77d42eaSMatthew Ahrens /* currently free, want to be allocated */ 2006a2cdcdd2SPaul Dagnelie err = dmu_object_claim(rwa->os, drro->drr_object, 2007efb80947Sahrens drro->drr_type, drro->drr_blksz, 2008efb80947Sahrens drro->drr_bonustype, drro->drr_bonuslen, tx); 2009e77d42eaSMatthew Ahrens } else if (drro->drr_type != doi.doi_type || 2010e77d42eaSMatthew Ahrens drro->drr_blksz != doi.doi_data_block_size || 2011e77d42eaSMatthew Ahrens drro->drr_bonustype != doi.doi_bonus_type || 2012e77d42eaSMatthew Ahrens drro->drr_bonuslen != doi.doi_bonus_size) { 2013e77d42eaSMatthew Ahrens /* currently allocated, but with different properties */ 2014a2cdcdd2SPaul Dagnelie err = dmu_object_reclaim(rwa->os, drro->drr_object, 2015efb80947Sahrens drro->drr_type, drro->drr_blksz, 2016e77d42eaSMatthew Ahrens drro->drr_bonustype, drro->drr_bonuslen, tx); 2017efb80947Sahrens } 20183b2aab18SMatthew Ahrens if (err != 0) { 2019e77d42eaSMatthew Ahrens dmu_tx_commit(tx); 2020be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 20210a586ceaSMark Shellenbaum } 20222bf405a2SMark Maybee 2023a2cdcdd2SPaul Dagnelie dmu_object_set_checksum(rwa->os, drro->drr_object, 202498110f08SMatthew Ahrens drro->drr_checksumtype, tx); 2025a2cdcdd2SPaul Dagnelie dmu_object_set_compress(rwa->os, drro->drr_object, 202698110f08SMatthew Ahrens drro->drr_compress, tx); 2027efb80947Sahrens 2028adee0b6fSTim Haley if (data != NULL) { 2029efb80947Sahrens dmu_buf_t *db; 2030adee0b6fSTim Haley 2031a2cdcdd2SPaul Dagnelie VERIFY0(dmu_bonus_hold(rwa->os, drro->drr_object, FTAG, &db)); 2032efb80947Sahrens dmu_buf_will_dirty(db, tx); 2033efb80947Sahrens 20341934e92fSmaybee ASSERT3U(db->db_size, >=, drro->drr_bonuslen); 20351934e92fSmaybee bcopy(data, db->db_data, drro->drr_bonuslen); 2036a2cdcdd2SPaul Dagnelie if (rwa->byteswap) { 2037ad135b5dSChristopher Siden dmu_object_byteswap_t byteswap = 2038ad135b5dSChristopher Siden DMU_OT_BYTESWAP(drro->drr_bonustype); 2039ad135b5dSChristopher Siden dmu_ot_byteswap[byteswap].ob_func(db->db_data, 2040efb80947Sahrens drro->drr_bonuslen); 2041efb80947Sahrens } 2042efb80947Sahrens dmu_buf_rele(db, FTAG); 2043efb80947Sahrens } 2044efb80947Sahrens dmu_tx_commit(tx); 20459c3fd121SMatthew Ahrens 2046efb80947Sahrens return (0); 2047efb80947Sahrens } 2048efb80947Sahrens 2049efb80947Sahrens /* ARGSUSED */ 2050efb80947Sahrens static int 2051a2cdcdd2SPaul Dagnelie receive_freeobjects(struct receive_writer_arg *rwa, 2052efb80947Sahrens struct drr_freeobjects *drrfo) 2053efb80947Sahrens { 2054efb80947Sahrens uint64_t obj; 205568ecb2ecSPaul Dagnelie int next_err = 0; 2056efb80947Sahrens 2057efb80947Sahrens if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) 2058be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2059efb80947Sahrens 2060efb80947Sahrens for (obj = drrfo->drr_firstobj; 206168ecb2ecSPaul Dagnelie obj < drrfo->drr_firstobj + drrfo->drr_numobjs && next_err == 0; 206268ecb2ecSPaul Dagnelie next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { 2063efb80947Sahrens int err; 2064efb80947Sahrens 2065a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, obj, NULL) != 0) 2066efb80947Sahrens continue; 2067efb80947Sahrens 2068a2cdcdd2SPaul Dagnelie err = dmu_free_long_object(rwa->os, obj); 20693b2aab18SMatthew Ahrens if (err != 0) 2070efb80947Sahrens return (err); 2071efb80947Sahrens } 207268ecb2ecSPaul Dagnelie if (next_err != ESRCH) 207368ecb2ecSPaul Dagnelie return (next_err); 2074efb80947Sahrens return (0); 2075efb80947Sahrens } 2076efb80947Sahrens 2077efb80947Sahrens static int 2078a2cdcdd2SPaul Dagnelie receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw, 2079a2cdcdd2SPaul Dagnelie arc_buf_t *abuf) 2080efb80947Sahrens { 2081efb80947Sahrens dmu_tx_t *tx; 2082efb80947Sahrens int err; 2083efb80947Sahrens 2084efb80947Sahrens if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset || 2085ad135b5dSChristopher Siden !DMU_OT_IS_VALID(drrw->drr_type)) 2086be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2087efb80947Sahrens 20889c3fd121SMatthew Ahrens /* 20899c3fd121SMatthew Ahrens * For resuming to work, records must be in increasing order 20909c3fd121SMatthew Ahrens * by (object, offset). 20919c3fd121SMatthew Ahrens */ 20929c3fd121SMatthew Ahrens if (drrw->drr_object < rwa->last_object || 20939c3fd121SMatthew Ahrens (drrw->drr_object == rwa->last_object && 20949c3fd121SMatthew Ahrens drrw->drr_offset < rwa->last_offset)) { 20959c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 20969c3fd121SMatthew Ahrens } 20979c3fd121SMatthew Ahrens rwa->last_object = drrw->drr_object; 20989c3fd121SMatthew Ahrens rwa->last_offset = drrw->drr_offset; 20999c3fd121SMatthew Ahrens 2100a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrw->drr_object, NULL) != 0) 21018a904709SMatthew Ahrens return (SET_ERROR(EINVAL)); 21028a904709SMatthew Ahrens 2103a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 2104efb80947Sahrens 2105efb80947Sahrens dmu_tx_hold_write(tx, drrw->drr_object, 2106efb80947Sahrens drrw->drr_offset, drrw->drr_length); 2107efb80947Sahrens err = dmu_tx_assign(tx, TXG_WAIT); 21083b2aab18SMatthew Ahrens if (err != 0) { 2109efb80947Sahrens dmu_tx_abort(tx); 2110efb80947Sahrens return (err); 2111efb80947Sahrens } 2112a2cdcdd2SPaul Dagnelie if (rwa->byteswap) { 2113ad135b5dSChristopher Siden dmu_object_byteswap_t byteswap = 2114ad135b5dSChristopher Siden DMU_OT_BYTESWAP(drrw->drr_type); 211598110f08SMatthew Ahrens dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, 211698110f08SMatthew Ahrens drrw->drr_length); 2117ad135b5dSChristopher Siden } 211898110f08SMatthew Ahrens 211998110f08SMatthew Ahrens dmu_buf_t *bonus; 2120a2cdcdd2SPaul Dagnelie if (dmu_bonus_hold(rwa->os, drrw->drr_object, FTAG, &bonus) != 0) 212198110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 21228a904709SMatthew Ahrens dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx); 21239c3fd121SMatthew Ahrens 21249c3fd121SMatthew Ahrens /* 21259c3fd121SMatthew Ahrens * Note: If the receive fails, we want the resume stream to start 21269c3fd121SMatthew Ahrens * with the same record that we last successfully received (as opposed 21279c3fd121SMatthew Ahrens * to the next record), so that we can verify that we are 21289c3fd121SMatthew Ahrens * resuming from the correct location. 21299c3fd121SMatthew Ahrens */ 21309c3fd121SMatthew Ahrens save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); 2131efb80947Sahrens dmu_tx_commit(tx); 21328a904709SMatthew Ahrens dmu_buf_rele(bonus, FTAG); 21339c3fd121SMatthew Ahrens 2134efb80947Sahrens return (0); 2135efb80947Sahrens } 2136efb80947Sahrens 21379e69d7d0SLori Alt /* 21389e69d7d0SLori Alt * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed 21399e69d7d0SLori Alt * streams to refer to a copy of the data that is already on the 21409e69d7d0SLori Alt * system because it came in earlier in the stream. This function 21419e69d7d0SLori Alt * finds the earlier copy of the data, and uses that copy instead of 21429e69d7d0SLori Alt * data from the stream to fulfill this write. 21439e69d7d0SLori Alt */ 21449e69d7d0SLori Alt static int 2145a2cdcdd2SPaul Dagnelie receive_write_byref(struct receive_writer_arg *rwa, 2146a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwbr) 21479e69d7d0SLori Alt { 21489e69d7d0SLori Alt dmu_tx_t *tx; 21499e69d7d0SLori Alt int err; 21509e69d7d0SLori Alt guid_map_entry_t gmesrch; 21519e69d7d0SLori Alt guid_map_entry_t *gmep; 21525d7b4d43SMatthew Ahrens avl_index_t where; 21539e69d7d0SLori Alt objset_t *ref_os = NULL; 21549e69d7d0SLori Alt dmu_buf_t *dbp; 21559e69d7d0SLori Alt 21569e69d7d0SLori Alt if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) 2157be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 21589e69d7d0SLori Alt 21599e69d7d0SLori Alt /* 21609e69d7d0SLori Alt * If the GUID of the referenced dataset is different from the 21619e69d7d0SLori Alt * GUID of the target dataset, find the referenced dataset. 21629e69d7d0SLori Alt */ 21639e69d7d0SLori Alt if (drrwbr->drr_toguid != drrwbr->drr_refguid) { 21649e69d7d0SLori Alt gmesrch.guid = drrwbr->drr_refguid; 2165a2cdcdd2SPaul Dagnelie if ((gmep = avl_find(rwa->guid_to_ds_map, &gmesrch, 21669e69d7d0SLori Alt &where)) == NULL) { 2167be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 21689e69d7d0SLori Alt } 21699e69d7d0SLori Alt if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) 2170be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 21719e69d7d0SLori Alt } else { 2172a2cdcdd2SPaul Dagnelie ref_os = rwa->os; 21739e69d7d0SLori Alt } 21749e69d7d0SLori Alt 21755d7b4d43SMatthew Ahrens err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, 21765d7b4d43SMatthew Ahrens drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH); 21775d7b4d43SMatthew Ahrens if (err != 0) 21789e69d7d0SLori Alt return (err); 21799e69d7d0SLori Alt 2180a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 21819e69d7d0SLori Alt 21829e69d7d0SLori Alt dmu_tx_hold_write(tx, drrwbr->drr_object, 21839e69d7d0SLori Alt drrwbr->drr_offset, drrwbr->drr_length); 21849e69d7d0SLori Alt err = dmu_tx_assign(tx, TXG_WAIT); 21853b2aab18SMatthew Ahrens if (err != 0) { 21869e69d7d0SLori Alt dmu_tx_abort(tx); 21879e69d7d0SLori Alt return (err); 21889e69d7d0SLori Alt } 2189a2cdcdd2SPaul Dagnelie dmu_write(rwa->os, drrwbr->drr_object, 21909e69d7d0SLori Alt drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); 21919e69d7d0SLori Alt dmu_buf_rele(dbp, FTAG); 21929c3fd121SMatthew Ahrens 21939c3fd121SMatthew Ahrens /* See comment in restore_write. */ 21949c3fd121SMatthew Ahrens save_resume_state(rwa, drrwbr->drr_object, drrwbr->drr_offset, tx); 21959e69d7d0SLori Alt dmu_tx_commit(tx); 21969e69d7d0SLori Alt return (0); 21979e69d7d0SLori Alt } 21989e69d7d0SLori Alt 21995d7b4d43SMatthew Ahrens static int 2200a2cdcdd2SPaul Dagnelie receive_write_embedded(struct receive_writer_arg *rwa, 22019c3fd121SMatthew Ahrens struct drr_write_embedded *drrwe, void *data) 22025d7b4d43SMatthew Ahrens { 22035d7b4d43SMatthew Ahrens dmu_tx_t *tx; 22045d7b4d43SMatthew Ahrens int err; 22055d7b4d43SMatthew Ahrens 22069c3fd121SMatthew Ahrens if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) 22075d7b4d43SMatthew Ahrens return (EINVAL); 22085d7b4d43SMatthew Ahrens 22099c3fd121SMatthew Ahrens if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) 22105d7b4d43SMatthew Ahrens return (EINVAL); 22115d7b4d43SMatthew Ahrens 22129c3fd121SMatthew Ahrens if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) 22135d7b4d43SMatthew Ahrens return (EINVAL); 22149c3fd121SMatthew Ahrens if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) 22155d7b4d43SMatthew Ahrens return (EINVAL); 22165d7b4d43SMatthew Ahrens 2217a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 22185d7b4d43SMatthew Ahrens 22199c3fd121SMatthew Ahrens dmu_tx_hold_write(tx, drrwe->drr_object, 22209c3fd121SMatthew Ahrens drrwe->drr_offset, drrwe->drr_length); 22215d7b4d43SMatthew Ahrens err = dmu_tx_assign(tx, TXG_WAIT); 22225d7b4d43SMatthew Ahrens if (err != 0) { 22235d7b4d43SMatthew Ahrens dmu_tx_abort(tx); 22245d7b4d43SMatthew Ahrens return (err); 22255d7b4d43SMatthew Ahrens } 22265d7b4d43SMatthew Ahrens 22279c3fd121SMatthew Ahrens dmu_write_embedded(rwa->os, drrwe->drr_object, 22289c3fd121SMatthew Ahrens drrwe->drr_offset, data, drrwe->drr_etype, 22299c3fd121SMatthew Ahrens drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, 2230a2cdcdd2SPaul Dagnelie rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); 22315d7b4d43SMatthew Ahrens 22329c3fd121SMatthew Ahrens /* See comment in restore_write. */ 22339c3fd121SMatthew Ahrens save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); 22345d7b4d43SMatthew Ahrens dmu_tx_commit(tx); 22355d7b4d43SMatthew Ahrens return (0); 22365d7b4d43SMatthew Ahrens } 22375d7b4d43SMatthew Ahrens 22380a586ceaSMark Shellenbaum static int 2239a2cdcdd2SPaul Dagnelie receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, 2240a2cdcdd2SPaul Dagnelie void *data) 22410a586ceaSMark Shellenbaum { 22420a586ceaSMark Shellenbaum dmu_tx_t *tx; 22430a586ceaSMark Shellenbaum dmu_buf_t *db, *db_spill; 22440a586ceaSMark Shellenbaum int err; 22450a586ceaSMark Shellenbaum 22460a586ceaSMark Shellenbaum if (drrs->drr_length < SPA_MINBLOCKSIZE || 2247a2cdcdd2SPaul Dagnelie drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) 2248be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 22490a586ceaSMark Shellenbaum 2250a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) 2251be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 22520a586ceaSMark Shellenbaum 2253a2cdcdd2SPaul Dagnelie VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); 22540a586ceaSMark Shellenbaum if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) { 22550a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 22560a586ceaSMark Shellenbaum return (err); 22570a586ceaSMark Shellenbaum } 22580a586ceaSMark Shellenbaum 2259a2cdcdd2SPaul Dagnelie tx = dmu_tx_create(rwa->os); 22600a586ceaSMark Shellenbaum 22610a586ceaSMark Shellenbaum dmu_tx_hold_spill(tx, db->db_object); 22620a586ceaSMark Shellenbaum 22630a586ceaSMark Shellenbaum err = dmu_tx_assign(tx, TXG_WAIT); 22643b2aab18SMatthew Ahrens if (err != 0) { 22650a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 22660a586ceaSMark Shellenbaum dmu_buf_rele(db_spill, FTAG); 22670a586ceaSMark Shellenbaum dmu_tx_abort(tx); 22680a586ceaSMark Shellenbaum return (err); 22690a586ceaSMark Shellenbaum } 22700a586ceaSMark Shellenbaum dmu_buf_will_dirty(db_spill, tx); 22710a586ceaSMark Shellenbaum 22720a586ceaSMark Shellenbaum if (db_spill->db_size < drrs->drr_length) 22730a586ceaSMark Shellenbaum VERIFY(0 == dbuf_spill_set_blksz(db_spill, 22740a586ceaSMark Shellenbaum drrs->drr_length, tx)); 22750a586ceaSMark Shellenbaum bcopy(data, db_spill->db_data, drrs->drr_length); 22760a586ceaSMark Shellenbaum 22770a586ceaSMark Shellenbaum dmu_buf_rele(db, FTAG); 22780a586ceaSMark Shellenbaum dmu_buf_rele(db_spill, FTAG); 22790a586ceaSMark Shellenbaum 22800a586ceaSMark Shellenbaum dmu_tx_commit(tx); 22810a586ceaSMark Shellenbaum return (0); 22820a586ceaSMark Shellenbaum } 22830a586ceaSMark Shellenbaum 2284efb80947Sahrens /* ARGSUSED */ 2285efb80947Sahrens static int 2286a2cdcdd2SPaul Dagnelie receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) 2287efb80947Sahrens { 2288efb80947Sahrens int err; 2289efb80947Sahrens 2290efb80947Sahrens if (drrf->drr_length != -1ULL && 2291efb80947Sahrens drrf->drr_offset + drrf->drr_length < drrf->drr_offset) 2292be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2293efb80947Sahrens 2294a2cdcdd2SPaul Dagnelie if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) 2295be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2296efb80947Sahrens 2297a2cdcdd2SPaul Dagnelie err = dmu_free_long_range(rwa->os, drrf->drr_object, 2298efb80947Sahrens drrf->drr_offset, drrf->drr_length); 2299a2cdcdd2SPaul Dagnelie 2300efb80947Sahrens return (err); 2301efb80947Sahrens } 2302efb80947Sahrens 23033b2aab18SMatthew Ahrens /* used to destroy the drc_ds on error */ 23043b2aab18SMatthew Ahrens static void 23053b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) 23063b2aab18SMatthew Ahrens { 23079c3fd121SMatthew Ahrens if (drc->drc_resumable) { 23089c3fd121SMatthew Ahrens /* wait for our resume state to be written to disk */ 23099c3fd121SMatthew Ahrens txg_wait_synced(drc->drc_ds->ds_dir->dd_pool, 0); 23109c3fd121SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 23119c3fd121SMatthew Ahrens } else { 23129adfa60dSMatthew Ahrens char name[ZFS_MAX_DATASET_NAME_LEN]; 23139c3fd121SMatthew Ahrens dsl_dataset_name(drc->drc_ds, name); 23149c3fd121SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 23159c3fd121SMatthew Ahrens (void) dsl_destroy_head(name); 23169c3fd121SMatthew Ahrens } 23173b2aab18SMatthew Ahrens } 23183b2aab18SMatthew Ahrens 231998110f08SMatthew Ahrens static void 2320a2cdcdd2SPaul Dagnelie receive_cksum(struct receive_arg *ra, int len, void *buf) 232198110f08SMatthew Ahrens { 232298110f08SMatthew Ahrens if (ra->byteswap) { 232398110f08SMatthew Ahrens fletcher_4_incremental_byteswap(buf, len, &ra->cksum); 232498110f08SMatthew Ahrens } else { 232598110f08SMatthew Ahrens fletcher_4_incremental_native(buf, len, &ra->cksum); 232698110f08SMatthew Ahrens } 232798110f08SMatthew Ahrens } 232898110f08SMatthew Ahrens 232998110f08SMatthew Ahrens /* 2330a2cdcdd2SPaul Dagnelie * Read the payload into a buffer of size len, and update the current record's 2331a2cdcdd2SPaul Dagnelie * payload field. 2332a2cdcdd2SPaul Dagnelie * Allocate ra->next_rrd and read the next record's header into 2333a2cdcdd2SPaul Dagnelie * ra->next_rrd->header. 233498110f08SMatthew Ahrens * Verify checksum of payload and next record. 233598110f08SMatthew Ahrens */ 233698110f08SMatthew Ahrens static int 2337a2cdcdd2SPaul Dagnelie receive_read_payload_and_next_header(struct receive_arg *ra, int len, void *buf) 233898110f08SMatthew Ahrens { 233998110f08SMatthew Ahrens int err; 234098110f08SMatthew Ahrens 234198110f08SMatthew Ahrens if (len != 0) { 2342a2cdcdd2SPaul Dagnelie ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); 23439c3fd121SMatthew Ahrens err = receive_read(ra, len, buf); 234498110f08SMatthew Ahrens if (err != 0) 234598110f08SMatthew Ahrens return (err); 23469c3fd121SMatthew Ahrens receive_cksum(ra, len, buf); 23479c3fd121SMatthew Ahrens 23489c3fd121SMatthew Ahrens /* note: rrd is NULL when reading the begin record's payload */ 23499c3fd121SMatthew Ahrens if (ra->rrd != NULL) { 23509c3fd121SMatthew Ahrens ra->rrd->payload = buf; 23519c3fd121SMatthew Ahrens ra->rrd->payload_size = len; 23529c3fd121SMatthew Ahrens ra->rrd->bytes_read = ra->bytes_read; 23539c3fd121SMatthew Ahrens } 235498110f08SMatthew Ahrens } 235598110f08SMatthew Ahrens 235698110f08SMatthew Ahrens ra->prev_cksum = ra->cksum; 235798110f08SMatthew Ahrens 2358a2cdcdd2SPaul Dagnelie ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); 2359a2cdcdd2SPaul Dagnelie err = receive_read(ra, sizeof (ra->next_rrd->header), 2360a2cdcdd2SPaul Dagnelie &ra->next_rrd->header); 23619c3fd121SMatthew Ahrens ra->next_rrd->bytes_read = ra->bytes_read; 2362a2cdcdd2SPaul Dagnelie if (err != 0) { 2363a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2364a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 236598110f08SMatthew Ahrens return (err); 2366a2cdcdd2SPaul Dagnelie } 2367a2cdcdd2SPaul Dagnelie if (ra->next_rrd->header.drr_type == DRR_BEGIN) { 2368a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2369a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 237098110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 2371a2cdcdd2SPaul Dagnelie } 237298110f08SMatthew Ahrens 237398110f08SMatthew Ahrens /* 237498110f08SMatthew Ahrens * Note: checksum is of everything up to but not including the 237598110f08SMatthew Ahrens * checksum itself. 237698110f08SMatthew Ahrens */ 237798110f08SMatthew Ahrens ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 237898110f08SMatthew Ahrens ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); 2379a2cdcdd2SPaul Dagnelie receive_cksum(ra, 238098110f08SMatthew Ahrens offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), 2381a2cdcdd2SPaul Dagnelie &ra->next_rrd->header); 238298110f08SMatthew Ahrens 2383a2cdcdd2SPaul Dagnelie zio_cksum_t cksum_orig = 2384a2cdcdd2SPaul Dagnelie ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; 2385a2cdcdd2SPaul Dagnelie zio_cksum_t *cksump = 2386a2cdcdd2SPaul Dagnelie &ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; 238798110f08SMatthew Ahrens 238898110f08SMatthew Ahrens if (ra->byteswap) 2389a2cdcdd2SPaul Dagnelie byteswap_record(&ra->next_rrd->header); 239098110f08SMatthew Ahrens 239198110f08SMatthew Ahrens if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && 2392a2cdcdd2SPaul Dagnelie !ZIO_CHECKSUM_EQUAL(ra->cksum, *cksump)) { 2393a2cdcdd2SPaul Dagnelie kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); 2394a2cdcdd2SPaul Dagnelie ra->next_rrd = NULL; 239598110f08SMatthew Ahrens return (SET_ERROR(ECKSUM)); 2396a2cdcdd2SPaul Dagnelie } 239798110f08SMatthew Ahrens 2398a2cdcdd2SPaul Dagnelie receive_cksum(ra, sizeof (cksum_orig), &cksum_orig); 239998110f08SMatthew Ahrens 240098110f08SMatthew Ahrens return (0); 240198110f08SMatthew Ahrens } 240298110f08SMatthew Ahrens 240368ecb2ecSPaul Dagnelie static void 240468ecb2ecSPaul Dagnelie objlist_create(struct objlist *list) 240568ecb2ecSPaul Dagnelie { 240668ecb2ecSPaul Dagnelie list_create(&list->list, sizeof (struct receive_objnode), 240768ecb2ecSPaul Dagnelie offsetof(struct receive_objnode, node)); 240868ecb2ecSPaul Dagnelie list->last_lookup = 0; 240968ecb2ecSPaul Dagnelie } 241068ecb2ecSPaul Dagnelie 241168ecb2ecSPaul Dagnelie static void 241268ecb2ecSPaul Dagnelie objlist_destroy(struct objlist *list) 241368ecb2ecSPaul Dagnelie { 241468ecb2ecSPaul Dagnelie for (struct receive_objnode *n = list_remove_head(&list->list); 241568ecb2ecSPaul Dagnelie n != NULL; n = list_remove_head(&list->list)) { 241668ecb2ecSPaul Dagnelie kmem_free(n, sizeof (*n)); 241768ecb2ecSPaul Dagnelie } 241868ecb2ecSPaul Dagnelie list_destroy(&list->list); 241968ecb2ecSPaul Dagnelie } 242068ecb2ecSPaul Dagnelie 242168ecb2ecSPaul Dagnelie /* 242268ecb2ecSPaul Dagnelie * This function looks through the objlist to see if the specified object number 242368ecb2ecSPaul Dagnelie * is contained in the objlist. In the process, it will remove all object 242468ecb2ecSPaul Dagnelie * numbers in the list that are smaller than the specified object number. Thus, 242568ecb2ecSPaul Dagnelie * any lookup of an object number smaller than a previously looked up object 242668ecb2ecSPaul Dagnelie * number will always return false; therefore, all lookups should be done in 242768ecb2ecSPaul Dagnelie * ascending order. 242868ecb2ecSPaul Dagnelie */ 242968ecb2ecSPaul Dagnelie static boolean_t 243068ecb2ecSPaul Dagnelie objlist_exists(struct objlist *list, uint64_t object) 243168ecb2ecSPaul Dagnelie { 243268ecb2ecSPaul Dagnelie struct receive_objnode *node = list_head(&list->list); 243368ecb2ecSPaul Dagnelie ASSERT3U(object, >=, list->last_lookup); 243468ecb2ecSPaul Dagnelie list->last_lookup = object; 243568ecb2ecSPaul Dagnelie while (node != NULL && node->object < object) { 243668ecb2ecSPaul Dagnelie VERIFY3P(node, ==, list_remove_head(&list->list)); 243768ecb2ecSPaul Dagnelie kmem_free(node, sizeof (*node)); 243868ecb2ecSPaul Dagnelie node = list_head(&list->list); 243968ecb2ecSPaul Dagnelie } 244068ecb2ecSPaul Dagnelie return (node != NULL && node->object == object); 244168ecb2ecSPaul Dagnelie } 244268ecb2ecSPaul Dagnelie 244368ecb2ecSPaul Dagnelie /* 244468ecb2ecSPaul Dagnelie * The objlist is a list of object numbers stored in ascending order. However, 244568ecb2ecSPaul Dagnelie * the insertion of new object numbers does not seek out the correct location to 244668ecb2ecSPaul Dagnelie * store a new object number; instead, it appends it to the list for simplicity. 244768ecb2ecSPaul Dagnelie * Thus, any users must take care to only insert new object numbers in ascending 244868ecb2ecSPaul Dagnelie * order. 244968ecb2ecSPaul Dagnelie */ 245068ecb2ecSPaul Dagnelie static void 245168ecb2ecSPaul Dagnelie objlist_insert(struct objlist *list, uint64_t object) 245268ecb2ecSPaul Dagnelie { 245368ecb2ecSPaul Dagnelie struct receive_objnode *node = kmem_zalloc(sizeof (*node), KM_SLEEP); 245468ecb2ecSPaul Dagnelie node->object = object; 245568ecb2ecSPaul Dagnelie #ifdef ZFS_DEBUG 245668ecb2ecSPaul Dagnelie struct receive_objnode *last_object = list_tail(&list->list); 245768ecb2ecSPaul Dagnelie uint64_t last_objnum = (last_object != NULL ? last_object->object : 0); 245868ecb2ecSPaul Dagnelie ASSERT3U(node->object, >, last_objnum); 245968ecb2ecSPaul Dagnelie #endif 246068ecb2ecSPaul Dagnelie list_insert_tail(&list->list, node); 246168ecb2ecSPaul Dagnelie } 246268ecb2ecSPaul Dagnelie 2463a2cdcdd2SPaul Dagnelie /* 2464a2cdcdd2SPaul Dagnelie * Issue the prefetch reads for any necessary indirect blocks. 2465a2cdcdd2SPaul Dagnelie * 2466a2cdcdd2SPaul Dagnelie * We use the object ignore list to tell us whether or not to issue prefetches 2467a2cdcdd2SPaul Dagnelie * for a given object. We do this for both correctness (in case the blocksize 2468a2cdcdd2SPaul Dagnelie * of an object has changed) and performance (if the object doesn't exist, don't 2469a2cdcdd2SPaul Dagnelie * needlessly try to issue prefetches). We also trim the list as we go through 2470a2cdcdd2SPaul Dagnelie * the stream to prevent it from growing to an unbounded size. 2471a2cdcdd2SPaul Dagnelie * 2472a2cdcdd2SPaul Dagnelie * The object numbers within will always be in sorted order, and any write 2473a2cdcdd2SPaul Dagnelie * records we see will also be in sorted order, but they're not sorted with 2474a2cdcdd2SPaul Dagnelie * respect to each other (i.e. we can get several object records before 2475a2cdcdd2SPaul Dagnelie * receiving each object's write records). As a result, once we've reached a 2476a2cdcdd2SPaul Dagnelie * given object number, we can safely remove any reference to lower object 2477a2cdcdd2SPaul Dagnelie * numbers in the ignore list. In practice, we receive up to 32 object records 2478a2cdcdd2SPaul Dagnelie * before receiving write records, so the list can have up to 32 nodes in it. 2479a2cdcdd2SPaul Dagnelie */ 2480a2cdcdd2SPaul Dagnelie /* ARGSUSED */ 2481a2cdcdd2SPaul Dagnelie static void 2482a2cdcdd2SPaul Dagnelie receive_read_prefetch(struct receive_arg *ra, 2483a2cdcdd2SPaul Dagnelie uint64_t object, uint64_t offset, uint64_t length) 2484a2cdcdd2SPaul Dagnelie { 248568ecb2ecSPaul Dagnelie if (!objlist_exists(&ra->ignore_objlist, object)) { 2486a2cdcdd2SPaul Dagnelie dmu_prefetch(ra->os, object, 1, offset, length, 2487a2cdcdd2SPaul Dagnelie ZIO_PRIORITY_SYNC_READ); 2488a2cdcdd2SPaul Dagnelie } 2489a2cdcdd2SPaul Dagnelie } 2490a2cdcdd2SPaul Dagnelie 2491a2cdcdd2SPaul Dagnelie /* 2492a2cdcdd2SPaul Dagnelie * Read records off the stream, issuing any necessary prefetches. 2493a2cdcdd2SPaul Dagnelie */ 249498110f08SMatthew Ahrens static int 2495a2cdcdd2SPaul Dagnelie receive_read_record(struct receive_arg *ra) 249698110f08SMatthew Ahrens { 249798110f08SMatthew Ahrens int err; 249898110f08SMatthew Ahrens 2499a2cdcdd2SPaul Dagnelie switch (ra->rrd->header.drr_type) { 250098110f08SMatthew Ahrens case DRR_OBJECT: 250198110f08SMatthew Ahrens { 2502a2cdcdd2SPaul Dagnelie struct drr_object *drro = &ra->rrd->header.drr_u.drr_object; 2503a2cdcdd2SPaul Dagnelie uint32_t size = P2ROUNDUP(drro->drr_bonuslen, 8); 2504a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(size, KM_SLEEP); 2505a2cdcdd2SPaul Dagnelie dmu_object_info_t doi; 2506a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, size, buf); 2507a2cdcdd2SPaul Dagnelie if (err != 0) { 2508a2cdcdd2SPaul Dagnelie kmem_free(buf, size); 250998110f08SMatthew Ahrens return (err); 2510a2cdcdd2SPaul Dagnelie } 2511a2cdcdd2SPaul Dagnelie err = dmu_object_info(ra->os, drro->drr_object, &doi); 2512a2cdcdd2SPaul Dagnelie /* 2513a2cdcdd2SPaul Dagnelie * See receive_read_prefetch for an explanation why we're 2514a2cdcdd2SPaul Dagnelie * storing this object in the ignore_obj_list. 2515a2cdcdd2SPaul Dagnelie */ 2516a2cdcdd2SPaul Dagnelie if (err == ENOENT || 2517a2cdcdd2SPaul Dagnelie (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { 251868ecb2ecSPaul Dagnelie objlist_insert(&ra->ignore_objlist, drro->drr_object); 2519a2cdcdd2SPaul Dagnelie err = 0; 2520a2cdcdd2SPaul Dagnelie } 2521a2cdcdd2SPaul Dagnelie return (err); 252298110f08SMatthew Ahrens } 252398110f08SMatthew Ahrens case DRR_FREEOBJECTS: 252498110f08SMatthew Ahrens { 2525a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2526a2cdcdd2SPaul Dagnelie return (err); 252798110f08SMatthew Ahrens } 252898110f08SMatthew Ahrens case DRR_WRITE: 252998110f08SMatthew Ahrens { 2530a2cdcdd2SPaul Dagnelie struct drr_write *drrw = &ra->rrd->header.drr_u.drr_write; 253198110f08SMatthew Ahrens arc_buf_t *abuf = arc_loan_buf(dmu_objset_spa(ra->os), 253298110f08SMatthew Ahrens drrw->drr_length); 253398110f08SMatthew Ahrens 2534a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 253598110f08SMatthew Ahrens drrw->drr_length, abuf->b_data); 2536a2cdcdd2SPaul Dagnelie if (err != 0) { 253798110f08SMatthew Ahrens dmu_return_arcbuf(abuf); 2538a2cdcdd2SPaul Dagnelie return (err); 2539a2cdcdd2SPaul Dagnelie } 2540a2cdcdd2SPaul Dagnelie ra->rrd->write_buf = abuf; 2541a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrw->drr_object, drrw->drr_offset, 2542a2cdcdd2SPaul Dagnelie drrw->drr_length); 254398110f08SMatthew Ahrens return (err); 254498110f08SMatthew Ahrens } 254598110f08SMatthew Ahrens case DRR_WRITE_BYREF: 254698110f08SMatthew Ahrens { 2547a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwb = 2548a2cdcdd2SPaul Dagnelie &ra->rrd->header.drr_u.drr_write_byref; 2549a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2550a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrwb->drr_object, drrwb->drr_offset, 2551a2cdcdd2SPaul Dagnelie drrwb->drr_length); 2552a2cdcdd2SPaul Dagnelie return (err); 255398110f08SMatthew Ahrens } 255498110f08SMatthew Ahrens case DRR_WRITE_EMBEDDED: 255598110f08SMatthew Ahrens { 255698110f08SMatthew Ahrens struct drr_write_embedded *drrwe = 2557a2cdcdd2SPaul Dagnelie &ra->rrd->header.drr_u.drr_write_embedded; 2558a2cdcdd2SPaul Dagnelie uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); 2559a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(size, KM_SLEEP); 2560a2cdcdd2SPaul Dagnelie 2561a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, size, buf); 2562a2cdcdd2SPaul Dagnelie if (err != 0) { 2563a2cdcdd2SPaul Dagnelie kmem_free(buf, size); 256498110f08SMatthew Ahrens return (err); 2565a2cdcdd2SPaul Dagnelie } 2566a2cdcdd2SPaul Dagnelie 2567a2cdcdd2SPaul Dagnelie receive_read_prefetch(ra, drrwe->drr_object, drrwe->drr_offset, 2568a2cdcdd2SPaul Dagnelie drrwe->drr_length); 2569a2cdcdd2SPaul Dagnelie return (err); 257098110f08SMatthew Ahrens } 257198110f08SMatthew Ahrens case DRR_FREE: 257298110f08SMatthew Ahrens { 2573a2cdcdd2SPaul Dagnelie /* 2574a2cdcdd2SPaul Dagnelie * It might be beneficial to prefetch indirect blocks here, but 2575a2cdcdd2SPaul Dagnelie * we don't really have the data to decide for sure. 2576a2cdcdd2SPaul Dagnelie */ 2577a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, 0, NULL); 2578a2cdcdd2SPaul Dagnelie return (err); 257998110f08SMatthew Ahrens } 258098110f08SMatthew Ahrens case DRR_END: 258198110f08SMatthew Ahrens { 2582a2cdcdd2SPaul Dagnelie struct drr_end *drre = &ra->rrd->header.drr_u.drr_end; 258398110f08SMatthew Ahrens if (!ZIO_CHECKSUM_EQUAL(ra->prev_cksum, drre->drr_checksum)) 25849c3fd121SMatthew Ahrens return (SET_ERROR(ECKSUM)); 258598110f08SMatthew Ahrens return (0); 258698110f08SMatthew Ahrens } 258798110f08SMatthew Ahrens case DRR_SPILL: 258898110f08SMatthew Ahrens { 2589a2cdcdd2SPaul Dagnelie struct drr_spill *drrs = &ra->rrd->header.drr_u.drr_spill; 2590a2cdcdd2SPaul Dagnelie void *buf = kmem_zalloc(drrs->drr_length, KM_SLEEP); 2591a2cdcdd2SPaul Dagnelie err = receive_read_payload_and_next_header(ra, drrs->drr_length, 2592a2cdcdd2SPaul Dagnelie buf); 259398110f08SMatthew Ahrens if (err != 0) 2594a2cdcdd2SPaul Dagnelie kmem_free(buf, drrs->drr_length); 2595a2cdcdd2SPaul Dagnelie return (err); 259698110f08SMatthew Ahrens } 259798110f08SMatthew Ahrens default: 259898110f08SMatthew Ahrens return (SET_ERROR(EINVAL)); 259998110f08SMatthew Ahrens } 260098110f08SMatthew Ahrens } 260198110f08SMatthew Ahrens 26023cb34c60Sahrens /* 2603a2cdcdd2SPaul Dagnelie * Commit the records to the pool. 2604a2cdcdd2SPaul Dagnelie */ 2605a2cdcdd2SPaul Dagnelie static int 2606a2cdcdd2SPaul Dagnelie receive_process_record(struct receive_writer_arg *rwa, 2607a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd) 2608a2cdcdd2SPaul Dagnelie { 2609a2cdcdd2SPaul Dagnelie int err; 2610a2cdcdd2SPaul Dagnelie 26119c3fd121SMatthew Ahrens /* Processing in order, therefore bytes_read should be increasing. */ 26129c3fd121SMatthew Ahrens ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); 26139c3fd121SMatthew Ahrens rwa->bytes_read = rrd->bytes_read; 26149c3fd121SMatthew Ahrens 2615a2cdcdd2SPaul Dagnelie switch (rrd->header.drr_type) { 2616a2cdcdd2SPaul Dagnelie case DRR_OBJECT: 2617a2cdcdd2SPaul Dagnelie { 2618a2cdcdd2SPaul Dagnelie struct drr_object *drro = &rrd->header.drr_u.drr_object; 2619a2cdcdd2SPaul Dagnelie err = receive_object(rwa, drro, rrd->payload); 2620a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2621a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2622a2cdcdd2SPaul Dagnelie return (err); 2623a2cdcdd2SPaul Dagnelie } 2624a2cdcdd2SPaul Dagnelie case DRR_FREEOBJECTS: 2625a2cdcdd2SPaul Dagnelie { 2626a2cdcdd2SPaul Dagnelie struct drr_freeobjects *drrfo = 2627a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_freeobjects; 2628a2cdcdd2SPaul Dagnelie return (receive_freeobjects(rwa, drrfo)); 2629a2cdcdd2SPaul Dagnelie } 2630a2cdcdd2SPaul Dagnelie case DRR_WRITE: 2631a2cdcdd2SPaul Dagnelie { 2632a2cdcdd2SPaul Dagnelie struct drr_write *drrw = &rrd->header.drr_u.drr_write; 2633a2cdcdd2SPaul Dagnelie err = receive_write(rwa, drrw, rrd->write_buf); 2634a2cdcdd2SPaul Dagnelie /* if receive_write() is successful, it consumes the arc_buf */ 2635a2cdcdd2SPaul Dagnelie if (err != 0) 2636a2cdcdd2SPaul Dagnelie dmu_return_arcbuf(rrd->write_buf); 2637a2cdcdd2SPaul Dagnelie rrd->write_buf = NULL; 2638a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2639a2cdcdd2SPaul Dagnelie return (err); 2640a2cdcdd2SPaul Dagnelie } 2641a2cdcdd2SPaul Dagnelie case DRR_WRITE_BYREF: 2642a2cdcdd2SPaul Dagnelie { 2643a2cdcdd2SPaul Dagnelie struct drr_write_byref *drrwbr = 2644a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_write_byref; 2645a2cdcdd2SPaul Dagnelie return (receive_write_byref(rwa, drrwbr)); 2646a2cdcdd2SPaul Dagnelie } 2647a2cdcdd2SPaul Dagnelie case DRR_WRITE_EMBEDDED: 2648a2cdcdd2SPaul Dagnelie { 2649a2cdcdd2SPaul Dagnelie struct drr_write_embedded *drrwe = 2650a2cdcdd2SPaul Dagnelie &rrd->header.drr_u.drr_write_embedded; 2651a2cdcdd2SPaul Dagnelie err = receive_write_embedded(rwa, drrwe, rrd->payload); 2652a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2653a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2654a2cdcdd2SPaul Dagnelie return (err); 2655a2cdcdd2SPaul Dagnelie } 2656a2cdcdd2SPaul Dagnelie case DRR_FREE: 2657a2cdcdd2SPaul Dagnelie { 2658a2cdcdd2SPaul Dagnelie struct drr_free *drrf = &rrd->header.drr_u.drr_free; 2659a2cdcdd2SPaul Dagnelie return (receive_free(rwa, drrf)); 2660a2cdcdd2SPaul Dagnelie } 2661a2cdcdd2SPaul Dagnelie case DRR_SPILL: 2662a2cdcdd2SPaul Dagnelie { 2663a2cdcdd2SPaul Dagnelie struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; 2664a2cdcdd2SPaul Dagnelie err = receive_spill(rwa, drrs, rrd->payload); 2665a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2666a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2667a2cdcdd2SPaul Dagnelie return (err); 2668a2cdcdd2SPaul Dagnelie } 2669a2cdcdd2SPaul Dagnelie default: 2670a2cdcdd2SPaul Dagnelie return (SET_ERROR(EINVAL)); 2671a2cdcdd2SPaul Dagnelie } 2672a2cdcdd2SPaul Dagnelie } 2673a2cdcdd2SPaul Dagnelie 2674a2cdcdd2SPaul Dagnelie /* 2675a2cdcdd2SPaul Dagnelie * dmu_recv_stream's worker thread; pull records off the queue, and then call 2676a2cdcdd2SPaul Dagnelie * receive_process_record When we're done, signal the main thread and exit. 2677a2cdcdd2SPaul Dagnelie */ 2678a2cdcdd2SPaul Dagnelie static void 2679a2cdcdd2SPaul Dagnelie receive_writer_thread(void *arg) 2680a2cdcdd2SPaul Dagnelie { 2681a2cdcdd2SPaul Dagnelie struct receive_writer_arg *rwa = arg; 2682a2cdcdd2SPaul Dagnelie struct receive_record_arg *rrd; 2683a2cdcdd2SPaul Dagnelie for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; 2684a2cdcdd2SPaul Dagnelie rrd = bqueue_dequeue(&rwa->q)) { 2685a2cdcdd2SPaul Dagnelie /* 2686a2cdcdd2SPaul Dagnelie * If there's an error, the main thread will stop putting things 2687a2cdcdd2SPaul Dagnelie * on the queue, but we need to clear everything in it before we 2688a2cdcdd2SPaul Dagnelie * can exit. 2689a2cdcdd2SPaul Dagnelie */ 2690a2cdcdd2SPaul Dagnelie if (rwa->err == 0) { 2691a2cdcdd2SPaul Dagnelie rwa->err = receive_process_record(rwa, rrd); 2692a2cdcdd2SPaul Dagnelie } else if (rrd->write_buf != NULL) { 2693a2cdcdd2SPaul Dagnelie dmu_return_arcbuf(rrd->write_buf); 2694a2cdcdd2SPaul Dagnelie rrd->write_buf = NULL; 2695a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2696a2cdcdd2SPaul Dagnelie } else if (rrd->payload != NULL) { 2697a2cdcdd2SPaul Dagnelie kmem_free(rrd->payload, rrd->payload_size); 2698a2cdcdd2SPaul Dagnelie rrd->payload = NULL; 2699a2cdcdd2SPaul Dagnelie } 2700a2cdcdd2SPaul Dagnelie kmem_free(rrd, sizeof (*rrd)); 2701a2cdcdd2SPaul Dagnelie } 2702a2cdcdd2SPaul Dagnelie kmem_free(rrd, sizeof (*rrd)); 2703a2cdcdd2SPaul Dagnelie mutex_enter(&rwa->mutex); 2704a2cdcdd2SPaul Dagnelie rwa->done = B_TRUE; 2705a2cdcdd2SPaul Dagnelie cv_signal(&rwa->cv); 2706a2cdcdd2SPaul Dagnelie mutex_exit(&rwa->mutex); 2707a2cdcdd2SPaul Dagnelie } 2708a2cdcdd2SPaul Dagnelie 27099c3fd121SMatthew Ahrens static int 27109c3fd121SMatthew Ahrens resume_check(struct receive_arg *ra, nvlist_t *begin_nvl) 27119c3fd121SMatthew Ahrens { 27129c3fd121SMatthew Ahrens uint64_t val; 27139c3fd121SMatthew Ahrens objset_t *mos = dmu_objset_pool(ra->os)->dp_meta_objset; 27149c3fd121SMatthew Ahrens uint64_t dsobj = dmu_objset_id(ra->os); 27159c3fd121SMatthew Ahrens uint64_t resume_obj, resume_off; 27169c3fd121SMatthew Ahrens 27179c3fd121SMatthew Ahrens if (nvlist_lookup_uint64(begin_nvl, 27189c3fd121SMatthew Ahrens "resume_object", &resume_obj) != 0 || 27199c3fd121SMatthew Ahrens nvlist_lookup_uint64(begin_nvl, 27209c3fd121SMatthew Ahrens "resume_offset", &resume_off) != 0) { 27219c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 27229c3fd121SMatthew Ahrens } 27239c3fd121SMatthew Ahrens VERIFY0(zap_lookup(mos, dsobj, 27249c3fd121SMatthew Ahrens DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); 27259c3fd121SMatthew Ahrens if (resume_obj != val) 27269c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 27279c3fd121SMatthew Ahrens VERIFY0(zap_lookup(mos, dsobj, 27289c3fd121SMatthew Ahrens DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); 27299c3fd121SMatthew Ahrens if (resume_off != val) 27309c3fd121SMatthew Ahrens return (SET_ERROR(EINVAL)); 27319c3fd121SMatthew Ahrens 27329c3fd121SMatthew Ahrens return (0); 27339c3fd121SMatthew Ahrens } 27349c3fd121SMatthew Ahrens 2735a2cdcdd2SPaul Dagnelie /* 2736a2cdcdd2SPaul Dagnelie * Read in the stream's records, one by one, and apply them to the pool. There 2737a2cdcdd2SPaul Dagnelie * are two threads involved; the thread that calls this function will spin up a 2738a2cdcdd2SPaul Dagnelie * worker thread, read the records off the stream one by one, and issue 2739a2cdcdd2SPaul Dagnelie * prefetches for any necessary indirect blocks. It will then push the records 2740a2cdcdd2SPaul Dagnelie * onto an internal blocking queue. The worker thread will pull the records off 2741a2cdcdd2SPaul Dagnelie * the queue, and actually write the data into the DMU. This way, the worker 2742a2cdcdd2SPaul Dagnelie * thread doesn't have to wait for reads to complete, since everything it needs 2743a2cdcdd2SPaul Dagnelie * (the indirect blocks) will be prefetched. 2744a2cdcdd2SPaul Dagnelie * 27453cb34c60Sahrens * NB: callers *must* call dmu_recv_end() if this succeeds. 27463cb34c60Sahrens */ 2747efb80947Sahrens int 2748c99e4bdcSChris Kirby dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp, 2749c99e4bdcSChris Kirby int cleanup_fd, uint64_t *action_handlep) 2750efb80947Sahrens { 275198110f08SMatthew Ahrens int err = 0; 2752a2cdcdd2SPaul Dagnelie struct receive_arg ra = { 0 }; 2753a2cdcdd2SPaul Dagnelie struct receive_writer_arg rwa = { 0 }; 27549e69d7d0SLori Alt int featureflags; 27559c3fd121SMatthew Ahrens nvlist_t *begin_nvl = NULL; 2756efb80947Sahrens 27573b2aab18SMatthew Ahrens ra.byteswap = drc->drc_byteswap; 27583b2aab18SMatthew Ahrens ra.cksum = drc->drc_cksum; 27593cb34c60Sahrens ra.vp = vp; 27603cb34c60Sahrens ra.voff = *voffp; 27619c3fd121SMatthew Ahrens 27629c3fd121SMatthew Ahrens if (dsl_dataset_is_zapified(drc->drc_ds)) { 27639c3fd121SMatthew Ahrens (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, 27649c3fd121SMatthew Ahrens drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, 27659c3fd121SMatthew Ahrens sizeof (ra.bytes_read), 1, &ra.bytes_read); 27669c3fd121SMatthew Ahrens } 27679c3fd121SMatthew Ahrens 276868ecb2ecSPaul Dagnelie objlist_create(&ra.ignore_objlist); 2769efb80947Sahrens 27703cb34c60Sahrens /* these were verified in dmu_recv_begin */ 27713b2aab18SMatthew Ahrens ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, 27729e69d7d0SLori Alt DMU_SUBSTREAM); 27733b2aab18SMatthew Ahrens ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); 2774efb80947Sahrens 2775efb80947Sahrens /* 2776efb80947Sahrens * Open the objset we are modifying. 2777efb80947Sahrens */ 277898110f08SMatthew Ahrens VERIFY0(dmu_objset_from_ds(drc->drc_ds, &ra.os)); 2779efb80947Sahrens 2780c1379625SJustin T. Gibbs ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); 2781efb80947Sahrens 27829e69d7d0SLori Alt featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); 27839e69d7d0SLori Alt 27849e69d7d0SLori Alt /* if this stream is dedup'ed, set up the avl tree for guid mapping */ 27859e69d7d0SLori Alt if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { 2786a7f53a56SChris Kirby minor_t minor; 2787a7f53a56SChris Kirby 2788c99e4bdcSChris Kirby if (cleanup_fd == -1) { 2789be6fd75aSMatthew Ahrens ra.err = SET_ERROR(EBADF); 2790c99e4bdcSChris Kirby goto out; 2791c99e4bdcSChris Kirby } 2792a7f53a56SChris Kirby ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor); 27933b2aab18SMatthew Ahrens if (ra.err != 0) { 2794a7f53a56SChris Kirby cleanup_fd = -1; 2795a7f53a56SChris Kirby goto out; 2796a7f53a56SChris Kirby } 2797a7f53a56SChris Kirby 2798c99e4bdcSChris Kirby if (*action_handlep == 0) { 2799a2cdcdd2SPaul Dagnelie rwa.guid_to_ds_map = 2800c99e4bdcSChris Kirby kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); 2801a2cdcdd2SPaul Dagnelie avl_create(rwa.guid_to_ds_map, guid_compare, 2802c99e4bdcSChris Kirby sizeof (guid_map_entry_t), 2803c99e4bdcSChris Kirby offsetof(guid_map_entry_t, avlnode)); 280498110f08SMatthew Ahrens err = zfs_onexit_add_cb(minor, 2805a2cdcdd2SPaul Dagnelie free_guid_map_onexit, rwa.guid_to_ds_map, 2806c99e4bdcSChris Kirby action_handlep); 28073b2aab18SMatthew Ahrens if (ra.err != 0) 2808c99e4bdcSChris Kirby goto out; 2809c99e4bdcSChris Kirby } else { 281098110f08SMatthew Ahrens err = zfs_onexit_cb_data(minor, *action_handlep, 2811a2cdcdd2SPaul Dagnelie (void **)&rwa.guid_to_ds_map); 28123b2aab18SMatthew Ahrens if (ra.err != 0) 2813c99e4bdcSChris Kirby goto out; 2814c99e4bdcSChris Kirby } 2815ec5cf9d5SAlexander Stetsenko 2816a2cdcdd2SPaul Dagnelie drc->drc_guid_to_ds_map = rwa.guid_to_ds_map; 28179e69d7d0SLori Alt } 28189e69d7d0SLori Alt 28199c3fd121SMatthew Ahrens uint32_t payloadlen = drc->drc_drr_begin->drr_payloadlen; 28209c3fd121SMatthew Ahrens void *payload = NULL; 28219c3fd121SMatthew Ahrens if (payloadlen != 0) 28229c3fd121SMatthew Ahrens payload = kmem_alloc(payloadlen, KM_SLEEP); 28239c3fd121SMatthew Ahrens 28249c3fd121SMatthew Ahrens err = receive_read_payload_and_next_header(&ra, payloadlen, payload); 28259c3fd121SMatthew Ahrens if (err != 0) { 28269c3fd121SMatthew Ahrens if (payloadlen != 0) 28279c3fd121SMatthew Ahrens kmem_free(payload, payloadlen); 282898110f08SMatthew Ahrens goto out; 28299c3fd121SMatthew Ahrens } 28309c3fd121SMatthew Ahrens if (payloadlen != 0) { 28319c3fd121SMatthew Ahrens err = nvlist_unpack(payload, payloadlen, &begin_nvl, KM_SLEEP); 28329c3fd121SMatthew Ahrens kmem_free(payload, payloadlen); 28339c3fd121SMatthew Ahrens if (err != 0) 28349c3fd121SMatthew Ahrens goto out; 28359c3fd121SMatthew Ahrens } 28369c3fd121SMatthew Ahrens 28379c3fd121SMatthew Ahrens if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { 28389c3fd121SMatthew Ahrens err = resume_check(&ra, begin_nvl); 28399c3fd121SMatthew Ahrens if (err != 0) 28409c3fd121SMatthew Ahrens goto out; 28419c3fd121SMatthew Ahrens } 284298110f08SMatthew Ahrens 2843a2cdcdd2SPaul Dagnelie (void) bqueue_init(&rwa.q, zfs_recv_queue_length, 2844a2cdcdd2SPaul Dagnelie offsetof(struct receive_record_arg, node)); 2845a2cdcdd2SPaul Dagnelie cv_init(&rwa.cv, NULL, CV_DEFAULT, NULL); 2846a2cdcdd2SPaul Dagnelie mutex_init(&rwa.mutex, NULL, MUTEX_DEFAULT, NULL); 2847a2cdcdd2SPaul Dagnelie rwa.os = ra.os; 2848a2cdcdd2SPaul Dagnelie rwa.byteswap = drc->drc_byteswap; 28499c3fd121SMatthew Ahrens rwa.resumable = drc->drc_resumable; 2850a2cdcdd2SPaul Dagnelie 2851a2cdcdd2SPaul Dagnelie (void) thread_create(NULL, 0, receive_writer_thread, &rwa, 0, curproc, 2852a2cdcdd2SPaul Dagnelie TS_RUN, minclsyspri); 2853a2cdcdd2SPaul Dagnelie /* 2854a2cdcdd2SPaul Dagnelie * We're reading rwa.err without locks, which is safe since we are the 2855a2cdcdd2SPaul Dagnelie * only reader, and the worker thread is the only writer. It's ok if we 2856a2cdcdd2SPaul Dagnelie * miss a write for an iteration or two of the loop, since the writer 2857a2cdcdd2SPaul Dagnelie * thread will keep freeing records we send it until we send it an eos 2858a2cdcdd2SPaul Dagnelie * marker. 2859a2cdcdd2SPaul Dagnelie * 2860a2cdcdd2SPaul Dagnelie * We can leave this loop in 3 ways: First, if rwa.err is 2861a2cdcdd2SPaul Dagnelie * non-zero. In that case, the writer thread will free the rrd we just 2862a2cdcdd2SPaul Dagnelie * pushed. Second, if we're interrupted; in that case, either it's the 2863a2cdcdd2SPaul Dagnelie * first loop and ra.rrd was never allocated, or it's later, and ra.rrd 2864a2cdcdd2SPaul Dagnelie * has been handed off to the writer thread who will free it. Finally, 2865a2cdcdd2SPaul Dagnelie * if receive_read_record fails or we're at the end of the stream, then 2866a2cdcdd2SPaul Dagnelie * we free ra.rrd and exit. 2867a2cdcdd2SPaul Dagnelie */ 2868a2cdcdd2SPaul Dagnelie while (rwa.err == 0) { 2869efb80947Sahrens if (issig(JUSTLOOKING) && issig(FORREAL)) { 287098110f08SMatthew Ahrens err = SET_ERROR(EINTR); 287198110f08SMatthew Ahrens break; 2872efb80947Sahrens } 2873efb80947Sahrens 2874a2cdcdd2SPaul Dagnelie ASSERT3P(ra.rrd, ==, NULL); 2875a2cdcdd2SPaul Dagnelie ra.rrd = ra.next_rrd; 2876a2cdcdd2SPaul Dagnelie ra.next_rrd = NULL; 2877a2cdcdd2SPaul Dagnelie /* Allocates and loads header into ra.next_rrd */ 2878a2cdcdd2SPaul Dagnelie err = receive_read_record(&ra); 2879efb80947Sahrens 2880a2cdcdd2SPaul Dagnelie if (ra.rrd->header.drr_type == DRR_END || err != 0) { 2881a2cdcdd2SPaul Dagnelie kmem_free(ra.rrd, sizeof (*ra.rrd)); 2882a2cdcdd2SPaul Dagnelie ra.rrd = NULL; 28830a586ceaSMark Shellenbaum break; 2884a2cdcdd2SPaul Dagnelie } 2885a2cdcdd2SPaul Dagnelie 2886a2cdcdd2SPaul Dagnelie bqueue_enqueue(&rwa.q, ra.rrd, 2887a2cdcdd2SPaul Dagnelie sizeof (struct receive_record_arg) + ra.rrd->payload_size); 2888a2cdcdd2SPaul Dagnelie ra.rrd = NULL; 2889a2cdcdd2SPaul Dagnelie } 2890a2cdcdd2SPaul Dagnelie if (ra.next_rrd == NULL) 2891a2cdcdd2SPaul Dagnelie ra.next_rrd = kmem_zalloc(sizeof (*ra.next_rrd), KM_SLEEP); 2892a2cdcdd2SPaul Dagnelie ra.next_rrd->eos_marker = B_TRUE; 2893a2cdcdd2SPaul Dagnelie bqueue_enqueue(&rwa.q, ra.next_rrd, 1); 2894a2cdcdd2SPaul Dagnelie 2895a2cdcdd2SPaul Dagnelie mutex_enter(&rwa.mutex); 2896a2cdcdd2SPaul Dagnelie while (!rwa.done) { 2897a2cdcdd2SPaul Dagnelie cv_wait(&rwa.cv, &rwa.mutex); 2898efb80947Sahrens } 2899a2cdcdd2SPaul Dagnelie mutex_exit(&rwa.mutex); 2900a2cdcdd2SPaul Dagnelie 2901a2cdcdd2SPaul Dagnelie cv_destroy(&rwa.cv); 2902a2cdcdd2SPaul Dagnelie mutex_destroy(&rwa.mutex); 2903a2cdcdd2SPaul Dagnelie bqueue_destroy(&rwa.q); 2904a2cdcdd2SPaul Dagnelie if (err == 0) 2905a2cdcdd2SPaul Dagnelie err = rwa.err; 2906efb80947Sahrens 2907efb80947Sahrens out: 29089c3fd121SMatthew Ahrens nvlist_free(begin_nvl); 2909a7f53a56SChris Kirby if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) 2910a7f53a56SChris Kirby zfs_onexit_fd_rele(cleanup_fd); 2911a7f53a56SChris Kirby 291298110f08SMatthew Ahrens if (err != 0) { 2913efb80947Sahrens /* 29149c3fd121SMatthew Ahrens * Clean up references. If receive is not resumable, 29159c3fd121SMatthew Ahrens * destroy what we created, so we don't leave it in 29169c3fd121SMatthew Ahrens * the inconsistent state. 2917efb80947Sahrens */ 29183b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(drc); 2919efb80947Sahrens } 2920efb80947Sahrens 29213cb34c60Sahrens *voffp = ra.voff; 292268ecb2ecSPaul Dagnelie objlist_destroy(&ra.ignore_objlist); 292398110f08SMatthew Ahrens return (err); 2924efb80947Sahrens } 2925f18faf3fSek 29263cb34c60Sahrens static int 29273b2aab18SMatthew Ahrens dmu_recv_end_check(void *arg, dmu_tx_t *tx) 29283cb34c60Sahrens { 29293b2aab18SMatthew Ahrens dmu_recv_cookie_t *drc = arg; 29303b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 29313b2aab18SMatthew Ahrens int error; 29323b2aab18SMatthew Ahrens 29333b2aab18SMatthew Ahrens ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); 29343cb34c60Sahrens 29353b2aab18SMatthew Ahrens if (!drc->drc_newfs) { 29363b2aab18SMatthew Ahrens dsl_dataset_t *origin_head; 29373b2aab18SMatthew Ahrens 29383b2aab18SMatthew Ahrens error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); 29393b2aab18SMatthew Ahrens if (error != 0) 29403b2aab18SMatthew Ahrens return (error); 294134f2f8cfSMatthew Ahrens if (drc->drc_force) { 294234f2f8cfSMatthew Ahrens /* 294334f2f8cfSMatthew Ahrens * We will destroy any snapshots in tofs (i.e. before 294434f2f8cfSMatthew Ahrens * origin_head) that are after the origin (which is 294534f2f8cfSMatthew Ahrens * the snap before drc_ds, because drc_ds can not 294634f2f8cfSMatthew Ahrens * have any snaps of its own). 294734f2f8cfSMatthew Ahrens */ 2948c1379625SJustin T. Gibbs uint64_t obj; 2949c1379625SJustin T. Gibbs 2950c1379625SJustin T. Gibbs obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 2951c1379625SJustin T. Gibbs while (obj != 2952c1379625SJustin T. Gibbs dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 295334f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 295434f2f8cfSMatthew Ahrens error = dsl_dataset_hold_obj(dp, obj, FTAG, 295534f2f8cfSMatthew Ahrens &snap); 295634f2f8cfSMatthew Ahrens if (error != 0) 2957beddaa9cSAndriy Gapon break; 295834f2f8cfSMatthew Ahrens if (snap->ds_dir != origin_head->ds_dir) 295934f2f8cfSMatthew Ahrens error = SET_ERROR(EINVAL); 296034f2f8cfSMatthew Ahrens if (error == 0) { 296134f2f8cfSMatthew Ahrens error = dsl_destroy_snapshot_check_impl( 296234f2f8cfSMatthew Ahrens snap, B_FALSE); 296334f2f8cfSMatthew Ahrens } 2964c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 296534f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 296634f2f8cfSMatthew Ahrens if (error != 0) 2967beddaa9cSAndriy Gapon break; 2968beddaa9cSAndriy Gapon } 2969beddaa9cSAndriy Gapon if (error != 0) { 2970beddaa9cSAndriy Gapon dsl_dataset_rele(origin_head, FTAG); 2971beddaa9cSAndriy Gapon return (error); 297234f2f8cfSMatthew Ahrens } 297334f2f8cfSMatthew Ahrens } 29743b2aab18SMatthew Ahrens error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, 297591948b51SKeith M Wesolowski origin_head, drc->drc_force, drc->drc_owner, tx); 29763b2aab18SMatthew Ahrens if (error != 0) { 29773b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 29783b2aab18SMatthew Ahrens return (error); 29793b2aab18SMatthew Ahrens } 29803b2aab18SMatthew Ahrens error = dsl_dataset_snapshot_check_impl(origin_head, 2981a2afb611SJerry Jelinek drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 29823b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 29833b2aab18SMatthew Ahrens if (error != 0) 29843b2aab18SMatthew Ahrens return (error); 29853b2aab18SMatthew Ahrens 29863b2aab18SMatthew Ahrens error = dsl_destroy_head_check_impl(drc->drc_ds, 1); 29873b2aab18SMatthew Ahrens } else { 29883b2aab18SMatthew Ahrens error = dsl_dataset_snapshot_check_impl(drc->drc_ds, 2989a2afb611SJerry Jelinek drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); 29903b2aab18SMatthew Ahrens } 29913b2aab18SMatthew Ahrens return (error); 29923cb34c60Sahrens } 29933cb34c60Sahrens 29943cb34c60Sahrens static void 29953b2aab18SMatthew Ahrens dmu_recv_end_sync(void *arg, dmu_tx_t *tx) 29963cb34c60Sahrens { 29973b2aab18SMatthew Ahrens dmu_recv_cookie_t *drc = arg; 29983b2aab18SMatthew Ahrens dsl_pool_t *dp = dmu_tx_pool(tx); 29993b2aab18SMatthew Ahrens 30003b2aab18SMatthew Ahrens spa_history_log_internal_ds(drc->drc_ds, "finish receiving", 30013b2aab18SMatthew Ahrens tx, "snap=%s", drc->drc_tosnap); 30023b2aab18SMatthew Ahrens 30033b2aab18SMatthew Ahrens if (!drc->drc_newfs) { 30043b2aab18SMatthew Ahrens dsl_dataset_t *origin_head; 30053b2aab18SMatthew Ahrens 30063b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, 30073b2aab18SMatthew Ahrens &origin_head)); 300834f2f8cfSMatthew Ahrens 300934f2f8cfSMatthew Ahrens if (drc->drc_force) { 301034f2f8cfSMatthew Ahrens /* 301134f2f8cfSMatthew Ahrens * Destroy any snapshots of drc_tofs (origin_head) 301234f2f8cfSMatthew Ahrens * after the origin (the snap before drc_ds). 301334f2f8cfSMatthew Ahrens */ 3014c1379625SJustin T. Gibbs uint64_t obj; 3015c1379625SJustin T. Gibbs 3016c1379625SJustin T. Gibbs obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; 3017c1379625SJustin T. Gibbs while (obj != 3018c1379625SJustin T. Gibbs dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { 301934f2f8cfSMatthew Ahrens dsl_dataset_t *snap; 302034f2f8cfSMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, 302134f2f8cfSMatthew Ahrens &snap)); 302234f2f8cfSMatthew Ahrens ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); 3023c1379625SJustin T. Gibbs obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 302434f2f8cfSMatthew Ahrens dsl_destroy_snapshot_sync_impl(snap, 302534f2f8cfSMatthew Ahrens B_FALSE, tx); 302634f2f8cfSMatthew Ahrens dsl_dataset_rele(snap, FTAG); 302734f2f8cfSMatthew Ahrens } 302834f2f8cfSMatthew Ahrens } 302934f2f8cfSMatthew Ahrens VERIFY3P(drc->drc_ds->ds_prev, ==, 303034f2f8cfSMatthew Ahrens origin_head->ds_prev); 303134f2f8cfSMatthew Ahrens 30323b2aab18SMatthew Ahrens dsl_dataset_clone_swap_sync_impl(drc->drc_ds, 30333b2aab18SMatthew Ahrens origin_head, tx); 30343b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(origin_head, 30353b2aab18SMatthew Ahrens drc->drc_tosnap, tx); 30363b2aab18SMatthew Ahrens 30373b2aab18SMatthew Ahrens /* set snapshot's creation time and guid */ 30383b2aab18SMatthew Ahrens dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); 3039c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = 30403b2aab18SMatthew Ahrens drc->drc_drrb->drr_creation_time; 3041c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_guid = 30423b2aab18SMatthew Ahrens drc->drc_drrb->drr_toguid; 3043c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= 30443b2aab18SMatthew Ahrens ~DS_FLAG_INCONSISTENT; 30453b2aab18SMatthew Ahrens 30463b2aab18SMatthew Ahrens dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 3047c1379625SJustin T. Gibbs dsl_dataset_phys(origin_head)->ds_flags &= 3048c1379625SJustin T. Gibbs ~DS_FLAG_INCONSISTENT; 30493b2aab18SMatthew Ahrens 30503b2aab18SMatthew Ahrens dsl_dataset_rele(origin_head, FTAG); 30513b2aab18SMatthew Ahrens dsl_destroy_head_sync_impl(drc->drc_ds, tx); 305291948b51SKeith M Wesolowski 305391948b51SKeith M Wesolowski if (drc->drc_owner != NULL) 305491948b51SKeith M Wesolowski VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); 30553b2aab18SMatthew Ahrens } else { 30563b2aab18SMatthew Ahrens dsl_dataset_t *ds = drc->drc_ds; 30573cb34c60Sahrens 30583b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); 30593cb34c60Sahrens 30603b2aab18SMatthew Ahrens /* set snapshot's creation time and guid */ 30613b2aab18SMatthew Ahrens dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 3062c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_creation_time = 30633b2aab18SMatthew Ahrens drc->drc_drrb->drr_creation_time; 3064c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_guid = 3065c1379625SJustin T. Gibbs drc->drc_drrb->drr_toguid; 3066c1379625SJustin T. Gibbs dsl_dataset_phys(ds->ds_prev)->ds_flags &= 3067c1379625SJustin T. Gibbs ~DS_FLAG_INCONSISTENT; 30683cb34c60Sahrens 30693b2aab18SMatthew Ahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 3070c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; 30719c3fd121SMatthew Ahrens if (dsl_dataset_has_resume_receive_state(ds)) { 30729c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 30739c3fd121SMatthew Ahrens DS_FIELD_RESUME_FROMGUID, tx); 30749c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 30759c3fd121SMatthew Ahrens DS_FIELD_RESUME_OBJECT, tx); 30769c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 30779c3fd121SMatthew Ahrens DS_FIELD_RESUME_OFFSET, tx); 30789c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 30799c3fd121SMatthew Ahrens DS_FIELD_RESUME_BYTES, tx); 30809c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 30819c3fd121SMatthew Ahrens DS_FIELD_RESUME_TOGUID, tx); 30829c3fd121SMatthew Ahrens (void) zap_remove(dp->dp_meta_objset, ds->ds_object, 30839c3fd121SMatthew Ahrens DS_FIELD_RESUME_TONAME, tx); 30849c3fd121SMatthew Ahrens } 30853b2aab18SMatthew Ahrens } 3086c1379625SJustin T. Gibbs drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; 30873b2aab18SMatthew Ahrens /* 30883b2aab18SMatthew Ahrens * Release the hold from dmu_recv_begin. This must be done before 30893b2aab18SMatthew Ahrens * we return to open context, so that when we free the dataset's dnode, 30903b2aab18SMatthew Ahrens * we can evict its bonus buffer. 30913b2aab18SMatthew Ahrens */ 30923b2aab18SMatthew Ahrens dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); 30933b2aab18SMatthew Ahrens drc->drc_ds = NULL; 30943cb34c60Sahrens } 30953cb34c60Sahrens 3096ec5cf9d5SAlexander Stetsenko static int 30973b2aab18SMatthew Ahrens add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) 3098ec5cf9d5SAlexander Stetsenko { 30993b2aab18SMatthew Ahrens dsl_pool_t *dp; 3100ec5cf9d5SAlexander Stetsenko dsl_dataset_t *snapds; 3101ec5cf9d5SAlexander Stetsenko guid_map_entry_t *gmep; 3102ec5cf9d5SAlexander Stetsenko int err; 3103ec5cf9d5SAlexander Stetsenko 3104ec5cf9d5SAlexander Stetsenko ASSERT(guid_map != NULL); 3105ec5cf9d5SAlexander Stetsenko 31063b2aab18SMatthew Ahrens err = dsl_pool_hold(name, FTAG, &dp); 31073b2aab18SMatthew Ahrens if (err != 0) 31083b2aab18SMatthew Ahrens return (err); 3109de8d9cffSMatthew Ahrens gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); 3110de8d9cffSMatthew Ahrens err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds); 3111ec5cf9d5SAlexander Stetsenko if (err == 0) { 3112c1379625SJustin T. Gibbs gmep->guid = dsl_dataset_phys(snapds)->ds_guid; 3113ec5cf9d5SAlexander Stetsenko gmep->gme_ds = snapds; 3114ec5cf9d5SAlexander Stetsenko avl_add(guid_map, gmep); 31153b2aab18SMatthew Ahrens dsl_dataset_long_hold(snapds, gmep); 3116de8d9cffSMatthew Ahrens } else { 3117de8d9cffSMatthew Ahrens kmem_free(gmep, sizeof (*gmep)); 3118ec5cf9d5SAlexander Stetsenko } 3119ec5cf9d5SAlexander Stetsenko 31203b2aab18SMatthew Ahrens dsl_pool_rele(dp, FTAG); 3121ec5cf9d5SAlexander Stetsenko return (err); 3122ec5cf9d5SAlexander Stetsenko } 3123ec5cf9d5SAlexander Stetsenko 31243b2aab18SMatthew Ahrens static int dmu_recv_end_modified_blocks = 3; 31253b2aab18SMatthew Ahrens 3126ae46e4c7SMatthew Ahrens static int 3127ae46e4c7SMatthew Ahrens dmu_recv_existing_end(dmu_recv_cookie_t *drc) 3128f18faf3fSek { 31293b2aab18SMatthew Ahrens int error; 31303cb34c60Sahrens 31313b2aab18SMatthew Ahrens #ifdef _KERNEL 31323b2aab18SMatthew Ahrens /* 31333b2aab18SMatthew Ahrens * We will be destroying the ds; make sure its origin is unmounted if 31343b2aab18SMatthew Ahrens * necessary. 31353b2aab18SMatthew Ahrens */ 31369adfa60dSMatthew Ahrens char name[ZFS_MAX_DATASET_NAME_LEN]; 31373b2aab18SMatthew Ahrens dsl_dataset_name(drc->drc_ds, name); 31383b2aab18SMatthew Ahrens zfs_destroy_unmount_origin(name); 31393b2aab18SMatthew Ahrens #endif 31403cb34c60Sahrens 31413b2aab18SMatthew Ahrens error = dsl_sync_task(drc->drc_tofs, 31423b2aab18SMatthew Ahrens dmu_recv_end_check, dmu_recv_end_sync, drc, 31437d46dc6cSMatthew Ahrens dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL); 31443cb34c60Sahrens 31453b2aab18SMatthew Ahrens if (error != 0) 31463b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(drc); 31473b2aab18SMatthew Ahrens return (error); 3148f18faf3fSek } 3149ae46e4c7SMatthew Ahrens 3150ae46e4c7SMatthew Ahrens static int 3151ae46e4c7SMatthew Ahrens dmu_recv_new_end(dmu_recv_cookie_t *drc) 3152ae46e4c7SMatthew Ahrens { 31533b2aab18SMatthew Ahrens int error; 3154ae46e4c7SMatthew Ahrens 31553b2aab18SMatthew Ahrens error = dsl_sync_task(drc->drc_tofs, 31563b2aab18SMatthew Ahrens dmu_recv_end_check, dmu_recv_end_sync, drc, 31577d46dc6cSMatthew Ahrens dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL); 3158ae46e4c7SMatthew Ahrens 31593b2aab18SMatthew Ahrens if (error != 0) { 31603b2aab18SMatthew Ahrens dmu_recv_cleanup_ds(drc); 31613b2aab18SMatthew Ahrens } else if (drc->drc_guid_to_ds_map != NULL) { 31623b2aab18SMatthew Ahrens (void) add_ds_to_guidmap(drc->drc_tofs, 31633b2aab18SMatthew Ahrens drc->drc_guid_to_ds_map, 31643b2aab18SMatthew Ahrens drc->drc_newsnapobj); 3165ae46e4c7SMatthew Ahrens } 31663b2aab18SMatthew Ahrens return (error); 3167ae46e4c7SMatthew Ahrens } 3168ae46e4c7SMatthew Ahrens 3169ae46e4c7SMatthew Ahrens int 317091948b51SKeith M Wesolowski dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) 3171ae46e4c7SMatthew Ahrens { 317291948b51SKeith M Wesolowski drc->drc_owner = owner; 317391948b51SKeith M Wesolowski 31743b2aab18SMatthew Ahrens if (drc->drc_newfs) 3175ae46e4c7SMatthew Ahrens return (dmu_recv_new_end(drc)); 31763b2aab18SMatthew Ahrens else 31773b2aab18SMatthew Ahrens return (dmu_recv_existing_end(drc)); 3178ae46e4c7SMatthew Ahrens } 31792f3d8780SMatthew Ahrens 31802f3d8780SMatthew Ahrens /* 31812f3d8780SMatthew Ahrens * Return TRUE if this objset is currently being received into. 31822f3d8780SMatthew Ahrens */ 31832f3d8780SMatthew Ahrens boolean_t 31842f3d8780SMatthew Ahrens dmu_objset_is_receiving(objset_t *os) 31852f3d8780SMatthew Ahrens { 31862f3d8780SMatthew Ahrens return (os->os_dsl_dataset != NULL && 31872f3d8780SMatthew Ahrens os->os_dsl_dataset->ds_owner == dmu_recv_tag); 31882f3d8780SMatthew Ahrens } 3189