/* * difffile.c - DIFF file handling source code. Read and write diff files. * * Copyright (c) 2001-2006, NLnet Labs. All rights reserved. * * See LICENSE for the license. * */ #include "config.h" #include #include #include #include #include #include "difffile.h" #include "xfrd-disk.h" #include "util.h" #include "packet.h" #include "rdata.h" #include "udb.h" #include "udbzone.h" #include "nsec3.h" #include "nsd.h" #include "rrl.h" static int write_64(FILE *out, uint64_t val) { return write_data(out, &val, sizeof(val)); } static int write_32(FILE *out, uint32_t val) { val = htonl(val); return write_data(out, &val, sizeof(val)); } static int write_8(FILE *out, uint8_t val) { return write_data(out, &val, sizeof(val)); } static int write_str(FILE *out, const char* str) { uint32_t len = strlen(str); if(!write_32(out, len)) return 0; return write_data(out, str, len); } void diff_write_packet(const char* zone, const char* pat, uint32_t old_serial, uint32_t new_serial, uint32_t seq_nr, uint8_t* data, size_t len, struct nsd* nsd, uint64_t filenumber) { FILE* df = xfrd_open_xfrfile(nsd, filenumber, seq_nr?"a":"w"); if(!df) { log_msg(LOG_ERR, "could not open transfer %s file %lld: %s", zone, (long long)filenumber, strerror(errno)); return; } /* if first part, first write the header */ if(seq_nr == 0) { struct timeval tv; if (gettimeofday(&tv, NULL) != 0) { log_msg(LOG_ERR, "could not get timestamp for %s: %s", zone, strerror(errno)); } if(!write_32(df, DIFF_PART_XFRF) || !write_8(df, 0) /* notcommitted(yet) */ || !write_32(df, 0) /* numberofparts when done */ || !write_64(df, (uint64_t) tv.tv_sec) || !write_32(df, (uint32_t) tv.tv_usec) || !write_32(df, old_serial) || !write_32(df, new_serial) || !write_64(df, (uint64_t) tv.tv_sec) || !write_32(df, (uint32_t) tv.tv_usec) || !write_str(df, zone) || !write_str(df, pat)) { log_msg(LOG_ERR, "could not write transfer %s file %lld: %s", zone, (long long)filenumber, strerror(errno)); fclose(df); return; } } if(!write_32(df, DIFF_PART_XXFR) || !write_32(df, len) || !write_data(df, data, len) || !write_32(df, len)) { log_msg(LOG_ERR, "could not write transfer %s file %lld: %s", zone, (long long)filenumber, strerror(errno)); } fclose(df); } void diff_write_commit(const char* zone, uint32_t old_serial, uint32_t new_serial, uint32_t num_parts, uint8_t commit, const char* log_str, struct nsd* nsd, uint64_t filenumber) { struct timeval tv; FILE* df; if (gettimeofday(&tv, NULL) != 0) { log_msg(LOG_ERR, "could not set timestamp for %s: %s", zone, strerror(errno)); } /* overwrite the first part of the file with 'committed = 1', * as well as the end_time and number of parts. * also write old_serial and new_serial, so that a bad file mixup * will result in unusable serial numbers. */ df = xfrd_open_xfrfile(nsd, filenumber, "r+"); if(!df) { log_msg(LOG_ERR, "could not open transfer %s file %lld: %s", zone, (long long)filenumber, strerror(errno)); return; } if(!write_32(df, DIFF_PART_XFRF) || !write_8(df, commit) /* committed */ || !write_32(df, num_parts) || !write_64(df, (uint64_t) tv.tv_sec) || !write_32(df, (uint32_t) tv.tv_usec) || !write_32(df, old_serial) || !write_32(df, new_serial)) { log_msg(LOG_ERR, "could not write transfer %s file %lld: %s", zone, (long long)filenumber, strerror(errno)); fclose(df); return; } /* append the log_str to the end of the file */ if(fseek(df, 0, SEEK_END) == -1) { log_msg(LOG_ERR, "could not fseek transfer %s file %lld: %s", zone, (long long)filenumber, strerror(errno)); fclose(df); return; } if(!write_str(df, log_str)) { log_msg(LOG_ERR, "could not write transfer %s file %lld: %s", zone, (long long)filenumber, strerror(errno)); fclose(df); return; } fflush(df); fclose(df); } int diff_read_64(FILE *in, uint64_t* result) { if (fread(result, sizeof(*result), 1, in) == 1) { return 1; } else { return 0; } } int diff_read_32(FILE *in, uint32_t* result) { if (fread(result, sizeof(*result), 1, in) == 1) { *result = ntohl(*result); return 1; } else { return 0; } } int diff_read_8(FILE *in, uint8_t* result) { if (fread(result, sizeof(*result), 1, in) == 1) { return 1; } else { return 0; } } int diff_read_str(FILE* in, char* buf, size_t len) { uint32_t disklen; if(!diff_read_32(in, &disklen)) return 0; if(disklen >= len) return 0; if(fread(buf, disklen, 1, in) != 1) return 0; buf[disklen] = 0; return 1; } static void add_rdata_to_recyclebin(namedb_type* db, rr_type* rr) { /* add rdatas to recycle bin. */ size_t i; for(i=0; irdata_count; i++) { if(!rdata_atom_is_domain(rr->type, i)) region_recycle(db->region, rr->rdatas[i].data, rdata_atom_size(rr->rdatas[i]) + sizeof(uint16_t)); } region_recycle(db->region, rr->rdatas, sizeof(rdata_atom_type)*rr->rdata_count); } /* this routine determines if below a domain there exist names with * data (is_existing) or no names below the domain have data. */ static int has_data_below(domain_type* top) { domain_type* d = top; assert(d != NULL); /* in the canonical ordering subdomains are after this name */ d = domain_next(d); while(d != NULL && domain_is_subdomain(d, top)) { if(d->is_existing) return 1; d = domain_next(d); } return 0; } /** check if domain with 0 rrsets has become empty (nonexist) */ static domain_type* rrset_zero_nonexist_check(domain_type* domain, domain_type* ce) { /* is the node now an empty node (completely deleted) */ if(domain->rrsets == 0) { /* if there is no data below it, it becomes non existing. also empty nonterminals above it become nonexisting */ /* check for data below this node. */ if(!has_data_below(domain)) { /* nonexist this domain and all parent empty nonterminals */ domain_type* p = domain; while(p != NULL && p->rrsets == 0) { if(p == ce || has_data_below(p)) return p; p->is_existing = 0; /* fixup wildcard child of parent */ if(p->parent && p->parent->wildcard_child_closest_match == p) p->parent->wildcard_child_closest_match = domain_previous_existing_child(p); p = p->parent; } } } return NULL; } /** remove rrset. Adjusts zone params. Does not remove domain */ static void rrset_delete(namedb_type* db, domain_type* domain, rrset_type* rrset) { int i; /* find previous */ rrset_type** pp = &domain->rrsets; while(*pp && *pp != rrset) { pp = &( (*pp)->next ); } if(!*pp) { /* rrset does not exist for domain */ return; } *pp = rrset->next; DEBUG(DEBUG_XFRD,2, (LOG_INFO, "delete rrset of %s type %s", domain_to_string(domain), rrtype_to_string(rrset_rrtype(rrset)))); /* is this a SOA rrset ? */ if(rrset->zone->soa_rrset == rrset) { rrset->zone->soa_rrset = 0; } if(rrset->zone->ns_rrset == rrset) { rrset->zone->ns_rrset = 0; } if(domain == rrset->zone->apex && rrset_rrtype(rrset) == TYPE_RRSIG) { for (i = 0; i < rrset->rr_count; ++i) { if(rr_rrsig_type_covered(&rrset->rrs[i])==TYPE_DNSKEY) { rrset->zone->is_secure = 0; break; } } } /* recycle the memory space of the rrset */ for (i = 0; i < rrset->rr_count; ++i) add_rdata_to_recyclebin(db, &rrset->rrs[i]); region_recycle(db->region, rrset->rrs, sizeof(rr_type) * rrset->rr_count); rrset->rr_count = 0; region_recycle(db->region, rrset, sizeof(rrset_type)); } static int rdatas_equal(rdata_atom_type *a, rdata_atom_type *b, int num, uint16_t type, int* rdnum, char** reason) { int k, start, end; start = 0; end = num; /** * SOA RDATA comparisons in XFR are more lenient, * only serial rdata is checked. **/ if (type == TYPE_SOA) { start = 2; end = 3; } for(k = start; k < end; k++) { if(rdata_atom_is_domain(type, k)) { if(dname_compare(domain_dname(a[k].domain), domain_dname(b[k].domain))!=0) { *rdnum = k; *reason = "dname data"; return 0; } } else if(rdata_atom_is_literal_domain(type, k)) { /* literal dname, but compare case insensitive */ if(a[k].data[0] != b[k].data[0]) { *rdnum = k; *reason = "literal dname len"; return 0; /* uncompressed len must be equal*/ } if(!dname_equal_nocase((uint8_t*)(a[k].data+1), (uint8_t*)(b[k].data+1), a[k].data[0])) { *rdnum = k; *reason = "literal dname data"; return 0; } } else { /* check length */ if(a[k].data[0] != b[k].data[0]) { *rdnum = k; *reason = "rdata len"; return 0; } /* check data */ if(memcmp(a[k].data+1, b[k].data+1, a[k].data[0])!=0) { *rdnum = k; *reason = "rdata data"; return 0; } } } return 1; } static void debug_find_rr_num(rrset_type* rrset, uint16_t type, uint16_t klass, rdata_atom_type *rdatas, ssize_t rdata_num) { int i, rd; char* reason = ""; for(i=0; i < rrset->rr_count; ++i) { if (rrset->rrs[i].type != type) { log_msg(LOG_WARNING, "diff: RR <%s, %s> does not match " "RR num %d type %s", dname_to_string(domain_dname(rrset->rrs[i].owner),0), rrtype_to_string(type), i, rrtype_to_string(rrset->rrs[i].type)); } if (rrset->rrs[i].klass != klass) { log_msg(LOG_WARNING, "diff: RR <%s, %s> class %d " "does not match RR num %d class %d", dname_to_string(domain_dname(rrset->rrs[i].owner),0), rrtype_to_string(type), klass, i, rrset->rrs[i].klass); } if (rrset->rrs[i].rdata_count != rdata_num) { log_msg(LOG_WARNING, "diff: RR <%s, %s> rdlen %u " "does not match RR num %d rdlen %d", dname_to_string(domain_dname(rrset->rrs[i].owner),0), rrtype_to_string(type), (unsigned) rdata_num, i, (unsigned) rrset->rrs[i].rdata_count); } if (!rdatas_equal(rdatas, rrset->rrs[i].rdatas, rdata_num, type, &rd, &reason)) { log_msg(LOG_WARNING, "diff: RR <%s, %s> rdata element " "%d differs from RR num %d rdata (%s)", dname_to_string(domain_dname(rrset->rrs[i].owner),0), rrtype_to_string(type), rd, i, reason); } } } static int find_rr_num(rrset_type* rrset, uint16_t type, uint16_t klass, rdata_atom_type *rdatas, ssize_t rdata_num, int add) { int i, rd; char* reason; for(i=0; i < rrset->rr_count; ++i) { if(rrset->rrs[i].type == type && rrset->rrs[i].klass == klass && rrset->rrs[i].rdata_count == rdata_num && rdatas_equal(rdatas, rrset->rrs[i].rdatas, rdata_num, type, &rd, &reason)) { return i; } } /* this is odd. Log why rr cannot be found. */ if (!add) { debug_find_rr_num(rrset, type, klass, rdatas, rdata_num); } return -1; } #ifdef NSEC3 /* see if nsec3 deletion triggers need action */ static void nsec3_delete_rr_trigger(namedb_type* db, rr_type* rr, zone_type* zone, udb_ptr* udbz) { /* the RR has not actually been deleted yet, so we can inspect it */ if(!zone->nsec3_param) return; /* see if the domain was an NSEC3-domain in the chain, but no longer */ if(rr->type == TYPE_NSEC3 && rr->owner->nsec3 && rr->owner->nsec3->nsec3_node.key && nsec3_rr_uses_params(rr, zone) && nsec3_in_chain_count(rr->owner, zone) <= 1) { domain_type* prev = nsec3_chain_find_prev(zone, rr->owner); /* remove from prehash because no longer an NSEC3 domain */ if(domain_is_prehash(db->domains, rr->owner)) prehash_del(db->domains, rr->owner); /* fixup the last in the zone */ if(rr->owner == zone->nsec3_last) zone->nsec3_last = prev; /* unlink from the nsec3tree */ zone_del_domain_in_hash_tree(zone->nsec3tree, &rr->owner->nsec3->nsec3_node); /* add previous NSEC3 to the prehash list */ if(prev && prev != rr->owner) prehash_add(db->domains, prev); else nsec3_clear_precompile(db, zone); /* this domain becomes ordinary data domain: done later */ } /* see if the rr was NSEC3PARAM that we were using */ else if(rr->type == TYPE_NSEC3PARAM && rr == zone->nsec3_param) { /* clear trees, wipe hashes, wipe precompile */ nsec3_clear_precompile(db, zone); /* pick up new nsec3param (from udb, or avoid deleted rr) */ nsec3_find_zone_param(db, zone, udbz, rr, 0); /* if no more NSEC3, done */ if(!zone->nsec3_param) return; nsec3_precompile_newparam(db, zone); } } /* see if nsec3 prehash can be removed with new rrset content */ static void nsec3_rrsets_changed_remove_prehash(domain_type* domain, zone_type* zone) { /* deletion of rrset already done, we can check if conditions apply */ /* see if the domain is no longer precompiled */ /* it has a hash_node, but no longer fulfills conditions */ if(nsec3_domain_part_of_zone(domain, zone) && domain->nsec3 && domain->nsec3->hash_wc && domain->nsec3->hash_wc->hash.node.key && !nsec3_condition_hash(domain, zone)) { /* remove precompile */ domain->nsec3->nsec3_cover = NULL; domain->nsec3->nsec3_wcard_child_cover = NULL; domain->nsec3->nsec3_is_exact = 0; /* remove it from the hash tree */ zone_del_domain_in_hash_tree(zone->hashtree, &domain->nsec3->hash_wc->hash.node); zone_del_domain_in_hash_tree(zone->wchashtree, &domain->nsec3->hash_wc->wc.node); } if(domain != zone->apex && domain->nsec3 && domain->nsec3->ds_parent_hash && domain->nsec3->ds_parent_hash->node.key && (!domain->parent || nsec3_domain_part_of_zone(domain->parent, zone)) && !nsec3_condition_dshash(domain, zone)) { /* remove precompile */ domain->nsec3->nsec3_ds_parent_cover = NULL; domain->nsec3->nsec3_ds_parent_is_exact = 0; /* remove it from the hash tree */ zone_del_domain_in_hash_tree(zone->dshashtree, &domain->nsec3->ds_parent_hash->node); } } /* see if domain needs to get precompiled info */ static void nsec3_rrsets_changed_add_prehash(namedb_type* db, domain_type* domain, zone_type* zone) { if(!zone->nsec3_param) return; if((!domain->nsec3 || !domain->nsec3->hash_wc || !domain->nsec3->hash_wc->hash.node.key) && nsec3_condition_hash(domain, zone)) { region_type* tmpregion = region_create(xalloc, free); nsec3_precompile_domain(db, domain, zone, tmpregion); region_destroy(tmpregion); } if((!domain->nsec3 || !domain->nsec3->ds_parent_hash || !domain->nsec3->ds_parent_hash->node.key) && nsec3_condition_dshash(domain, zone)) { nsec3_precompile_domain_ds(db, domain, zone); } } /* see if nsec3 rrset-deletion triggers need action */ static void nsec3_delete_rrset_trigger(namedb_type* db, domain_type* domain, zone_type* zone, uint16_t type) { if(!zone->nsec3_param) return; nsec3_rrsets_changed_remove_prehash(domain, zone); /* for type nsec3, or a delegation, the domain may have become a * 'normal' domain with its remaining data now */ if(type == TYPE_NSEC3 || type == TYPE_NS || type == TYPE_DS) nsec3_rrsets_changed_add_prehash(db, domain, zone); /* for type DNAME or a delegation, obscured data may be revealed */ if(type == TYPE_NS || type == TYPE_DS || type == TYPE_DNAME) { /* walk over subdomains and check them each */ domain_type *d; for(d=domain_next(domain); d && domain_is_subdomain(d, domain); d=domain_next(d)) { nsec3_rrsets_changed_add_prehash(db, d, zone); } } } /* see if nsec3 addition triggers need action */ static void nsec3_add_rr_trigger(namedb_type* db, rr_type* rr, zone_type* zone, udb_ptr* udbz) { /* the RR has been added in full, also to UDB (and thus NSEC3PARAM * in the udb has been adjusted) */ if(zone->nsec3_param && rr->type == TYPE_NSEC3 && (!rr->owner->nsec3 || !rr->owner->nsec3->nsec3_node.key) && nsec3_rr_uses_params(rr, zone)) { /* added NSEC3 into the chain */ nsec3_precompile_nsec3rr(db, rr->owner, zone); /* the domain has become an NSEC3-domain, if it was precompiled * previously, remove that, neatly done in routine above */ nsec3_rrsets_changed_remove_prehash(rr->owner, zone); /* set this NSEC3 to prehash */ prehash_add(db->domains, rr->owner); } else if(!zone->nsec3_param && rr->type == TYPE_NSEC3PARAM) { /* see if this means NSEC3 chain can be used */ nsec3_find_zone_param(db, zone, udbz, NULL, 0); if(!zone->nsec3_param) return; nsec3_zone_trees_create(db->region, zone); nsec3_precompile_newparam(db, zone); } } /* see if nsec3 rrset-addition triggers need action */ static void nsec3_add_rrset_trigger(namedb_type* db, domain_type* domain, zone_type* zone, uint16_t type) { /* the rrset has been added so we can inspect it */ if(!zone->nsec3_param) return; /* because the rrset is added we can check conditions easily. * check if domain needs to become precompiled now */ nsec3_rrsets_changed_add_prehash(db, domain, zone); /* if a delegation, it changes from normal name to unhashed referral */ if(type == TYPE_NS || type == TYPE_DS) { nsec3_rrsets_changed_remove_prehash(domain, zone); } /* if delegation or DNAME added, then some RRs may get obscured */ if(type == TYPE_NS || type == TYPE_DS || type == TYPE_DNAME) { /* walk over subdomains and check them each */ domain_type *d; for(d=domain_next(domain); d && domain_is_subdomain(d, domain); d=domain_next(d)) { nsec3_rrsets_changed_remove_prehash(d, zone); } } } #endif /* NSEC3 */ /* fixup usage lower for domain names in the rdata */ static void rr_lower_usage(namedb_type* db, rr_type* rr) { unsigned i; for(i=0; irdata_count; i++) { if(rdata_atom_is_domain(rr->type, i)) { assert(rdata_atom_domain(rr->rdatas[i])->usage > 0); rdata_atom_domain(rr->rdatas[i])->usage --; if(rdata_atom_domain(rr->rdatas[i])->usage == 0) domain_table_deldomain(db, rdata_atom_domain(rr->rdatas[i])); } } } static void rrset_lower_usage(namedb_type* db, rrset_type* rrset) { unsigned i; for(i=0; irr_count; i++) rr_lower_usage(db, &rrset->rrs[i]); } int delete_RR(namedb_type* db, const dname_type* dname, uint16_t type, uint16_t klass, buffer_type* packet, size_t rdatalen, zone_type *zone, region_type* temp_region, udb_ptr* udbz, int* softfail) { domain_type *domain; rrset_type *rrset; domain = domain_table_find(db->domains, dname); if(!domain) { log_msg(LOG_WARNING, "diff: domain %s does not exist", dname_to_string(dname,0)); buffer_skip(packet, rdatalen); *softfail = 1; return 1; /* not fatal error */ } rrset = domain_find_rrset(domain, zone, type); if(!rrset) { log_msg(LOG_WARNING, "diff: rrset %s does not exist", dname_to_string(dname,0)); buffer_skip(packet, rdatalen); *softfail = 1; return 1; /* not fatal error */ } else { /* find the RR in the rrset */ domain_table_type *temptable; rdata_atom_type *rdatas; ssize_t rdata_num; int rrnum; temptable = domain_table_create(temp_region); /* This will ensure that the dnames in rdata are * normalized, conform RFC 4035, section 6.2 */ rdata_num = rdata_wireformat_to_rdata_atoms( temp_region, temptable, type, rdatalen, packet, &rdatas); if(rdata_num == -1) { log_msg(LOG_ERR, "diff: bad rdata for %s", dname_to_string(dname,0)); return 0; } rrnum = find_rr_num(rrset, type, klass, rdatas, rdata_num, 0); if(rrnum == -1 && type == TYPE_SOA && domain == zone->apex && rrset->rr_count != 0) rrnum = 0; /* replace existing SOA if no match */ if(rrnum == -1) { log_msg(LOG_WARNING, "diff: RR <%s, %s> does not exist", dname_to_string(dname,0), rrtype_to_string(type)); *softfail = 1; return 1; /* not fatal error */ } /* delete the normalized RR from the udb */ if(db->udb) udb_del_rr(db->udb, udbz, &rrset->rrs[rrnum]); #ifdef NSEC3 /* process triggers for RR deletions */ nsec3_delete_rr_trigger(db, &rrset->rrs[rrnum], zone, udbz); #endif /* lower usage (possibly deleting other domains, and thus * invalidating the current RR's domain pointers) */ rr_lower_usage(db, &rrset->rrs[rrnum]); if(rrset->rr_count == 1) { /* delete entire rrset */ rrset_delete(db, domain, rrset); /* check if domain is now nonexisting (or parents) */ rrset_zero_nonexist_check(domain, NULL); #ifdef NSEC3 /* cleanup nsec3 */ nsec3_delete_rrset_trigger(db, domain, zone, type); #endif /* see if the domain can be deleted (and inspect parents) */ domain_table_deldomain(db, domain); } else { /* swap out the bad RR and decrease the count */ rr_type* rrs_orig = rrset->rrs; add_rdata_to_recyclebin(db, &rrset->rrs[rrnum]); if(rrnum < rrset->rr_count-1) rrset->rrs[rrnum] = rrset->rrs[rrset->rr_count-1]; memset(&rrset->rrs[rrset->rr_count-1], 0, sizeof(rr_type)); /* realloc the rrs array one smaller */ rrset->rrs = region_alloc_array_init(db->region, rrs_orig, (rrset->rr_count-1), sizeof(rr_type)); if(!rrset->rrs) { log_msg(LOG_ERR, "out of memory, %s:%d", __FILE__, __LINE__); exit(1); } region_recycle(db->region, rrs_orig, sizeof(rr_type) * rrset->rr_count); #ifdef NSEC3 if(type == TYPE_NSEC3PARAM && zone->nsec3_param) { /* fixup nsec3_param pointer to same RR */ assert(zone->nsec3_param >= rrs_orig && zone->nsec3_param <= rrs_orig+rrset->rr_count); /* last moved to rrnum, others at same index*/ if(zone->nsec3_param == &rrs_orig[ rrset->rr_count-1]) zone->nsec3_param = &rrset->rrs[rrnum]; else zone->nsec3_param = (void*)zone->nsec3_param -(void*)rrs_orig + (void*)rrset->rrs; } #endif /* NSEC3 */ rrset->rr_count --; #ifdef NSEC3 /* for type nsec3, the domain may have become a * 'normal' domain with its remaining data now */ if(type == TYPE_NSEC3) nsec3_rrsets_changed_add_prehash(db, domain, zone); #endif /* NSEC3 */ } } return 1; } int add_RR(namedb_type* db, const dname_type* dname, uint16_t type, uint16_t klass, uint32_t ttl, buffer_type* packet, size_t rdatalen, zone_type *zone, udb_ptr* udbz, int* softfail) { domain_type* domain; rrset_type* rrset; rdata_atom_type *rdatas; rr_type *rrs_old; ssize_t rdata_num; int rrnum; #ifdef NSEC3 int rrset_added = 0; #endif domain = domain_table_find(db->domains, dname); if(!domain) { /* create the domain */ domain = domain_table_insert(db->domains, dname); } rrset = domain_find_rrset(domain, zone, type); if(!rrset) { /* create the rrset */ rrset = region_alloc(db->region, sizeof(rrset_type)); if(!rrset) { log_msg(LOG_ERR, "out of memory, %s:%d", __FILE__, __LINE__); exit(1); } rrset->zone = zone; rrset->rrs = 0; rrset->rr_count = 0; domain_add_rrset(domain, rrset); #ifdef NSEC3 rrset_added = 1; #endif } /* dnames in rdata are normalized, conform RFC 4035, * Section 6.2 */ rdata_num = rdata_wireformat_to_rdata_atoms( db->region, db->domains, type, rdatalen, packet, &rdatas); if(rdata_num == -1) { log_msg(LOG_ERR, "diff: bad rdata for %s", dname_to_string(dname,0)); return 0; } rrnum = find_rr_num(rrset, type, klass, rdatas, rdata_num, 1); if(rrnum != -1) { DEBUG(DEBUG_XFRD, 2, (LOG_ERR, "diff: RR <%s, %s> already exists", dname_to_string(dname,0), rrtype_to_string(type))); /* ignore already existing RR: lenient accepting of messages */ *softfail = 1; return 1; } if(rrset->rr_count == 65535) { log_msg(LOG_ERR, "diff: too many RRs at %s", dname_to_string(dname,0)); return 0; } /* re-alloc the rrs and add the new */ rrs_old = rrset->rrs; rrset->rrs = region_alloc_array(db->region, (rrset->rr_count+1), sizeof(rr_type)); if(!rrset->rrs) { log_msg(LOG_ERR, "out of memory, %s:%d", __FILE__, __LINE__); exit(1); } if(rrs_old) memcpy(rrset->rrs, rrs_old, rrset->rr_count * sizeof(rr_type)); region_recycle(db->region, rrs_old, sizeof(rr_type) * rrset->rr_count); rrset->rr_count ++; rrset->rrs[rrset->rr_count - 1].owner = domain; rrset->rrs[rrset->rr_count - 1].rdatas = rdatas; rrset->rrs[rrset->rr_count - 1].ttl = ttl; rrset->rrs[rrset->rr_count - 1].type = type; rrset->rrs[rrset->rr_count - 1].klass = klass; rrset->rrs[rrset->rr_count - 1].rdata_count = rdata_num; /* see if it is a SOA */ if(domain == zone->apex) { apex_rrset_checks(db, rrset, domain); #ifdef NSEC3 if(type == TYPE_NSEC3PARAM && zone->nsec3_param) { /* the pointer just changed, fix it up to point * to the same record */ assert(zone->nsec3_param >= rrs_old && zone->nsec3_param < rrs_old+rrset->rr_count); /* in this order to make sure no overflow/underflow*/ zone->nsec3_param = (void*)zone->nsec3_param - (void*)rrs_old + (void*)rrset->rrs; } #endif /* NSEC3 */ } /* write the just-normalized RR to the udb */ if(db->udb) { if(!udb_write_rr(db->udb, udbz, &rrset->rrs[rrset->rr_count - 1])) { log_msg(LOG_ERR, "could not add RR to nsd.db, disk-space?"); return 0; } } #ifdef NSEC3 if(rrset_added) { domain_type* p = domain->parent; nsec3_add_rrset_trigger(db, domain, zone, type); /* go up and process (possibly created) empty nonterminals, * until we hit the apex or root */ while(p && p->rrsets == NULL && !p->is_apex) { nsec3_rrsets_changed_add_prehash(db, p, zone); p = p->parent; } } nsec3_add_rr_trigger(db, &rrset->rrs[rrset->rr_count - 1], zone, udbz); #endif /* NSEC3 */ return 1; } static zone_type* find_or_create_zone(namedb_type* db, const dname_type* zone_name, struct nsd_options* opt, const char* zstr, const char* patname) { zone_type* zone; struct zone_options* zopt; zone = namedb_find_zone(db, zone_name); if(zone) { return zone; } zopt = zone_options_find(opt, zone_name); if(!zopt) { /* if _implicit_ then insert as _part_of_config */ if(strncmp(patname, PATTERN_IMPLICIT_MARKER, strlen(PATTERN_IMPLICIT_MARKER)) == 0) { zopt = zone_options_create(opt->region); if(!zopt) return 0; zopt->part_of_config = 1; zopt->name = region_strdup(opt->region, zstr); zopt->pattern = pattern_options_find(opt, patname); if(!zopt->name || !zopt->pattern) return 0; if(!nsd_options_insert_zone(opt, zopt)) { log_msg(LOG_ERR, "bad domain name or duplicate zone '%s' " "pattern %s", zstr, patname); } } else { /* create zone : presumably already added to zonelist * by xfrd, who wrote the AXFR or IXFR to disk, so we only * need to add it to our config. * This process does not need linesize and offset zonelist */ zopt = zone_list_zone_insert(opt, zstr, patname, 0, 0); if(!zopt) return 0; } } zone = namedb_zone_create(db, zone_name, zopt); return zone; } void delete_zone_rrs(namedb_type* db, zone_type* zone) { rrset_type *rrset; domain_type *domain = zone->apex, *next; int nonexist_check = 0; /* go through entire tree below the zone apex (incl subzones) */ while(domain && domain_is_subdomain(domain, zone->apex)) { DEBUG(DEBUG_XFRD,2, (LOG_INFO, "delete zone visit %s", domain_to_string(domain))); /* delete all rrsets of the zone */ while((rrset = domain_find_any_rrset(domain, zone))) { /* lower usage can delete other domains */ rrset_lower_usage(db, rrset); /* rrset del does not delete our domain(yet) */ rrset_delete(db, domain, rrset); /* no rrset_zero_nonexist_check, do that later */ if(domain->rrsets == 0) nonexist_check = 1; } /* the delete upcoming could delete parents, but nothing next * or after the domain so store next ptr */ next = domain_next(domain); /* see if the domain can be deleted (and inspect parents) */ domain_table_deldomain(db, domain); domain = next; } /* check if data deletions have created nonexisting domain entries, * but after deleting domains so the checks are faster */ if(nonexist_check) { domain_type* ce = NULL; /* for speeding up has_data_below */ DEBUG(DEBUG_XFRD, 1, (LOG_INFO, "axfrdel: zero rrset check")); domain = zone->apex; while(domain && domain_is_subdomain(domain, zone->apex)) { /* the interesting domains should be existing==1 * and rrsets==0, speeding up out processing of * sub-zones, since we only spuriously check empty * nonterminals */ if(domain->is_existing) ce = rrset_zero_nonexist_check(domain, ce); domain = domain_next(domain); } } DEBUG(DEBUG_XFRD, 1, (LOG_INFO, "axfrdel: recyclebin holds %lu bytes", (unsigned long) region_get_recycle_size(db->region))); #ifndef NDEBUG if(nsd_debug_level >= 2) region_log_stats(db->region); #endif assert(zone->soa_rrset == 0); /* keep zone->soa_nx_rrset alloced: it is reused */ assert(zone->ns_rrset == 0); assert(zone->is_secure == 0); } /* return value 0: syntaxerror,badIXFR, 1:OK, 2:done_and_skip_it */ static int apply_ixfr(namedb_type* db, FILE *in, const char* zone, uint32_t serialno, struct nsd_options* opt, uint32_t seq_nr, uint32_t seq_total, int* is_axfr, int* delete_mode, int* rr_count, udb_ptr* udbz, struct zone** zone_res, const char* patname, int* bytes, int* softfail) { uint32_t msglen, checklen, pkttype; int qcount, ancount, counter; buffer_type* packet; region_type* region; int i; uint16_t rrlen; const dname_type *dname_zone, *dname; zone_type* zone_db; /* note that errors could not really happen due to format of the * packet since xfrd has checked all dnames and RRs before commit, * this is why the errors are fatal (exit process), it must be * something internal or a bad disk or something. */ /* read ixfr packet RRs and apply to in memory db */ if(!diff_read_32(in, &pkttype) || pkttype != DIFF_PART_XXFR) { log_msg(LOG_ERR, "could not read type or wrong type"); return 0; } if(!diff_read_32(in, &msglen)) { log_msg(LOG_ERR, "could not read len"); return 0; } if(msglen < QHEADERSZ) { log_msg(LOG_ERR, "msg too short"); return 0; } region = region_create(xalloc, free); if(!region) { log_msg(LOG_ERR, "out of memory"); return 0; } packet = buffer_create(region, QIOBUFSZ); if(msglen > QIOBUFSZ) { log_msg(LOG_ERR, "msg too long"); region_destroy(region); return 0; } buffer_clear(packet); if(fread(buffer_begin(packet), msglen, 1, in) != 1) { log_msg(LOG_ERR, "short fread: %s", strerror(errno)); region_destroy(region); return 0; } buffer_set_limit(packet, msglen); /* see if check on data fails: checks that we are not reading * random garbage */ if(!diff_read_32(in, &checklen) || checklen != msglen) { log_msg(LOG_ERR, "transfer part has incorrect checkvalue"); return 0; } *bytes += msglen; dname_zone = dname_parse(region, zone); zone_db = find_or_create_zone(db, dname_zone, opt, zone, patname); if(!zone_db) { log_msg(LOG_ERR, "could not create zone %s %s", zone, patname); region_destroy(region); return 0; } *zone_res = zone_db; /* only answer section is really used, question, additional and authority section RRs are skipped */ qcount = QDCOUNT(packet); ancount = ANCOUNT(packet); buffer_skip(packet, QHEADERSZ); /* qcount should be 0 or 1 really, ancount limited by 64k packet */ if(qcount > 64 || ancount > 65530) { log_msg(LOG_ERR, "RR count impossibly high"); region_destroy(region); return 0; } /* skip queries */ for(i=0; insec3_param = NULL; #endif delete_zone_rrs(db, zone_db); if(db->udb) udb_zone_clear(db->udb, udbz); /* add everything else (incl end SOA) */ *delete_mode = 0; *is_axfr = 1; DEBUG(DEBUG_XFRD,2, (LOG_INFO, "diff: %s sawAXFR count %d, ax %d, delmode %d", dname_to_string(dname_zone, 0), *rr_count, *is_axfr, *delete_mode)); } if(*rr_count == 1 && type == TYPE_SOA) { /* if the serial no of the SOA equals the serialno, then AXFR */ size_t bufpos = buffer_position(packet); uint32_t thisserial; if(!packet_skip_dname(packet) || !packet_skip_dname(packet) || buffer_remaining(packet) < sizeof(uint32_t)*5) { log_msg(LOG_ERR, "bad xfr SOA RR formerr."); region_destroy(region); return 0; } thisserial = buffer_read_u32(packet); if(thisserial == serialno) { /* AXFR */ #ifdef NSEC3 nsec3_clear_precompile(db, zone_db); zone_db->nsec3_param = NULL; #endif delete_zone_rrs(db, zone_db); if(db->udb) udb_zone_clear(db->udb, udbz); *delete_mode = 0; *is_axfr = 1; } /* must have stuff in memory for a successful IXFR, * the serial number of the SOA has been checked * previously (by check_for_bad_serial) if it exists */ if(!*is_axfr && !domain_find_rrset(zone_db->apex, zone_db, TYPE_SOA)) { log_msg(LOG_ERR, "%s SOA serial %u is not " "in memory, skip IXFR", zone, serialno); region_destroy(region); /* break out and stop the IXFR, ignore it */ return 2; } buffer_set_position(packet, bufpos); } if(type == TYPE_SOA && !*is_axfr) { /* switch from delete-part to add-part and back again, just before soa - so it gets deleted and added too */ /* this means we switch to delete mode for the final SOA */ *delete_mode = !*delete_mode; DEBUG(DEBUG_XFRD,2, (LOG_INFO, "diff: %s IXFRswapdel count %d, ax %d, delmode %d", dname_to_string(dname_zone, 0), *rr_count, *is_axfr, *delete_mode)); } if(type == TYPE_TSIG || type == TYPE_OPT) { /* ignore pseudo RRs */ buffer_skip(packet, rrlen); continue; } DEBUG(DEBUG_XFRD,2, (LOG_INFO, "xfr %s RR dname is %s type %s", *delete_mode?"del":"add", dname_to_string(dname,0), rrtype_to_string(type))); if(*delete_mode) { /* delete this rr */ if(!*is_axfr && type == TYPE_SOA && counter==ancount-1 && seq_nr == seq_total-1) { continue; /* do not delete final SOA RR for IXFR */ } if(!delete_RR(db, dname, type, klass, packet, rrlen, zone_db, region, udbz, softfail)) { region_destroy(region); return 0; } } else { /* add this rr */ if(!add_RR(db, dname, type, klass, ttl, packet, rrlen, zone_db, udbz, softfail)) { region_destroy(region); return 0; } } } region_destroy(region); return 1; } static int check_for_bad_serial(namedb_type* db, const char* zone_str, uint32_t old_serial) { /* see if serial OK with in-memory serial */ domain_type* domain; region_type* region = region_create(xalloc, free); const dname_type* zone_name = dname_parse(region, zone_str); zone_type* zone = 0; domain = domain_table_find(db->domains, zone_name); if(domain) zone = domain_find_zone(db, domain); if(zone && zone->apex == domain && zone->soa_rrset && old_serial) { uint32_t memserial; memcpy(&memserial, rdata_atom_data(zone->soa_rrset->rrs[0].rdatas[2]), sizeof(uint32_t)); if(old_serial != ntohl(memserial)) { region_destroy(region); return 1; } } region_destroy(region); return 0; } static int apply_ixfr_for_zone(nsd_type* nsd, zone_type* zonedb, FILE* in, struct nsd_options* opt, udb_base* taskudb, udb_ptr* last_task, uint32_t xfrfilenr) { char zone_buf[3072]; char log_buf[5120]; char patname_buf[2048]; uint32_t old_serial, new_serial, num_parts, type; uint64_t time_end_0, time_start_0; uint32_t time_end_1, time_start_1; uint8_t committed; uint32_t i; int num_bytes = 0; assert(zonedb); /* read zone name and serial */ if(!diff_read_32(in, &type)) { log_msg(LOG_ERR, "diff file too short"); return 0; } if(type != DIFF_PART_XFRF) { log_msg(LOG_ERR, "xfr file has wrong format"); return 0; } /* committed and num_parts are first because they need to be * updated once the rest is written. The log buf is not certain * until its done, so at end of file. The patname is in case a * new zone is created, we know what the options-pattern is */ if(!diff_read_8(in, &committed) || !diff_read_32(in, &num_parts) || !diff_read_64(in, &time_end_0) || !diff_read_32(in, &time_end_1) || !diff_read_32(in, &old_serial) || !diff_read_32(in, &new_serial) || !diff_read_64(in, &time_start_0) || !diff_read_32(in, &time_start_1) || !diff_read_str(in, zone_buf, sizeof(zone_buf)) || !diff_read_str(in, patname_buf, sizeof(patname_buf))) { log_msg(LOG_ERR, "diff file bad commit part"); return 0; } /* has been read in completely */ if(strcmp(zone_buf, domain_to_string(zonedb->apex)) != 0) { log_msg(LOG_ERR, "file %s does not match task %s", zone_buf, domain_to_string(zonedb->apex)); return 0; } if(!committed) { log_msg(LOG_ERR, "diff file %s was not committed", zone_buf); return 0; } if(num_parts == 0) { log_msg(LOG_ERR, "diff file %s was not completed", zone_buf); return 0; } if(check_for_bad_serial(nsd->db, zone_buf, old_serial)) { DEBUG(DEBUG_XFRD,1, (LOG_ERR, "skipping diff file commit with bad serial")); return 1; } if(committed) { int is_axfr=0, delete_mode=0, rr_count=0, softfail=0; const dname_type* apex = domain_dname_const(zonedb->apex); udb_ptr z; DEBUG(DEBUG_XFRD,1, (LOG_INFO, "processing xfr: %s", zone_buf)); memset(&z, 0, sizeof(z)); /* if udb==NULL, have &z defined */ if(nsd->db->udb) { if(udb_base_get_userflags(nsd->db->udb) != 0) { log_msg(LOG_ERR, "database corrupted, cannot update"); xfrd_unlink_xfrfile(nsd, xfrfilenr); exit(1); } /* all parts were checked by xfrd before commit */ if(!udb_zone_search(nsd->db->udb, &z, dname_name(apex), apex->name_size)) { /* create it */ if(!udb_zone_create(nsd->db->udb, &z, dname_name(apex), apex->name_size)) { /* out of disk space perhaps */ log_msg(LOG_ERR, "could not udb_create_zone " "%s, disk space full?", zone_buf); return 0; } } /* set the udb dirty until we are finished applying changes */ udb_base_set_userflags(nsd->db->udb, 1); } /* read and apply all of the parts */ for(i=0; idb, in, zone_buf, new_serial, opt, i, num_parts, &is_axfr, &delete_mode, &rr_count, (nsd->db->udb?&z:NULL), &zonedb, patname_buf, &num_bytes, &softfail); assert(zonedb); if(ret == 0) { log_msg(LOG_ERR, "bad ixfr packet part %d in diff file for %s", (int)i, zone_buf); xfrd_unlink_xfrfile(nsd, xfrfilenr); /* the udb is still dirty, it is bad */ exit(1); } else if(ret == 2) { break; } } if(nsd->db->udb) udb_base_set_userflags(nsd->db->udb, 0); /* read the final log_str: but do not fail on it */ if(!diff_read_str(in, log_buf, sizeof(log_buf))) { log_msg(LOG_ERR, "could not read log for transfer %s", zone_buf); snprintf(log_buf, sizeof(log_buf), "error reading log"); } #ifdef NSEC3 if(zonedb) prehash_zone(nsd->db, zonedb); #endif /* NSEC3 */ zonedb->is_changed = 1; if(nsd->db->udb) { assert(z.base); ZONE(&z)->is_changed = 1; ZONE(&z)->mtime = time_end_0; ZONE(&z)->mtime_nsec = time_end_1*1000; udb_zone_set_log_str(nsd->db->udb, &z, log_buf); udb_zone_set_file_str(nsd->db->udb, &z, NULL); udb_ptr_unlink(&z, nsd->db->udb); } else { zonedb->mtime.tv_sec = time_end_0; zonedb->mtime.tv_nsec = time_end_1*1000; if(zonedb->logstr) region_recycle(nsd->db->region, zonedb->logstr, strlen(zonedb->logstr)+1); zonedb->logstr = region_strdup(nsd->db->region, log_buf); if(zonedb->filename) region_recycle(nsd->db->region, zonedb->filename, strlen(zonedb->filename)+1); zonedb->filename = NULL; } if(softfail && taskudb && !is_axfr) { log_msg(LOG_ERR, "Failed to apply IXFR cleanly " "(deletes nonexistent RRs, adds existing RRs). " "Zone %s contents is different from master, " "starting AXFR. Transfer %s", zone_buf, log_buf); /* add/del failures in IXFR, get an AXFR */ task_new_soainfo(taskudb, last_task, zonedb, 1); } else { if(taskudb) task_new_soainfo(taskudb, last_task, zonedb, 0); } if(1 <= verbosity) { double elapsed = (double)(time_end_0 - time_start_0)+ (double)((double)time_end_1 -(double)time_start_1) / 1000000.0; VERBOSITY(1, (LOG_INFO, "zone %s %s of %d bytes in %g seconds", zone_buf, log_buf, num_bytes, elapsed)); } } else { DEBUG(DEBUG_XFRD,1, (LOG_INFO, "skipping xfr: %s", log_buf)); } return 1; } struct udb_base* task_file_create(const char* file) { return udb_base_create_new(file, &namedb_walkfunc, NULL); } static int task_create_new_elem(struct udb_base* udb, udb_ptr* last, udb_ptr* e, size_t sz, const dname_type* zname) { if(!udb_ptr_alloc_space(e, udb, udb_chunk_type_task, sz)) { return 0; } if(udb_ptr_is_null(last)) { udb_base_set_userdata(udb, e->data); } else { udb_rptr_set_ptr(&TASKLIST(last)->next, udb, e); } udb_ptr_set_ptr(last, udb, e); /* fill in tasklist item */ udb_rel_ptr_init(&TASKLIST(e)->next); TASKLIST(e)->size = sz; TASKLIST(e)->oldserial = 0; TASKLIST(e)->newserial = 0; TASKLIST(e)->yesno = 0; if(zname) { memmove(TASKLIST(e)->zname, zname, dname_total_size(zname)); } return 1; } void task_new_soainfo(struct udb_base* udb, udb_ptr* last, struct zone* z, int gone) { /* calculate size */ udb_ptr e; size_t sz; const dname_type* apex, *ns, *em; if(!z || !z->apex || !domain_dname(z->apex)) return; /* safety check */ DEBUG(DEBUG_IPC,1, (LOG_INFO, "nsd: add soa info for zone %s", domain_to_string(z->apex))); apex = domain_dname(z->apex); sz = sizeof(struct task_list_d) + dname_total_size(apex); if(z->soa_rrset && !gone) { ns = domain_dname(rdata_atom_domain( z->soa_rrset->rrs[0].rdatas[0])); em = domain_dname(rdata_atom_domain( z->soa_rrset->rrs[0].rdatas[1])); sz += sizeof(uint32_t)*6 + sizeof(uint8_t)*2 + ns->name_size + em->name_size; } else { ns = 0; em = 0; } /* create new task_list item */ if(!task_create_new_elem(udb, last, &e, sz, apex)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add SOAINFO"); return; } TASKLIST(&e)->task_type = task_soa_info; if(z->soa_rrset && !gone) { uint32_t ttl = htonl(z->soa_rrset->rrs[0].ttl); uint8_t* p = (uint8_t*)TASKLIST(&e)->zname; p += dname_total_size(apex); memmove(p, &ttl, sizeof(uint32_t)); p += sizeof(uint32_t); memmove(p, &ns->name_size, sizeof(uint8_t)); p += sizeof(uint8_t); memmove(p, dname_name(ns), ns->name_size); p += ns->name_size; memmove(p, &em->name_size, sizeof(uint8_t)); p += sizeof(uint8_t); memmove(p, dname_name(em), em->name_size); p += em->name_size; memmove(p, rdata_atom_data(z->soa_rrset->rrs[0].rdatas[2]), sizeof(uint32_t)); p += sizeof(uint32_t); memmove(p, rdata_atom_data(z->soa_rrset->rrs[0].rdatas[3]), sizeof(uint32_t)); p += sizeof(uint32_t); memmove(p, rdata_atom_data(z->soa_rrset->rrs[0].rdatas[4]), sizeof(uint32_t)); p += sizeof(uint32_t); memmove(p, rdata_atom_data(z->soa_rrset->rrs[0].rdatas[5]), sizeof(uint32_t)); p += sizeof(uint32_t); memmove(p, rdata_atom_data(z->soa_rrset->rrs[0].rdatas[6]), sizeof(uint32_t)); } udb_ptr_unlink(&e, udb); } void task_process_sync(struct udb_base* taskudb) { /* need to sync before other process uses the mmap? */ DEBUG(DEBUG_IPC,1, (LOG_INFO, "task procsync %s size %d", taskudb->fname, (int)taskudb->base_size)); (void)taskudb; } void task_remap(struct udb_base* taskudb) { DEBUG(DEBUG_IPC,1, (LOG_INFO, "task remap %s size %d", taskudb->fname, (int)taskudb->glob_data->fsize)); udb_base_remap_process(taskudb); } void task_clear(struct udb_base* taskudb) { udb_ptr t, n; udb_ptr_new(&t, taskudb, udb_base_get_userdata(taskudb)); udb_base_set_userdata(taskudb, 0); udb_ptr_init(&n, taskudb); while(!udb_ptr_is_null(&t)) { udb_ptr_set_rptr(&n, taskudb, &TASKLIST(&t)->next); udb_rptr_zero(&TASKLIST(&t)->next, taskudb); udb_ptr_free_space(&t, taskudb, TASKLIST(&t)->size); udb_ptr_set_ptr(&t, taskudb, &n); } udb_ptr_unlink(&t, taskudb); udb_ptr_unlink(&n, taskudb); } void task_new_expire(struct udb_base* udb, udb_ptr* last, const struct dname* z, int expired) { udb_ptr e; if(!z) return; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add expire info for zone %s", dname_to_string(z,NULL))); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d)+ dname_total_size(z), z)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add expire"); return; } TASKLIST(&e)->task_type = task_expire; TASKLIST(&e)->yesno = expired; udb_ptr_unlink(&e, udb); } void task_new_check_zonefiles(udb_base* udb, udb_ptr* last, const dname_type* zone) { udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task checkzonefiles")); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d) + (zone?dname_total_size(zone):0), zone)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add check_zones"); return; } TASKLIST(&e)->task_type = task_check_zonefiles; TASKLIST(&e)->yesno = (zone!=NULL); udb_ptr_unlink(&e, udb); } void task_new_write_zonefiles(udb_base* udb, udb_ptr* last, const dname_type* zone) { udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task writezonefiles")); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d) + (zone?dname_total_size(zone):0), zone)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add writezones"); return; } TASKLIST(&e)->task_type = task_write_zonefiles; TASKLIST(&e)->yesno = (zone!=NULL); udb_ptr_unlink(&e, udb); } void task_new_set_verbosity(udb_base* udb, udb_ptr* last, int v) { udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task set_verbosity")); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d), NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add set_v"); return; } TASKLIST(&e)->task_type = task_set_verbosity; TASKLIST(&e)->yesno = v; udb_ptr_unlink(&e, udb); } #ifdef BIND8_STATS void* task_new_stat_info(udb_base* udb, udb_ptr* last, struct nsdst* stat, size_t child_count) { void* p; udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task stat_info")); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d)+ sizeof(*stat) + sizeof(stc_type)*child_count, NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add stati"); return NULL; } TASKLIST(&e)->task_type = task_stat_info; p = TASKLIST(&e)->zname; memcpy(p, stat, sizeof(*stat)); udb_ptr_unlink(&e, udb); return (char*)p + sizeof(*stat); } #endif /* BIND8_STATS */ void task_new_add_zone(udb_base* udb, udb_ptr* last, const char* zone, const char* pattern, unsigned zonestatid) { size_t zlen = strlen(zone); size_t plen = strlen(pattern); void *p; udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task addzone %s %s", zone, pattern)); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d)+ zlen + 1 + plen + 1, NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add addz"); return; } TASKLIST(&e)->task_type = task_add_zone; TASKLIST(&e)->yesno = zonestatid; p = TASKLIST(&e)->zname; memcpy(p, zone, zlen+1); memmove((char*)p+zlen+1, pattern, plen+1); udb_ptr_unlink(&e, udb); } void task_new_del_zone(udb_base* udb, udb_ptr* last, const dname_type* dname) { udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task delzone %s", dname_to_string(dname, 0))); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d) +dname_total_size(dname), dname)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add delz"); return; } TASKLIST(&e)->task_type = task_del_zone; udb_ptr_unlink(&e, udb); } void task_new_add_key(udb_base* udb, udb_ptr* last, struct key_options* key) { char* p; udb_ptr e; assert(key->name && key->algorithm && key->secret); DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task addkey")); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d) +strlen(key->name)+1+strlen(key->algorithm)+1+ strlen(key->secret)+1, NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add addk"); return; } TASKLIST(&e)->task_type = task_add_key; p = (char*)TASKLIST(&e)->zname; memmove(p, key->name, strlen(key->name)+1); p+=strlen(key->name)+1; memmove(p, key->algorithm, strlen(key->algorithm)+1); p+=strlen(key->algorithm)+1; memmove(p, key->secret, strlen(key->secret)+1); udb_ptr_unlink(&e, udb); } void task_new_del_key(udb_base* udb, udb_ptr* last, const char* name) { char* p; udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task delkey")); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d) +strlen(name)+1, NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add delk"); return; } TASKLIST(&e)->task_type = task_del_key; p = (char*)TASKLIST(&e)->zname; memmove(p, name, strlen(name)+1); udb_ptr_unlink(&e, udb); } void task_new_add_pattern(udb_base* udb, udb_ptr* last, struct pattern_options* p) { region_type* temp; buffer_type* buffer; udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task addpattern %s", p->pname)); temp = region_create(xalloc, free); buffer = buffer_create(temp, 4096); pattern_options_marshal(buffer, p); buffer_flip(buffer); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d) + buffer_limit(buffer), NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add addp"); region_destroy(temp); return; } TASKLIST(&e)->task_type = task_add_pattern; TASKLIST(&e)->yesno = buffer_limit(buffer); memmove(TASKLIST(&e)->zname, buffer_begin(buffer), buffer_limit(buffer)); udb_ptr_unlink(&e, udb); region_destroy(temp); } void task_new_del_pattern(udb_base* udb, udb_ptr* last, const char* name) { char* p; udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task delpattern %s", name)); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d) +strlen(name)+1, NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add delp"); return; } TASKLIST(&e)->task_type = task_del_pattern; p = (char*)TASKLIST(&e)->zname; memmove(p, name, strlen(name)+1); udb_ptr_unlink(&e, udb); } void task_new_opt_change(udb_base* udb, udb_ptr* last, struct nsd_options* opt) { udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task opt_change")); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d), NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add o_c"); return; } TASKLIST(&e)->task_type = task_opt_change; #ifdef RATELIMIT TASKLIST(&e)->oldserial = opt->rrl_ratelimit; TASKLIST(&e)->newserial = opt->rrl_whitelist_ratelimit; TASKLIST(&e)->yesno = (uint64_t) opt->rrl_slip; #else (void)opt; #endif udb_ptr_unlink(&e, udb); } void task_new_zonestat_inc(udb_base* udb, udb_ptr* last, unsigned sz) { udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task zonestat_inc")); if(sz == 0) return; /* no need to decrease to 0 */ if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d), NULL)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add z_i"); return; } TASKLIST(&e)->task_type = task_zonestat_inc; TASKLIST(&e)->oldserial = (uint32_t)sz; udb_ptr_unlink(&e, udb); } int task_new_apply_xfr(udb_base* udb, udb_ptr* last, const dname_type* dname, uint32_t old_serial, uint32_t new_serial, uint64_t filenumber) { udb_ptr e; DEBUG(DEBUG_IPC,1, (LOG_INFO, "add task apply_xfr")); if(!task_create_new_elem(udb, last, &e, sizeof(struct task_list_d) +dname_total_size(dname), dname)) { log_msg(LOG_ERR, "tasklist: out of space, cannot add applyxfr"); return 0; } TASKLIST(&e)->oldserial = old_serial; TASKLIST(&e)->newserial = new_serial; TASKLIST(&e)->yesno = filenumber; TASKLIST(&e)->task_type = task_apply_xfr; udb_ptr_unlink(&e, udb); return 1; } void task_process_expire(namedb_type* db, struct task_list_d* task) { uint8_t ok; zone_type* z = namedb_find_zone(db, task->zname); assert(task->task_type == task_expire); if(!z) { DEBUG(DEBUG_IPC, 1, (LOG_WARNING, "zone %s %s but not in zonetree", dname_to_string(task->zname, NULL), task->yesno?"expired":"unexpired")); return; } DEBUG(DEBUG_IPC,1, (LOG_INFO, "xfrd: expire task zone %s %s", dname_to_string(task->zname,0), task->yesno?"expired":"unexpired")); /* find zone, set expire flag */ ok = !task->yesno; /* only update zone->is_ok if needed to minimize copy-on-write * of memory pages shared after fork() */ if(ok && !z->is_ok) z->is_ok = 1; else if(!ok && z->is_ok) z->is_ok = 0; } static void task_process_set_verbosity(struct task_list_d* task) { DEBUG(DEBUG_IPC,1, (LOG_INFO, "verbosity task %d", (int)task->yesno)); verbosity = task->yesno; } static void task_process_checkzones(struct nsd* nsd, udb_base* udb, udb_ptr* last_task, struct task_list_d* task) { /* on SIGHUP check if zone-text-files changed and if so, * reread. When from xfrd-reload, no need to fstat the files */ if(task->yesno) { struct zone_options* zo = zone_options_find(nsd->options, task->zname); if(zo) namedb_check_zonefile(nsd, udb, last_task, zo); } else { /* check all zones */ namedb_check_zonefiles(nsd, nsd->options, udb, last_task); } } static void task_process_writezones(struct nsd* nsd, struct task_list_d* task) { if(task->yesno) { struct zone_options* zo = zone_options_find(nsd->options, task->zname); if(zo) namedb_write_zonefile(nsd, zo); } else { namedb_write_zonefiles(nsd, nsd->options); } } static void task_process_add_zone(struct nsd* nsd, udb_base* udb, udb_ptr* last_task, struct task_list_d* task) { zone_type* z; const dname_type* zdname; const char* zname = (const char*)task->zname; const char* pname = zname + strlen(zname)+1; DEBUG(DEBUG_IPC,1, (LOG_INFO, "addzone task %s %s", zname, pname)); zdname = dname_parse(nsd->db->region, zname); if(!zdname) { log_msg(LOG_ERR, "can not parse zone name %s", zname); return; } /* create zone */ z = find_or_create_zone(nsd->db, zdname, nsd->options, zname, pname); if(!z) { region_recycle(nsd->db->region, (void*)zdname, dname_total_size(zdname)); log_msg(LOG_ERR, "can not add zone %s %s", zname, pname); return; } z->zonestatid = (unsigned)task->yesno; /* if zone is empty, attempt to read the zonefile from disk (if any) */ if(!z->soa_rrset && z->opts->pattern->zonefile) { namedb_read_zonefile(nsd, z, udb, last_task); } } static void task_process_del_zone(struct nsd* nsd, struct task_list_d* task) { zone_type* zone; struct zone_options* zopt; DEBUG(DEBUG_IPC,1, (LOG_INFO, "delzone task %s", dname_to_string( task->zname, NULL))); zone = namedb_find_zone(nsd->db, task->zname); if(!zone) return; #ifdef NSEC3 nsec3_clear_precompile(nsd->db, zone); zone->nsec3_param = NULL; #endif delete_zone_rrs(nsd->db, zone); if(nsd->db->udb) { udb_ptr udbz; if(udb_zone_search(nsd->db->udb, &udbz, dname_name(task->zname), task->zname->name_size)) { udb_zone_delete(nsd->db->udb, &udbz); udb_ptr_unlink(&udbz, nsd->db->udb); } } /* remove from zonetree, apex, soa */ zopt = zone->opts; namedb_zone_delete(nsd->db, zone); /* remove from options (zone_list already edited by xfrd) */ zone_options_delete(nsd->options, zopt); } static void task_process_add_key(struct nsd* nsd, struct task_list_d* task) { struct key_options key; key.name = (char*)task->zname; DEBUG(DEBUG_IPC,1, (LOG_INFO, "addkey task %s", key.name)); key.algorithm = key.name + strlen(key.name)+1; key.secret = key.algorithm + strlen(key.algorithm)+1; key_options_add_modify(nsd->options, &key); memset(key.secret, 0xdd, strlen(key.secret)); /* wipe secret */ } static void task_process_del_key(struct nsd* nsd, struct task_list_d* task) { char* name = (char*)task->zname; DEBUG(DEBUG_IPC,1, (LOG_INFO, "delkey task %s", name)); /* this is reload and nothing is using the TSIG key right now */ key_options_remove(nsd->options, name); } static void task_process_add_pattern(struct nsd* nsd, struct task_list_d* task) { region_type* temp = region_create(xalloc, free); buffer_type buffer; struct pattern_options *pat; buffer_create_from(&buffer, task->zname, task->yesno); pat = pattern_options_unmarshal(temp, &buffer); DEBUG(DEBUG_IPC,1, (LOG_INFO, "addpattern task %s", pat->pname)); pattern_options_add_modify(nsd->options, pat); region_destroy(temp); } static void task_process_del_pattern(struct nsd* nsd, struct task_list_d* task) { char* name = (char*)task->zname; DEBUG(DEBUG_IPC,1, (LOG_INFO, "delpattern task %s", name)); pattern_options_remove(nsd->options, name); } static void task_process_opt_change(struct nsd* nsd, struct task_list_d* task) { DEBUG(DEBUG_IPC,1, (LOG_INFO, "optchange task")); #ifdef RATELIMIT nsd->options->rrl_ratelimit = task->oldserial; nsd->options->rrl_whitelist_ratelimit = task->newserial; nsd->options->rrl_slip = task->yesno; rrl_set_limit(nsd->options->rrl_ratelimit, nsd->options->rrl_whitelist_ratelimit, nsd->options->rrl_slip); #else (void)nsd; (void)task; #endif } #ifdef USE_ZONE_STATS static void task_process_zonestat_inc(struct nsd* nsd, udb_base* udb, udb_ptr *last_task, struct task_list_d* task) { DEBUG(DEBUG_IPC,1, (LOG_INFO, "zonestat_inc task %u", (unsigned)task->oldserial)); nsd->zonestatdesired = (unsigned)task->oldserial; /* send echo to xfrd to increment on its end */ task_new_zonestat_inc(udb, last_task, nsd->zonestatdesired); } #endif static void task_process_apply_xfr(struct nsd* nsd, udb_base* udb, udb_ptr *last_task, udb_ptr* task) { /* we have to use an udb_ptr task here, because the apply_xfr procedure * appends soa_info which may remap and change the pointer. */ zone_type* zone; FILE* df; DEBUG(DEBUG_IPC,1, (LOG_INFO, "applyxfr task %s", dname_to_string( TASKLIST(task)->zname, NULL))); zone = namedb_find_zone(nsd->db, TASKLIST(task)->zname); if(!zone) { /* assume the zone has been deleted and a zone transfer was * still waiting to be processed */ xfrd_unlink_xfrfile(nsd, TASKLIST(task)->yesno); return; } /* apply the XFR */ /* oldserial, newserial, yesno is filenumber */ df = xfrd_open_xfrfile(nsd, TASKLIST(task)->yesno, "r"); if(!df) { /* could not open file to update */ /* there is no reply to xfrd failed-update, * because xfrd has a scan for apply-failures. */ xfrd_unlink_xfrfile(nsd, TASKLIST(task)->yesno); return; } /* read and apply zone transfer */ if(!apply_ixfr_for_zone(nsd, zone, df, nsd->options, udb, last_task, TASKLIST(task)->yesno)) { /* there is no reply to xfrd failed-update, * because xfrd has a scan for apply-failures. */ } fclose(df); xfrd_unlink_xfrfile(nsd, TASKLIST(task)->yesno); } void task_process_in_reload(struct nsd* nsd, udb_base* udb, udb_ptr *last_task, udb_ptr* task) { switch(TASKLIST(task)->task_type) { case task_expire: task_process_expire(nsd->db, TASKLIST(task)); break; case task_check_zonefiles: task_process_checkzones(nsd, udb, last_task, TASKLIST(task)); break; case task_write_zonefiles: task_process_writezones(nsd, TASKLIST(task)); break; case task_set_verbosity: task_process_set_verbosity(TASKLIST(task)); break; case task_add_zone: task_process_add_zone(nsd, udb, last_task, TASKLIST(task)); break; case task_del_zone: task_process_del_zone(nsd, TASKLIST(task)); break; case task_add_key: task_process_add_key(nsd, TASKLIST(task)); break; case task_del_key: task_process_del_key(nsd, TASKLIST(task)); break; case task_add_pattern: task_process_add_pattern(nsd, TASKLIST(task)); break; case task_del_pattern: task_process_del_pattern(nsd, TASKLIST(task)); break; case task_opt_change: task_process_opt_change(nsd, TASKLIST(task)); break; #ifdef USE_ZONE_STATS case task_zonestat_inc: task_process_zonestat_inc(nsd, udb, last_task, TASKLIST(task)); break; #endif case task_apply_xfr: task_process_apply_xfr(nsd, udb, last_task, task); break; default: log_msg(LOG_WARNING, "unhandled task in reload type %d", (int)TASKLIST(task)->task_type); break; } udb_ptr_free_space(task, udb, TASKLIST(task)->size); }