Index: stable/11/cddl/contrib/opensolaris/lib/libzfs/common/libzfs_sendrecv.c =================================================================== --- stable/11/cddl/contrib/opensolaris/lib/libzfs/common/libzfs_sendrecv.c (revision 352375) +++ stable/11/cddl/contrib/opensolaris/lib/libzfs/common/libzfs_sendrecv.c (revision 352376) @@ -1,3827 +1,3834 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2015 by Delphix. All rights reserved. * Copyright (c) 2012, Joyent, Inc. All rights reserved. * Copyright (c) 2012 Pawel Jakub Dawidek. All rights reserved. * Copyright (c) 2013 Steven Hartland. All rights reserved. * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright 2016 Igor Kozhukhov */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "zfs_namecheck.h" #include "zfs_prop.h" #include "zfs_fletcher.h" #include "libzfs_impl.h" #include #include #include #include #ifdef __FreeBSD__ extern int zfs_ioctl_version; #endif /* in libzfs_dataset.c */ extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *); /* We need to use something for ENODATA. */ #define ENODATA EIDRM static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *, recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int, uint64_t *, const char *); static int guid_to_name(libzfs_handle_t *, const char *, uint64_t, boolean_t, char *); static const zio_cksum_t zero_cksum = { 0 }; typedef struct dedup_arg { int inputfd; int outputfd; libzfs_handle_t *dedup_hdl; } dedup_arg_t; typedef struct progress_arg { zfs_handle_t *pa_zhp; int pa_fd; boolean_t pa_parsable; boolean_t pa_astitle; uint64_t pa_size; } progress_arg_t; typedef struct dataref { uint64_t ref_guid; uint64_t ref_object; uint64_t ref_offset; } dataref_t; typedef struct dedup_entry { struct dedup_entry *dde_next; zio_cksum_t dde_chksum; uint64_t dde_prop; dataref_t dde_ref; } dedup_entry_t; #define MAX_DDT_PHYSMEM_PERCENT 20 #define SMALLEST_POSSIBLE_MAX_DDT_MB 128 typedef struct dedup_table { dedup_entry_t **dedup_hash_array; umem_cache_t *ddecache; uint64_t max_ddt_size; /* max dedup table size in bytes */ uint64_t cur_ddt_size; /* current dedup table size in bytes */ uint64_t ddt_count; int numhashbits; boolean_t ddt_full; } dedup_table_t; static int high_order_bit(uint64_t n) { int count; for (count = 0; n != 0; count++) n >>= 1; return (count); } static size_t ssread(void *buf, size_t len, FILE *stream) { size_t outlen; if ((outlen = fread(buf, len, 1, stream)) == 0) return (0); return (outlen); } static void ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp, zio_cksum_t *cs, uint64_t prop, dataref_t *dr) { dedup_entry_t *dde; if (ddt->cur_ddt_size >= ddt->max_ddt_size) { if (ddt->ddt_full == B_FALSE) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Dedup table full. Deduplication will continue " "with existing table entries")); ddt->ddt_full = B_TRUE; } return; } if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT)) != NULL) { assert(*ddepp == NULL); dde->dde_next = NULL; dde->dde_chksum = *cs; dde->dde_prop = prop; dde->dde_ref = *dr; *ddepp = dde; ddt->cur_ddt_size += sizeof (dedup_entry_t); ddt->ddt_count++; } } /* * Using the specified dedup table, do a lookup for an entry with * the checksum cs. If found, return the block's reference info * in *dr. Otherwise, insert a new entry in the dedup table, using * the reference information specified by *dr. * * return value: true - entry was found * false - entry was not found */ static boolean_t ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs, uint64_t prop, dataref_t *dr) { uint32_t hashcode; dedup_entry_t **ddepp; hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits); for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL; ddepp = &((*ddepp)->dde_next)) { if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) && (*ddepp)->dde_prop == prop) { *dr = (*ddepp)->dde_ref; return (B_TRUE); } } ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr); return (B_FALSE); } static int dump_record(dmu_replay_record_t *drr, void *payload, int payload_len, zio_cksum_t *zc, int outfd) { ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); (void) fletcher_4_incremental_native(drr, offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc); if (drr->drr_type != DRR_BEGIN) { ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u. drr_checksum.drr_checksum)); drr->drr_u.drr_checksum.drr_checksum = *zc; } (void) fletcher_4_incremental_native( &drr->drr_u.drr_checksum.drr_checksum, sizeof (zio_cksum_t), zc); if (write(outfd, drr, sizeof (*drr)) == -1) return (errno); if (payload_len != 0) { (void) fletcher_4_incremental_native(payload, payload_len, zc); if (write(outfd, payload, payload_len) == -1) return (errno); } return (0); } /* * This function is started in a separate thread when the dedup option * has been requested. The main send thread determines the list of * snapshots to be included in the send stream and makes the ioctl calls * for each one. But instead of having the ioctl send the output to the * the output fd specified by the caller of zfs_send()), the * ioctl is told to direct the output to a pipe, which is read by the * alternate thread running THIS function. This function does the * dedup'ing by: * 1. building a dedup table (the DDT) * 2. doing checksums on each data block and inserting a record in the DDT * 3. looking for matching checksums, and * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever * a duplicate block is found. * The output of this function then goes to the output fd requested * by the caller of zfs_send(). */ static void * cksummer(void *arg) { dedup_arg_t *dda = arg; char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE); dmu_replay_record_t thedrr; dmu_replay_record_t *drr = &thedrr; FILE *ofp; int outfd; dedup_table_t ddt; zio_cksum_t stream_cksum; uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE); uint64_t numbuckets; ddt.max_ddt_size = MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100, SMALLEST_POSSIBLE_MAX_DDT_MB << 20); numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t)); /* * numbuckets must be a power of 2. Increase number to * a power of 2 if necessary. */ if (!ISP2(numbuckets)) numbuckets = 1 << high_order_bit(numbuckets); ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *)); ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0, NULL, NULL, NULL, NULL, NULL, 0); ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *); ddt.numhashbits = high_order_bit(numbuckets) - 1; ddt.ddt_full = B_FALSE; outfd = dda->outputfd; ofp = fdopen(dda->inputfd, "r"); while (ssread(drr, sizeof (*drr), ofp) != 0) { /* * kernel filled in checksum, we are going to write same * record, but need to regenerate checksum. */ if (drr->drr_type != DRR_BEGIN) { bzero(&drr->drr_u.drr_checksum.drr_checksum, sizeof (drr->drr_u.drr_checksum.drr_checksum)); } switch (drr->drr_type) { case DRR_BEGIN: { struct drr_begin *drrb = &drr->drr_u.drr_begin; int fflags; int sz = 0; ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0); ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); /* set the DEDUP feature flag for this stream */ fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); fflags |= (DMU_BACKUP_FEATURE_DEDUP | DMU_BACKUP_FEATURE_DEDUPPROPS); DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags); if (drr->drr_payloadlen != 0) { sz = drr->drr_payloadlen; if (sz > SPA_MAXBLOCKSIZE) { buf = zfs_realloc(dda->dedup_hdl, buf, SPA_MAXBLOCKSIZE, sz); } (void) ssread(buf, sz, ofp); if (ferror(stdin)) perror("fread"); } if (dump_record(drr, buf, sz, &stream_cksum, outfd) != 0) goto out; break; } case DRR_END: { struct drr_end *drre = &drr->drr_u.drr_end; /* use the recalculated checksum */ drre->drr_checksum = stream_cksum; if (dump_record(drr, NULL, 0, &stream_cksum, outfd) != 0) goto out; break; } case DRR_OBJECT: { struct drr_object *drro = &drr->drr_u.drr_object; if (drro->drr_bonuslen > 0) { (void) ssread(buf, P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), ofp); } if (dump_record(drr, buf, P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8), &stream_cksum, outfd) != 0) goto out; break; } case DRR_SPILL: { struct drr_spill *drrs = &drr->drr_u.drr_spill; (void) ssread(buf, drrs->drr_length, ofp); if (dump_record(drr, buf, drrs->drr_length, &stream_cksum, outfd) != 0) goto out; break; } case DRR_FREEOBJECTS: { if (dump_record(drr, NULL, 0, &stream_cksum, outfd) != 0) goto out; break; } case DRR_WRITE: { struct drr_write *drrw = &drr->drr_u.drr_write; dataref_t dataref; uint64_t payload_size; payload_size = DRR_WRITE_PAYLOAD_SIZE(drrw); (void) ssread(buf, payload_size, ofp); /* * Use the existing checksum if it's dedup-capable, * else calculate a SHA256 checksum for it. */ if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum, zero_cksum) || !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) { SHA256_CTX ctx; zio_cksum_t tmpsha256; SHA256Init(&ctx); SHA256Update(&ctx, buf, payload_size); SHA256Final(&tmpsha256, &ctx); drrw->drr_key.ddk_cksum.zc_word[0] = BE_64(tmpsha256.zc_word[0]); drrw->drr_key.ddk_cksum.zc_word[1] = BE_64(tmpsha256.zc_word[1]); drrw->drr_key.ddk_cksum.zc_word[2] = BE_64(tmpsha256.zc_word[2]); drrw->drr_key.ddk_cksum.zc_word[3] = BE_64(tmpsha256.zc_word[3]); drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256; drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP; } dataref.ref_guid = drrw->drr_toguid; dataref.ref_object = drrw->drr_object; dataref.ref_offset = drrw->drr_offset; if (ddt_update(dda->dedup_hdl, &ddt, &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop, &dataref)) { dmu_replay_record_t wbr_drr = {0}; struct drr_write_byref *wbr_drrr = &wbr_drr.drr_u.drr_write_byref; /* block already present in stream */ wbr_drr.drr_type = DRR_WRITE_BYREF; wbr_drrr->drr_object = drrw->drr_object; wbr_drrr->drr_offset = drrw->drr_offset; wbr_drrr->drr_length = drrw->drr_logical_size; wbr_drrr->drr_toguid = drrw->drr_toguid; wbr_drrr->drr_refguid = dataref.ref_guid; wbr_drrr->drr_refobject = dataref.ref_object; wbr_drrr->drr_refoffset = dataref.ref_offset; wbr_drrr->drr_checksumtype = drrw->drr_checksumtype; wbr_drrr->drr_checksumflags = drrw->drr_checksumtype; wbr_drrr->drr_key.ddk_cksum = drrw->drr_key.ddk_cksum; wbr_drrr->drr_key.ddk_prop = drrw->drr_key.ddk_prop; if (dump_record(&wbr_drr, NULL, 0, &stream_cksum, outfd) != 0) goto out; } else { /* block not previously seen */ if (dump_record(drr, buf, payload_size, &stream_cksum, outfd) != 0) goto out; } break; } case DRR_WRITE_EMBEDDED: { struct drr_write_embedded *drrwe = &drr->drr_u.drr_write_embedded; (void) ssread(buf, P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp); if (dump_record(drr, buf, P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), &stream_cksum, outfd) != 0) goto out; break; } case DRR_FREE: { if (dump_record(drr, NULL, 0, &stream_cksum, outfd) != 0) goto out; break; } default: (void) fprintf(stderr, "INVALID record type 0x%x\n", drr->drr_type); /* should never happen, so assert */ assert(B_FALSE); } } out: umem_cache_destroy(ddt.ddecache); free(ddt.dedup_hash_array); free(buf); (void) fclose(ofp); return (NULL); } /* * Routines for dealing with the AVL tree of fs-nvlists */ typedef struct fsavl_node { avl_node_t fn_node; nvlist_t *fn_nvfs; char *fn_snapname; uint64_t fn_guid; } fsavl_node_t; static int fsavl_compare(const void *arg1, const void *arg2) { const fsavl_node_t *fn1 = (const fsavl_node_t *)arg1; const fsavl_node_t *fn2 = (const fsavl_node_t *)arg2; return (AVL_CMP(fn1->fn_guid, fn2->fn_guid)); } /* * Given the GUID of a snapshot, find its containing filesystem and * (optionally) name. */ static nvlist_t * fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname) { fsavl_node_t fn_find; fsavl_node_t *fn; fn_find.fn_guid = snapguid; fn = avl_find(avl, &fn_find, NULL); if (fn) { if (snapname) *snapname = fn->fn_snapname; return (fn->fn_nvfs); } return (NULL); } static void fsavl_destroy(avl_tree_t *avl) { fsavl_node_t *fn; void *cookie; if (avl == NULL) return; cookie = NULL; while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL) free(fn); avl_destroy(avl); free(avl); } /* * Given an nvlist, produce an avl tree of snapshots, ordered by guid */ static avl_tree_t * fsavl_create(nvlist_t *fss) { avl_tree_t *fsavl; nvpair_t *fselem = NULL; if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL) return (NULL); avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t), offsetof(fsavl_node_t, fn_node)); while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) { nvlist_t *nvfs, *snaps; nvpair_t *snapelem = NULL; VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); while ((snapelem = nvlist_next_nvpair(snaps, snapelem)) != NULL) { fsavl_node_t *fn; uint64_t guid; VERIFY(0 == nvpair_value_uint64(snapelem, &guid)); if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) { fsavl_destroy(fsavl); return (NULL); } fn->fn_nvfs = nvfs; fn->fn_snapname = nvpair_name(snapelem); fn->fn_guid = guid; /* * Note: if there are multiple snaps with the * same GUID, we ignore all but one. */ if (avl_find(fsavl, fn, NULL) == NULL) avl_add(fsavl, fn); else free(fn); } } return (fsavl); } /* * Routines for dealing with the giant nvlist of fs-nvlists, etc. */ typedef struct send_data { /* * assigned inside every recursive call, * restored from *_save on return: * * guid of fromsnap snapshot in parent dataset * txg of fromsnap snapshot in current dataset * txg of tosnap snapshot in current dataset */ uint64_t parent_fromsnap_guid; uint64_t fromsnap_txg; uint64_t tosnap_txg; /* the nvlists get accumulated during depth-first traversal */ nvlist_t *parent_snaps; nvlist_t *fss; nvlist_t *snapprops; /* send-receive configuration, does not change during traversal */ const char *fsname; const char *fromsnap; const char *tosnap; boolean_t recursive; boolean_t verbose; /* * The header nvlist is of the following format: * { * "tosnap" -> string * "fromsnap" -> string (if incremental) * "fss" -> { * id -> { * * "name" -> string (full name; for debugging) * "parentfromsnap" -> number (guid of fromsnap in parent) * * "props" -> { name -> value (only if set here) } * "snaps" -> { name (lastname) -> number (guid) } * "snapprops" -> { name (lastname) -> { name -> value } } * * "origin" -> number (guid) (if clone) * "sent" -> boolean (not on-disk) * } * } * } * */ } send_data_t; static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv); static int send_iterate_snap(zfs_handle_t *zhp, void *arg) { send_data_t *sd = arg; uint64_t guid = zhp->zfs_dmustats.dds_guid; uint64_t txg = zhp->zfs_dmustats.dds_creation_txg; char *snapname; nvlist_t *nv; snapname = strrchr(zhp->zfs_name, '@')+1; if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) { if (sd->verbose) { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "skipping snapshot %s because it was created " "after the destination snapshot (%s)\n"), zhp->zfs_name, sd->tosnap); } zfs_close(zhp); return (0); } VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid)); /* * NB: if there is no fromsnap here (it's a newly created fs in * an incremental replication), we will substitute the tosnap. */ if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) || (sd->parent_fromsnap_guid == 0 && sd->tosnap && strcmp(snapname, sd->tosnap) == 0)) { sd->parent_fromsnap_guid = guid; } VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); send_iterate_prop(zhp, nv); VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv)); nvlist_free(nv); zfs_close(zhp); return (0); } static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv) { nvpair_t *elem = NULL; while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) { char *propname = nvpair_name(elem); zfs_prop_t prop = zfs_name_to_prop(propname); nvlist_t *propnv; if (!zfs_prop_user(propname)) { /* * Realistically, this should never happen. However, * we want the ability to add DSL properties without * needing to make incompatible version changes. We * need to ignore unknown properties to allow older * software to still send datasets containing these * properties, with the unknown properties elided. */ if (prop == ZPROP_INVAL) continue; if (zfs_prop_readonly(prop)) continue; } verify(nvpair_value_nvlist(elem, &propnv) == 0); if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION || prop == ZFS_PROP_REFQUOTA || prop == ZFS_PROP_REFRESERVATION) { char *source; uint64_t value; verify(nvlist_lookup_uint64(propnv, ZPROP_VALUE, &value) == 0); if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) continue; /* * May have no source before SPA_VERSION_RECVD_PROPS, * but is still modifiable. */ if (nvlist_lookup_string(propnv, ZPROP_SOURCE, &source) == 0) { if ((strcmp(source, zhp->zfs_name) != 0) && (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0)) continue; } } else { char *source; if (nvlist_lookup_string(propnv, ZPROP_SOURCE, &source) != 0) continue; if ((strcmp(source, zhp->zfs_name) != 0) && (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0)) continue; } if (zfs_prop_user(propname) || zfs_prop_get_type(prop) == PROP_TYPE_STRING) { char *value; verify(nvlist_lookup_string(propnv, ZPROP_VALUE, &value) == 0); VERIFY(0 == nvlist_add_string(nv, propname, value)); } else { uint64_t value; verify(nvlist_lookup_uint64(propnv, ZPROP_VALUE, &value) == 0); VERIFY(0 == nvlist_add_uint64(nv, propname, value)); } } } /* * returns snapshot creation txg * and returns 0 if the snapshot does not exist */ static uint64_t get_snap_txg(libzfs_handle_t *hdl, const char *fs, const char *snap) { char name[ZFS_MAX_DATASET_NAME_LEN]; uint64_t txg = 0; if (fs == NULL || fs[0] == '\0' || snap == NULL || snap[0] == '\0') return (txg); (void) snprintf(name, sizeof (name), "%s@%s", fs, snap); if (zfs_dataset_exists(hdl, name, ZFS_TYPE_SNAPSHOT)) { zfs_handle_t *zhp = zfs_open(hdl, name, ZFS_TYPE_SNAPSHOT); if (zhp != NULL) { txg = zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG); zfs_close(zhp); } } return (txg); } /* * recursively generate nvlists describing datasets. See comment * for the data structure send_data_t above for description of contents * of the nvlist. */ static int send_iterate_fs(zfs_handle_t *zhp, void *arg) { send_data_t *sd = arg; nvlist_t *nvfs, *nv; int rv = 0; uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid; uint64_t fromsnap_txg_save = sd->fromsnap_txg; uint64_t tosnap_txg_save = sd->tosnap_txg; uint64_t txg = zhp->zfs_dmustats.dds_creation_txg; uint64_t guid = zhp->zfs_dmustats.dds_guid; uint64_t fromsnap_txg, tosnap_txg; char guidstring[64]; fromsnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->fromsnap); if (fromsnap_txg != 0) sd->fromsnap_txg = fromsnap_txg; tosnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->tosnap); if (tosnap_txg != 0) sd->tosnap_txg = tosnap_txg; /* * on the send side, if the current dataset does not have tosnap, * perform two additional checks: * * - skip sending the current dataset if it was created later than * the parent tosnap * - return error if the current dataset was created earlier than * the parent tosnap */ if (sd->tosnap != NULL && tosnap_txg == 0) { if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) { if (sd->verbose) { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "skipping dataset %s: snapshot %s does " "not exist\n"), zhp->zfs_name, sd->tosnap); } } else { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "cannot send %s@%s%s: snapshot %s@%s does not " "exist\n"), sd->fsname, sd->tosnap, sd->recursive ? dgettext(TEXT_DOMAIN, " recursively") : "", zhp->zfs_name, sd->tosnap); rv = -1; } goto out; } VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0)); VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name)); VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap", sd->parent_fromsnap_guid)); if (zhp->zfs_dmustats.dds_origin[0]) { zfs_handle_t *origin = zfs_open(zhp->zfs_hdl, zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT); if (origin == NULL) { rv = -1; goto out; } VERIFY(0 == nvlist_add_uint64(nvfs, "origin", origin->zfs_dmustats.dds_guid)); } /* iterate over props */ VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0)); send_iterate_prop(zhp, nv); VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv)); nvlist_free(nv); /* iterate over snaps, and set sd->parent_fromsnap_guid */ sd->parent_fromsnap_guid = 0; VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0)); VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0)); (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd); VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps)); VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops)); nvlist_free(sd->parent_snaps); nvlist_free(sd->snapprops); /* add this fs to nvlist */ (void) snprintf(guidstring, sizeof (guidstring), "0x%llx", (longlong_t)guid); VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs)); nvlist_free(nvfs); /* iterate over children */ if (sd->recursive) rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd); out: sd->parent_fromsnap_guid = parent_fromsnap_guid_save; sd->fromsnap_txg = fromsnap_txg_save; sd->tosnap_txg = tosnap_txg_save; zfs_close(zhp); return (rv); } static int gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap, const char *tosnap, boolean_t recursive, boolean_t verbose, nvlist_t **nvlp, avl_tree_t **avlp) { zfs_handle_t *zhp; send_data_t sd = { 0 }; int error; zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); if (zhp == NULL) return (EZFS_BADTYPE); VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0)); sd.fsname = fsname; sd.fromsnap = fromsnap; sd.tosnap = tosnap; sd.recursive = recursive; sd.verbose = verbose; if ((error = send_iterate_fs(zhp, &sd)) != 0) { nvlist_free(sd.fss); if (avlp != NULL) *avlp = NULL; *nvlp = NULL; return (error); } if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) { nvlist_free(sd.fss); *nvlp = NULL; return (EZFS_NOMEM); } *nvlp = sd.fss; return (0); } /* * Routines specific to "zfs send" */ typedef struct send_dump_data { /* these are all just the short snapname (the part after the @) */ const char *fromsnap; const char *tosnap; char prevsnap[ZFS_MAX_DATASET_NAME_LEN]; uint64_t prevsnap_obj; boolean_t seenfrom, seento, replicate, doall, fromorigin; boolean_t verbose, dryrun, parsable, progress, embed_data, std_out; boolean_t progressastitle; boolean_t large_block, compress; int outfd; boolean_t err; nvlist_t *fss; nvlist_t *snapholds; avl_tree_t *fsavl; snapfilter_cb_t *filter_cb; void *filter_cb_arg; nvlist_t *debugnv; char holdtag[ZFS_MAX_DATASET_NAME_LEN]; int cleanup_fd; uint64_t size; } send_dump_data_t; static int estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj, boolean_t fromorigin, enum lzc_send_flags flags, uint64_t *sizep) { zfs_cmd_t zc = { 0 }; libzfs_handle_t *hdl = zhp->zfs_hdl; assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); assert(fromsnap_obj == 0 || !fromorigin); (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); zc.zc_obj = fromorigin; zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); zc.zc_fromobj = fromsnap_obj; zc.zc_guid = 1; /* estimate flag */ zc.zc_flags = flags; if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { char errbuf[1024]; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "warning: cannot estimate space for '%s'"), zhp->zfs_name); switch (errno) { case EXDEV: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "not an earlier snapshot from the same fs")); return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); case ENOENT: if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_SNAPSHOT)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "incremental source (@%s) does not exist"), zc.zc_value); } return (zfs_error(hdl, EZFS_NOENT, errbuf)); case EDQUOT: case EFBIG: case EIO: case ENOLINK: case ENOSPC: case ENXIO: case EPIPE: case ERANGE: case EFAULT: case EROFS: zfs_error_aux(hdl, strerror(errno)); return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); default: return (zfs_standard_error(hdl, errno, errbuf)); } } *sizep = zc.zc_objset_type; return (0); } /* * Dumps a backup of the given snapshot (incremental from fromsnap if it's not * NULL) to the file descriptor specified by outfd. */ static int dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj, boolean_t fromorigin, int outfd, enum lzc_send_flags flags, nvlist_t *debugnv) { zfs_cmd_t zc = { 0 }; libzfs_handle_t *hdl = zhp->zfs_hdl; nvlist_t *thisdbg; assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); assert(fromsnap_obj == 0 || !fromorigin); (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); zc.zc_cookie = outfd; zc.zc_obj = fromorigin; zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); zc.zc_fromobj = fromsnap_obj; zc.zc_flags = flags; VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0)); if (fromsnap && fromsnap[0] != '\0') { VERIFY(0 == nvlist_add_string(thisdbg, "fromsnap", fromsnap)); } if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) { char errbuf[1024]; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "warning: cannot send '%s'"), zhp->zfs_name); VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno)); if (debugnv) { VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg)); } nvlist_free(thisdbg); switch (errno) { case EXDEV: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "not an earlier snapshot from the same fs")); return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); case ENOENT: if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_SNAPSHOT)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "incremental source (@%s) does not exist"), zc.zc_value); } return (zfs_error(hdl, EZFS_NOENT, errbuf)); case EDQUOT: case EFBIG: case EIO: case ENOLINK: case ENOSPC: #ifdef illumos case ENOSTR: #endif case ENXIO: case EPIPE: case ERANGE: case EFAULT: case EROFS: zfs_error_aux(hdl, strerror(errno)); return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); default: return (zfs_standard_error(hdl, errno, errbuf)); } } if (debugnv) VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg)); nvlist_free(thisdbg); return (0); } static void gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd) { assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); /* * zfs_send() only sets snapholds for sends that need them, * e.g. replication and doall. */ if (sdd->snapholds == NULL) return; fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag); } static void * send_progress_thread(void *arg) { progress_arg_t *pa = arg; zfs_cmd_t zc = { 0 }; zfs_handle_t *zhp = pa->pa_zhp; libzfs_handle_t *hdl = zhp->zfs_hdl; unsigned long long bytes, total; char buf[16]; time_t t; struct tm *tm; (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name)); if (!pa->pa_parsable && !pa->pa_astitle) (void) fprintf(stderr, "TIME SENT SNAPSHOT\n"); /* * Print the progress from ZFS_IOC_SEND_PROGRESS every second. */ for (;;) { (void) sleep(1); zc.zc_cookie = pa->pa_fd; if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0) return ((void *)-1); (void) time(&t); tm = localtime(&t); bytes = zc.zc_cookie; if (pa->pa_astitle) { int pct; if (pa->pa_size > bytes) pct = 100 * bytes / pa->pa_size; else pct = 100; setproctitle("sending %s (%d%%: %llu/%llu)", zhp->zfs_name, pct, bytes, pa->pa_size); } else if (pa->pa_parsable) { (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n", tm->tm_hour, tm->tm_min, tm->tm_sec, bytes, zhp->zfs_name); } else { zfs_nicenum(bytes, buf, sizeof (buf)); (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n", tm->tm_hour, tm->tm_min, tm->tm_sec, buf, zhp->zfs_name); } } } static void send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap, uint64_t size, boolean_t parsable) { if (parsable) { if (fromsnap != NULL) { (void) fprintf(fout, "incremental\t%s\t%s", fromsnap, tosnap); } else { (void) fprintf(fout, "full\t%s", tosnap); } } else { if (fromsnap != NULL) { if (strchr(fromsnap, '@') == NULL && strchr(fromsnap, '#') == NULL) { (void) fprintf(fout, dgettext(TEXT_DOMAIN, "send from @%s to %s"), fromsnap, tosnap); } else { (void) fprintf(fout, dgettext(TEXT_DOMAIN, "send from %s to %s"), fromsnap, tosnap); } } else { (void) fprintf(fout, dgettext(TEXT_DOMAIN, "full send of %s"), tosnap); } } if (size != 0) { if (parsable) { (void) fprintf(fout, "\t%llu", (longlong_t)size); } else { char buf[16]; zfs_nicenum(size, buf, sizeof (buf)); (void) fprintf(fout, dgettext(TEXT_DOMAIN, " estimated size is %s"), buf); } } (void) fprintf(fout, "\n"); } static int dump_snapshot(zfs_handle_t *zhp, void *arg) { send_dump_data_t *sdd = arg; progress_arg_t pa = { 0 }; pthread_t tid; char *thissnap; enum lzc_send_flags flags = 0; int err; boolean_t isfromsnap, istosnap, fromorigin; boolean_t exclude = B_FALSE; FILE *fout = sdd->std_out ? stdout : stderr; uint64_t size = 0; err = 0; thissnap = strchr(zhp->zfs_name, '@') + 1; isfromsnap = (sdd->fromsnap != NULL && strcmp(sdd->fromsnap, thissnap) == 0); if (!sdd->seenfrom && isfromsnap) { gather_holds(zhp, sdd); sdd->seenfrom = B_TRUE; (void) strcpy(sdd->prevsnap, thissnap); sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); zfs_close(zhp); return (0); } if (sdd->seento || !sdd->seenfrom) { zfs_close(zhp); return (0); } istosnap = (strcmp(sdd->tosnap, thissnap) == 0); if (istosnap) sdd->seento = B_TRUE; if (sdd->large_block) flags |= LZC_SEND_FLAG_LARGE_BLOCK; if (sdd->embed_data) flags |= LZC_SEND_FLAG_EMBED_DATA; if (sdd->compress) flags |= LZC_SEND_FLAG_COMPRESS; if (!sdd->doall && !isfromsnap && !istosnap) { if (sdd->replicate) { char *snapname; nvlist_t *snapprops; /* * Filter out all intermediate snapshots except origin * snapshots needed to replicate clones. */ nvlist_t *nvfs = fsavl_find(sdd->fsavl, zhp->zfs_dmustats.dds_guid, &snapname); VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snapprops", &snapprops)); VERIFY(0 == nvlist_lookup_nvlist(snapprops, thissnap, &snapprops)); exclude = !nvlist_exists(snapprops, "is_clone_origin"); } else { exclude = B_TRUE; } } /* * If a filter function exists, call it to determine whether * this snapshot will be sent. */ if (exclude || (sdd->filter_cb != NULL && sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) { /* * This snapshot is filtered out. Don't send it, and don't * set prevsnap_obj, so it will be as if this snapshot didn't * exist, and the next accepted snapshot will be sent as * an incremental from the last accepted one, or as the * first (and full) snapshot in the case of a replication, * non-incremental send. */ zfs_close(zhp); return (0); } gather_holds(zhp, sdd); fromorigin = sdd->prevsnap[0] == '\0' && (sdd->fromorigin || sdd->replicate); if (sdd->progress && sdd->dryrun) { (void) estimate_ioctl(zhp, sdd->prevsnap_obj, fromorigin, flags, &size); sdd->size += size; } if (sdd->verbose) { send_print_verbose(fout, zhp->zfs_name, sdd->prevsnap[0] ? sdd->prevsnap : NULL, size, sdd->parsable); } if (!sdd->dryrun) { /* * If progress reporting is requested, spawn a new thread to * poll ZFS_IOC_SEND_PROGRESS at a regular interval. */ if (sdd->progress) { pa.pa_zhp = zhp; pa.pa_fd = sdd->outfd; pa.pa_parsable = sdd->parsable; pa.pa_size = sdd->size; pa.pa_astitle = sdd->progressastitle; if ((err = pthread_create(&tid, NULL, send_progress_thread, &pa)) != 0) { zfs_close(zhp); return (err); } } err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj, fromorigin, sdd->outfd, flags, sdd->debugnv); if (sdd->progress) { (void) pthread_cancel(tid); (void) pthread_join(tid, NULL); } } (void) strcpy(sdd->prevsnap, thissnap); sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); zfs_close(zhp); return (err); } static int dump_filesystem(zfs_handle_t *zhp, void *arg) { int rv = 0; send_dump_data_t *sdd = arg; boolean_t missingfrom = B_FALSE; zfs_cmd_t zc = { 0 }; (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", zhp->zfs_name, sdd->tosnap); if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "WARNING: could not send %s@%s: does not exist\n"), zhp->zfs_name, sdd->tosnap); sdd->err = B_TRUE; return (0); } if (sdd->replicate && sdd->fromsnap) { /* * If this fs does not have fromsnap, and we're doing * recursive, we need to send a full stream from the * beginning (or an incremental from the origin if this * is a clone). If we're doing non-recursive, then let * them get the error. */ (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", zhp->zfs_name, sdd->fromsnap); if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) { missingfrom = B_TRUE; } } sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0; sdd->prevsnap_obj = 0; if (sdd->fromsnap == NULL || missingfrom) sdd->seenfrom = B_TRUE; rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg); if (!sdd->seenfrom) { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "WARNING: could not send %s@%s:\n" "incremental source (%s@%s) does not exist\n"), zhp->zfs_name, sdd->tosnap, zhp->zfs_name, sdd->fromsnap); sdd->err = B_TRUE; } else if (!sdd->seento) { if (sdd->fromsnap) { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "WARNING: could not send %s@%s:\n" "incremental source (%s@%s) " "is not earlier than it\n"), zhp->zfs_name, sdd->tosnap, zhp->zfs_name, sdd->fromsnap); } else { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "WARNING: " "could not send %s@%s: does not exist\n"), zhp->zfs_name, sdd->tosnap); } sdd->err = B_TRUE; } return (rv); } static int dump_filesystems(zfs_handle_t *rzhp, void *arg) { send_dump_data_t *sdd = arg; nvpair_t *fspair; boolean_t needagain, progress; if (!sdd->replicate) return (dump_filesystem(rzhp, sdd)); /* Mark the clone origin snapshots. */ for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; fspair = nvlist_next_nvpair(sdd->fss, fspair)) { nvlist_t *nvfs; uint64_t origin_guid = 0; VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs)); (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid); if (origin_guid != 0) { char *snapname; nvlist_t *origin_nv = fsavl_find(sdd->fsavl, origin_guid, &snapname); if (origin_nv != NULL) { nvlist_t *snapprops; VERIFY(0 == nvlist_lookup_nvlist(origin_nv, "snapprops", &snapprops)); VERIFY(0 == nvlist_lookup_nvlist(snapprops, snapname, &snapprops)); VERIFY(0 == nvlist_add_boolean( snapprops, "is_clone_origin")); } } } again: needagain = progress = B_FALSE; for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; fspair = nvlist_next_nvpair(sdd->fss, fspair)) { nvlist_t *fslist, *parent_nv; char *fsname; zfs_handle_t *zhp; int err; uint64_t origin_guid = 0; uint64_t parent_guid = 0; VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); if (nvlist_lookup_boolean(fslist, "sent") == 0) continue; VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0); (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid); (void) nvlist_lookup_uint64(fslist, "parentfromsnap", &parent_guid); if (parent_guid != 0) { parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL); if (!nvlist_exists(parent_nv, "sent")) { /* parent has not been sent; skip this one */ needagain = B_TRUE; continue; } } if (origin_guid != 0) { nvlist_t *origin_nv = fsavl_find(sdd->fsavl, origin_guid, NULL); if (origin_nv != NULL && !nvlist_exists(origin_nv, "sent")) { /* * origin has not been sent yet; * skip this clone. */ needagain = B_TRUE; continue; } } zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET); if (zhp == NULL) return (-1); err = dump_filesystem(zhp, sdd); VERIFY(nvlist_add_boolean(fslist, "sent") == 0); progress = B_TRUE; zfs_close(zhp); if (err) return (err); } if (needagain) { assert(progress); goto again; } /* clean out the sent flags in case we reuse this fss */ for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair; fspair = nvlist_next_nvpair(sdd->fss, fspair)) { nvlist_t *fslist; VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0); (void) nvlist_remove_all(fslist, "sent"); } return (0); } nvlist_t * zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token) { unsigned int version; int nread; unsigned long long checksum, packed_len; /* * Decode token header, which is: * -- * Note that the only supported token version is 1. */ nread = sscanf(token, "%u-%llx-%llx-", &version, &checksum, &packed_len); if (nread != 3) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "resume token is corrupt (invalid format)")); return (NULL); } if (version != ZFS_SEND_RESUME_TOKEN_VERSION) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "resume token is corrupt (invalid version %u)"), version); return (NULL); } /* convert hexadecimal representation to binary */ token = strrchr(token, '-') + 1; int len = strlen(token) / 2; unsigned char *compressed = zfs_alloc(hdl, len); for (int i = 0; i < len; i++) { nread = sscanf(token + i * 2, "%2hhx", compressed + i); if (nread != 1) { free(compressed); zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "resume token is corrupt " "(payload is not hex-encoded)")); return (NULL); } } /* verify checksum */ zio_cksum_t cksum; fletcher_4_native(compressed, len, NULL, &cksum); if (cksum.zc_word[0] != checksum) { free(compressed); zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "resume token is corrupt (incorrect checksum)")); return (NULL); } /* uncompress */ void *packed = zfs_alloc(hdl, packed_len); uLongf packed_len_long = packed_len; if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK || packed_len_long != packed_len) { free(packed); free(compressed); zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "resume token is corrupt (decompression failed)")); return (NULL); } /* unpack nvlist */ nvlist_t *nv; int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP); free(packed); free(compressed); if (error != 0) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "resume token is corrupt (nvlist_unpack failed)")); return (NULL); } return (nv); } int zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd, const char *resume_token) { char errbuf[1024]; char *toname; char *fromname = NULL; uint64_t resumeobj, resumeoff, toguid, fromguid, bytes; zfs_handle_t *zhp; int error = 0; char name[ZFS_MAX_DATASET_NAME_LEN]; enum lzc_send_flags lzc_flags = 0; uint64_t size = 0; FILE *fout = (flags->verbose && flags->dryrun) ? stdout : stderr; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot resume send")); nvlist_t *resume_nvl = zfs_send_resume_token_to_nvlist(hdl, resume_token); if (resume_nvl == NULL) { /* * zfs_error_aux has already been set by * zfs_send_resume_token_to_nvlist */ return (zfs_error(hdl, EZFS_FAULT, errbuf)); } if (flags->verbose) { (void) fprintf(fout, dgettext(TEXT_DOMAIN, "resume token contents:\n")); nvlist_print(fout, resume_nvl); } if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 || nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 || nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 || nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 || nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "resume token is corrupt")); return (zfs_error(hdl, EZFS_FAULT, errbuf)); } fromguid = 0; (void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid); if (flags->largeblock || nvlist_exists(resume_nvl, "largeblockok")) lzc_flags |= LZC_SEND_FLAG_LARGE_BLOCK; if (flags->embed_data || nvlist_exists(resume_nvl, "embedok")) lzc_flags |= LZC_SEND_FLAG_EMBED_DATA; if (flags->compress || nvlist_exists(resume_nvl, "compressok")) lzc_flags |= LZC_SEND_FLAG_COMPRESS; if (guid_to_name(hdl, toname, toguid, B_FALSE, name) != 0) { if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' is no longer the same snapshot used in " "the initial send"), toname); } else { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' used in the initial send no longer exists"), toname); } return (zfs_error(hdl, EZFS_BADPATH, errbuf)); } zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); if (zhp == NULL) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "unable to access '%s'"), name); return (zfs_error(hdl, EZFS_BADPATH, errbuf)); } if (fromguid != 0) { if (guid_to_name(hdl, toname, fromguid, B_TRUE, name) != 0) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "incremental source %#llx no longer exists"), (longlong_t)fromguid); return (zfs_error(hdl, EZFS_BADPATH, errbuf)); } fromname = name; } if (flags->progress) { error = lzc_send_space(zhp->zfs_name, fromname, lzc_flags, &size); if (error == 0) size = MAX(0, (int64_t)(size - bytes)); } if (flags->verbose) { send_print_verbose(fout, zhp->zfs_name, fromname, size, flags->parsable); } if (!flags->dryrun) { progress_arg_t pa = { 0 }; pthread_t tid; /* * If progress reporting is requested, spawn a new thread to * poll ZFS_IOC_SEND_PROGRESS at a regular interval. */ if (flags->progress) { pa.pa_zhp = zhp; pa.pa_fd = outfd; pa.pa_parsable = flags->parsable; pa.pa_size = size; pa.pa_astitle = flags->progressastitle; error = pthread_create(&tid, NULL, send_progress_thread, &pa); if (error != 0) { zfs_close(zhp); return (error); } } error = lzc_send_resume(zhp->zfs_name, fromname, outfd, lzc_flags, resumeobj, resumeoff); if (flags->progress) { (void) pthread_cancel(tid); (void) pthread_join(tid, NULL); } char errbuf[1024]; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "warning: cannot send '%s'"), zhp->zfs_name); zfs_close(zhp); switch (error) { case 0: return (0); case EXDEV: case ENOENT: case EDQUOT: case EFBIG: case EIO: case ENOLINK: case ENOSPC: #ifdef illumos case ENOSTR: #endif case ENXIO: case EPIPE: case ERANGE: case EFAULT: case EROFS: zfs_error_aux(hdl, strerror(errno)); return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); default: return (zfs_standard_error(hdl, errno, errbuf)); } } zfs_close(zhp); return (error); } /* * Generate a send stream for the dataset identified by the argument zhp. * * The content of the send stream is the snapshot identified by * 'tosnap'. Incremental streams are requested in two ways: * - from the snapshot identified by "fromsnap" (if non-null) or * - from the origin of the dataset identified by zhp, which must * be a clone. In this case, "fromsnap" is null and "fromorigin" * is TRUE. * * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM) * if "replicate" is set. If "doall" is set, dump all the intermediate * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall" * case too. If "props" is set, send properties. */ int zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, sendflags_t *flags, int outfd, snapfilter_cb_t filter_func, void *cb_arg, nvlist_t **debugnvp) { char errbuf[1024]; send_dump_data_t sdd = { 0 }; int err = 0; nvlist_t *fss = NULL; avl_tree_t *fsavl = NULL; static uint64_t holdseq; int spa_version; pthread_t tid = 0; int pipefd[2]; dedup_arg_t dda = { 0 }; int featureflags = 0; FILE *fout; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot send '%s'"), zhp->zfs_name); if (fromsnap && fromsnap[0] == '\0') { zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN, "zero-length incremental source")); return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf)); } if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) { uint64_t version; version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION); if (version >= ZPL_VERSION_SA) { featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; } } if (flags->dedup && !flags->dryrun) { featureflags |= (DMU_BACKUP_FEATURE_DEDUP | DMU_BACKUP_FEATURE_DEDUPPROPS); if ((err = pipe(pipefd)) != 0) { zfs_error_aux(zhp->zfs_hdl, strerror(errno)); return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED, errbuf)); } dda.outputfd = outfd; dda.inputfd = pipefd[1]; dda.dedup_hdl = zhp->zfs_hdl; if ((err = pthread_create(&tid, NULL, cksummer, &dda)) != 0) { (void) close(pipefd[0]); (void) close(pipefd[1]); zfs_error_aux(zhp->zfs_hdl, strerror(errno)); return (zfs_error(zhp->zfs_hdl, EZFS_THREADCREATEFAILED, errbuf)); } } if (flags->replicate || flags->doall || flags->props) { dmu_replay_record_t drr = { 0 }; char *packbuf = NULL; size_t buflen = 0; zio_cksum_t zc = { 0 }; if (flags->replicate || flags->props) { nvlist_t *hdrnv; VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0)); if (fromsnap) { VERIFY(0 == nvlist_add_string(hdrnv, "fromsnap", fromsnap)); } VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap)); if (!flags->replicate) { VERIFY(0 == nvlist_add_boolean(hdrnv, "not_recursive")); } err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name, fromsnap, tosnap, flags->replicate, flags->verbose, &fss, &fsavl); if (err) goto err_out; VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss)); err = nvlist_pack(hdrnv, &packbuf, &buflen, NV_ENCODE_XDR, 0); if (debugnvp) *debugnvp = hdrnv; else nvlist_free(hdrnv); if (err) goto stderr_out; } if (!flags->dryrun) { /* write first begin record */ drr.drr_type = DRR_BEGIN; drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin. drr_versioninfo, DMU_COMPOUNDSTREAM); DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin. drr_versioninfo, featureflags); (void) snprintf(drr.drr_u.drr_begin.drr_toname, sizeof (drr.drr_u.drr_begin.drr_toname), "%s@%s", zhp->zfs_name, tosnap); drr.drr_payloadlen = buflen; err = dump_record(&drr, packbuf, buflen, &zc, outfd); free(packbuf); if (err != 0) goto stderr_out; /* write end record */ bzero(&drr, sizeof (drr)); drr.drr_type = DRR_END; drr.drr_u.drr_end.drr_checksum = zc; err = write(outfd, &drr, sizeof (drr)); if (err == -1) { err = errno; goto stderr_out; } err = 0; } } /* dump each stream */ sdd.fromsnap = fromsnap; sdd.tosnap = tosnap; if (tid != 0) sdd.outfd = pipefd[0]; else sdd.outfd = outfd; sdd.replicate = flags->replicate; sdd.doall = flags->doall; sdd.fromorigin = flags->fromorigin; sdd.fss = fss; sdd.fsavl = fsavl; sdd.verbose = flags->verbose; sdd.parsable = flags->parsable; sdd.progress = flags->progress; sdd.progressastitle = flags->progressastitle; sdd.dryrun = flags->dryrun; sdd.large_block = flags->largeblock; sdd.embed_data = flags->embed_data; sdd.compress = flags->compress; sdd.filter_cb = filter_func; sdd.filter_cb_arg = cb_arg; if (debugnvp) sdd.debugnv = *debugnvp; if (sdd.verbose && sdd.dryrun) sdd.std_out = B_TRUE; fout = sdd.std_out ? stdout : stderr; /* * Some flags require that we place user holds on the datasets that are * being sent so they don't get destroyed during the send. We can skip * this step if the pool is imported read-only since the datasets cannot * be destroyed. */ if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp), ZPOOL_PROP_READONLY, NULL) && zfs_spa_version(zhp, &spa_version) == 0 && spa_version >= SPA_VERSION_USERREFS && (flags->doall || flags->replicate)) { ++holdseq; (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag), ".send-%d-%llu", getpid(), (u_longlong_t)holdseq); sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); if (sdd.cleanup_fd < 0) { err = errno; goto stderr_out; } sdd.snapholds = fnvlist_alloc(); } else { sdd.cleanup_fd = -1; sdd.snapholds = NULL; } if (flags->progress || sdd.snapholds != NULL) { /* * Do a verbose no-op dry run to get all the verbose output * or to gather snapshot hold's before generating any data, * then do a non-verbose real run to generate the streams. */ sdd.dryrun = B_TRUE; err = dump_filesystems(zhp, &sdd); if (err != 0) goto stderr_out; if (flags->verbose) { if (flags->parsable) { (void) fprintf(fout, "size\t%llu\n", (longlong_t)sdd.size); } else { char buf[16]; zfs_nicenum(sdd.size, buf, sizeof (buf)); (void) fprintf(fout, dgettext(TEXT_DOMAIN, "total estimated size is %s\n"), buf); } } /* Ensure no snaps found is treated as an error. */ if (!sdd.seento) { err = ENOENT; goto err_out; } /* Skip the second run if dryrun was requested. */ if (flags->dryrun) goto err_out; if (sdd.snapholds != NULL) { err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds); if (err != 0) goto stderr_out; fnvlist_free(sdd.snapholds); sdd.snapholds = NULL; } sdd.dryrun = B_FALSE; sdd.verbose = B_FALSE; } err = dump_filesystems(zhp, &sdd); fsavl_destroy(fsavl); nvlist_free(fss); /* Ensure no snaps found is treated as an error. */ if (err == 0 && !sdd.seento) err = ENOENT; if (tid != 0) { if (err != 0) (void) pthread_cancel(tid); (void) close(pipefd[0]); (void) pthread_join(tid, NULL); } if (sdd.cleanup_fd != -1) { VERIFY(0 == close(sdd.cleanup_fd)); sdd.cleanup_fd = -1; } if (!flags->dryrun && (flags->replicate || flags->doall || flags->props)) { /* * write final end record. NB: want to do this even if * there was some error, because it might not be totally * failed. */ dmu_replay_record_t drr = { 0 }; drr.drr_type = DRR_END; if (write(outfd, &drr, sizeof (drr)) == -1) { return (zfs_standard_error(zhp->zfs_hdl, errno, errbuf)); } } return (err || sdd.err); stderr_out: err = zfs_standard_error(zhp->zfs_hdl, err, errbuf); err_out: fsavl_destroy(fsavl); nvlist_free(fss); fnvlist_free(sdd.snapholds); if (sdd.cleanup_fd != -1) VERIFY(0 == close(sdd.cleanup_fd)); if (tid != 0) { (void) pthread_cancel(tid); (void) close(pipefd[0]); (void) pthread_join(tid, NULL); } return (err); } int zfs_send_one(zfs_handle_t *zhp, const char *from, int fd, enum lzc_send_flags flags) { int err; libzfs_handle_t *hdl = zhp->zfs_hdl; char errbuf[1024]; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "warning: cannot send '%s'"), zhp->zfs_name); err = lzc_send(zhp->zfs_name, from, fd, flags); if (err != 0) { switch (errno) { case EXDEV: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "not an earlier snapshot from the same fs")); return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf)); case ENOENT: case ESRCH: if (lzc_exists(zhp->zfs_name)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "incremental source (%s) does not exist"), from); } return (zfs_error(hdl, EZFS_NOENT, errbuf)); case EBUSY: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "target is busy; if a filesystem, " "it must not be mounted")); return (zfs_error(hdl, EZFS_BUSY, errbuf)); case EDQUOT: case EFBIG: case EIO: case ENOLINK: case ENOSPC: #ifdef illumos case ENOSTR: #endif case ENXIO: case EPIPE: case ERANGE: case EFAULT: case EROFS: zfs_error_aux(hdl, strerror(errno)); return (zfs_error(hdl, EZFS_BADBACKUP, errbuf)); default: return (zfs_standard_error(hdl, errno, errbuf)); } } return (err != 0); } /* * Routines specific to "zfs recv" */ static int recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen, boolean_t byteswap, zio_cksum_t *zc) { char *cp = buf; int rv; int len = ilen; assert(ilen <= SPA_MAXBLOCKSIZE); do { rv = read(fd, cp, len); cp += rv; len -= rv; } while (rv > 0); if (rv < 0 || len != 0) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "failed to read from stream")); return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN, "cannot receive"))); } if (zc) { if (byteswap) (void) fletcher_4_incremental_byteswap(buf, ilen, zc); else (void) fletcher_4_incremental_native(buf, ilen, zc); } return (0); } static int recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp, boolean_t byteswap, zio_cksum_t *zc) { char *buf; int err; buf = zfs_alloc(hdl, len); if (buf == NULL) return (ENOMEM); err = recv_read(hdl, fd, buf, len, byteswap, zc); if (err != 0) { free(buf); return (err); } err = nvlist_unpack(buf, len, nvp, 0); free(buf); if (err != 0) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " "stream (malformed nvlist)")); return (EINVAL); } return (0); } static int recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname, int baselen, char *newname, recvflags_t *flags) { static int seq; int err; prop_changelist_t *clp; zfs_handle_t *zhp; zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); if (zhp == NULL) return (-1); clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, flags->force ? MS_FORCE : 0); zfs_close(zhp); if (clp == NULL) return (-1); err = changelist_prefix(clp); if (err) return (err); if (tryname) { (void) strcpy(newname, tryname); if (flags->verbose) { (void) printf("attempting rename %s to %s\n", name, newname); } err = lzc_rename(name, newname); if (err == 0) changelist_rename(clp, name, tryname); } else { err = ENOENT; } if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) { seq++; (void) snprintf(newname, ZFS_MAX_DATASET_NAME_LEN, "%.*srecv-%u-%u", baselen, name, getpid(), seq); if (flags->verbose) { (void) printf("failed - trying rename %s to %s\n", name, newname); } err = lzc_rename(name, newname); if (err == 0) changelist_rename(clp, name, newname); if (err && flags->verbose) { (void) printf("failed (%u) - " "will try again on next pass\n", errno); } err = EAGAIN; } else if (flags->verbose) { if (err == 0) (void) printf("success\n"); else (void) printf("failed (%u)\n", errno); } (void) changelist_postfix(clp); changelist_free(clp); return (err); } static int recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen, char *newname, recvflags_t *flags) { int err = 0; prop_changelist_t *clp; zfs_handle_t *zhp; boolean_t defer = B_FALSE; int spa_version; zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET); if (zhp == NULL) return (-1); clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, flags->force ? MS_FORCE : 0); if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT && zfs_spa_version(zhp, &spa_version) == 0 && spa_version >= SPA_VERSION_USERREFS) defer = B_TRUE; zfs_close(zhp); if (clp == NULL) return (-1); err = changelist_prefix(clp); if (err) return (err); if (flags->verbose) (void) printf("attempting destroy %s\n", name); if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) { nvlist_t *nv = fnvlist_alloc(); fnvlist_add_boolean(nv, name); err = lzc_destroy_snaps(nv, defer, NULL); fnvlist_free(nv); } else { err = lzc_destroy(name); } if (err == 0) { if (flags->verbose) (void) printf("success\n"); changelist_remove(clp, name); } (void) changelist_postfix(clp); changelist_free(clp); /* * Deferred destroy might destroy the snapshot or only mark it to be * destroyed later, and it returns success in either case. */ if (err != 0 || (defer && zfs_dataset_exists(hdl, name, ZFS_TYPE_SNAPSHOT))) { err = recv_rename(hdl, name, NULL, baselen, newname, flags); } return (err); } typedef struct guid_to_name_data { uint64_t guid; boolean_t bookmark_ok; char *name; char *skip; } guid_to_name_data_t; static int guid_to_name_cb(zfs_handle_t *zhp, void *arg) { guid_to_name_data_t *gtnd = arg; const char *slash; int err; if (gtnd->skip != NULL && (slash = strrchr(zhp->zfs_name, '/')) != NULL && strcmp(slash + 1, gtnd->skip) == 0) { zfs_close(zhp); return (0); } if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid) { (void) strcpy(gtnd->name, zhp->zfs_name); zfs_close(zhp); return (EEXIST); } err = zfs_iter_children(zhp, guid_to_name_cb, gtnd); if (err != EEXIST && gtnd->bookmark_ok) err = zfs_iter_bookmarks(zhp, guid_to_name_cb, gtnd); zfs_close(zhp); return (err); } /* * Attempt to find the local dataset associated with this guid. In the case of * multiple matches, we attempt to find the "best" match by searching * progressively larger portions of the hierarchy. This allows one to send a * tree of datasets individually and guarantee that we will find the source * guid within that hierarchy, even if there are multiple matches elsewhere. */ static int guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid, boolean_t bookmark_ok, char *name) { char pname[ZFS_MAX_DATASET_NAME_LEN]; guid_to_name_data_t gtnd; gtnd.guid = guid; gtnd.bookmark_ok = bookmark_ok; gtnd.name = name; gtnd.skip = NULL; /* * Search progressively larger portions of the hierarchy, starting * with the filesystem specified by 'parent'. This will * select the "most local" version of the origin snapshot in the case * that there are multiple matching snapshots in the system. */ (void) strlcpy(pname, parent, sizeof (pname)); char *cp = strrchr(pname, '@'); if (cp == NULL) cp = strchr(pname, '\0'); for (; cp != NULL; cp = strrchr(pname, '/')) { /* Chop off the last component and open the parent */ *cp = '\0'; zfs_handle_t *zhp = make_dataset_handle(hdl, pname); if (zhp == NULL) continue; int err = guid_to_name_cb(zfs_handle_dup(zhp), >nd); if (err != EEXIST) err = zfs_iter_children(zhp, guid_to_name_cb, >nd); if (err != EEXIST && bookmark_ok) err = zfs_iter_bookmarks(zhp, guid_to_name_cb, >nd); zfs_close(zhp); if (err == EEXIST) return (0); /* * Remember the last portion of the dataset so we skip it next * time through (as we've already searched that portion of the * hierarchy). */ gtnd.skip = strrchr(pname, '/') + 1; } return (ENOENT); } /* * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if * guid1 is after guid2. */ static int created_before(libzfs_handle_t *hdl, avl_tree_t *avl, uint64_t guid1, uint64_t guid2) { nvlist_t *nvfs; char *fsname, *snapname; char buf[ZFS_MAX_DATASET_NAME_LEN]; int rv; zfs_handle_t *guid1hdl, *guid2hdl; uint64_t create1, create2; if (guid2 == 0) return (0); if (guid1 == 0) return (1); nvfs = fsavl_find(avl, guid1, &snapname); VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); if (guid1hdl == NULL) return (-1); nvfs = fsavl_find(avl, guid2, &snapname); VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname); guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT); if (guid2hdl == NULL) { zfs_close(guid1hdl); return (-1); } create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG); create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG); if (create1 < create2) rv = -1; else if (create1 > create2) rv = +1; else rv = 0; zfs_close(guid1hdl); zfs_close(guid2hdl); return (rv); } static int recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs, recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl, nvlist_t *renamed) { nvlist_t *local_nv, *deleted = NULL; avl_tree_t *local_avl; nvpair_t *fselem, *nextfselem; char *fromsnap; char newname[ZFS_MAX_DATASET_NAME_LEN]; char guidname[32]; int error; boolean_t needagain, progress, recursive; char *s1, *s2; VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap)); recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == ENOENT); if (flags->dryrun) return (0); again: needagain = progress = B_FALSE; VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0)); if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL, recursive, B_FALSE, &local_nv, &local_avl)) != 0) return (error); /* * Process deletes and renames */ for (fselem = nvlist_next_nvpair(local_nv, NULL); fselem; fselem = nextfselem) { nvlist_t *nvfs, *snaps; nvlist_t *stream_nvfs = NULL; nvpair_t *snapelem, *nextsnapelem; uint64_t fromguid = 0; uint64_t originguid = 0; uint64_t stream_originguid = 0; uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid; char *fsname, *stream_fsname; nextfselem = nvlist_next_nvpair(local_nv, fselem); VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs)); VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps)); VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname)); VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap", &parent_fromsnap_guid)); (void) nvlist_lookup_uint64(nvfs, "origin", &originguid); /* * First find the stream's fs, so we can check for * a different origin (due to "zfs promote") */ for (snapelem = nvlist_next_nvpair(snaps, NULL); snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) { uint64_t thisguid; VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); stream_nvfs = fsavl_find(stream_avl, thisguid, NULL); if (stream_nvfs != NULL) break; } /* check for promote */ (void) nvlist_lookup_uint64(stream_nvfs, "origin", &stream_originguid); if (stream_nvfs && originguid != stream_originguid) { switch (created_before(hdl, local_avl, stream_originguid, originguid)) { case 1: { /* promote it! */ zfs_cmd_t zc = { 0 }; nvlist_t *origin_nvfs; char *origin_fsname; if (flags->verbose) (void) printf("promoting %s\n", fsname); origin_nvfs = fsavl_find(local_avl, originguid, NULL); VERIFY(0 == nvlist_lookup_string(origin_nvfs, "name", &origin_fsname)); (void) strlcpy(zc.zc_value, origin_fsname, sizeof (zc.zc_value)); (void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_name)); error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc); if (error == 0) progress = B_TRUE; break; } default: break; case -1: fsavl_destroy(local_avl); nvlist_free(local_nv); return (-1); } /* * We had/have the wrong origin, therefore our * list of snapshots is wrong. Need to handle * them on the next pass. */ needagain = B_TRUE; continue; } for (snapelem = nvlist_next_nvpair(snaps, NULL); snapelem; snapelem = nextsnapelem) { uint64_t thisguid; char *stream_snapname; nvlist_t *found, *props; nextsnapelem = nvlist_next_nvpair(snaps, snapelem); VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid)); found = fsavl_find(stream_avl, thisguid, &stream_snapname); /* check for delete */ if (found == NULL) { char name[ZFS_MAX_DATASET_NAME_LEN]; if (!flags->force) continue; (void) snprintf(name, sizeof (name), "%s@%s", fsname, nvpair_name(snapelem)); error = recv_destroy(hdl, name, strlen(fsname)+1, newname, flags); if (error) needagain = B_TRUE; else progress = B_TRUE; sprintf(guidname, "%" PRIu64, thisguid); nvlist_add_boolean(deleted, guidname); continue; } stream_nvfs = found; if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops", &props) && 0 == nvlist_lookup_nvlist(props, stream_snapname, &props)) { zfs_cmd_t zc = { 0 }; zc.zc_cookie = B_TRUE; /* received */ (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s", fsname, nvpair_name(snapelem)); if (zcmd_write_src_nvlist(hdl, &zc, props) == 0) { (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc); zcmd_free_nvlists(&zc); } } /* check for different snapname */ if (strcmp(nvpair_name(snapelem), stream_snapname) != 0) { char name[ZFS_MAX_DATASET_NAME_LEN]; char tryname[ZFS_MAX_DATASET_NAME_LEN]; (void) snprintf(name, sizeof (name), "%s@%s", fsname, nvpair_name(snapelem)); (void) snprintf(tryname, sizeof (name), "%s@%s", fsname, stream_snapname); error = recv_rename(hdl, name, tryname, strlen(fsname)+1, newname, flags); if (error) needagain = B_TRUE; else progress = B_TRUE; } if (strcmp(stream_snapname, fromsnap) == 0) fromguid = thisguid; } /* check for delete */ if (stream_nvfs == NULL) { if (!flags->force) continue; error = recv_destroy(hdl, fsname, strlen(tofs)+1, newname, flags); if (error) needagain = B_TRUE; else progress = B_TRUE; sprintf(guidname, "%" PRIu64, parent_fromsnap_guid); nvlist_add_boolean(deleted, guidname); continue; } if (fromguid == 0) { if (flags->verbose) { (void) printf("local fs %s does not have " "fromsnap (%s in stream); must have " "been deleted locally; ignoring\n", fsname, fromsnap); } continue; } VERIFY(0 == nvlist_lookup_string(stream_nvfs, "name", &stream_fsname)); VERIFY(0 == nvlist_lookup_uint64(stream_nvfs, "parentfromsnap", &stream_parent_fromsnap_guid)); s1 = strrchr(fsname, '/'); s2 = strrchr(stream_fsname, '/'); /* * Check if we're going to rename based on parent guid change * and the current parent guid was also deleted. If it was then * rename will fail and is likely unneeded, so avoid this and * force an early retry to determine the new * parent_fromsnap_guid. */ if (stream_parent_fromsnap_guid != 0 && parent_fromsnap_guid != 0 && stream_parent_fromsnap_guid != parent_fromsnap_guid) { sprintf(guidname, "%" PRIu64, parent_fromsnap_guid); if (nvlist_exists(deleted, guidname)) { progress = B_TRUE; needagain = B_TRUE; goto doagain; } } /* * Check for rename. If the exact receive path is specified, it * does not count as a rename, but we still need to check the * datasets beneath it. */ if ((stream_parent_fromsnap_guid != 0 && parent_fromsnap_guid != 0 && stream_parent_fromsnap_guid != parent_fromsnap_guid) || ((flags->isprefix || strcmp(tofs, fsname) != 0) && (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) { nvlist_t *parent; char tryname[ZFS_MAX_DATASET_NAME_LEN]; parent = fsavl_find(local_avl, stream_parent_fromsnap_guid, NULL); /* * NB: parent might not be found if we used the * tosnap for stream_parent_fromsnap_guid, * because the parent is a newly-created fs; * we'll be able to rename it after we recv the * new fs. */ if (parent != NULL) { char *pname; VERIFY(0 == nvlist_lookup_string(parent, "name", &pname)); (void) snprintf(tryname, sizeof (tryname), "%s%s", pname, strrchr(stream_fsname, '/')); } else { tryname[0] = '\0'; if (flags->verbose) { (void) printf("local fs %s new parent " "not found\n", fsname); } } newname[0] = '\0'; error = recv_rename(hdl, fsname, tryname, strlen(tofs)+1, newname, flags); if (renamed != NULL && newname[0] != '\0') { VERIFY(0 == nvlist_add_boolean(renamed, newname)); } if (error) needagain = B_TRUE; else progress = B_TRUE; } } doagain: fsavl_destroy(local_avl); nvlist_free(local_nv); nvlist_free(deleted); if (needagain && progress) { /* do another pass to fix up temporary names */ if (flags->verbose) (void) printf("another pass:\n"); goto again; } return (needagain); } static int zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname, recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc, char **top_zfs, int cleanup_fd, uint64_t *action_handlep) { nvlist_t *stream_nv = NULL; avl_tree_t *stream_avl = NULL; char *fromsnap = NULL; char *sendsnap = NULL; char *cp; char tofs[ZFS_MAX_DATASET_NAME_LEN]; char sendfs[ZFS_MAX_DATASET_NAME_LEN]; char errbuf[1024]; dmu_replay_record_t drre; int error; boolean_t anyerr = B_FALSE; boolean_t softerr = B_FALSE; boolean_t recursive; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot receive")); assert(drr->drr_type == DRR_BEGIN); assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC); assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) == DMU_COMPOUNDSTREAM); /* * Read in the nvlist from the stream. */ if (drr->drr_payloadlen != 0) { error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen, &stream_nv, flags->byteswap, zc); if (error) { error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); goto out; } } recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == ENOENT); if (recursive && strchr(destname, '@')) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot specify snapshot name for multi-snapshot stream")); error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); goto out; } /* * Read in the end record and verify checksum. */ if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre), flags->byteswap, NULL))) goto out; if (flags->byteswap) { drre.drr_type = BSWAP_32(drre.drr_type); drre.drr_u.drr_end.drr_checksum.zc_word[0] = BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]); drre.drr_u.drr_end.drr_checksum.zc_word[1] = BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]); drre.drr_u.drr_end.drr_checksum.zc_word[2] = BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]); drre.drr_u.drr_end.drr_checksum.zc_word[3] = BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]); } if (drre.drr_type != DRR_END) { error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); goto out; } if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "incorrect header checksum")); error = zfs_error(hdl, EZFS_BADSTREAM, errbuf); goto out; } (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap); if (drr->drr_payloadlen != 0) { nvlist_t *stream_fss; VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss", &stream_fss)); if ((stream_avl = fsavl_create(stream_fss)) == NULL) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "couldn't allocate avl tree")); error = zfs_error(hdl, EZFS_NOMEM, errbuf); goto out; } if (fromsnap != NULL && recursive) { nvlist_t *renamed = NULL; nvpair_t *pair = NULL; (void) strlcpy(tofs, destname, sizeof (tofs)); if (flags->isprefix) { struct drr_begin *drrb = &drr->drr_u.drr_begin; int i; if (flags->istail) { cp = strrchr(drrb->drr_toname, '/'); if (cp == NULL) { (void) strlcat(tofs, "/", sizeof (tofs)); i = 0; } else { i = (cp - drrb->drr_toname); } } else { i = strcspn(drrb->drr_toname, "/@"); } /* zfs_receive_one() will create_parents() */ (void) strlcat(tofs, &drrb->drr_toname[i], sizeof (tofs)); *strchr(tofs, '@') = '\0'; } if (!flags->dryrun && !flags->nomount) { VERIFY(0 == nvlist_alloc(&renamed, NV_UNIQUE_NAME, 0)); } softerr = recv_incremental_replication(hdl, tofs, flags, stream_nv, stream_avl, renamed); /* Unmount renamed filesystems before receiving. */ while ((pair = nvlist_next_nvpair(renamed, pair)) != NULL) { zfs_handle_t *zhp; prop_changelist_t *clp = NULL; zhp = zfs_open(hdl, nvpair_name(pair), ZFS_TYPE_FILESYSTEM); if (zhp != NULL) { clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT, 0, 0); zfs_close(zhp); if (clp != NULL) { softerr |= changelist_prefix(clp); changelist_free(clp); } } } nvlist_free(renamed); } } /* * Get the fs specified by the first path in the stream (the top level * specified by 'zfs send') and pass it to each invocation of * zfs_receive_one(). */ (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname, sizeof (sendfs)); if ((cp = strchr(sendfs, '@')) != NULL) { *cp = '\0'; /* * Find the "sendsnap", the final snapshot in a replication * stream. zfs_receive_one() handles certain errors * differently, depending on if the contained stream is the * last one or not. */ sendsnap = (cp + 1); } /* Finally, receive each contained stream */ do { /* * we should figure out if it has a recoverable * error, in which case do a recv_skip() and drive on. * Note, if we fail due to already having this guid, * zfs_receive_one() will take care of it (ie, * recv_skip() and return 0). */ error = zfs_receive_impl(hdl, destname, NULL, flags, fd, sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd, action_handlep, sendsnap); if (error == ENODATA) { error = 0; break; } anyerr |= error; } while (error == 0); if (drr->drr_payloadlen != 0 && recursive && fromsnap != NULL) { /* * Now that we have the fs's they sent us, try the * renames again. */ softerr = recv_incremental_replication(hdl, tofs, flags, stream_nv, stream_avl, NULL); } out: fsavl_destroy(stream_avl); nvlist_free(stream_nv); if (softerr) error = -2; if (anyerr) error = -1; return (error); } static void trunc_prop_errs(int truncated) { ASSERT(truncated != 0); if (truncated == 1) (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "1 more property could not be set\n")); else (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "%d more properties could not be set\n"), truncated); } static int recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap) { dmu_replay_record_t *drr; void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE); char errbuf[1024]; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot receive:")); /* XXX would be great to use lseek if possible... */ drr = buf; while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t), byteswap, NULL) == 0) { if (byteswap) drr->drr_type = BSWAP_32(drr->drr_type); switch (drr->drr_type) { case DRR_BEGIN: if (drr->drr_payloadlen != 0) { (void) recv_read(hdl, fd, buf, drr->drr_payloadlen, B_FALSE, NULL); } break; case DRR_END: free(buf); return (0); case DRR_OBJECT: if (byteswap) { drr->drr_u.drr_object.drr_bonuslen = BSWAP_32(drr->drr_u.drr_object. drr_bonuslen); } (void) recv_read(hdl, fd, buf, P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8), B_FALSE, NULL); break; case DRR_WRITE: if (byteswap) { drr->drr_u.drr_write.drr_logical_size = BSWAP_64( drr->drr_u.drr_write.drr_logical_size); drr->drr_u.drr_write.drr_compressed_size = BSWAP_64( drr->drr_u.drr_write.drr_compressed_size); } uint64_t payload_size = DRR_WRITE_PAYLOAD_SIZE(&drr->drr_u.drr_write); (void) recv_read(hdl, fd, buf, payload_size, B_FALSE, NULL); break; case DRR_SPILL: if (byteswap) { drr->drr_u.drr_spill.drr_length = BSWAP_64(drr->drr_u.drr_spill.drr_length); } (void) recv_read(hdl, fd, buf, drr->drr_u.drr_spill.drr_length, B_FALSE, NULL); break; case DRR_WRITE_EMBEDDED: if (byteswap) { drr->drr_u.drr_write_embedded.drr_psize = BSWAP_32(drr->drr_u.drr_write_embedded. drr_psize); } (void) recv_read(hdl, fd, buf, P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize, 8), B_FALSE, NULL); break; case DRR_WRITE_BYREF: case DRR_FREEOBJECTS: case DRR_FREE: break; default: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid record type")); return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); } } free(buf); return (-1); } static void recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap, boolean_t resumable) { char target_fs[ZFS_MAX_DATASET_NAME_LEN]; zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "checksum mismatch or incomplete stream")); if (!resumable) return; (void) strlcpy(target_fs, target_snap, sizeof (target_fs)); *strchr(target_fs, '@') = '\0'; zfs_handle_t *zhp = zfs_open(hdl, target_fs, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); if (zhp == NULL) return; char token_buf[ZFS_MAXPROPLEN]; int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN, token_buf, sizeof (token_buf), NULL, NULL, 0, B_TRUE); if (error == 0) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "checksum mismatch or incomplete stream.\n" "Partially received snapshot is saved.\n" "A resuming stream can be generated on the sending " "system by running:\n" " zfs send -t %s"), token_buf); } zfs_close(zhp); } /* * Restores a backup of tosnap from the file descriptor specified by infd. */ static int zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap, const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr, dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, uint64_t *action_handlep, const char *finalsnap) { zfs_cmd_t zc = { 0 }; time_t begin_time; int ioctl_err, ioctl_errno, err; char *cp; struct drr_begin *drrb = &drr->drr_u.drr_begin; char errbuf[1024]; char prop_errbuf[1024]; const char *chopprefix; boolean_t newfs = B_FALSE; boolean_t stream_wantsnewfs; uint64_t parent_snapguid = 0; prop_changelist_t *clp = NULL; nvlist_t *snapprops_nvlist = NULL; zprop_errflags_t prop_errflags; boolean_t recursive; char *snapname = NULL; begin_time = time(NULL); (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot receive")); recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") == ENOENT); if (stream_avl != NULL) { nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid, &snapname); nvlist_t *props; int ret; (void) nvlist_lookup_uint64(fs, "parentfromsnap", &parent_snapguid); err = nvlist_lookup_nvlist(fs, "props", &props); if (err) VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0)); if (flags->canmountoff) { VERIFY(0 == nvlist_add_uint64(props, zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0)); } ret = zcmd_write_src_nvlist(hdl, &zc, props); if (err) nvlist_free(props); if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) { VERIFY(0 == nvlist_lookup_nvlist(props, snapname, &snapprops_nvlist)); } if (ret != 0) return (-1); } cp = NULL; /* * Determine how much of the snapshot name stored in the stream * we are going to tack on to the name they specified on the * command line, and how much we are going to chop off. * * If they specified a snapshot, chop the entire name stored in * the stream. */ if (flags->istail) { /* * A filesystem was specified with -e. We want to tack on only * the tail of the sent snapshot path. */ if (strchr(tosnap, '@')) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " "argument - snapshot not allowed with -e")); return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); } chopprefix = strrchr(sendfs, '/'); if (chopprefix == NULL) { /* * The tail is the poolname, so we need to * prepend a path separator. */ int len = strlen(drrb->drr_toname); cp = malloc(len + 2); cp[0] = '/'; (void) strcpy(&cp[1], drrb->drr_toname); chopprefix = cp; } else { chopprefix = drrb->drr_toname + (chopprefix - sendfs); } } else if (flags->isprefix) { /* * A filesystem was specified with -d. We want to tack on * everything but the first element of the sent snapshot path * (all but the pool name). */ if (strchr(tosnap, '@')) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " "argument - snapshot not allowed with -d")); return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); } chopprefix = strchr(drrb->drr_toname, '/'); if (chopprefix == NULL) chopprefix = strchr(drrb->drr_toname, '@'); } else if (strchr(tosnap, '@') == NULL) { /* * If a filesystem was specified without -d or -e, we want to * tack on everything after the fs specified by 'zfs send'. */ chopprefix = drrb->drr_toname + strlen(sendfs); } else { /* A snapshot was specified as an exact path (no -d or -e). */ if (recursive) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot specify snapshot name for multi-snapshot " "stream")); return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); } chopprefix = drrb->drr_toname + strlen(drrb->drr_toname); } ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname); ASSERT(chopprefix > drrb->drr_toname); ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname)); ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' || chopprefix[0] == '\0'); /* * Determine name of destination snapshot, store in zc_value. */ (void) strcpy(zc.zc_value, tosnap); (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value)); #ifdef __FreeBSD__ if (zfs_ioctl_version == ZFS_IOCVER_UNDEF) zfs_ioctl_version = get_zfs_ioctl_version(); /* * For forward compatibility hide tosnap in zc_value */ if (zfs_ioctl_version < ZFS_IOCVER_LZC) (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap); #endif free(cp); if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) { zcmd_free_nvlists(&zc); return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf)); } /* * Determine the name of the origin snapshot, store in zc_string. */ if (originsnap) { (void) strncpy(zc.zc_string, originsnap, sizeof (zc.zc_string)); if (flags->verbose) (void) printf("using provided clone origin %s\n", zc.zc_string); } else if (drrb->drr_flags & DRR_FLAG_CLONE) { if (guid_to_name(hdl, zc.zc_value, drrb->drr_fromguid, B_FALSE, zc.zc_string) != 0) { zcmd_free_nvlists(&zc); zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "local origin for clone %s does not exist"), zc.zc_value); return (zfs_error(hdl, EZFS_NOENT, errbuf)); } if (flags->verbose) (void) printf("found clone origin %s\n", zc.zc_string); } boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & DMU_BACKUP_FEATURE_RESUMING; stream_wantsnewfs = (drrb->drr_fromguid == 0 || (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming; if (stream_wantsnewfs) { /* * if the parent fs does not exist, look for it based on * the parent snap GUID */ (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot receive new filesystem stream")); (void) strcpy(zc.zc_name, zc.zc_value); cp = strrchr(zc.zc_name, '/'); if (cp) *cp = '\0'; if (cp && !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { char suffix[ZFS_MAX_DATASET_NAME_LEN]; (void) strcpy(suffix, strrchr(zc.zc_value, '/')); if (guid_to_name(hdl, zc.zc_name, parent_snapguid, B_FALSE, zc.zc_value) == 0) { *strchr(zc.zc_value, '@') = '\0'; (void) strcat(zc.zc_value, suffix); } } } else { /* - * if the fs does not exist, look for it based on the - * fromsnap GUID + * If the fs does not exist, look for it based on the + * fromsnap GUID. */ - (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, - "cannot receive incremental stream")); + if (resuming) { + (void) snprintf(errbuf, sizeof (errbuf), + dgettext(TEXT_DOMAIN, + "cannot receive resume stream")); + } else { + (void) snprintf(errbuf, sizeof (errbuf), + dgettext(TEXT_DOMAIN, + "cannot receive incremental stream")); + } (void) strcpy(zc.zc_name, zc.zc_value); *strchr(zc.zc_name, '@') = '\0'; /* * If the exact receive path was specified and this is the * topmost path in the stream, then if the fs does not exist we * should look no further. */ if ((flags->isprefix || (*(chopprefix = drrb->drr_toname + strlen(sendfs)) != '\0' && *chopprefix != '@')) && !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { char snap[ZFS_MAX_DATASET_NAME_LEN]; (void) strcpy(snap, strchr(zc.zc_value, '@')); if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid, B_FALSE, zc.zc_value) == 0) { *strchr(zc.zc_value, '@') = '\0'; (void) strcat(zc.zc_value, snap); } } } (void) strcpy(zc.zc_name, zc.zc_value); *strchr(zc.zc_name, '@') = '\0'; if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) { zfs_handle_t *zhp; /* * Destination fs exists. It must be one of these cases: * - an incremental send stream * - the stream specifies a new fs (full stream or clone) * and they want us to blow away the existing fs (and * have therefore specified -F and removed any snapshots) * - we are resuming a failed receive. */ if (stream_wantsnewfs) { if (!flags->force) { zcmd_free_nvlists(&zc); zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "destination '%s' exists\n" "must specify -F to overwrite it"), zc.zc_name); return (zfs_error(hdl, EZFS_EXISTS, errbuf)); } if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT, &zc) == 0) { zcmd_free_nvlists(&zc); zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "destination has snapshots (eg. %s)\n" "must destroy them to overwrite it"), zc.zc_name); return (zfs_error(hdl, EZFS_EXISTS, errbuf)); } } if ((zhp = zfs_open(hdl, zc.zc_name, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) { zcmd_free_nvlists(&zc); return (-1); } if (stream_wantsnewfs && zhp->zfs_dmustats.dds_origin[0]) { zcmd_free_nvlists(&zc); zfs_close(zhp); zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "destination '%s' is a clone\n" "must destroy it to overwrite it"), zc.zc_name); return (zfs_error(hdl, EZFS_EXISTS, errbuf)); } if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM && stream_wantsnewfs) { /* We can't do online recv in this case */ clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0); if (clp == NULL) { zfs_close(zhp); zcmd_free_nvlists(&zc); return (-1); } if (changelist_prefix(clp) != 0) { changelist_free(clp); zfs_close(zhp); zcmd_free_nvlists(&zc); return (-1); } } /* * If we are resuming a newfs, set newfs here so that we will * mount it if the recv succeeds this time. We can tell * that it was a newfs on the first recv because the fs * itself will be inconsistent (if the fs existed when we * did the first recv, we would have received it into * .../%recv). */ if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT)) newfs = B_TRUE; zfs_close(zhp); } else { /* * Destination filesystem does not exist. Therefore we better * be creating a new filesystem (either from a full backup, or * a clone). It would therefore be invalid if the user * specified only the pool name (i.e. if the destination name * contained no slash character). */ if (!stream_wantsnewfs || (cp = strrchr(zc.zc_name, '/')) == NULL) { zcmd_free_nvlists(&zc); zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "destination '%s' does not exist"), zc.zc_name); return (zfs_error(hdl, EZFS_NOENT, errbuf)); } /* * Trim off the final dataset component so we perform the * recvbackup ioctl to the filesystems's parent. */ *cp = '\0'; if (flags->isprefix && !flags->istail && !flags->dryrun && create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) { zcmd_free_nvlists(&zc); return (zfs_error(hdl, EZFS_BADRESTORE, errbuf)); } newfs = B_TRUE; } zc.zc_begin_record = *drr_noswap; zc.zc_cookie = infd; zc.zc_guid = flags->force; zc.zc_resumable = flags->resumable; if (flags->verbose) { (void) printf("%s %s stream of %s into %s\n", flags->dryrun ? "would receive" : "receiving", drrb->drr_fromguid ? "incremental" : "full", drrb->drr_toname, zc.zc_value); (void) fflush(stdout); } if (flags->dryrun) { zcmd_free_nvlists(&zc); return (recv_skip(hdl, infd, flags->byteswap)); } zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf; zc.zc_nvlist_dst_size = sizeof (prop_errbuf); zc.zc_cleanup_fd = cleanup_fd; zc.zc_action_handle = *action_handlep; err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc); ioctl_errno = errno; prop_errflags = (zprop_errflags_t)zc.zc_obj; if (err == 0) { nvlist_t *prop_errors; VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, zc.zc_nvlist_dst_size, &prop_errors, 0)); nvpair_t *prop_err = NULL; while ((prop_err = nvlist_next_nvpair(prop_errors, prop_err)) != NULL) { char tbuf[1024]; zfs_prop_t prop; int intval; prop = zfs_name_to_prop(nvpair_name(prop_err)); (void) nvpair_value_int32(prop_err, &intval); if (strcmp(nvpair_name(prop_err), ZPROP_N_MORE_ERRORS) == 0) { trunc_prop_errs(intval); break; } else if (snapname == NULL || finalsnap == NULL || strcmp(finalsnap, snapname) == 0 || strcmp(nvpair_name(prop_err), zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) { /* * Skip the special case of, for example, * "refquota", errors on intermediate * snapshots leading up to a final one. * That's why we have all of the checks above. * * See zfs_ioctl.c's extract_delay_props() for * a list of props which can fail on * intermediate snapshots, but shouldn't * affect the overall receive. */ (void) snprintf(tbuf, sizeof (tbuf), dgettext(TEXT_DOMAIN, "cannot receive %s property on %s"), nvpair_name(prop_err), zc.zc_name); zfs_setprop_error(hdl, prop, intval, tbuf); } } nvlist_free(prop_errors); } zc.zc_nvlist_dst = 0; zc.zc_nvlist_dst_size = 0; zcmd_free_nvlists(&zc); if (err == 0 && snapprops_nvlist) { zfs_cmd_t zc2 = { 0 }; (void) strcpy(zc2.zc_name, zc.zc_value); zc2.zc_cookie = B_TRUE; /* received */ if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) { (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2); zcmd_free_nvlists(&zc2); } } if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) { /* * It may be that this snapshot already exists, * in which case we want to consume & ignore it * rather than failing. */ avl_tree_t *local_avl; nvlist_t *local_nv, *fs; cp = strchr(zc.zc_value, '@'); /* * XXX Do this faster by just iterating over snaps in * this fs. Also if zc_value does not exist, we will * get a strange "does not exist" error message. */ *cp = '\0'; if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE, B_FALSE, &local_nv, &local_avl) == 0) { *cp = '@'; fs = fsavl_find(local_avl, drrb->drr_toguid, NULL); fsavl_destroy(local_avl); nvlist_free(local_nv); if (fs != NULL) { if (flags->verbose) { (void) printf("snap %s already exists; " "ignoring\n", zc.zc_value); } err = ioctl_err = recv_skip(hdl, infd, flags->byteswap); } } *cp = '@'; } if (ioctl_err != 0) { switch (ioctl_errno) { case ENODEV: cp = strchr(zc.zc_value, '@'); *cp = '\0'; zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "most recent snapshot of %s does not\n" "match incremental source"), zc.zc_value); (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); *cp = '@'; break; case ETXTBSY: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "destination %s has been modified\n" "since most recent snapshot"), zc.zc_name); (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf); break; case EEXIST: cp = strchr(zc.zc_value, '@'); if (newfs) { /* it's the containing fs that exists */ *cp = '\0'; } zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "destination already exists")); (void) zfs_error_fmt(hdl, EZFS_EXISTS, dgettext(TEXT_DOMAIN, "cannot restore to %s"), zc.zc_value); *cp = '@'; break; case EINVAL: (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); break; case ECKSUM: recv_ecksum_set_aux(hdl, zc.zc_value, flags->resumable); (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf); break; case ENOTSUP: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be upgraded to receive this stream.")); (void) zfs_error(hdl, EZFS_BADVERSION, errbuf); break; case EDQUOT: zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "destination %s space quota exceeded"), zc.zc_name); (void) zfs_error(hdl, EZFS_NOSPC, errbuf); break; default: (void) zfs_standard_error(hdl, ioctl_errno, errbuf); } } /* * Mount the target filesystem (if created). Also mount any * children of the target filesystem if we did a replication * receive (indicated by stream_avl being non-NULL). */ cp = strchr(zc.zc_value, '@'); if (cp && (ioctl_err == 0 || !newfs)) { zfs_handle_t *h; *cp = '\0'; h = zfs_open(hdl, zc.zc_value, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME); if (h != NULL) { if (h->zfs_type == ZFS_TYPE_VOLUME) { *cp = '@'; } else if (newfs || stream_avl) { /* * Track the first/top of hierarchy fs, * for mounting and sharing later. */ if (top_zfs && *top_zfs == NULL) *top_zfs = zfs_strdup(hdl, zc.zc_value); } zfs_close(h); } *cp = '@'; } if (clp) { if (!flags->nomount) err |= changelist_postfix(clp); changelist_free(clp); } if (prop_errflags & ZPROP_ERR_NOCLEAR) { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " "failed to clear unreceived properties on %s"), zc.zc_name); (void) fprintf(stderr, "\n"); } if (prop_errflags & ZPROP_ERR_NORESTORE) { (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: " "failed to restore original properties on %s"), zc.zc_name); (void) fprintf(stderr, "\n"); } if (err || ioctl_err) return (-1); *action_handlep = zc.zc_action_handle; if (flags->verbose) { char buf1[64]; char buf2[64]; uint64_t bytes = zc.zc_cookie; time_t delta = time(NULL) - begin_time; if (delta == 0) delta = 1; zfs_nicenum(bytes, buf1, sizeof (buf1)); zfs_nicenum(bytes/delta, buf2, sizeof (buf1)); (void) printf("received %sB stream in %lu seconds (%sB/sec)\n", buf1, delta, buf2); } return (0); } static int zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap, const char *originsnap, recvflags_t *flags, int infd, const char *sendfs, nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd, uint64_t *action_handlep, const char *finalsnap) { int err; dmu_replay_record_t drr, drr_noswap; struct drr_begin *drrb = &drr.drr_u.drr_begin; char errbuf[1024]; zio_cksum_t zcksum = { 0 }; uint64_t featureflags; int hdrtype; (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN, "cannot receive")); if (flags->isprefix && !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs " "(%s) does not exist"), tosnap); return (zfs_error(hdl, EZFS_NOENT, errbuf)); } if (originsnap && !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs " "(%s) does not exist"), originsnap); return (zfs_error(hdl, EZFS_NOENT, errbuf)); } /* read in the BEGIN record */ if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE, &zcksum))) return (err); if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) { /* It's the double end record at the end of a package */ return (ENODATA); } /* the kernel needs the non-byteswapped begin record */ drr_noswap = drr; flags->byteswap = B_FALSE; if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { /* * We computed the checksum in the wrong byteorder in * recv_read() above; do it again correctly. */ bzero(&zcksum, sizeof (zio_cksum_t)); (void) fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum); flags->byteswap = B_TRUE; drr.drr_type = BSWAP_32(drr.drr_type); drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen); drrb->drr_magic = BSWAP_64(drrb->drr_magic); drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo); drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time); drrb->drr_type = BSWAP_32(drrb->drr_type); drrb->drr_flags = BSWAP_32(drrb->drr_flags); drrb->drr_toguid = BSWAP_64(drrb->drr_toguid); drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid); } if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " "stream (bad magic number)")); return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); } featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo); if (!DMU_STREAM_SUPPORTED(featureflags) || (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "stream has unsupported feature, feature flags = %lx"), featureflags); return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); } if (strchr(drrb->drr_toname, '@') == NULL) { zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid " "stream (bad snapshot name)")); return (zfs_error(hdl, EZFS_BADSTREAM, errbuf)); } if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) { char nonpackage_sendfs[ZFS_MAX_DATASET_NAME_LEN]; if (sendfs == NULL) { /* * We were not called from zfs_receive_package(). Get * the fs specified by 'zfs send'. */ char *cp; (void) strlcpy(nonpackage_sendfs, drr.drr_u.drr_begin.drr_toname, sizeof (nonpackage_sendfs)); if ((cp = strchr(nonpackage_sendfs, '@')) != NULL) *cp = '\0'; sendfs = nonpackage_sendfs; VERIFY(finalsnap == NULL); } return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags, &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd, action_handlep, finalsnap)); } else { assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_COMPOUNDSTREAM); return (zfs_receive_package(hdl, infd, tosnap, flags, &drr, &zcksum, top_zfs, cleanup_fd, action_handlep)); } } /* * Restores a backup of tosnap from the file descriptor specified by infd. * Return 0 on total success, -2 if some things couldn't be * destroyed/renamed/promoted, -1 if some things couldn't be received. * (-1 will override -2, if -1 and the resumable flag was specified the * transfer can be resumed if the sending side supports it). */ int zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props, recvflags_t *flags, int infd, avl_tree_t *stream_avl) { char *top_zfs = NULL; int err; int cleanup_fd; uint64_t action_handle = 0; char *originsnap = NULL; if (props) { err = nvlist_lookup_string(props, "origin", &originsnap); if (err && err != ENOENT) return (err); } cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL); VERIFY(cleanup_fd >= 0); err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL, stream_avl, &top_zfs, cleanup_fd, &action_handle, NULL); VERIFY(0 == close(cleanup_fd)); if (err == 0 && !flags->nomount && top_zfs) { zfs_handle_t *zhp; prop_changelist_t *clp; zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM); if (zhp != NULL) { clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT, CL_GATHER_MOUNT_ALWAYS, 0); zfs_close(zhp); if (clp != NULL) { /* mount and share received datasets */ err = changelist_postfix(clp); changelist_free(clp); } } if (zhp == NULL || clp == NULL || err) err = -1; } if (top_zfs) free(top_zfs); return (err); } Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_send.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_send.c (revision 352375) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_send.c (revision 352376) @@ -1,3435 +1,3439 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2011, 2015 by Delphix. All rights reserved. * Copyright (c) 2014, Joyent, Inc. All rights reserved. * Copyright (c) 2012, Martin Matuska . All rights reserved. * Copyright 2014 HybridCluster. All rights reserved. * Copyright 2016 RackTop Systems. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef __FreeBSD__ #undef dump_write #define dump_write dmu_dump_write #endif /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */ int zfs_send_corrupt_data = B_FALSE; int zfs_send_queue_length = 16 * 1024 * 1024; int zfs_recv_queue_length = 16 * 1024 * 1024; /* Set this tunable to FALSE to disable setting of DRR_FLAG_FREERECORDS */ int zfs_send_set_freerecords_bit = B_TRUE; #ifdef _KERNEL TUNABLE_INT("vfs.zfs.send_set_freerecords_bit", &zfs_send_set_freerecords_bit); #endif static char *dmu_recv_tag = "dmu_recv_tag"; const char *recv_clone_name = "%recv"; /* * Use this to override the recordsize calculation for fast zfs send estimates. */ uint64_t zfs_override_estimate_recordsize = 0; #define BP_SPAN(datablkszsec, indblkshift, level) \ (((uint64_t)datablkszsec) << (SPA_MINBLOCKSHIFT + \ (level) * (indblkshift - SPA_BLKPTRSHIFT))) static void byteswap_record(dmu_replay_record_t *drr); struct send_thread_arg { bqueue_t q; dsl_dataset_t *ds; /* Dataset to traverse */ uint64_t fromtxg; /* Traverse from this txg */ int flags; /* flags to pass to traverse_dataset */ int error_code; boolean_t cancel; zbookmark_phys_t resume; }; struct send_block_record { boolean_t eos_marker; /* Marks the end of the stream */ blkptr_t bp; zbookmark_phys_t zb; uint8_t indblkshift; uint16_t datablkszsec; bqueue_node_t ln; }; static int dump_bytes(dmu_sendarg_t *dsp, void *buf, int len) { dsl_dataset_t *ds = dmu_objset_ds(dsp->dsa_os); struct uio auio; struct iovec aiov; /* * The code does not rely on this (len being a multiple of 8). We keep * this assertion because of the corresponding assertion in * receive_read(). Keeping this assertion ensures that we do not * inadvertently break backwards compatibility (causing the assertion * in receive_read() to trigger on old software). * * Removing the assertions could be rolled into a new feature that uses * data that isn't 8-byte aligned; if the assertions were removed, a * feature flag would have to be added. */ ASSERT0(len % 8); aiov.iov_base = buf; aiov.iov_len = len; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_resid = len; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_WRITE; auio.uio_offset = (off_t)-1; auio.uio_td = dsp->dsa_td; #ifdef _KERNEL if (dsp->dsa_fp->f_type == DTYPE_VNODE) bwillwrite(); dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0, dsp->dsa_td); #else fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__); dsp->dsa_err = EOPNOTSUPP; #endif mutex_enter(&ds->ds_sendstream_lock); *dsp->dsa_off += len; mutex_exit(&ds->ds_sendstream_lock); return (dsp->dsa_err); } /* * For all record types except BEGIN, fill in the checksum (overlaid in * drr_u.drr_checksum.drr_checksum). The checksum verifies everything * up to the start of the checksum itself. */ static int dump_record(dmu_sendarg_t *dsp, void *payload, int payload_len) { ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); (void) fletcher_4_incremental_native(dsp->dsa_drr, offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), &dsp->dsa_zc); if (dsp->dsa_drr->drr_type == DRR_BEGIN) { dsp->dsa_sent_begin = B_TRUE; } else { ASSERT(ZIO_CHECKSUM_IS_ZERO(&dsp->dsa_drr->drr_u. drr_checksum.drr_checksum)); dsp->dsa_drr->drr_u.drr_checksum.drr_checksum = dsp->dsa_zc; } if (dsp->dsa_drr->drr_type == DRR_END) { dsp->dsa_sent_end = B_TRUE; } (void) fletcher_4_incremental_native(&dsp->dsa_drr-> drr_u.drr_checksum.drr_checksum, sizeof (zio_cksum_t), &dsp->dsa_zc); if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0) return (SET_ERROR(EINTR)); if (payload_len != 0) { (void) fletcher_4_incremental_native(payload, payload_len, &dsp->dsa_zc); if (dump_bytes(dsp, payload, payload_len) != 0) return (SET_ERROR(EINTR)); } return (0); } /* * Fill in the drr_free struct, or perform aggregation if the previous record is * also a free record, and the two are adjacent. * * Note that we send free records even for a full send, because we want to be * able to receive a full send as a clone, which requires a list of all the free * and freeobject records that were generated on the source. */ static int dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, uint64_t length) { struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free); /* * When we receive a free record, dbuf_free_range() assumes * that the receiving system doesn't have any dbufs in the range * being freed. This is always true because there is a one-record * constraint: we only send one WRITE record for any given * object,offset. We know that the one-record constraint is * true because we always send data in increasing order by * object,offset. * * If the increasing-order constraint ever changes, we should find * another way to assert that the one-record constraint is still * satisfied. */ ASSERT(object > dsp->dsa_last_data_object || (object == dsp->dsa_last_data_object && offset > dsp->dsa_last_data_offset)); if (length != -1ULL && offset + length < offset) length = -1ULL; /* * If there is a pending op, but it's not PENDING_FREE, push it out, * since free block aggregation can only be done for blocks of the * same type (i.e., DRR_FREE records can only be aggregated with * other DRR_FREE records. DRR_FREEOBJECTS records can only be * aggregated with other DRR_FREEOBJECTS records. */ if (dsp->dsa_pending_op != PENDING_NONE && dsp->dsa_pending_op != PENDING_FREE) { if (dump_record(dsp, NULL, 0) != 0) return (SET_ERROR(EINTR)); dsp->dsa_pending_op = PENDING_NONE; } if (dsp->dsa_pending_op == PENDING_FREE) { /* * There should never be a PENDING_FREE if length is -1 * (because dump_dnode is the only place where this * function is called with a -1, and only after flushing * any pending record). */ ASSERT(length != -1ULL); /* * Check to see whether this free block can be aggregated * with pending one. */ if (drrf->drr_object == object && drrf->drr_offset + drrf->drr_length == offset) { drrf->drr_length += length; return (0); } else { /* not a continuation. Push out pending record */ if (dump_record(dsp, NULL, 0) != 0) return (SET_ERROR(EINTR)); dsp->dsa_pending_op = PENDING_NONE; } } /* create a FREE record and make it pending */ bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); dsp->dsa_drr->drr_type = DRR_FREE; drrf->drr_object = object; drrf->drr_offset = offset; drrf->drr_length = length; drrf->drr_toguid = dsp->dsa_toguid; if (length == -1ULL) { if (dump_record(dsp, NULL, 0) != 0) return (SET_ERROR(EINTR)); } else { dsp->dsa_pending_op = PENDING_FREE; } return (0); } static int dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type, uint64_t object, uint64_t offset, int lsize, int psize, const blkptr_t *bp, void *data) { uint64_t payload_size; struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write); /* * We send data in increasing object, offset order. * See comment in dump_free() for details. */ ASSERT(object > dsp->dsa_last_data_object || (object == dsp->dsa_last_data_object && offset > dsp->dsa_last_data_offset)); dsp->dsa_last_data_object = object; dsp->dsa_last_data_offset = offset + lsize - 1; /* * If there is any kind of pending aggregation (currently either * a grouping of free objects or free blocks), push it out to * the stream, since aggregation can't be done across operations * of different types. */ if (dsp->dsa_pending_op != PENDING_NONE) { if (dump_record(dsp, NULL, 0) != 0) return (SET_ERROR(EINTR)); dsp->dsa_pending_op = PENDING_NONE; } /* write a WRITE record */ bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); dsp->dsa_drr->drr_type = DRR_WRITE; drrw->drr_object = object; drrw->drr_type = type; drrw->drr_offset = offset; drrw->drr_toguid = dsp->dsa_toguid; drrw->drr_logical_size = lsize; /* only set the compression fields if the buf is compressed */ if (lsize != psize) { ASSERT(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED); ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(!BP_SHOULD_BYTESWAP(bp)); ASSERT(!DMU_OT_IS_METADATA(BP_GET_TYPE(bp))); ASSERT3U(BP_GET_COMPRESS(bp), !=, ZIO_COMPRESS_OFF); ASSERT3S(psize, >, 0); ASSERT3S(lsize, >=, psize); drrw->drr_compressiontype = BP_GET_COMPRESS(bp); drrw->drr_compressed_size = psize; payload_size = drrw->drr_compressed_size; } else { payload_size = drrw->drr_logical_size; } if (bp == NULL || BP_IS_EMBEDDED(bp)) { /* * There's no pre-computed checksum for partial-block * writes or embedded BP's, so (like * fletcher4-checkummed blocks) userland will have to * compute a dedup-capable checksum itself. */ drrw->drr_checksumtype = ZIO_CHECKSUM_OFF; } else { drrw->drr_checksumtype = BP_GET_CHECKSUM(bp); if (zio_checksum_table[drrw->drr_checksumtype].ci_flags & ZCHECKSUM_FLAG_DEDUP) drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP; DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp)); DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp)); DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp)); drrw->drr_key.ddk_cksum = bp->blk_cksum; } if (dump_record(dsp, data, payload_size) != 0) return (SET_ERROR(EINTR)); return (0); } static int dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp) { char buf[BPE_PAYLOAD_SIZE]; struct drr_write_embedded *drrw = &(dsp->dsa_drr->drr_u.drr_write_embedded); if (dsp->dsa_pending_op != PENDING_NONE) { if (dump_record(dsp, NULL, 0) != 0) return (EINTR); dsp->dsa_pending_op = PENDING_NONE; } ASSERT(BP_IS_EMBEDDED(bp)); bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED; drrw->drr_object = object; drrw->drr_offset = offset; drrw->drr_length = blksz; drrw->drr_toguid = dsp->dsa_toguid; drrw->drr_compression = BP_GET_COMPRESS(bp); drrw->drr_etype = BPE_GET_ETYPE(bp); drrw->drr_lsize = BPE_GET_LSIZE(bp); drrw->drr_psize = BPE_GET_PSIZE(bp); decode_embedded_bp_compressed(bp, buf); if (dump_record(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0) return (EINTR); return (0); } static int dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data) { struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill); if (dsp->dsa_pending_op != PENDING_NONE) { if (dump_record(dsp, NULL, 0) != 0) return (SET_ERROR(EINTR)); dsp->dsa_pending_op = PENDING_NONE; } /* write a SPILL record */ bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); dsp->dsa_drr->drr_type = DRR_SPILL; drrs->drr_object = object; drrs->drr_length = blksz; drrs->drr_toguid = dsp->dsa_toguid; if (dump_record(dsp, data, blksz) != 0) return (SET_ERROR(EINTR)); return (0); } static int dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs) { struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects); /* * If there is a pending op, but it's not PENDING_FREEOBJECTS, * push it out, since free block aggregation can only be done for * blocks of the same type (i.e., DRR_FREE records can only be * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records * can only be aggregated with other DRR_FREEOBJECTS records. */ if (dsp->dsa_pending_op != PENDING_NONE && dsp->dsa_pending_op != PENDING_FREEOBJECTS) { if (dump_record(dsp, NULL, 0) != 0) return (SET_ERROR(EINTR)); dsp->dsa_pending_op = PENDING_NONE; } if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) { /* * See whether this free object array can be aggregated * with pending one */ if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) { drrfo->drr_numobjs += numobjs; return (0); } else { /* can't be aggregated. Push out pending record */ if (dump_record(dsp, NULL, 0) != 0) return (SET_ERROR(EINTR)); dsp->dsa_pending_op = PENDING_NONE; } } /* write a FREEOBJECTS record */ bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); dsp->dsa_drr->drr_type = DRR_FREEOBJECTS; drrfo->drr_firstobj = firstobj; drrfo->drr_numobjs = numobjs; drrfo->drr_toguid = dsp->dsa_toguid; dsp->dsa_pending_op = PENDING_FREEOBJECTS; return (0); } static int dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp) { struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object); if (object < dsp->dsa_resume_object) { /* * Note: when resuming, we will visit all the dnodes in * the block of dnodes that we are resuming from. In * this case it's unnecessary to send the dnodes prior to * the one we are resuming from. We should be at most one * block's worth of dnodes behind the resume point. */ ASSERT3U(dsp->dsa_resume_object - object, <, 1 << (DNODE_BLOCK_SHIFT - DNODE_SHIFT)); return (0); } if (dnp == NULL || dnp->dn_type == DMU_OT_NONE) return (dump_freeobjects(dsp, object, 1)); if (dsp->dsa_pending_op != PENDING_NONE) { if (dump_record(dsp, NULL, 0) != 0) return (SET_ERROR(EINTR)); dsp->dsa_pending_op = PENDING_NONE; } /* write an OBJECT record */ bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t)); dsp->dsa_drr->drr_type = DRR_OBJECT; drro->drr_object = object; drro->drr_type = dnp->dn_type; drro->drr_bonustype = dnp->dn_bonustype; drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; drro->drr_bonuslen = dnp->dn_bonuslen; drro->drr_checksumtype = dnp->dn_checksum; drro->drr_compress = dnp->dn_compress; drro->drr_toguid = dsp->dsa_toguid; if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE) drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE; if (dump_record(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0) { return (SET_ERROR(EINTR)); } /* Free anything past the end of the file. */ if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) * (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0) return (SET_ERROR(EINTR)); if (dsp->dsa_err != 0) return (SET_ERROR(EINTR)); return (0); } static boolean_t backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp) { if (!BP_IS_EMBEDDED(bp)) return (B_FALSE); /* * Compression function must be legacy, or explicitly enabled. */ if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS && !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LZ4))) return (B_FALSE); /* * Embed type must be explicitly enabled. */ switch (BPE_GET_ETYPE(bp)) { case BP_EMBEDDED_TYPE_DATA: if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) return (B_TRUE); break; default: return (B_FALSE); } return (B_FALSE); } /* * This is the callback function to traverse_dataset that acts as the worker * thread for dmu_send_impl. */ /*ARGSUSED*/ static int send_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, const zbookmark_phys_t *zb, const struct dnode_phys *dnp, void *arg) { struct send_thread_arg *sta = arg; struct send_block_record *record; uint64_t record_size; int err = 0; ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || zb->zb_object >= sta->resume.zb_object); if (sta->cancel) return (SET_ERROR(EINTR)); if (bp == NULL) { ASSERT3U(zb->zb_level, ==, ZB_DNODE_LEVEL); return (0); } else if (zb->zb_level < 0) { return (0); } record = kmem_zalloc(sizeof (struct send_block_record), KM_SLEEP); record->eos_marker = B_FALSE; record->bp = *bp; record->zb = *zb; record->indblkshift = dnp->dn_indblkshift; record->datablkszsec = dnp->dn_datablkszsec; record_size = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT; bqueue_enqueue(&sta->q, record, record_size); return (err); } /* * This function kicks off the traverse_dataset. It also handles setting the * error code of the thread in case something goes wrong, and pushes the End of * Stream record when the traverse_dataset call has finished. If there is no * dataset to traverse, the thread immediately pushes End of Stream marker. */ static void send_traverse_thread(void *arg) { struct send_thread_arg *st_arg = arg; int err; struct send_block_record *data; if (st_arg->ds != NULL) { err = traverse_dataset_resume(st_arg->ds, st_arg->fromtxg, &st_arg->resume, st_arg->flags, send_cb, st_arg); if (err != EINTR) st_arg->error_code = err; } data = kmem_zalloc(sizeof (*data), KM_SLEEP); data->eos_marker = B_TRUE; bqueue_enqueue(&st_arg->q, data, 1); thread_exit(); } /* * This function actually handles figuring out what kind of record needs to be * dumped, reading the data (which has hopefully been prefetched), and calling * the appropriate helper function. */ static int do_dump(dmu_sendarg_t *dsa, struct send_block_record *data) { dsl_dataset_t *ds = dmu_objset_ds(dsa->dsa_os); const blkptr_t *bp = &data->bp; const zbookmark_phys_t *zb = &data->zb; uint8_t indblkshift = data->indblkshift; uint16_t dblkszsec = data->datablkszsec; spa_t *spa = ds->ds_dir->dd_pool->dp_spa; dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE; int err = 0; ASSERT3U(zb->zb_level, >=, 0); ASSERT(zb->zb_object == DMU_META_DNODE_OBJECT || zb->zb_object >= dsa->dsa_resume_object); if (zb->zb_object != DMU_META_DNODE_OBJECT && DMU_OBJECT_IS_SPECIAL(zb->zb_object)) { return (0); } else if (BP_IS_HOLE(bp) && zb->zb_object == DMU_META_DNODE_OBJECT) { uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT; err = dump_freeobjects(dsa, dnobj, span >> DNODE_SHIFT); } else if (BP_IS_HOLE(bp)) { uint64_t span = BP_SPAN(dblkszsec, indblkshift, zb->zb_level); uint64_t offset = zb->zb_blkid * span; err = dump_free(dsa, zb->zb_object, offset, span); } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) { return (0); } else if (type == DMU_OT_DNODE) { int blksz = BP_GET_LSIZE(bp); arc_flags_t aflags = ARC_FLAG_WAIT; arc_buf_t *abuf; ASSERT0(zb->zb_level); if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &aflags, zb) != 0) return (SET_ERROR(EIO)); dnode_phys_t *blk = abuf->b_data; uint64_t dnobj = zb->zb_blkid * (blksz >> DNODE_SHIFT); for (int i = 0; i < blksz >> DNODE_SHIFT; i++) { err = dump_dnode(dsa, dnobj + i, blk + i); if (err != 0) break; } arc_buf_destroy(abuf, &abuf); } else if (type == DMU_OT_SA) { arc_flags_t aflags = ARC_FLAG_WAIT; arc_buf_t *abuf; int blksz = BP_GET_LSIZE(bp); if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &aflags, zb) != 0) return (SET_ERROR(EIO)); err = dump_spill(dsa, zb->zb_object, blksz, abuf->b_data); arc_buf_destroy(abuf, &abuf); } else if (backup_do_embed(dsa, bp)) { /* it's an embedded level-0 block of a regular object */ int blksz = dblkszsec << SPA_MINBLOCKSHIFT; ASSERT0(zb->zb_level); err = dump_write_embedded(dsa, zb->zb_object, zb->zb_blkid * blksz, blksz, bp); } else { /* it's a level-0 block of a regular object */ arc_flags_t aflags = ARC_FLAG_WAIT; arc_buf_t *abuf; int blksz = dblkszsec << SPA_MINBLOCKSHIFT; uint64_t offset; /* * If we have large blocks stored on disk but the send flags * don't allow us to send large blocks, we split the data from * the arc buf into chunks. */ boolean_t split_large_blocks = blksz > SPA_OLD_MAXBLOCKSIZE && !(dsa->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS); /* * We should only request compressed data from the ARC if all * the following are true: * - stream compression was requested * - we aren't splitting large blocks into smaller chunks * - the data won't need to be byteswapped before sending * - this isn't an embedded block * - this isn't metadata (if receiving on a different endian * system it can be byteswapped more easily) */ boolean_t request_compressed = (dsa->dsa_featureflags & DMU_BACKUP_FEATURE_COMPRESSED) && !split_large_blocks && !BP_SHOULD_BYTESWAP(bp) && !BP_IS_EMBEDDED(bp) && !DMU_OT_IS_METADATA(BP_GET_TYPE(bp)); ASSERT0(zb->zb_level); ASSERT(zb->zb_object > dsa->dsa_resume_object || (zb->zb_object == dsa->dsa_resume_object && zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); ASSERT0(zb->zb_level); ASSERT(zb->zb_object > dsa->dsa_resume_object || (zb->zb_object == dsa->dsa_resume_object && zb->zb_blkid * blksz >= dsa->dsa_resume_offset)); ASSERT3U(blksz, ==, BP_GET_LSIZE(bp)); enum zio_flag zioflags = ZIO_FLAG_CANFAIL; if (request_compressed) zioflags |= ZIO_FLAG_RAW; if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ, zioflags, &aflags, zb) != 0) { if (zfs_send_corrupt_data) { /* Send a block filled with 0x"zfs badd bloc" */ abuf = arc_alloc_buf(spa, &abuf, ARC_BUFC_DATA, blksz); uint64_t *ptr; for (ptr = abuf->b_data; (char *)ptr < (char *)abuf->b_data + blksz; ptr++) *ptr = 0x2f5baddb10cULL; } else { return (SET_ERROR(EIO)); } } offset = zb->zb_blkid * blksz; if (split_large_blocks) { ASSERT3U(arc_get_compression(abuf), ==, ZIO_COMPRESS_OFF); char *buf = abuf->b_data; while (blksz > 0 && err == 0) { int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE); err = dump_write(dsa, type, zb->zb_object, offset, n, n, NULL, buf); offset += n; buf += n; blksz -= n; } } else { err = dump_write(dsa, type, zb->zb_object, offset, blksz, arc_buf_size(abuf), bp, abuf->b_data); } arc_buf_destroy(abuf, &abuf); } ASSERT(err == 0 || err == EINTR); return (err); } /* * Pop the new data off the queue, and free the old data. */ static struct send_block_record * get_next_record(bqueue_t *bq, struct send_block_record *data) { struct send_block_record *tmp = bqueue_dequeue(bq); kmem_free(data, sizeof (*data)); return (tmp); } /* * Actually do the bulk of the work in a zfs send. * * Note: Releases dp using the specified tag. */ static int dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *to_ds, zfs_bookmark_phys_t *ancestor_zb, boolean_t is_clone, boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, int outfd, uint64_t resumeobj, uint64_t resumeoff, #ifdef illumos vnode_t *vp, offset_t *off) #else struct file *fp, offset_t *off) #endif { objset_t *os; dmu_replay_record_t *drr; dmu_sendarg_t *dsp; int err; uint64_t fromtxg = 0; uint64_t featureflags = 0; struct send_thread_arg to_arg = { 0 }; err = dmu_objset_from_ds(to_ds, &os); if (err != 0) { dsl_pool_rele(dp, tag); return (err); } drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP); drr->drr_type = DRR_BEGIN; drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC; DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo, DMU_SUBSTREAM); #ifdef _KERNEL if (dmu_objset_type(os) == DMU_OST_ZFS) { uint64_t version; if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) { kmem_free(drr, sizeof (dmu_replay_record_t)); dsl_pool_rele(dp, tag); return (SET_ERROR(EINVAL)); } if (version >= ZPL_VERSION_SA) { featureflags |= DMU_BACKUP_FEATURE_SA_SPILL; } } #endif if (large_block_ok && to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS]) featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS; if (embedok && spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) { featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA; if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) featureflags |= DMU_BACKUP_FEATURE_LZ4; } if (compressok) { featureflags |= DMU_BACKUP_FEATURE_COMPRESSED; } if ((featureflags & (DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_COMPRESSED)) != 0 && spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) { featureflags |= DMU_BACKUP_FEATURE_LZ4; } if (resumeobj != 0 || resumeoff != 0) { featureflags |= DMU_BACKUP_FEATURE_RESUMING; } DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo, featureflags); drr->drr_u.drr_begin.drr_creation_time = dsl_dataset_phys(to_ds)->ds_creation_time; drr->drr_u.drr_begin.drr_type = dmu_objset_type(os); if (is_clone) drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE; drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(to_ds)->ds_guid; if (dsl_dataset_phys(to_ds)->ds_flags & DS_FLAG_CI_DATASET) drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA; if (zfs_send_set_freerecords_bit) drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_FREERECORDS; if (ancestor_zb != NULL) { drr->drr_u.drr_begin.drr_fromguid = ancestor_zb->zbm_guid; fromtxg = ancestor_zb->zbm_creation_txg; } dsl_dataset_name(to_ds, drr->drr_u.drr_begin.drr_toname); if (!to_ds->ds_is_snapshot) { (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--", sizeof (drr->drr_u.drr_begin.drr_toname)); } dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP); dsp->dsa_drr = drr; dsp->dsa_outfd = outfd; dsp->dsa_proc = curproc; dsp->dsa_td = curthread; dsp->dsa_fp = fp; dsp->dsa_os = os; dsp->dsa_off = off; dsp->dsa_toguid = dsl_dataset_phys(to_ds)->ds_guid; dsp->dsa_pending_op = PENDING_NONE; dsp->dsa_featureflags = featureflags; dsp->dsa_resume_object = resumeobj; dsp->dsa_resume_offset = resumeoff; mutex_enter(&to_ds->ds_sendstream_lock); list_insert_head(&to_ds->ds_sendstreams, dsp); mutex_exit(&to_ds->ds_sendstream_lock); dsl_dataset_long_hold(to_ds, FTAG); dsl_pool_rele(dp, tag); void *payload = NULL; size_t payload_len = 0; if (resumeobj != 0 || resumeoff != 0) { dmu_object_info_t to_doi; err = dmu_object_info(os, resumeobj, &to_doi); if (err != 0) goto out; SET_BOOKMARK(&to_arg.resume, to_ds->ds_object, resumeobj, 0, resumeoff / to_doi.doi_data_block_size); nvlist_t *nvl = fnvlist_alloc(); fnvlist_add_uint64(nvl, "resume_object", resumeobj); fnvlist_add_uint64(nvl, "resume_offset", resumeoff); payload = fnvlist_pack(nvl, &payload_len); drr->drr_payloadlen = payload_len; fnvlist_free(nvl); } err = dump_record(dsp, payload, payload_len); fnvlist_pack_free(payload, payload_len); if (err != 0) { err = dsp->dsa_err; goto out; } err = bqueue_init(&to_arg.q, zfs_send_queue_length, offsetof(struct send_block_record, ln)); to_arg.error_code = 0; to_arg.cancel = B_FALSE; to_arg.ds = to_ds; to_arg.fromtxg = fromtxg; to_arg.flags = TRAVERSE_PRE | TRAVERSE_PREFETCH; (void) thread_create(NULL, 0, send_traverse_thread, &to_arg, 0, &p0, TS_RUN, minclsyspri); struct send_block_record *to_data; to_data = bqueue_dequeue(&to_arg.q); while (!to_data->eos_marker && err == 0) { err = do_dump(dsp, to_data); to_data = get_next_record(&to_arg.q, to_data); if (issig(JUSTLOOKING) && issig(FORREAL)) err = EINTR; } if (err != 0) { to_arg.cancel = B_TRUE; while (!to_data->eos_marker) { to_data = get_next_record(&to_arg.q, to_data); } } kmem_free(to_data, sizeof (*to_data)); bqueue_destroy(&to_arg.q); if (err == 0 && to_arg.error_code != 0) err = to_arg.error_code; if (err != 0) goto out; if (dsp->dsa_pending_op != PENDING_NONE) if (dump_record(dsp, NULL, 0) != 0) err = SET_ERROR(EINTR); if (err != 0) { if (err == EINTR && dsp->dsa_err != 0) err = dsp->dsa_err; goto out; } bzero(drr, sizeof (dmu_replay_record_t)); drr->drr_type = DRR_END; drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc; drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid; if (dump_record(dsp, NULL, 0) != 0) err = dsp->dsa_err; out: mutex_enter(&to_ds->ds_sendstream_lock); list_remove(&to_ds->ds_sendstreams, dsp); mutex_exit(&to_ds->ds_sendstream_lock); VERIFY(err != 0 || (dsp->dsa_sent_begin && dsp->dsa_sent_end)); kmem_free(drr, sizeof (dmu_replay_record_t)); kmem_free(dsp, sizeof (dmu_sendarg_t)); dsl_dataset_long_rele(to_ds, FTAG); return (err); } int dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap, boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, #ifdef illumos int outfd, vnode_t *vp, offset_t *off) #else int outfd, struct file *fp, offset_t *off) #endif { dsl_pool_t *dp; dsl_dataset_t *ds; dsl_dataset_t *fromds = NULL; int err; err = dsl_pool_hold(pool, FTAG, &dp); if (err != 0) return (err); err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds); if (err != 0) { dsl_pool_rele(dp, FTAG); return (err); } if (fromsnap != 0) { zfs_bookmark_phys_t zb; boolean_t is_clone; err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds); if (err != 0) { dsl_dataset_rele(ds, FTAG); dsl_pool_rele(dp, FTAG); return (err); } if (!dsl_dataset_is_before(ds, fromds, 0)) err = SET_ERROR(EXDEV); zb.zbm_creation_time = dsl_dataset_phys(fromds)->ds_creation_time; zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; is_clone = (fromds->ds_dir != ds->ds_dir); dsl_dataset_rele(fromds, FTAG); err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok, large_block_ok, compressok, outfd, 0, 0, fp, off); } else { err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok, large_block_ok, compressok, outfd, 0, 0, fp, off); } dsl_dataset_rele(ds, FTAG); return (err); } int dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok, boolean_t large_block_ok, boolean_t compressok, int outfd, uint64_t resumeobj, uint64_t resumeoff, #ifdef illumos vnode_t *vp, offset_t *off) #else struct file *fp, offset_t *off) #endif { dsl_pool_t *dp; dsl_dataset_t *ds; int err; boolean_t owned = B_FALSE; if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL) return (SET_ERROR(EINVAL)); err = dsl_pool_hold(tosnap, FTAG, &dp); if (err != 0) return (err); if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) { /* * We are sending a filesystem or volume. Ensure * that it doesn't change by owning the dataset. */ err = dsl_dataset_own(dp, tosnap, FTAG, &ds); owned = B_TRUE; } else { err = dsl_dataset_hold(dp, tosnap, FTAG, &ds); } if (err != 0) { dsl_pool_rele(dp, FTAG); return (err); } if (fromsnap != NULL) { zfs_bookmark_phys_t zb; boolean_t is_clone = B_FALSE; int fsnamelen = strchr(tosnap, '@') - tosnap; /* * If the fromsnap is in a different filesystem, then * mark the send stream as a clone. */ if (strncmp(tosnap, fromsnap, fsnamelen) != 0 || (fromsnap[fsnamelen] != '@' && fromsnap[fsnamelen] != '#')) { is_clone = B_TRUE; } if (strchr(fromsnap, '@')) { dsl_dataset_t *fromds; err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds); if (err == 0) { if (!dsl_dataset_is_before(ds, fromds, 0)) err = SET_ERROR(EXDEV); zb.zbm_creation_time = dsl_dataset_phys(fromds)->ds_creation_time; zb.zbm_creation_txg = dsl_dataset_phys(fromds)->ds_creation_txg; zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid; is_clone = (ds->ds_dir != fromds->ds_dir); dsl_dataset_rele(fromds, FTAG); } } else { err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb); } if (err != 0) { dsl_dataset_rele(ds, FTAG); dsl_pool_rele(dp, FTAG); return (err); } err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok, large_block_ok, compressok, outfd, resumeobj, resumeoff, fp, off); } else { err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok, large_block_ok, compressok, outfd, resumeobj, resumeoff, fp, off); } if (owned) dsl_dataset_disown(ds, FTAG); else dsl_dataset_rele(ds, FTAG); return (err); } static int dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t uncompressed, uint64_t compressed, boolean_t stream_compressed, uint64_t *sizep) { int err = 0; uint64_t size; /* * Assume that space (both on-disk and in-stream) is dominated by * data. We will adjust for indirect blocks and the copies property, * but ignore per-object space used (eg, dnodes and DRR_OBJECT records). */ uint64_t recordsize; uint64_t record_count; objset_t *os; VERIFY0(dmu_objset_from_ds(ds, &os)); /* Assume all (uncompressed) blocks are recordsize. */ if (zfs_override_estimate_recordsize != 0) { recordsize = zfs_override_estimate_recordsize; } else if (os->os_phys->os_type == DMU_OST_ZVOL) { err = dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &recordsize); } else { err = dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), &recordsize); } if (err != 0) return (err); record_count = uncompressed / recordsize; /* * If we're estimating a send size for a compressed stream, use the * compressed data size to estimate the stream size. Otherwise, use the * uncompressed data size. */ size = stream_compressed ? compressed : uncompressed; /* * Subtract out approximate space used by indirect blocks. * Assume most space is used by data blocks (non-indirect, non-dnode). * Assume no ditto blocks or internal fragmentation. * * Therefore, space used by indirect blocks is sizeof(blkptr_t) per * block. */ size -= record_count * sizeof (blkptr_t); /* Add in the space for the record associated with each block. */ size += record_count * sizeof (dmu_replay_record_t); *sizep = size; return (0); } int dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, boolean_t stream_compressed, uint64_t *sizep) { dsl_pool_t *dp = ds->ds_dir->dd_pool; int err; uint64_t uncomp, comp; ASSERT(dsl_pool_config_held(dp)); /* tosnap must be a snapshot */ if (!ds->ds_is_snapshot) return (SET_ERROR(EINVAL)); /* fromsnap, if provided, must be a snapshot */ if (fromds != NULL && !fromds->ds_is_snapshot) return (SET_ERROR(EINVAL)); /* * fromsnap must be an earlier snapshot from the same fs as tosnap, * or the origin's fs. */ if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0)) return (SET_ERROR(EXDEV)); /* Get compressed and uncompressed size estimates of changed data. */ if (fromds == NULL) { uncomp = dsl_dataset_phys(ds)->ds_uncompressed_bytes; comp = dsl_dataset_phys(ds)->ds_compressed_bytes; } else { uint64_t used; err = dsl_dataset_space_written(fromds, ds, &used, &comp, &uncomp); if (err != 0) return (err); } err = dmu_adjust_send_estimate_for_indirects(ds, uncomp, comp, stream_compressed, sizep); /* * Add the size of the BEGIN and END records to the estimate. */ *sizep += 2 * sizeof (dmu_replay_record_t); return (err); } struct calculate_send_arg { uint64_t uncompressed; uint64_t compressed; }; /* * Simple callback used to traverse the blocks of a snapshot and sum their * uncompressed and compressed sizes. */ /* ARGSUSED */ static int dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) { struct calculate_send_arg *space = arg; if (bp != NULL && !BP_IS_HOLE(bp)) { space->uncompressed += BP_GET_UCSIZE(bp); space->compressed += BP_GET_PSIZE(bp); } return (0); } /* * Given a desination snapshot and a TXG, calculate the approximate size of a * send stream sent from that TXG. from_txg may be zero, indicating that the * whole snapshot will be sent. */ int dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg, boolean_t stream_compressed, uint64_t *sizep) { dsl_pool_t *dp = ds->ds_dir->dd_pool; int err; struct calculate_send_arg size = { 0 }; ASSERT(dsl_pool_config_held(dp)); /* tosnap must be a snapshot */ if (!ds->ds_is_snapshot) return (SET_ERROR(EINVAL)); /* verify that from_txg is before the provided snapshot was taken */ if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) { return (SET_ERROR(EXDEV)); } /* * traverse the blocks of the snapshot with birth times after * from_txg, summing their uncompressed size */ err = traverse_dataset(ds, from_txg, TRAVERSE_POST, dmu_calculate_send_traversal, &size); if (err) return (err); err = dmu_adjust_send_estimate_for_indirects(ds, size.uncompressed, size.compressed, stream_compressed, sizep); return (err); } typedef struct dmu_recv_begin_arg { const char *drba_origin; dmu_recv_cookie_t *drba_cookie; cred_t *drba_cred; uint64_t drba_snapobj; } dmu_recv_begin_arg_t; static int recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds, uint64_t fromguid) { uint64_t val; int error; dsl_pool_t *dp = ds->ds_dir->dd_pool; - /* temporary clone name must not exist */ + /* Temporary clone name must not exist. */ error = zap_lookup(dp->dp_meta_objset, dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name, 8, 1, &val); if (error != ENOENT) - return (error == 0 ? EBUSY : error); + return (error == 0 ? SET_ERROR(EBUSY) : error); - /* new snapshot name must not exist */ + /* Resume state must not be set. */ + if (dsl_dataset_has_resume_receive_state(ds)) + return (SET_ERROR(EBUSY)); + + /* New snapshot name must not exist. */ error = zap_lookup(dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap, 8, 1, &val); if (error != ENOENT) - return (error == 0 ? EEXIST : error); + return (error == 0 ? SET_ERROR(EEXIST) : error); /* * Check snapshot limit before receiving. We'll recheck again at the * end, but might as well abort before receiving if we're already over * the limit. * * Note that we do not check the file system limit with * dsl_dir_fscount_check because the temporary %clones don't count * against that limit. */ error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); if (error != 0) return (error); if (fromguid != 0) { dsl_dataset_t *snap; uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; /* Find snapshot in this dir that matches fromguid. */ while (obj != 0) { error = dsl_dataset_hold_obj(dp, obj, FTAG, &snap); if (error != 0) return (SET_ERROR(ENODEV)); if (snap->ds_dir != ds->ds_dir) { dsl_dataset_rele(snap, FTAG); return (SET_ERROR(ENODEV)); } if (dsl_dataset_phys(snap)->ds_guid == fromguid) break; obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; dsl_dataset_rele(snap, FTAG); } if (obj == 0) return (SET_ERROR(ENODEV)); if (drba->drba_cookie->drc_force) { drba->drba_snapobj = obj; } else { /* * If we are not forcing, there must be no * changes since fromsnap. */ if (dsl_dataset_modified_since_snap(ds, snap)) { dsl_dataset_rele(snap, FTAG); return (SET_ERROR(ETXTBSY)); } drba->drba_snapobj = ds->ds_prev->ds_object; } dsl_dataset_rele(snap, FTAG); } else { /* if full, then must be forced */ if (!drba->drba_cookie->drc_force) return (SET_ERROR(EEXIST)); /* start from $ORIGIN@$ORIGIN, if supported */ drba->drba_snapobj = dp->dp_origin_snap != NULL ? dp->dp_origin_snap->ds_object : 0; } return (0); } static int dmu_recv_begin_check(void *arg, dmu_tx_t *tx) { dmu_recv_begin_arg_t *drba = arg; dsl_pool_t *dp = dmu_tx_pool(tx); struct drr_begin *drrb = drba->drba_cookie->drc_drrb; uint64_t fromguid = drrb->drr_fromguid; int flags = drrb->drr_flags; int error; uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); dsl_dataset_t *ds; const char *tofs = drba->drba_cookie->drc_tofs; /* already checked */ ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); ASSERT(!(featureflags & DMU_BACKUP_FEATURE_RESUMING)); if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_COMPOUNDSTREAM || drrb->drr_type >= DMU_OST_NUMTYPES || ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL)) return (SET_ERROR(EINVAL)); /* Verify pool version supports SA if SA_SPILL feature set */ if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && spa_version(dp->dp_spa) < SPA_VERSION_SA) return (SET_ERROR(ENOTSUP)); if (drba->drba_cookie->drc_resumable && !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EXTENSIBLE_DATASET)) return (SET_ERROR(ENOTSUP)); /* * The receiving code doesn't know how to translate a WRITE_EMBEDDED * record to a plain WRITE record, so the pool must have the * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED * records. Same with WRITE_EMBEDDED records that use LZ4 compression. */ if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) return (SET_ERROR(ENOTSUP)); if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) return (SET_ERROR(ENOTSUP)); /* * The receiving code doesn't know how to translate large blocks * to smaller ones, so the pool must have the LARGE_BLOCKS * feature enabled if the stream has LARGE_BLOCKS. */ if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) && !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) return (SET_ERROR(ENOTSUP)); error = dsl_dataset_hold(dp, tofs, FTAG, &ds); if (error == 0) { /* target fs already exists; recv into temp clone */ /* Can't recv a clone into an existing fs */ if (flags & DRR_FLAG_CLONE || drba->drba_origin) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } error = recv_begin_check_existing_impl(drba, ds, fromguid); dsl_dataset_rele(ds, FTAG); } else if (error == ENOENT) { /* target fs does not exist; must be a full backup or clone */ char buf[ZFS_MAX_DATASET_NAME_LEN]; /* * If it's a non-clone incremental, we are missing the * target fs, so fail the recv. */ if (fromguid != 0 && !(flags & DRR_FLAG_CLONE || drba->drba_origin)) return (SET_ERROR(ENOENT)); /* * If we're receiving a full send as a clone, and it doesn't * contain all the necessary free records and freeobject * records, reject it. */ if (fromguid == 0 && drba->drba_origin && !(flags & DRR_FLAG_FREERECORDS)) return (SET_ERROR(EINVAL)); /* Open the parent of tofs */ ASSERT3U(strlen(tofs), <, sizeof (buf)); (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1); error = dsl_dataset_hold(dp, buf, FTAG, &ds); if (error != 0) return (error); /* * Check filesystem and snapshot limits before receiving. We'll * recheck snapshot limits again at the end (we create the * filesystems and increment those counts during begin_sync). */ error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } if (drba->drba_origin != NULL) { dsl_dataset_t *origin; error = dsl_dataset_hold(dp, drba->drba_origin, FTAG, &origin); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } if (!origin->ds_is_snapshot) { dsl_dataset_rele(origin, FTAG); dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } if (dsl_dataset_phys(origin)->ds_guid != fromguid && fromguid != 0) { dsl_dataset_rele(origin, FTAG); dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENODEV)); } dsl_dataset_rele(origin, FTAG); } dsl_dataset_rele(ds, FTAG); error = 0; } return (error); } static void dmu_recv_begin_sync(void *arg, dmu_tx_t *tx) { dmu_recv_begin_arg_t *drba = arg; dsl_pool_t *dp = dmu_tx_pool(tx); objset_t *mos = dp->dp_meta_objset; struct drr_begin *drrb = drba->drba_cookie->drc_drrb; const char *tofs = drba->drba_cookie->drc_tofs; dsl_dataset_t *ds, *newds; uint64_t dsobj; int error; uint64_t crflags = 0; if (drrb->drr_flags & DRR_FLAG_CI_DATA) crflags |= DS_FLAG_CI_DATASET; error = dsl_dataset_hold(dp, tofs, FTAG, &ds); if (error == 0) { /* create temporary clone */ dsl_dataset_t *snap = NULL; if (drba->drba_snapobj != 0) { VERIFY0(dsl_dataset_hold_obj(dp, drba->drba_snapobj, FTAG, &snap)); } dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name, snap, crflags, drba->drba_cred, tx); if (drba->drba_snapobj != 0) dsl_dataset_rele(snap, FTAG); dsl_dataset_rele(ds, FTAG); } else { dsl_dir_t *dd; const char *tail; dsl_dataset_t *origin = NULL; VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail)); if (drba->drba_origin != NULL) { VERIFY0(dsl_dataset_hold(dp, drba->drba_origin, FTAG, &origin)); } /* Create new dataset. */ dsobj = dsl_dataset_create_sync(dd, strrchr(tofs, '/') + 1, origin, crflags, drba->drba_cred, tx); if (origin != NULL) dsl_dataset_rele(origin, FTAG); dsl_dir_rele(dd, FTAG); drba->drba_cookie->drc_newfs = B_TRUE; } VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds)); if (drba->drba_cookie->drc_resumable) { dsl_dataset_zapify(newds, tx); if (drrb->drr_fromguid != 0) { VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_FROMGUID, 8, 1, &drrb->drr_fromguid, tx)); } VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TOGUID, 8, 1, &drrb->drr_toguid, tx)); VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_TONAME, 1, strlen(drrb->drr_toname) + 1, drrb->drr_toname, tx)); uint64_t one = 1; uint64_t zero = 0; VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OBJECT, 8, 1, &one, tx)); VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_OFFSET, 8, 1, &zero, tx)); VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_BYTES, 8, 1, &zero, tx)); if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & DMU_BACKUP_FEATURE_LARGE_BLOCKS) { VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_LARGEBLOCK, 8, 1, &one, tx)); } if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & DMU_BACKUP_FEATURE_EMBED_DATA) { VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_EMBEDOK, 8, 1, &one, tx)); } if (DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) & DMU_BACKUP_FEATURE_COMPRESSED) { VERIFY0(zap_add(mos, dsobj, DS_FIELD_RESUME_COMPRESSOK, 8, 1, &one, tx)); } } dmu_buf_will_dirty(newds->ds_dbuf, tx); dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT; /* * If we actually created a non-clone, we need to create the * objset in our new dataset. */ rrw_enter(&newds->ds_bp_rwlock, RW_READER, FTAG); if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) { (void) dmu_objset_create_impl(dp->dp_spa, newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx); } rrw_exit(&newds->ds_bp_rwlock, FTAG); drba->drba_cookie->drc_ds = newds; spa_history_log_internal_ds(newds, "receive", tx, ""); } static int dmu_recv_resume_begin_check(void *arg, dmu_tx_t *tx) { dmu_recv_begin_arg_t *drba = arg; dsl_pool_t *dp = dmu_tx_pool(tx); struct drr_begin *drrb = drba->drba_cookie->drc_drrb; int error; uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo); dsl_dataset_t *ds; const char *tofs = drba->drba_cookie->drc_tofs; /* already checked */ ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC); ASSERT(featureflags & DMU_BACKUP_FEATURE_RESUMING); if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_COMPOUNDSTREAM || drrb->drr_type >= DMU_OST_NUMTYPES) return (SET_ERROR(EINVAL)); /* Verify pool version supports SA if SA_SPILL feature set */ if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) && spa_version(dp->dp_spa) < SPA_VERSION_SA) return (SET_ERROR(ENOTSUP)); /* * The receiving code doesn't know how to translate a WRITE_EMBEDDED * record to a plain WRITE record, so the pool must have the * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED * records. Same with WRITE_EMBEDDED records that use LZ4 compression. */ if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) && !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) return (SET_ERROR(ENOTSUP)); if ((featureflags & DMU_BACKUP_FEATURE_LZ4) && !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS)) return (SET_ERROR(ENOTSUP)); /* 6 extra bytes for /%recv */ char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; (void) snprintf(recvname, sizeof (recvname), "%s/%s", tofs, recv_clone_name); if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { /* %recv does not exist; continue in tofs */ error = dsl_dataset_hold(dp, tofs, FTAG, &ds); if (error != 0) return (error); } /* check that ds is marked inconsistent */ if (!DS_IS_INCONSISTENT(ds)) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } /* check that there is resuming data, and that the toguid matches */ if (!dsl_dataset_is_zapified(ds)) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } uint64_t val; error = zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val); if (error != 0 || drrb->drr_toguid != val) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } /* * Check if the receive is still running. If so, it will be owned. * Note that nothing else can own the dataset (e.g. after the receive * fails) because it will be marked inconsistent. */ if (dsl_dataset_has_owner(ds)) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EBUSY)); } /* There should not be any snapshots of this fs yet. */ if (ds->ds_prev != NULL && ds->ds_prev->ds_dir == ds->ds_dir) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } /* * Note: resume point will be checked when we process the first WRITE * record. */ /* check that the origin matches */ val = 0; (void) zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val); if (drrb->drr_fromguid != val) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } dsl_dataset_rele(ds, FTAG); return (0); } static void dmu_recv_resume_begin_sync(void *arg, dmu_tx_t *tx) { dmu_recv_begin_arg_t *drba = arg; dsl_pool_t *dp = dmu_tx_pool(tx); const char *tofs = drba->drba_cookie->drc_tofs; dsl_dataset_t *ds; uint64_t dsobj; /* 6 extra bytes for /%recv */ char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; (void) snprintf(recvname, sizeof (recvname), "%s/%s", tofs, recv_clone_name); if (dsl_dataset_hold(dp, recvname, FTAG, &ds) != 0) { /* %recv does not exist; continue in tofs */ VERIFY0(dsl_dataset_hold(dp, tofs, FTAG, &ds)); drba->drba_cookie->drc_newfs = B_TRUE; } /* clear the inconsistent flag so that we can own it */ ASSERT(DS_IS_INCONSISTENT(ds)); dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; dsobj = ds->ds_object; dsl_dataset_rele(ds, FTAG); VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &ds)); dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT; rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); ASSERT(!BP_IS_HOLE(dsl_dataset_get_blkptr(ds))); rrw_exit(&ds->ds_bp_rwlock, FTAG); drba->drba_cookie->drc_ds = ds; spa_history_log_internal_ds(ds, "resume receive", tx, ""); } /* * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin() * succeeds; otherwise we will leak the holds on the datasets. */ int dmu_recv_begin(char *tofs, char *tosnap, dmu_replay_record_t *drr_begin, boolean_t force, boolean_t resumable, char *origin, dmu_recv_cookie_t *drc) { dmu_recv_begin_arg_t drba = { 0 }; bzero(drc, sizeof (dmu_recv_cookie_t)); drc->drc_drr_begin = drr_begin; drc->drc_drrb = &drr_begin->drr_u.drr_begin; drc->drc_tosnap = tosnap; drc->drc_tofs = tofs; drc->drc_force = force; drc->drc_resumable = resumable; drc->drc_cred = CRED(); drc->drc_clone = (origin != NULL); if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) { drc->drc_byteswap = B_TRUE; (void) fletcher_4_incremental_byteswap(drr_begin, sizeof (dmu_replay_record_t), &drc->drc_cksum); byteswap_record(drr_begin); } else if (drc->drc_drrb->drr_magic == DMU_BACKUP_MAGIC) { (void) fletcher_4_incremental_native(drr_begin, sizeof (dmu_replay_record_t), &drc->drc_cksum); } else { return (SET_ERROR(EINVAL)); } drba.drba_origin = origin; drba.drba_cookie = drc; drba.drba_cred = CRED(); if (DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo) & DMU_BACKUP_FEATURE_RESUMING) { return (dsl_sync_task(tofs, dmu_recv_resume_begin_check, dmu_recv_resume_begin_sync, &drba, 5, ZFS_SPACE_CHECK_NORMAL)); } else { return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync, &drba, 5, ZFS_SPACE_CHECK_NORMAL)); } } struct receive_record_arg { dmu_replay_record_t header; void *payload; /* Pointer to a buffer containing the payload */ /* * If the record is a write, pointer to the arc_buf_t containing the * payload. */ arc_buf_t *write_buf; int payload_size; uint64_t bytes_read; /* bytes read from stream when record created */ boolean_t eos_marker; /* Marks the end of the stream */ bqueue_node_t node; }; struct receive_writer_arg { objset_t *os; boolean_t byteswap; bqueue_t q; /* * These three args are used to signal to the main thread that we're * done. */ kmutex_t mutex; kcondvar_t cv; boolean_t done; int err; /* A map from guid to dataset to help handle dedup'd streams. */ avl_tree_t *guid_to_ds_map; boolean_t resumable; uint64_t last_object; uint64_t last_offset; uint64_t max_object; /* highest object ID referenced in stream */ uint64_t bytes_read; /* bytes read when current record created */ }; struct objlist { list_t list; /* List of struct receive_objnode. */ /* * Last object looked up. Used to assert that objects are being looked * up in ascending order. */ uint64_t last_lookup; }; struct receive_objnode { list_node_t node; uint64_t object; }; struct receive_arg { objset_t *os; kthread_t *td; struct file *fp; uint64_t voff; /* The current offset in the stream */ uint64_t bytes_read; /* * A record that has had its payload read in, but hasn't yet been handed * off to the worker thread. */ struct receive_record_arg *rrd; /* A record that has had its header read in, but not its payload. */ struct receive_record_arg *next_rrd; zio_cksum_t cksum; zio_cksum_t prev_cksum; int err; boolean_t byteswap; /* Sorted list of objects not to issue prefetches for. */ struct objlist ignore_objlist; }; typedef struct guid_map_entry { uint64_t guid; dsl_dataset_t *gme_ds; avl_node_t avlnode; } guid_map_entry_t; static int guid_compare(const void *arg1, const void *arg2) { const guid_map_entry_t *gmep1 = (const guid_map_entry_t *)arg1; const guid_map_entry_t *gmep2 = (const guid_map_entry_t *)arg2; return (AVL_CMP(gmep1->guid, gmep2->guid)); } static void free_guid_map_onexit(void *arg) { avl_tree_t *ca = arg; void *cookie = NULL; guid_map_entry_t *gmep; while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) { dsl_dataset_long_rele(gmep->gme_ds, gmep); dsl_dataset_rele(gmep->gme_ds, gmep); kmem_free(gmep, sizeof (guid_map_entry_t)); } avl_destroy(ca); kmem_free(ca, sizeof (avl_tree_t)); } static int restore_bytes(struct receive_arg *ra, void *buf, int len, off_t off, ssize_t *resid) { struct uio auio; struct iovec aiov; int error; aiov.iov_base = buf; aiov.iov_len = len; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_resid = len; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_offset = off; auio.uio_td = ra->td; #ifdef _KERNEL error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td); #else fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__); error = EOPNOTSUPP; #endif *resid = auio.uio_resid; return (error); } static int receive_read(struct receive_arg *ra, int len, void *buf) { int done = 0; /* * The code doesn't rely on this (lengths being multiples of 8). See * comment in dump_bytes. */ ASSERT0(len % 8); while (done < len) { ssize_t resid; ra->err = restore_bytes(ra, buf + done, len - done, ra->voff, &resid); if (resid == len - done) { /* * Note: ECKSUM indicates that the receive * was interrupted and can potentially be resumed. */ ra->err = SET_ERROR(ECKSUM); } ra->voff += len - done - resid; done = len - resid; if (ra->err != 0) return (ra->err); } ra->bytes_read += len; ASSERT3U(done, ==, len); return (0); } static void byteswap_record(dmu_replay_record_t *drr) { #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X)) #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X)) drr->drr_type = BSWAP_32(drr->drr_type); drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen); switch (drr->drr_type) { case DRR_BEGIN: DO64(drr_begin.drr_magic); DO64(drr_begin.drr_versioninfo); DO64(drr_begin.drr_creation_time); DO32(drr_begin.drr_type); DO32(drr_begin.drr_flags); DO64(drr_begin.drr_toguid); DO64(drr_begin.drr_fromguid); break; case DRR_OBJECT: DO64(drr_object.drr_object); DO32(drr_object.drr_type); DO32(drr_object.drr_bonustype); DO32(drr_object.drr_blksz); DO32(drr_object.drr_bonuslen); DO64(drr_object.drr_toguid); break; case DRR_FREEOBJECTS: DO64(drr_freeobjects.drr_firstobj); DO64(drr_freeobjects.drr_numobjs); DO64(drr_freeobjects.drr_toguid); break; case DRR_WRITE: DO64(drr_write.drr_object); DO32(drr_write.drr_type); DO64(drr_write.drr_offset); DO64(drr_write.drr_logical_size); DO64(drr_write.drr_toguid); ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write.drr_key.ddk_cksum); DO64(drr_write.drr_key.ddk_prop); DO64(drr_write.drr_compressed_size); break; case DRR_WRITE_BYREF: DO64(drr_write_byref.drr_object); DO64(drr_write_byref.drr_offset); DO64(drr_write_byref.drr_length); DO64(drr_write_byref.drr_toguid); DO64(drr_write_byref.drr_refguid); DO64(drr_write_byref.drr_refobject); DO64(drr_write_byref.drr_refoffset); ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_write_byref. drr_key.ddk_cksum); DO64(drr_write_byref.drr_key.ddk_prop); break; case DRR_WRITE_EMBEDDED: DO64(drr_write_embedded.drr_object); DO64(drr_write_embedded.drr_offset); DO64(drr_write_embedded.drr_length); DO64(drr_write_embedded.drr_toguid); DO32(drr_write_embedded.drr_lsize); DO32(drr_write_embedded.drr_psize); break; case DRR_FREE: DO64(drr_free.drr_object); DO64(drr_free.drr_offset); DO64(drr_free.drr_length); DO64(drr_free.drr_toguid); break; case DRR_SPILL: DO64(drr_spill.drr_object); DO64(drr_spill.drr_length); DO64(drr_spill.drr_toguid); break; case DRR_END: DO64(drr_end.drr_toguid); ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_end.drr_checksum); break; } if (drr->drr_type != DRR_BEGIN) { ZIO_CHECKSUM_BSWAP(&drr->drr_u.drr_checksum.drr_checksum); } #undef DO64 #undef DO32 } static inline uint8_t deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size) { if (bonus_type == DMU_OT_SA) { return (1); } else { return (1 + ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT)); } } static void save_resume_state(struct receive_writer_arg *rwa, uint64_t object, uint64_t offset, dmu_tx_t *tx) { int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; if (!rwa->resumable) return; /* * We use ds_resume_bytes[] != 0 to indicate that we need to * update this on disk, so it must not be 0. */ ASSERT(rwa->bytes_read != 0); /* * We only resume from write records, which have a valid * (non-meta-dnode) object number. */ ASSERT(object != 0); /* * For resuming to work correctly, we must receive records in order, * sorted by object,offset. This is checked by the callers, but * assert it here for good measure. */ ASSERT3U(object, >=, rwa->os->os_dsl_dataset->ds_resume_object[txgoff]); ASSERT(object != rwa->os->os_dsl_dataset->ds_resume_object[txgoff] || offset >= rwa->os->os_dsl_dataset->ds_resume_offset[txgoff]); ASSERT3U(rwa->bytes_read, >=, rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff]); rwa->os->os_dsl_dataset->ds_resume_object[txgoff] = object; rwa->os->os_dsl_dataset->ds_resume_offset[txgoff] = offset; rwa->os->os_dsl_dataset->ds_resume_bytes[txgoff] = rwa->bytes_read; } static int receive_object(struct receive_writer_arg *rwa, struct drr_object *drro, void *data) { dmu_object_info_t doi; dmu_tx_t *tx; uint64_t object; int err; if (drro->drr_type == DMU_OT_NONE || !DMU_OT_IS_VALID(drro->drr_type) || !DMU_OT_IS_VALID(drro->drr_bonustype) || drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS || drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS || P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) || drro->drr_blksz < SPA_MINBLOCKSIZE || drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) || drro->drr_bonuslen > DN_MAX_BONUSLEN) { return (SET_ERROR(EINVAL)); } err = dmu_object_info(rwa->os, drro->drr_object, &doi); if (err != 0 && err != ENOENT) return (SET_ERROR(EINVAL)); object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT; if (drro->drr_object > rwa->max_object) rwa->max_object = drro->drr_object; /* * If we are losing blkptrs or changing the block size this must * be a new file instance. We must clear out the previous file * contents before we can change this type of metadata in the dnode. */ if (err == 0) { int nblkptr; nblkptr = deduce_nblkptr(drro->drr_bonustype, drro->drr_bonuslen); if (drro->drr_blksz != doi.doi_data_block_size || nblkptr < doi.doi_nblkptr) { err = dmu_free_long_range(rwa->os, drro->drr_object, 0, DMU_OBJECT_END); if (err != 0) return (SET_ERROR(EINVAL)); } } tx = dmu_tx_create(rwa->os); dmu_tx_hold_bonus(tx, object); err = dmu_tx_assign(tx, TXG_WAIT); if (err != 0) { dmu_tx_abort(tx); return (err); } if (object == DMU_NEW_OBJECT) { /* currently free, want to be allocated */ err = dmu_object_claim(rwa->os, drro->drr_object, drro->drr_type, drro->drr_blksz, drro->drr_bonustype, drro->drr_bonuslen, tx); } else if (drro->drr_type != doi.doi_type || drro->drr_blksz != doi.doi_data_block_size || drro->drr_bonustype != doi.doi_bonus_type || drro->drr_bonuslen != doi.doi_bonus_size) { /* currently allocated, but with different properties */ err = dmu_object_reclaim(rwa->os, drro->drr_object, drro->drr_type, drro->drr_blksz, drro->drr_bonustype, drro->drr_bonuslen, tx); } if (err != 0) { dmu_tx_commit(tx); return (SET_ERROR(EINVAL)); } dmu_object_set_checksum(rwa->os, drro->drr_object, drro->drr_checksumtype, tx); dmu_object_set_compress(rwa->os, drro->drr_object, drro->drr_compress, tx); if (data != NULL) { dmu_buf_t *db; VERIFY0(dmu_bonus_hold(rwa->os, drro->drr_object, FTAG, &db)); dmu_buf_will_dirty(db, tx); ASSERT3U(db->db_size, >=, drro->drr_bonuslen); bcopy(data, db->db_data, drro->drr_bonuslen); if (rwa->byteswap) { dmu_object_byteswap_t byteswap = DMU_OT_BYTESWAP(drro->drr_bonustype); dmu_ot_byteswap[byteswap].ob_func(db->db_data, drro->drr_bonuslen); } dmu_buf_rele(db, FTAG); } dmu_tx_commit(tx); return (0); } /* ARGSUSED */ static int receive_freeobjects(struct receive_writer_arg *rwa, struct drr_freeobjects *drrfo) { uint64_t obj; int next_err = 0; if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj) return (SET_ERROR(EINVAL)); for (obj = drrfo->drr_firstobj; obj < drrfo->drr_firstobj + drrfo->drr_numobjs && next_err == 0; next_err = dmu_object_next(rwa->os, &obj, FALSE, 0)) { int err; if (dmu_object_info(rwa->os, obj, NULL) != 0) continue; err = dmu_free_long_object(rwa->os, obj); if (err != 0) return (err); if (obj > rwa->max_object) rwa->max_object = obj; } if (next_err != ESRCH) return (next_err); return (0); } static int receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw, arc_buf_t *abuf) { dmu_tx_t *tx; int err; if (drrw->drr_offset + drrw->drr_logical_size < drrw->drr_offset || !DMU_OT_IS_VALID(drrw->drr_type)) return (SET_ERROR(EINVAL)); /* * For resuming to work, records must be in increasing order * by (object, offset). */ if (drrw->drr_object < rwa->last_object || (drrw->drr_object == rwa->last_object && drrw->drr_offset < rwa->last_offset)) { return (SET_ERROR(EINVAL)); } rwa->last_object = drrw->drr_object; rwa->last_offset = drrw->drr_offset; if (rwa->last_object > rwa->max_object) rwa->max_object = rwa->last_object; if (dmu_object_info(rwa->os, drrw->drr_object, NULL) != 0) return (SET_ERROR(EINVAL)); tx = dmu_tx_create(rwa->os); dmu_tx_hold_write(tx, drrw->drr_object, drrw->drr_offset, drrw->drr_logical_size); err = dmu_tx_assign(tx, TXG_WAIT); if (err != 0) { dmu_tx_abort(tx); return (err); } if (rwa->byteswap) { dmu_object_byteswap_t byteswap = DMU_OT_BYTESWAP(drrw->drr_type); dmu_ot_byteswap[byteswap].ob_func(abuf->b_data, DRR_WRITE_PAYLOAD_SIZE(drrw)); } /* use the bonus buf to look up the dnode in dmu_assign_arcbuf */ dmu_buf_t *bonus; if (dmu_bonus_hold(rwa->os, drrw->drr_object, FTAG, &bonus) != 0) return (SET_ERROR(EINVAL)); dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx); /* * Note: If the receive fails, we want the resume stream to start * with the same record that we last successfully received (as opposed * to the next record), so that we can verify that we are * resuming from the correct location. */ save_resume_state(rwa, drrw->drr_object, drrw->drr_offset, tx); dmu_tx_commit(tx); dmu_buf_rele(bonus, FTAG); return (0); } /* * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed * streams to refer to a copy of the data that is already on the * system because it came in earlier in the stream. This function * finds the earlier copy of the data, and uses that copy instead of * data from the stream to fulfill this write. */ static int receive_write_byref(struct receive_writer_arg *rwa, struct drr_write_byref *drrwbr) { dmu_tx_t *tx; int err; guid_map_entry_t gmesrch; guid_map_entry_t *gmep; avl_index_t where; objset_t *ref_os = NULL; dmu_buf_t *dbp; if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset) return (SET_ERROR(EINVAL)); /* * If the GUID of the referenced dataset is different from the * GUID of the target dataset, find the referenced dataset. */ if (drrwbr->drr_toguid != drrwbr->drr_refguid) { gmesrch.guid = drrwbr->drr_refguid; if ((gmep = avl_find(rwa->guid_to_ds_map, &gmesrch, &where)) == NULL) { return (SET_ERROR(EINVAL)); } if (dmu_objset_from_ds(gmep->gme_ds, &ref_os)) return (SET_ERROR(EINVAL)); } else { ref_os = rwa->os; } if (drrwbr->drr_object > rwa->max_object) rwa->max_object = drrwbr->drr_object; err = dmu_buf_hold(ref_os, drrwbr->drr_refobject, drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH); if (err != 0) return (err); tx = dmu_tx_create(rwa->os); dmu_tx_hold_write(tx, drrwbr->drr_object, drrwbr->drr_offset, drrwbr->drr_length); err = dmu_tx_assign(tx, TXG_WAIT); if (err != 0) { dmu_tx_abort(tx); return (err); } dmu_write(rwa->os, drrwbr->drr_object, drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx); dmu_buf_rele(dbp, FTAG); /* See comment in restore_write. */ save_resume_state(rwa, drrwbr->drr_object, drrwbr->drr_offset, tx); dmu_tx_commit(tx); return (0); } static int receive_write_embedded(struct receive_writer_arg *rwa, struct drr_write_embedded *drrwe, void *data) { dmu_tx_t *tx; int err; if (drrwe->drr_offset + drrwe->drr_length < drrwe->drr_offset) return (EINVAL); if (drrwe->drr_psize > BPE_PAYLOAD_SIZE) return (EINVAL); if (drrwe->drr_etype >= NUM_BP_EMBEDDED_TYPES) return (EINVAL); if (drrwe->drr_compression >= ZIO_COMPRESS_FUNCTIONS) return (EINVAL); if (drrwe->drr_object > rwa->max_object) rwa->max_object = drrwe->drr_object; tx = dmu_tx_create(rwa->os); dmu_tx_hold_write(tx, drrwe->drr_object, drrwe->drr_offset, drrwe->drr_length); err = dmu_tx_assign(tx, TXG_WAIT); if (err != 0) { dmu_tx_abort(tx); return (err); } dmu_write_embedded(rwa->os, drrwe->drr_object, drrwe->drr_offset, data, drrwe->drr_etype, drrwe->drr_compression, drrwe->drr_lsize, drrwe->drr_psize, rwa->byteswap ^ ZFS_HOST_BYTEORDER, tx); /* See comment in restore_write. */ save_resume_state(rwa, drrwe->drr_object, drrwe->drr_offset, tx); dmu_tx_commit(tx); return (0); } static int receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs, void *data) { dmu_tx_t *tx; dmu_buf_t *db, *db_spill; int err; if (drrs->drr_length < SPA_MINBLOCKSIZE || drrs->drr_length > spa_maxblocksize(dmu_objset_spa(rwa->os))) return (SET_ERROR(EINVAL)); if (dmu_object_info(rwa->os, drrs->drr_object, NULL) != 0) return (SET_ERROR(EINVAL)); if (drrs->drr_object > rwa->max_object) rwa->max_object = drrs->drr_object; VERIFY0(dmu_bonus_hold(rwa->os, drrs->drr_object, FTAG, &db)); if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) { dmu_buf_rele(db, FTAG); return (err); } tx = dmu_tx_create(rwa->os); dmu_tx_hold_spill(tx, db->db_object); err = dmu_tx_assign(tx, TXG_WAIT); if (err != 0) { dmu_buf_rele(db, FTAG); dmu_buf_rele(db_spill, FTAG); dmu_tx_abort(tx); return (err); } dmu_buf_will_dirty(db_spill, tx); if (db_spill->db_size < drrs->drr_length) VERIFY(0 == dbuf_spill_set_blksz(db_spill, drrs->drr_length, tx)); bcopy(data, db_spill->db_data, drrs->drr_length); dmu_buf_rele(db, FTAG); dmu_buf_rele(db_spill, FTAG); dmu_tx_commit(tx); return (0); } /* ARGSUSED */ static int receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf) { int err; if (drrf->drr_length != -1ULL && drrf->drr_offset + drrf->drr_length < drrf->drr_offset) return (SET_ERROR(EINVAL)); if (dmu_object_info(rwa->os, drrf->drr_object, NULL) != 0) return (SET_ERROR(EINVAL)); if (drrf->drr_object > rwa->max_object) rwa->max_object = drrf->drr_object; err = dmu_free_long_range(rwa->os, drrf->drr_object, drrf->drr_offset, drrf->drr_length); return (err); } /* used to destroy the drc_ds on error */ static void dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc) { if (drc->drc_resumable) { /* wait for our resume state to be written to disk */ txg_wait_synced(drc->drc_ds->ds_dir->dd_pool, 0); dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); } else { char name[ZFS_MAX_DATASET_NAME_LEN]; dsl_dataset_name(drc->drc_ds, name); dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); (void) dsl_destroy_head(name); } } static void receive_cksum(struct receive_arg *ra, int len, void *buf) { if (ra->byteswap) { (void) fletcher_4_incremental_byteswap(buf, len, &ra->cksum); } else { (void) fletcher_4_incremental_native(buf, len, &ra->cksum); } } /* * Read the payload into a buffer of size len, and update the current record's * payload field. * Allocate ra->next_rrd and read the next record's header into * ra->next_rrd->header. * Verify checksum of payload and next record. */ static int receive_read_payload_and_next_header(struct receive_arg *ra, int len, void *buf) { int err; if (len != 0) { ASSERT3U(len, <=, SPA_MAXBLOCKSIZE); err = receive_read(ra, len, buf); if (err != 0) return (err); receive_cksum(ra, len, buf); /* note: rrd is NULL when reading the begin record's payload */ if (ra->rrd != NULL) { ra->rrd->payload = buf; ra->rrd->payload_size = len; ra->rrd->bytes_read = ra->bytes_read; } } ra->prev_cksum = ra->cksum; ra->next_rrd = kmem_zalloc(sizeof (*ra->next_rrd), KM_SLEEP); err = receive_read(ra, sizeof (ra->next_rrd->header), &ra->next_rrd->header); ra->next_rrd->bytes_read = ra->bytes_read; if (err != 0) { kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); ra->next_rrd = NULL; return (err); } if (ra->next_rrd->header.drr_type == DRR_BEGIN) { kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); ra->next_rrd = NULL; return (SET_ERROR(EINVAL)); } /* * Note: checksum is of everything up to but not including the * checksum itself. */ ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t)); receive_cksum(ra, offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), &ra->next_rrd->header); zio_cksum_t cksum_orig = ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; zio_cksum_t *cksump = &ra->next_rrd->header.drr_u.drr_checksum.drr_checksum; if (ra->byteswap) byteswap_record(&ra->next_rrd->header); if ((!ZIO_CHECKSUM_IS_ZERO(cksump)) && !ZIO_CHECKSUM_EQUAL(ra->cksum, *cksump)) { kmem_free(ra->next_rrd, sizeof (*ra->next_rrd)); ra->next_rrd = NULL; return (SET_ERROR(ECKSUM)); } receive_cksum(ra, sizeof (cksum_orig), &cksum_orig); return (0); } static void objlist_create(struct objlist *list) { list_create(&list->list, sizeof (struct receive_objnode), offsetof(struct receive_objnode, node)); list->last_lookup = 0; } static void objlist_destroy(struct objlist *list) { for (struct receive_objnode *n = list_remove_head(&list->list); n != NULL; n = list_remove_head(&list->list)) { kmem_free(n, sizeof (*n)); } list_destroy(&list->list); } /* * This function looks through the objlist to see if the specified object number * is contained in the objlist. In the process, it will remove all object * numbers in the list that are smaller than the specified object number. Thus, * any lookup of an object number smaller than a previously looked up object * number will always return false; therefore, all lookups should be done in * ascending order. */ static boolean_t objlist_exists(struct objlist *list, uint64_t object) { struct receive_objnode *node = list_head(&list->list); ASSERT3U(object, >=, list->last_lookup); list->last_lookup = object; while (node != NULL && node->object < object) { VERIFY3P(node, ==, list_remove_head(&list->list)); kmem_free(node, sizeof (*node)); node = list_head(&list->list); } return (node != NULL && node->object == object); } /* * The objlist is a list of object numbers stored in ascending order. However, * the insertion of new object numbers does not seek out the correct location to * store a new object number; instead, it appends it to the list for simplicity. * Thus, any users must take care to only insert new object numbers in ascending * order. */ static void objlist_insert(struct objlist *list, uint64_t object) { struct receive_objnode *node = kmem_zalloc(sizeof (*node), KM_SLEEP); node->object = object; #ifdef ZFS_DEBUG struct receive_objnode *last_object = list_tail(&list->list); uint64_t last_objnum = (last_object != NULL ? last_object->object : 0); ASSERT3U(node->object, >, last_objnum); #endif list_insert_tail(&list->list, node); } /* * Issue the prefetch reads for any necessary indirect blocks. * * We use the object ignore list to tell us whether or not to issue prefetches * for a given object. We do this for both correctness (in case the blocksize * of an object has changed) and performance (if the object doesn't exist, don't * needlessly try to issue prefetches). We also trim the list as we go through * the stream to prevent it from growing to an unbounded size. * * The object numbers within will always be in sorted order, and any write * records we see will also be in sorted order, but they're not sorted with * respect to each other (i.e. we can get several object records before * receiving each object's write records). As a result, once we've reached a * given object number, we can safely remove any reference to lower object * numbers in the ignore list. In practice, we receive up to 32 object records * before receiving write records, so the list can have up to 32 nodes in it. */ /* ARGSUSED */ static void receive_read_prefetch(struct receive_arg *ra, uint64_t object, uint64_t offset, uint64_t length) { if (!objlist_exists(&ra->ignore_objlist, object)) { dmu_prefetch(ra->os, object, 1, offset, length, ZIO_PRIORITY_SYNC_READ); } } /* * Read records off the stream, issuing any necessary prefetches. */ static int receive_read_record(struct receive_arg *ra) { int err; switch (ra->rrd->header.drr_type) { case DRR_OBJECT: { struct drr_object *drro = &ra->rrd->header.drr_u.drr_object; uint32_t size = P2ROUNDUP(drro->drr_bonuslen, 8); void *buf = kmem_zalloc(size, KM_SLEEP); dmu_object_info_t doi; err = receive_read_payload_and_next_header(ra, size, buf); if (err != 0) { kmem_free(buf, size); return (err); } err = dmu_object_info(ra->os, drro->drr_object, &doi); /* * See receive_read_prefetch for an explanation why we're * storing this object in the ignore_obj_list. */ if (err == ENOENT || (err == 0 && doi.doi_data_block_size != drro->drr_blksz)) { objlist_insert(&ra->ignore_objlist, drro->drr_object); err = 0; } return (err); } case DRR_FREEOBJECTS: { err = receive_read_payload_and_next_header(ra, 0, NULL); return (err); } case DRR_WRITE: { struct drr_write *drrw = &ra->rrd->header.drr_u.drr_write; arc_buf_t *abuf; boolean_t is_meta = DMU_OT_IS_METADATA(drrw->drr_type); if (DRR_WRITE_COMPRESSED(drrw)) { ASSERT3U(drrw->drr_compressed_size, >, 0); ASSERT3U(drrw->drr_logical_size, >=, drrw->drr_compressed_size); ASSERT(!is_meta); abuf = arc_loan_compressed_buf( dmu_objset_spa(ra->os), drrw->drr_compressed_size, drrw->drr_logical_size, drrw->drr_compressiontype); } else { abuf = arc_loan_buf(dmu_objset_spa(ra->os), is_meta, drrw->drr_logical_size); } err = receive_read_payload_and_next_header(ra, DRR_WRITE_PAYLOAD_SIZE(drrw), abuf->b_data); if (err != 0) { dmu_return_arcbuf(abuf); return (err); } ra->rrd->write_buf = abuf; receive_read_prefetch(ra, drrw->drr_object, drrw->drr_offset, drrw->drr_logical_size); return (err); } case DRR_WRITE_BYREF: { struct drr_write_byref *drrwb = &ra->rrd->header.drr_u.drr_write_byref; err = receive_read_payload_and_next_header(ra, 0, NULL); receive_read_prefetch(ra, drrwb->drr_object, drrwb->drr_offset, drrwb->drr_length); return (err); } case DRR_WRITE_EMBEDDED: { struct drr_write_embedded *drrwe = &ra->rrd->header.drr_u.drr_write_embedded; uint32_t size = P2ROUNDUP(drrwe->drr_psize, 8); void *buf = kmem_zalloc(size, KM_SLEEP); err = receive_read_payload_and_next_header(ra, size, buf); if (err != 0) { kmem_free(buf, size); return (err); } receive_read_prefetch(ra, drrwe->drr_object, drrwe->drr_offset, drrwe->drr_length); return (err); } case DRR_FREE: { /* * It might be beneficial to prefetch indirect blocks here, but * we don't really have the data to decide for sure. */ err = receive_read_payload_and_next_header(ra, 0, NULL); return (err); } case DRR_END: { struct drr_end *drre = &ra->rrd->header.drr_u.drr_end; if (!ZIO_CHECKSUM_EQUAL(ra->prev_cksum, drre->drr_checksum)) return (SET_ERROR(ECKSUM)); return (0); } case DRR_SPILL: { struct drr_spill *drrs = &ra->rrd->header.drr_u.drr_spill; void *buf = kmem_zalloc(drrs->drr_length, KM_SLEEP); err = receive_read_payload_and_next_header(ra, drrs->drr_length, buf); if (err != 0) kmem_free(buf, drrs->drr_length); return (err); } default: return (SET_ERROR(EINVAL)); } } /* * Commit the records to the pool. */ static int receive_process_record(struct receive_writer_arg *rwa, struct receive_record_arg *rrd) { int err; /* Processing in order, therefore bytes_read should be increasing. */ ASSERT3U(rrd->bytes_read, >=, rwa->bytes_read); rwa->bytes_read = rrd->bytes_read; switch (rrd->header.drr_type) { case DRR_OBJECT: { struct drr_object *drro = &rrd->header.drr_u.drr_object; err = receive_object(rwa, drro, rrd->payload); kmem_free(rrd->payload, rrd->payload_size); rrd->payload = NULL; return (err); } case DRR_FREEOBJECTS: { struct drr_freeobjects *drrfo = &rrd->header.drr_u.drr_freeobjects; return (receive_freeobjects(rwa, drrfo)); } case DRR_WRITE: { struct drr_write *drrw = &rrd->header.drr_u.drr_write; err = receive_write(rwa, drrw, rrd->write_buf); /* if receive_write() is successful, it consumes the arc_buf */ if (err != 0) dmu_return_arcbuf(rrd->write_buf); rrd->write_buf = NULL; rrd->payload = NULL; return (err); } case DRR_WRITE_BYREF: { struct drr_write_byref *drrwbr = &rrd->header.drr_u.drr_write_byref; return (receive_write_byref(rwa, drrwbr)); } case DRR_WRITE_EMBEDDED: { struct drr_write_embedded *drrwe = &rrd->header.drr_u.drr_write_embedded; err = receive_write_embedded(rwa, drrwe, rrd->payload); kmem_free(rrd->payload, rrd->payload_size); rrd->payload = NULL; return (err); } case DRR_FREE: { struct drr_free *drrf = &rrd->header.drr_u.drr_free; return (receive_free(rwa, drrf)); } case DRR_SPILL: { struct drr_spill *drrs = &rrd->header.drr_u.drr_spill; err = receive_spill(rwa, drrs, rrd->payload); kmem_free(rrd->payload, rrd->payload_size); rrd->payload = NULL; return (err); } default: return (SET_ERROR(EINVAL)); } } /* * dmu_recv_stream's worker thread; pull records off the queue, and then call * receive_process_record When we're done, signal the main thread and exit. */ static void receive_writer_thread(void *arg) { struct receive_writer_arg *rwa = arg; struct receive_record_arg *rrd; for (rrd = bqueue_dequeue(&rwa->q); !rrd->eos_marker; rrd = bqueue_dequeue(&rwa->q)) { /* * If there's an error, the main thread will stop putting things * on the queue, but we need to clear everything in it before we * can exit. */ if (rwa->err == 0) { rwa->err = receive_process_record(rwa, rrd); } else if (rrd->write_buf != NULL) { dmu_return_arcbuf(rrd->write_buf); rrd->write_buf = NULL; rrd->payload = NULL; } else if (rrd->payload != NULL) { kmem_free(rrd->payload, rrd->payload_size); rrd->payload = NULL; } kmem_free(rrd, sizeof (*rrd)); } kmem_free(rrd, sizeof (*rrd)); mutex_enter(&rwa->mutex); rwa->done = B_TRUE; cv_signal(&rwa->cv); mutex_exit(&rwa->mutex); thread_exit(); } static int resume_check(struct receive_arg *ra, nvlist_t *begin_nvl) { uint64_t val; objset_t *mos = dmu_objset_pool(ra->os)->dp_meta_objset; uint64_t dsobj = dmu_objset_id(ra->os); uint64_t resume_obj, resume_off; if (nvlist_lookup_uint64(begin_nvl, "resume_object", &resume_obj) != 0 || nvlist_lookup_uint64(begin_nvl, "resume_offset", &resume_off) != 0) { return (SET_ERROR(EINVAL)); } VERIFY0(zap_lookup(mos, dsobj, DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val)); if (resume_obj != val) return (SET_ERROR(EINVAL)); VERIFY0(zap_lookup(mos, dsobj, DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val)); if (resume_off != val) return (SET_ERROR(EINVAL)); return (0); } /* * Read in the stream's records, one by one, and apply them to the pool. There * are two threads involved; the thread that calls this function will spin up a * worker thread, read the records off the stream one by one, and issue * prefetches for any necessary indirect blocks. It will then push the records * onto an internal blocking queue. The worker thread will pull the records off * the queue, and actually write the data into the DMU. This way, the worker * thread doesn't have to wait for reads to complete, since everything it needs * (the indirect blocks) will be prefetched. * * NB: callers *must* call dmu_recv_end() if this succeeds. */ int dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp, int cleanup_fd, uint64_t *action_handlep) { int err = 0; struct receive_arg ra = { 0 }; struct receive_writer_arg rwa = { 0 }; int featureflags; nvlist_t *begin_nvl = NULL; ra.byteswap = drc->drc_byteswap; ra.cksum = drc->drc_cksum; ra.td = curthread; ra.fp = fp; ra.voff = *voffp; if (dsl_dataset_is_zapified(drc->drc_ds)) { (void) zap_lookup(drc->drc_ds->ds_dir->dd_pool->dp_meta_objset, drc->drc_ds->ds_object, DS_FIELD_RESUME_BYTES, sizeof (ra.bytes_read), 1, &ra.bytes_read); } objlist_create(&ra.ignore_objlist); /* these were verified in dmu_recv_begin */ ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==, DMU_SUBSTREAM); ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES); /* * Open the objset we are modifying. */ VERIFY0(dmu_objset_from_ds(drc->drc_ds, &ra.os)); ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT); featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo); /* if this stream is dedup'ed, set up the avl tree for guid mapping */ if (featureflags & DMU_BACKUP_FEATURE_DEDUP) { minor_t minor; if (cleanup_fd == -1) { ra.err = SET_ERROR(EBADF); goto out; } ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor); if (ra.err != 0) { cleanup_fd = -1; goto out; } if (*action_handlep == 0) { rwa.guid_to_ds_map = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP); avl_create(rwa.guid_to_ds_map, guid_compare, sizeof (guid_map_entry_t), offsetof(guid_map_entry_t, avlnode)); err = zfs_onexit_add_cb(minor, free_guid_map_onexit, rwa.guid_to_ds_map, action_handlep); if (ra.err != 0) goto out; } else { err = zfs_onexit_cb_data(minor, *action_handlep, (void **)&rwa.guid_to_ds_map); if (ra.err != 0) goto out; } drc->drc_guid_to_ds_map = rwa.guid_to_ds_map; } uint32_t payloadlen = drc->drc_drr_begin->drr_payloadlen; void *payload = NULL; if (payloadlen != 0) payload = kmem_alloc(payloadlen, KM_SLEEP); err = receive_read_payload_and_next_header(&ra, payloadlen, payload); if (err != 0) { if (payloadlen != 0) kmem_free(payload, payloadlen); goto out; } if (payloadlen != 0) { err = nvlist_unpack(payload, payloadlen, &begin_nvl, KM_SLEEP); kmem_free(payload, payloadlen); if (err != 0) goto out; } if (featureflags & DMU_BACKUP_FEATURE_RESUMING) { err = resume_check(&ra, begin_nvl); if (err != 0) goto out; } (void) bqueue_init(&rwa.q, zfs_recv_queue_length, offsetof(struct receive_record_arg, node)); cv_init(&rwa.cv, NULL, CV_DEFAULT, NULL); mutex_init(&rwa.mutex, NULL, MUTEX_DEFAULT, NULL); rwa.os = ra.os; rwa.byteswap = drc->drc_byteswap; rwa.resumable = drc->drc_resumable; (void) thread_create(NULL, 0, receive_writer_thread, &rwa, 0, &p0, TS_RUN, minclsyspri); /* * We're reading rwa.err without locks, which is safe since we are the * only reader, and the worker thread is the only writer. It's ok if we * miss a write for an iteration or two of the loop, since the writer * thread will keep freeing records we send it until we send it an eos * marker. * * We can leave this loop in 3 ways: First, if rwa.err is * non-zero. In that case, the writer thread will free the rrd we just * pushed. Second, if we're interrupted; in that case, either it's the * first loop and ra.rrd was never allocated, or it's later, and ra.rrd * has been handed off to the writer thread who will free it. Finally, * if receive_read_record fails or we're at the end of the stream, then * we free ra.rrd and exit. */ while (rwa.err == 0) { if (issig(JUSTLOOKING) && issig(FORREAL)) { err = SET_ERROR(EINTR); break; } ASSERT3P(ra.rrd, ==, NULL); ra.rrd = ra.next_rrd; ra.next_rrd = NULL; /* Allocates and loads header into ra.next_rrd */ err = receive_read_record(&ra); if (ra.rrd->header.drr_type == DRR_END || err != 0) { kmem_free(ra.rrd, sizeof (*ra.rrd)); ra.rrd = NULL; break; } bqueue_enqueue(&rwa.q, ra.rrd, sizeof (struct receive_record_arg) + ra.rrd->payload_size); ra.rrd = NULL; } if (ra.next_rrd == NULL) ra.next_rrd = kmem_zalloc(sizeof (*ra.next_rrd), KM_SLEEP); ra.next_rrd->eos_marker = B_TRUE; bqueue_enqueue(&rwa.q, ra.next_rrd, 1); mutex_enter(&rwa.mutex); while (!rwa.done) { cv_wait(&rwa.cv, &rwa.mutex); } mutex_exit(&rwa.mutex); /* * If we are receiving a full stream as a clone, all object IDs which * are greater than the maximum ID referenced in the stream are * by definition unused and must be freed. Note that it's possible that * we've resumed this send and the first record we received was the END * record. In that case, max_object would be 0, but we shouldn't start * freeing all objects from there; instead we should start from the * resumeobj. */ if (drc->drc_clone && drc->drc_drrb->drr_fromguid == 0) { uint64_t obj; if (nvlist_lookup_uint64(begin_nvl, "resume_object", &obj) != 0) obj = 0; if (rwa.max_object > obj) obj = rwa.max_object; obj++; int free_err = 0; int next_err = 0; while (next_err == 0) { free_err = dmu_free_long_object(rwa.os, obj); if (free_err != 0 && free_err != ENOENT) break; next_err = dmu_object_next(rwa.os, &obj, FALSE, 0); } if (err == 0) { if (free_err != 0 && free_err != ENOENT) err = free_err; else if (next_err != ESRCH) err = next_err; } } cv_destroy(&rwa.cv); mutex_destroy(&rwa.mutex); bqueue_destroy(&rwa.q); if (err == 0) err = rwa.err; out: nvlist_free(begin_nvl); if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1)) zfs_onexit_fd_rele(cleanup_fd); if (err != 0) { /* * Clean up references. If receive is not resumable, * destroy what we created, so we don't leave it in * the inconsistent state. */ dmu_recv_cleanup_ds(drc); } *voffp = ra.voff; objlist_destroy(&ra.ignore_objlist); return (err); } static int dmu_recv_end_check(void *arg, dmu_tx_t *tx) { dmu_recv_cookie_t *drc = arg; dsl_pool_t *dp = dmu_tx_pool(tx); int error; ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag); if (!drc->drc_newfs) { dsl_dataset_t *origin_head; error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head); if (error != 0) return (error); if (drc->drc_force) { /* * We will destroy any snapshots in tofs (i.e. before * origin_head) that are after the origin (which is * the snap before drc_ds, because drc_ds can not * have any snaps of its own). */ uint64_t obj; obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; while (obj != dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { dsl_dataset_t *snap; error = dsl_dataset_hold_obj(dp, obj, FTAG, &snap); if (error != 0) break; if (snap->ds_dir != origin_head->ds_dir) error = SET_ERROR(EINVAL); if (error == 0) { error = dsl_destroy_snapshot_check_impl( snap, B_FALSE); } obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; dsl_dataset_rele(snap, FTAG); if (error != 0) break; } if (error != 0) { dsl_dataset_rele(origin_head, FTAG); return (error); } } error = dsl_dataset_clone_swap_check_impl(drc->drc_ds, origin_head, drc->drc_force, drc->drc_owner, tx); if (error != 0) { dsl_dataset_rele(origin_head, FTAG); return (error); } error = dsl_dataset_snapshot_check_impl(origin_head, drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); dsl_dataset_rele(origin_head, FTAG); if (error != 0) return (error); error = dsl_destroy_head_check_impl(drc->drc_ds, 1); } else { error = dsl_dataset_snapshot_check_impl(drc->drc_ds, drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred); } return (error); } static void dmu_recv_end_sync(void *arg, dmu_tx_t *tx) { dmu_recv_cookie_t *drc = arg; dsl_pool_t *dp = dmu_tx_pool(tx); spa_history_log_internal_ds(drc->drc_ds, "finish receiving", tx, "snap=%s", drc->drc_tosnap); if (!drc->drc_newfs) { dsl_dataset_t *origin_head; VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head)); if (drc->drc_force) { /* * Destroy any snapshots of drc_tofs (origin_head) * after the origin (the snap before drc_ds). */ uint64_t obj; obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; while (obj != dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) { dsl_dataset_t *snap; VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &snap)); ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir); obj = dsl_dataset_phys(snap)->ds_prev_snap_obj; dsl_destroy_snapshot_sync_impl(snap, B_FALSE, tx); dsl_dataset_rele(snap, FTAG); } } VERIFY3P(drc->drc_ds->ds_prev, ==, origin_head->ds_prev); dsl_dataset_clone_swap_sync_impl(drc->drc_ds, origin_head, tx); dsl_dataset_snapshot_sync_impl(origin_head, drc->drc_tosnap, tx); /* set snapshot's creation time and guid */ dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx); dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time = drc->drc_drrb->drr_creation_time; dsl_dataset_phys(origin_head->ds_prev)->ds_guid = drc->drc_drrb->drr_toguid; dsl_dataset_phys(origin_head->ds_prev)->ds_flags &= ~DS_FLAG_INCONSISTENT; dmu_buf_will_dirty(origin_head->ds_dbuf, tx); dsl_dataset_phys(origin_head)->ds_flags &= ~DS_FLAG_INCONSISTENT; drc->drc_newsnapobj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj; dsl_dataset_rele(origin_head, FTAG); dsl_destroy_head_sync_impl(drc->drc_ds, tx); if (drc->drc_owner != NULL) VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner); } else { dsl_dataset_t *ds = drc->drc_ds; dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx); /* set snapshot's creation time and guid */ dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); dsl_dataset_phys(ds->ds_prev)->ds_creation_time = drc->drc_drrb->drr_creation_time; dsl_dataset_phys(ds->ds_prev)->ds_guid = drc->drc_drrb->drr_toguid; dsl_dataset_phys(ds->ds_prev)->ds_flags &= ~DS_FLAG_INCONSISTENT; dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT; if (dsl_dataset_has_resume_receive_state(ds)) { (void) zap_remove(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_FROMGUID, tx); (void) zap_remove(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OBJECT, tx); (void) zap_remove(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OFFSET, tx); (void) zap_remove(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_BYTES, tx); (void) zap_remove(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TOGUID, tx); (void) zap_remove(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TONAME, tx); } drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj; } /* * Release the hold from dmu_recv_begin. This must be done before * we return to open context, so that when we free the dataset's dnode, * we can evict its bonus buffer. */ dsl_dataset_disown(drc->drc_ds, dmu_recv_tag); drc->drc_ds = NULL; } static int add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj) { dsl_pool_t *dp; dsl_dataset_t *snapds; guid_map_entry_t *gmep; int err; ASSERT(guid_map != NULL); err = dsl_pool_hold(name, FTAG, &dp); if (err != 0) return (err); gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP); err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds); if (err == 0) { gmep->guid = dsl_dataset_phys(snapds)->ds_guid; gmep->gme_ds = snapds; avl_add(guid_map, gmep); dsl_dataset_long_hold(snapds, gmep); } else kmem_free(gmep, sizeof (*gmep)); dsl_pool_rele(dp, FTAG); return (err); } static int dmu_recv_end_modified_blocks = 3; static int dmu_recv_existing_end(dmu_recv_cookie_t *drc) { #ifdef _KERNEL /* * We will be destroying the ds; make sure its origin is unmounted if * necessary. */ char name[ZFS_MAX_DATASET_NAME_LEN]; dsl_dataset_name(drc->drc_ds, name); zfs_destroy_unmount_origin(name); #endif return (dsl_sync_task(drc->drc_tofs, dmu_recv_end_check, dmu_recv_end_sync, drc, dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); } static int dmu_recv_new_end(dmu_recv_cookie_t *drc) { return (dsl_sync_task(drc->drc_tofs, dmu_recv_end_check, dmu_recv_end_sync, drc, dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL)); } int dmu_recv_end(dmu_recv_cookie_t *drc, void *owner) { int error; drc->drc_owner = owner; if (drc->drc_newfs) error = dmu_recv_new_end(drc); else error = dmu_recv_existing_end(drc); if (error != 0) { dmu_recv_cleanup_ds(drc); } else if (drc->drc_guid_to_ds_map != NULL) { (void) add_ds_to_guidmap(drc->drc_tofs, drc->drc_guid_to_ds_map, drc->drc_newsnapobj); } return (error); } /* * Return TRUE if this objset is currently being received into. */ boolean_t dmu_objset_is_receiving(objset_t *os) { return (os->os_dsl_dataset != NULL && os->os_dsl_dataset->ds_owner == dmu_recv_tag); } Index: stable/11 =================================================================== --- stable/11 (revision 352375) +++ stable/11 (revision 352376) Property changes on: stable/11 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r351803