From 8daa83a594a2e98f39d764422bfbdbc62c9efd44 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Fri, 19 Apr 2024 19:20:00 +0200 Subject: Adding upstream version 2:4.20.0+dfsg. Signed-off-by: Daniel Baumann --- source3/lib/dbwrap/dbwrap_ctdb.c | 1985 +++++++++++++++++++++++++++++++++++++ source3/lib/dbwrap/dbwrap_ctdb.h | 40 + source3/lib/dbwrap/dbwrap_open.c | 197 ++++ source3/lib/dbwrap/dbwrap_open.h | 45 + source3/lib/dbwrap/dbwrap_watch.c | 1285 ++++++++++++++++++++++++ source3/lib/dbwrap/dbwrap_watch.h | 45 + 6 files changed, 3597 insertions(+) create mode 100644 source3/lib/dbwrap/dbwrap_ctdb.c create mode 100644 source3/lib/dbwrap/dbwrap_ctdb.h create mode 100644 source3/lib/dbwrap/dbwrap_open.c create mode 100644 source3/lib/dbwrap/dbwrap_open.h create mode 100644 source3/lib/dbwrap/dbwrap_watch.c create mode 100644 source3/lib/dbwrap/dbwrap_watch.h (limited to 'source3/lib/dbwrap') diff --git a/source3/lib/dbwrap/dbwrap_ctdb.c b/source3/lib/dbwrap/dbwrap_ctdb.c new file mode 100644 index 0000000..46165e8 --- /dev/null +++ b/source3/lib/dbwrap/dbwrap_ctdb.c @@ -0,0 +1,1985 @@ +/* + Unix SMB/CIFS implementation. + Database interface wrapper around ctdbd + Copyright (C) Volker Lendecke 2007-2009 + Copyright (C) Michael Adam 2009 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#include "includes.h" +#include "system/filesys.h" +#include "lib/tdb_wrap/tdb_wrap.h" +#include "util_tdb.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap/dbwrap_ctdb.h" +#include "dbwrap/dbwrap_rbt.h" +#include "lib/param/param.h" + +#include "ctdb/include/ctdb_protocol.h" +#include "ctdbd_conn.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap/dbwrap_private.h" +#include "dbwrap/dbwrap_ctdb.h" +#include "g_lock.h" +#include "messages.h" +#include "messages_ctdb.h" +#include "lib/cluster_support.h" +#include "lib/util/tevent_ntstatus.h" + +struct db_ctdb_transaction_handle { + struct db_ctdb_ctx *ctx; + /* + * we store the writes done under a transaction: + */ + struct ctdb_marshall_buffer *m_write; + uint32_t nesting; + bool nested_cancel; + char *lock_name; +}; + +struct db_ctdb_ctx { + struct db_context *db; + struct tdb_wrap *wtdb; + uint32_t db_id; + struct db_ctdb_transaction_handle *transaction; + struct g_lock_ctx *lock_ctx; + + /* thresholds for warning messages */ + int warn_unlock_msecs; + int warn_migrate_msecs; + int warn_migrate_attempts; + int warn_locktime_msecs; +}; + +struct db_ctdb_rec { + struct db_ctdb_ctx *ctdb_ctx; + struct ctdb_ltdb_header header; + struct timeval lock_time; +}; + +struct ctdb_async_ctx { + bool initialized; + struct ctdbd_connection *async_conn; +}; + +static struct ctdb_async_ctx ctdb_async_ctx; + +static int ctdb_async_ctx_init_internal(TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + bool reinit) +{ + int ret; + + if (reinit) { + TALLOC_FREE(ctdb_async_ctx.async_conn); + ctdb_async_ctx.initialized = false; + } + + if (ctdb_async_ctx.initialized) { + return 0; + } + + become_root(); + ret = ctdbd_init_async_connection( + mem_ctx, + lp_ctdbd_socket(), + lp_ctdb_timeout(), + &ctdb_async_ctx.async_conn); + unbecome_root(); + + if (ret != 0) { + DBG_ERR("ctdbd_init_async_connection(%s, timeout=%d) " + "failed: ret=%d %s\n", + lp_ctdbd_socket(), + lp_ctdb_timeout(), + ret, strerror(ret)); + return ret; + } + + SMB_ASSERT(ctdb_async_ctx.async_conn != NULL); + + ctdb_async_ctx.initialized = true; + return 0; +} + +static int ctdb_async_ctx_init(TALLOC_CTX *mem_ctx, struct tevent_context *ev) +{ + return ctdb_async_ctx_init_internal(mem_ctx, ev, false); +} + +int ctdb_async_ctx_reinit(TALLOC_CTX *mem_ctx, struct tevent_context *ev) +{ + return ctdb_async_ctx_init_internal(mem_ctx, ev, true); +} + +static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb) +{ + enum TDB_ERROR tret = tdb_error(tdb); + + return map_nt_error_from_tdb(tret); +} + +struct db_ctdb_ltdb_parse_state { + void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header, + TDB_DATA data, void *private_data); + void *private_data; +}; + +static int db_ctdb_ltdb_parser(TDB_DATA key, TDB_DATA data, + void *private_data) +{ + struct db_ctdb_ltdb_parse_state *state = + (struct db_ctdb_ltdb_parse_state *)private_data; + + if (data.dsize < sizeof(struct ctdb_ltdb_header)) { + return -1; + } + + state->parser( + key, (struct ctdb_ltdb_header *)data.dptr, + make_tdb_data(data.dptr + sizeof(struct ctdb_ltdb_header), + data.dsize - sizeof(struct ctdb_ltdb_header)), + state->private_data); + return 0; +} + +static NTSTATUS db_ctdb_ltdb_parse( + struct db_ctdb_ctx *db, TDB_DATA key, + void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header, + TDB_DATA data, void *private_data), + void *private_data) +{ + struct db_ctdb_ltdb_parse_state state; + int ret; + + state.parser = parser; + state.private_data = private_data; + + ret = tdb_parse_record(db->wtdb->tdb, key, db_ctdb_ltdb_parser, + &state); + if (ret == -1) { + return NT_STATUS_NOT_FOUND; + } + return NT_STATUS_OK; +} + +/* + * Store a record together with the ctdb record header + * in the local copy of the database. + */ +static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db, + TDB_DATA key, + struct ctdb_ltdb_header *header, + const TDB_DATA *dbufs, int num_dbufs) +{ + TDB_DATA recs[num_dbufs+1]; + int ret; + + recs[0] = (TDB_DATA) { .dptr = (uint8_t *)header, + .dsize = sizeof(struct ctdb_ltdb_header) }; + memcpy(&recs[1], dbufs, sizeof(TDB_DATA) * num_dbufs); + + ret = tdb_storev(db->wtdb->tdb, key, recs, num_dbufs + 1, TDB_REPLACE); + + return (ret == 0) ? NT_STATUS_OK + : tdb_error_to_ntstatus(db->wtdb->tdb); + +} + +/* + form a ctdb_rec_data record from a key/data pair + */ +static struct ctdb_rec_data_old *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid, + TDB_DATA key, + struct ctdb_ltdb_header *header, + TDB_DATA data) +{ + size_t length; + struct ctdb_rec_data_old *d; + + length = offsetof(struct ctdb_rec_data_old, data) + key.dsize + + data.dsize + sizeof(*header); + d = (struct ctdb_rec_data_old *)talloc_size(mem_ctx, length); + if (d == NULL) { + return NULL; + } + d->length = length; + d->reqid = reqid; + d->keylen = key.dsize; + memcpy(&d->data[0], key.dptr, key.dsize); + + d->datalen = data.dsize + sizeof(*header); + memcpy(&d->data[key.dsize], header, sizeof(*header)); + memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize); + return d; +} + + +/* helper function for marshalling multiple records */ +static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx, + struct ctdb_marshall_buffer *m, + uint32_t db_id, + uint32_t reqid, + TDB_DATA key, + struct ctdb_ltdb_header *header, + TDB_DATA data) +{ + struct ctdb_rec_data_old *r; + size_t m_size, r_size; + struct ctdb_marshall_buffer *m2 = NULL; + + r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data); + if (r == NULL) { + talloc_free(m); + return NULL; + } + + if (m == NULL) { + m = (struct ctdb_marshall_buffer *)talloc_zero_size( + mem_ctx, offsetof(struct ctdb_marshall_buffer, data)); + if (m == NULL) { + goto done; + } + m->db_id = db_id; + } + + m_size = talloc_get_size(m); + r_size = talloc_get_size(r); + + m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size( + mem_ctx, m, m_size + r_size); + if (m2 == NULL) { + talloc_free(m); + goto done; + } + + memcpy(m_size + (uint8_t *)m2, r, r_size); + + m2->count++; + +done: + talloc_free(r); + return m2; +} + +/* we've finished marshalling, return a data blob with the marshalled records */ +static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m) +{ + TDB_DATA data; + data.dptr = (uint8_t *)m; + data.dsize = talloc_get_size(m); + return data; +} + +/* + loop over a marshalling buffer + + - pass r==NULL to start + - loop the number of times indicated by m->count +*/ +static struct ctdb_rec_data_old *db_ctdb_marshall_loop_next_key( + struct ctdb_marshall_buffer *m, struct ctdb_rec_data_old *r, TDB_DATA *key) +{ + if (r == NULL) { + r = (struct ctdb_rec_data_old *)&m->data[0]; + } else { + r = (struct ctdb_rec_data_old *)(r->length + (uint8_t *)r); + } + + key->dptr = &r->data[0]; + key->dsize = r->keylen; + return r; +} + +static bool db_ctdb_marshall_buf_parse( + struct ctdb_rec_data_old *r, uint32_t *reqid, + struct ctdb_ltdb_header **header, TDB_DATA *data) +{ + if (r->datalen < sizeof(struct ctdb_ltdb_header)) { + return false; + } + + *reqid = r->reqid; + + data->dptr = &r->data[r->keylen] + sizeof(struct ctdb_ltdb_header); + data->dsize = r->datalen - sizeof(struct ctdb_ltdb_header); + + *header = (struct ctdb_ltdb_header *)&r->data[r->keylen]; + + return true; +} + +/** + * CTDB transaction destructor + */ +static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h) +{ + NTSTATUS status; + + status = g_lock_unlock(h->ctx->lock_ctx, + string_term_tdb_data(h->lock_name)); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0, ("g_lock_unlock failed for %s: %s\n", h->lock_name, + nt_errstr(status))); + return -1; + } + return 0; +} + +/** + * CTDB dbwrap API: transaction_start function + * starts a transaction on a persistent database + */ +static int db_ctdb_transaction_start(struct db_context *db) +{ + struct db_ctdb_transaction_handle *h; + NTSTATUS status; + struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data, + struct db_ctdb_ctx); + + if (!db->persistent) { + DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n", + ctx->db_id)); + return -1; + } + + if (ctx->transaction) { + ctx->transaction->nesting++; + DEBUG(5, (__location__ " transaction start on db 0x%08x: nesting %d -> %d\n", + ctx->db_id, ctx->transaction->nesting - 1, ctx->transaction->nesting)); + return 0; + } + + h = talloc_zero(db, struct db_ctdb_transaction_handle); + if (h == NULL) { + DEBUG(0,(__location__ " oom for transaction handle\n")); + return -1; + } + + h->ctx = ctx; + + h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x", + (unsigned int)ctx->db_id); + if (h->lock_name == NULL) { + DEBUG(0, ("talloc_asprintf failed\n")); + TALLOC_FREE(h); + return -1; + } + + /* + * Wait a day, i.e. forever... + */ + status = g_lock_lock(ctx->lock_ctx, string_term_tdb_data(h->lock_name), + G_LOCK_WRITE, timeval_set(86400, 0), NULL, NULL); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status))); + TALLOC_FREE(h); + return -1; + } + + talloc_set_destructor(h, db_ctdb_transaction_destructor); + + ctx->transaction = h; + + DEBUG(5,(__location__ " transaction started on db 0x%08x\n", ctx->db_id)); + + return 0; +} + +static bool parse_newest_in_marshall_buffer( + struct ctdb_marshall_buffer *buf, TDB_DATA key, + void (*parser)(TDB_DATA key, struct ctdb_ltdb_header *header, + TDB_DATA data, void *private_data), + void *private_data) +{ + struct ctdb_rec_data_old *rec = NULL; + struct ctdb_ltdb_header *h = NULL; + TDB_DATA data; + uint32_t i; + + if (buf == NULL) { + return false; + } + + /* + * Walk the list of records written during this + * transaction. If we want to read one we have already + * written, return the last written sample. Thus we do not do + * a "break;" for the first hit, this record might have been + * overwritten later. + */ + + for (i=0; icount; i++) { + TDB_DATA tkey; + uint32_t reqid; + + rec = db_ctdb_marshall_loop_next_key(buf, rec, &tkey); + if (rec == NULL) { + return false; + } + + if (!tdb_data_equal(key, tkey)) { + continue; + } + + if (!db_ctdb_marshall_buf_parse(rec, &reqid, &h, &data)) { + return false; + } + } + + if (h == NULL) { + return false; + } + + parser(key, h, data, private_data); + + return true; +} + +struct pull_newest_from_marshall_buffer_state { + struct ctdb_ltdb_header *pheader; + TALLOC_CTX *mem_ctx; + TDB_DATA *pdata; +}; + +static void pull_newest_from_marshall_buffer_parser( + TDB_DATA key, struct ctdb_ltdb_header *header, + TDB_DATA data, void *private_data) +{ + struct pull_newest_from_marshall_buffer_state *state = + (struct pull_newest_from_marshall_buffer_state *)private_data; + + if (state->pheader != NULL) { + memcpy(state->pheader, header, sizeof(*state->pheader)); + } + if (state->pdata != NULL) { + state->pdata->dsize = data.dsize; + state->pdata->dptr = (uint8_t *)talloc_memdup( + state->mem_ctx, data.dptr, data.dsize); + } +} + +static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf, + TDB_DATA key, + struct ctdb_ltdb_header *pheader, + TALLOC_CTX *mem_ctx, + TDB_DATA *pdata) +{ + struct pull_newest_from_marshall_buffer_state state; + + state.pheader = pheader; + state.mem_ctx = mem_ctx; + state.pdata = pdata; + + if (!parse_newest_in_marshall_buffer( + buf, key, pull_newest_from_marshall_buffer_parser, + &state)) { + return false; + } + if ((pdata != NULL) && (pdata->dsize != 0) && (pdata->dptr == NULL)) { + /* ENOMEM */ + return false; + } + return true; +} + +static NTSTATUS db_ctdb_storev_transaction(struct db_record *rec, + const TDB_DATA *dbufs, int num_dbufs, + int flag); +static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec); + +static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx, + TALLOC_CTX *mem_ctx, + TDB_DATA key) +{ + struct db_record *result; + TDB_DATA ctdb_data; + + if (!(result = talloc(mem_ctx, struct db_record))) { + DEBUG(0, ("talloc failed\n")); + return NULL; + } + + result->db = ctx->db; + result->private_data = ctx->transaction; + + result->key.dsize = key.dsize; + result->key.dptr = (uint8_t *)talloc_memdup(result, key.dptr, + key.dsize); + if (result->key.dptr == NULL) { + DEBUG(0, ("talloc failed\n")); + TALLOC_FREE(result); + return NULL; + } + + result->storev = db_ctdb_storev_transaction; + result->delete_rec = db_ctdb_delete_transaction; + + if (ctx->transaction == NULL) { + DEBUG(0, ("no transaction available\n")); + TALLOC_FREE(result); + return NULL; + } + if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key, + NULL, result, &result->value)) { + result->value_valid = true; + return result; + } + + ctdb_data = tdb_fetch(ctx->wtdb->tdb, key); + if (ctdb_data.dptr == NULL) { + /* create the record */ + result->value = tdb_null; + result->value_valid = true; + return result; + } + + result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header); + result->value.dptr = NULL; + + if ((result->value.dsize != 0) + && !(result->value.dptr = (uint8_t *)talloc_memdup( + result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header), + result->value.dsize))) { + DEBUG(0, ("talloc failed\n")); + TALLOC_FREE(result); + return NULL; + } + result->value_valid = true; + + SAFE_FREE(ctdb_data.dptr); + + return result; +} + +static int db_ctdb_record_destructor(struct db_record **recp) +{ + struct db_record *rec = talloc_get_type_abort(*recp, struct db_record); + struct db_ctdb_transaction_handle *h = talloc_get_type_abort( + rec->private_data, struct db_ctdb_transaction_handle); + int ret = h->ctx->db->transaction_commit(h->ctx->db); + if (ret != 0) { + DEBUG(0,(__location__ " transaction_commit failed\n")); + } + return 0; +} + +/* + auto-create a transaction for persistent databases + */ +static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx, + TALLOC_CTX *mem_ctx, + TDB_DATA key) +{ + int res; + struct db_record *rec, **recp; + + res = db_ctdb_transaction_start(ctx->db); + if (res == -1) { + return NULL; + } + + rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key); + if (rec == NULL) { + ctx->db->transaction_cancel(ctx->db); + return NULL; + } + + /* destroy this transaction when we release the lock */ + recp = talloc(rec, struct db_record *); + if (recp == NULL) { + ctx->db->transaction_cancel(ctx->db); + talloc_free(rec); + return NULL; + } + *recp = rec; + talloc_set_destructor(recp, db_ctdb_record_destructor); + return rec; +} + + +/* + stores a record inside a transaction + */ +static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h, + TDB_DATA key, TDB_DATA data) +{ + TALLOC_CTX *tmp_ctx = talloc_new(h); + TDB_DATA rec; + struct ctdb_ltdb_header header; + + ZERO_STRUCT(header); + + /* we need the header so we can update the RSN */ + + if (!pull_newest_from_marshall_buffer(h->m_write, key, &header, + NULL, NULL)) { + + rec = tdb_fetch(h->ctx->wtdb->tdb, key); + + if (rec.dptr != NULL) { + memcpy(&header, rec.dptr, + sizeof(struct ctdb_ltdb_header)); + rec.dsize -= sizeof(struct ctdb_ltdb_header); + + /* + * a special case, we are writing the same + * data that is there now + */ + if (data.dsize == rec.dsize && + memcmp(data.dptr, + rec.dptr + sizeof(struct ctdb_ltdb_header), + data.dsize) == 0) { + SAFE_FREE(rec.dptr); + talloc_free(tmp_ctx); + return NT_STATUS_OK; + } + } + SAFE_FREE(rec.dptr); + } + + header.dmaster = get_my_vnn(); + header.rsn++; + + h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data); + if (h->m_write == NULL) { + DEBUG(0,(__location__ " Failed to add to marshalling record\n")); + talloc_free(tmp_ctx); + return NT_STATUS_NO_MEMORY; + } + + talloc_free(tmp_ctx); + return NT_STATUS_OK; +} + + +/* + a record store inside a transaction + */ +static NTSTATUS db_ctdb_storev_transaction( + struct db_record *rec, const TDB_DATA *dbufs, int num_dbufs, int flag) +{ + struct db_ctdb_transaction_handle *h = talloc_get_type_abort( + rec->private_data, struct db_ctdb_transaction_handle); + NTSTATUS status; + TDB_DATA data = {0}; + + status = dbwrap_merge_dbufs(&data, rec, dbufs, num_dbufs); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + + status = db_ctdb_transaction_store(h, rec->key, data); + + TALLOC_FREE(data.dptr); + + return status; +} + +/* + a record delete inside a transaction + */ +static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec) +{ + struct db_ctdb_transaction_handle *h = talloc_get_type_abort( + rec->private_data, struct db_ctdb_transaction_handle); + NTSTATUS status; + + status = db_ctdb_transaction_store(h, rec->key, tdb_null); + return status; +} + +static void db_ctdb_fetch_db_seqnum_parser( + TDB_DATA key, struct ctdb_ltdb_header *header, + TDB_DATA data, void *private_data) +{ + uint64_t *seqnum = (uint64_t *)private_data; + + if (data.dsize != sizeof(uint64_t)) { + *seqnum = 0; + return; + } + memcpy(seqnum, data.dptr, sizeof(*seqnum)); +} + +/** + * Fetch the db sequence number of a persistent db directly from the db. + */ +static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db, + uint64_t *seqnum) +{ + NTSTATUS status; + TDB_DATA key; + + if (seqnum == NULL) { + return NT_STATUS_INVALID_PARAMETER; + } + + key = string_term_tdb_data(CTDB_DB_SEQNUM_KEY); + + status = db_ctdb_ltdb_parse( + db, key, db_ctdb_fetch_db_seqnum_parser, seqnum); + + if (NT_STATUS_IS_OK(status)) { + return NT_STATUS_OK; + } + if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) { + *seqnum = 0; + return NT_STATUS_OK; + } + return status; +} + +/** + * Store the database sequence number inside a transaction. + */ +static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h, + uint64_t seqnum) +{ + NTSTATUS status; + TDB_DATA key = string_term_tdb_data(CTDB_DB_SEQNUM_KEY); + TDB_DATA data = { .dptr=(uint8_t *)&seqnum, .dsize=sizeof(seqnum) }; + + status = db_ctdb_transaction_store(h, key, data); + + return status; +} + +/* + commit a transaction + */ +static int db_ctdb_transaction_commit(struct db_context *db) +{ + struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data, + struct db_ctdb_ctx); + NTSTATUS rets; + int32_t status; + struct db_ctdb_transaction_handle *h = ctx->transaction; + uint64_t old_seqnum, new_seqnum; + int ret; + + if (h == NULL) { + DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id)); + return -1; + } + + if (h->nested_cancel) { + db->transaction_cancel(db); + DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n")); + return -1; + } + + if (h->nesting != 0) { + h->nesting--; + DEBUG(5, (__location__ " transaction commit on db 0x%08x: nesting %d -> %d\n", + ctx->db_id, ctx->transaction->nesting + 1, ctx->transaction->nesting)); + return 0; + } + + if (h->m_write == NULL) { + /* + * No changes were made, so don't change the seqnum, + * don't push to other node, just exit with success. + */ + ret = 0; + goto done; + } + + DEBUG(5,(__location__ " transaction commit on db 0x%08x\n", ctx->db_id)); + + /* + * As the last db action before committing, bump the database sequence + * number. Note that this undoes all changes to the seqnum records + * performed under the transaction. This record is not meant to be + * modified by user interaction. It is for internal use only... + */ + rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum); + if (!NT_STATUS_IS_OK(rets)) { + DEBUG(1, (__location__ " failed to fetch the db sequence number " + "in transaction commit on db 0x%08x\n", ctx->db_id)); + ret = -1; + goto done; + } + + new_seqnum = old_seqnum + 1; + + rets = db_ctdb_store_db_seqnum(h, new_seqnum); + if (!NT_STATUS_IS_OK(rets)) { + DEBUG(1, (__location__ "failed to store the db sequence number " + " in transaction commit on db 0x%08x\n", ctx->db_id)); + ret = -1; + goto done; + } + +again: + /* tell ctdbd to commit to the other nodes */ + ret = ctdbd_control_local(messaging_ctdb_connection(), + CTDB_CONTROL_TRANS3_COMMIT, + h->ctx->db_id, 0, + db_ctdb_marshall_finish(h->m_write), + NULL, NULL, &status); + if ((ret != 0) || status != 0) { + /* + * The TRANS3_COMMIT control should only possibly fail when a + * recovery has been running concurrently. In any case, the db + * will be the same on all nodes, either the new copy or the + * old copy. This can be detected by comparing the old and new + * local sequence numbers. + */ + rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum); + if (!NT_STATUS_IS_OK(rets)) { + DEBUG(1, (__location__ " failed to refetch db sequence " + "number after failed TRANS3_COMMIT\n")); + ret = -1; + goto done; + } + + if (new_seqnum == old_seqnum) { + /* Recovery prevented all our changes: retry. */ + goto again; + } + if (new_seqnum != (old_seqnum + 1)) { + DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != " + "old_seqnum[%lu] + (0 or 1) after failed " + "TRANS3_COMMIT - this should not happen!\n", + (unsigned long)new_seqnum, + (unsigned long)old_seqnum)); + ret = -1; + goto done; + } + /* + * Recovery propagated our changes to all nodes, completing + * our commit for us - succeed. + */ + } + + ret = 0; + +done: + h->ctx->transaction = NULL; + talloc_free(h); + return ret; +} + + +/* + cancel a transaction + */ +static int db_ctdb_transaction_cancel(struct db_context *db) +{ + struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data, + struct db_ctdb_ctx); + struct db_ctdb_transaction_handle *h = ctx->transaction; + + if (h == NULL) { + DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id)); + return -1; + } + + if (h->nesting != 0) { + h->nesting--; + h->nested_cancel = true; + DEBUG(5, (__location__ " transaction cancel on db 0x%08x: nesting %d -> %d\n", + ctx->db_id, ctx->transaction->nesting + 1, ctx->transaction->nesting)); + return 0; + } + + DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id)); + + ctx->transaction = NULL; + talloc_free(h); + return 0; +} + + +static NTSTATUS db_ctdb_storev(struct db_record *rec, + const TDB_DATA *dbufs, int num_dbufs, int flag) +{ + struct db_ctdb_rec *crec = talloc_get_type_abort( + rec->private_data, struct db_ctdb_rec); + NTSTATUS status; + + status = db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), + dbufs, num_dbufs); + return status; +} + + + +static NTSTATUS db_ctdb_send_schedule_for_deletion(struct db_record *rec) +{ + NTSTATUS status = NT_STATUS_OK; + int ret; + struct ctdb_control_schedule_for_deletion *dd; + TDB_DATA indata; + int32_t cstatus; + struct db_ctdb_rec *crec = talloc_get_type_abort( + rec->private_data, struct db_ctdb_rec); + struct db_ctdb_ctx *ctx = crec->ctdb_ctx; + + indata.dsize = offsetof(struct ctdb_control_schedule_for_deletion, key) + rec->key.dsize; + indata.dptr = talloc_zero_array(crec, uint8_t, indata.dsize); + if (indata.dptr == NULL) { + DEBUG(0, (__location__ " talloc failed!\n")); + return NT_STATUS_NO_MEMORY; + } + + dd = (struct ctdb_control_schedule_for_deletion *)(void *)indata.dptr; + dd->db_id = ctx->db_id; + dd->hdr = crec->header; + dd->keylen = rec->key.dsize; + memcpy(dd->key, rec->key.dptr, rec->key.dsize); + + ret = ctdbd_control_local(messaging_ctdb_connection(), + CTDB_CONTROL_SCHEDULE_FOR_DELETION, + crec->ctdb_ctx->db_id, + CTDB_CTRL_FLAG_NOREPLY, /* flags */ + indata, + NULL, /* mem_ctx */ + NULL, /* outdata */ + &cstatus); + talloc_free(indata.dptr); + + if ((ret != 0) || cstatus != 0) { + DEBUG(1, (__location__ " Error sending local control " + "SCHEDULE_FOR_DELETION: %s, cstatus = %"PRIi32"\n", + strerror(ret), cstatus)); + if (ret != 0) { + status = map_nt_error_from_unix(ret); + } else { + status = NT_STATUS_UNSUCCESSFUL; + } + } + + return status; +} + +static NTSTATUS db_ctdb_delete(struct db_record *rec) +{ + NTSTATUS status; + + /* + * We have to store the header with empty data. TODO: Fix the + * tdb-level cleanup + */ + + status = db_ctdb_storev(rec, &tdb_null, 1, 0); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + + status = db_ctdb_send_schedule_for_deletion(rec); + return status; +} + +static int db_ctdb_record_destr(struct db_record* data) +{ + struct db_ctdb_rec *crec = talloc_get_type_abort( + data->private_data, struct db_ctdb_rec); + int threshold; + int ret; + struct timeval before; + double timediff; + + DEBUG(10, (DEBUGLEVEL > 10 + ? "Unlocking db %u key %s\n" + : "Unlocking db %u key %.20s\n", + (int)crec->ctdb_ctx->db_id, + hex_encode_talloc(data, (unsigned char *)data->key.dptr, + data->key.dsize))); + + before = timeval_current(); + + ret = tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key); + + timediff = timeval_elapsed(&before); + timediff *= 1000; /* get us milliseconds */ + + if (timediff > crec->ctdb_ctx->warn_unlock_msecs) { + char *key; + key = hex_encode_talloc(talloc_tos(), + (unsigned char *)data->key.dptr, + data->key.dsize); + DEBUG(0, ("tdb_chainunlock on db %s, key %s took %f milliseconds\n", + tdb_name(crec->ctdb_ctx->wtdb->tdb), key, + timediff)); + TALLOC_FREE(key); + } + + if (ret != 0) { + DEBUG(0, ("tdb_chainunlock failed\n")); + return -1; + } + + threshold = crec->ctdb_ctx->warn_locktime_msecs; + if (threshold != 0) { + timediff = timeval_elapsed(&crec->lock_time) * 1000; + if (timediff > threshold) { + const char *key; + + key = hex_encode_talloc(data, + (unsigned char *)data->key.dptr, + data->key.dsize); + DEBUG(0, ("Held tdb lock on db %s, key %s " + "%f milliseconds\n", + tdb_name(crec->ctdb_ctx->wtdb->tdb), + key, timediff)); + } + } + + return 0; +} + +/** + * Check whether we have a valid local copy of the given record, + * either for reading or for writing. + */ +static bool db_ctdb_can_use_local_hdr(const struct ctdb_ltdb_header *hdr, + uint32_t my_vnn, bool read_only) +{ + if (hdr->dmaster != my_vnn) { + /* If we're not dmaster, it must be r/o copy. */ + return read_only && (hdr->flags & CTDB_REC_RO_HAVE_READONLY); + } + + /* + * If we want write access, no one may have r/o copies. + */ + return read_only || !(hdr->flags & CTDB_REC_RO_HAVE_DELEGATIONS); +} + +static bool db_ctdb_can_use_local_copy(TDB_DATA ctdb_data, uint32_t my_vnn, + bool read_only) +{ + if (ctdb_data.dptr == NULL) { + return false; + } + + if (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) { + return false; + } + + return db_ctdb_can_use_local_hdr( + (struct ctdb_ltdb_header *)ctdb_data.dptr, my_vnn, read_only); +} + +static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx, + TALLOC_CTX *mem_ctx, + TDB_DATA key) +{ + struct db_record *result; + struct db_ctdb_rec *crec; + TDB_DATA ctdb_data; + int migrate_attempts; + struct timeval migrate_start; + struct timeval chainlock_start; + struct timeval ctdb_start_time; + double chainlock_time = 0; + double ctdb_time = 0; + int duration_msecs; + int lockret; + int ret; + + if (!(result = talloc(mem_ctx, struct db_record))) { + DEBUG(0, ("talloc failed\n")); + return NULL; + } + + if (!(crec = talloc_zero(result, struct db_ctdb_rec))) { + DEBUG(0, ("talloc failed\n")); + TALLOC_FREE(result); + return NULL; + } + + result->db = ctx->db; + result->private_data = (void *)crec; + crec->ctdb_ctx = ctx; + + result->key.dsize = key.dsize; + result->key.dptr = (uint8_t *)talloc_memdup(result, key.dptr, + key.dsize); + if (result->key.dptr == NULL) { + DEBUG(0, ("talloc failed\n")); + TALLOC_FREE(result); + return NULL; + } + + migrate_attempts = 0; + GetTimeOfDay(&migrate_start); + + /* + * Do a blocking lock on the record + */ +again: + + if (DEBUGLEVEL >= 10) { + char *keystr = hex_encode_talloc(result, key.dptr, key.dsize); + DEBUG(10, (DEBUGLEVEL > 10 + ? "Locking db %u key %s\n" + : "Locking db %u key %.20s\n", + (int)crec->ctdb_ctx->db_id, keystr)); + TALLOC_FREE(keystr); + } + + GetTimeOfDay(&chainlock_start); + lockret = tdb_chainlock(ctx->wtdb->tdb, key); + chainlock_time += timeval_elapsed(&chainlock_start); + + if (lockret != 0) { + DEBUG(3, ("tdb_chainlock failed\n")); + TALLOC_FREE(result); + return NULL; + } + + result->storev = db_ctdb_storev; + result->delete_rec = db_ctdb_delete; + talloc_set_destructor(result, db_ctdb_record_destr); + + ctdb_data = tdb_fetch(ctx->wtdb->tdb, key); + + /* + * See if we have a valid record and we are the dmaster. If so, we can + * take the shortcut and just return it. + */ + + if (!db_ctdb_can_use_local_copy(ctdb_data, get_my_vnn(), false)) { + SAFE_FREE(ctdb_data.dptr); + tdb_chainunlock(ctx->wtdb->tdb, key); + talloc_set_destructor(result, NULL); + + migrate_attempts += 1; + + DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %"PRIu32" " + "(%"PRIu32") %"PRIu32"\n", + ctdb_data.dptr, ctdb_data.dptr ? + ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : + UINT32_MAX, + get_my_vnn(), + ctdb_data.dptr ? + ((struct ctdb_ltdb_header *)ctdb_data.dptr)->flags : 0)); + + GetTimeOfDay(&ctdb_start_time); + ret = ctdbd_migrate(messaging_ctdb_connection(), ctx->db_id, + key); + ctdb_time += timeval_elapsed(&ctdb_start_time); + + if (ret != 0) { + DEBUG(5, ("ctdbd_migrate failed: %s\n", + strerror(ret))); + TALLOC_FREE(result); + return NULL; + } + /* now its migrated, try again */ + goto again; + } + + { + double duration; + duration = timeval_elapsed(&migrate_start); + + /* + * Convert the duration to milliseconds to avoid a + * floating-point division of + * lp_parm_int("migrate_duration") by 1000. + */ + duration_msecs = duration * 1000; + } + + if ((migrate_attempts > ctx->warn_migrate_attempts) || + (duration_msecs > ctx->warn_migrate_msecs)) { + int chain = 0; + + if (tdb_get_flags(ctx->wtdb->tdb) & TDB_INCOMPATIBLE_HASH) { + chain = tdb_jenkins_hash(&key) % + tdb_hash_size(ctx->wtdb->tdb); + } + + DEBUG(0, ("db_ctdb_fetch_locked for %s key %s, chain %d " + "needed %d attempts, %d milliseconds, " + "chainlock: %f ms, CTDB %f ms\n", + tdb_name(ctx->wtdb->tdb), + hex_encode_talloc(talloc_tos(), + (unsigned char *)key.dptr, + key.dsize), + chain, + migrate_attempts, duration_msecs, + chainlock_time * 1000.0, + ctdb_time * 1000.0)); + } + + GetTimeOfDay(&crec->lock_time); + + memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header)); + + result->value.dsize = ctdb_data.dsize - sizeof(crec->header); + result->value.dptr = NULL; + + if (result->value.dsize != 0) { + result->value.dptr = talloc_memdup( + result, ctdb_data.dptr + sizeof(crec->header), + result->value.dsize); + if (result->value.dptr == NULL) { + DBG_ERR("talloc failed\n"); + TALLOC_FREE(result); + return NULL; + } + } + result->value_valid = true; + + SAFE_FREE(ctdb_data.dptr); + + return result; +} + +static struct db_record *db_ctdb_fetch_locked(struct db_context *db, + TALLOC_CTX *mem_ctx, + TDB_DATA key) +{ + struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data, + struct db_ctdb_ctx); + + if (ctx->transaction != NULL) { + return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key); + } + + if (db->persistent) { + return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key); + } + + return fetch_locked_internal(ctx, mem_ctx, key); +} + +struct db_ctdb_parse_record_state { + void (*parser)(TDB_DATA key, TDB_DATA data, void *private_data); + void *private_data; + uint32_t my_vnn; + bool ask_for_readonly_copy; + bool done; + bool empty_record; +}; + +static void db_ctdb_parse_record_parser( + TDB_DATA key, struct ctdb_ltdb_header *header, + TDB_DATA data, void *private_data) +{ + struct db_ctdb_parse_record_state *state = + (struct db_ctdb_parse_record_state *)private_data; + state->parser(key, data, state->private_data); +} + +static void db_ctdb_parse_record_parser_nonpersistent( + TDB_DATA key, struct ctdb_ltdb_header *header, + TDB_DATA data, void *private_data) +{ + struct db_ctdb_parse_record_state *state = + (struct db_ctdb_parse_record_state *)private_data; + + if (db_ctdb_can_use_local_hdr(header, state->my_vnn, true)) { + /* + * A record consisting only of the ctdb header can be + * a validly created empty record or a tombstone + * record of a deleted record (not vacuumed yet). Mark + * it accordingly. + */ + state->empty_record = (data.dsize == 0); + if (!state->empty_record) { + state->parser(key, data, state->private_data); + } + state->done = true; + } else { + /* + * We found something in the db, so it seems that this record, + * while not usable locally right now, is popular. Ask for a + * R/O copy. + */ + state->ask_for_readonly_copy = true; + } +} + +static NTSTATUS db_ctdb_try_parse_local_record(struct db_ctdb_ctx *ctx, + TDB_DATA key, + struct db_ctdb_parse_record_state *state) +{ + NTSTATUS status; + + if (ctx->transaction != NULL) { + struct db_ctdb_transaction_handle *h = ctx->transaction; + bool found; + + /* + * Transactions only happen for persistent db's. + */ + + found = parse_newest_in_marshall_buffer( + h->m_write, key, db_ctdb_parse_record_parser, state); + + if (found) { + return NT_STATUS_OK; + } + } + + if (ctx->db->persistent) { + /* + * Persistent db, but not found in the transaction buffer + */ + return db_ctdb_ltdb_parse( + ctx, key, db_ctdb_parse_record_parser, state); + } + + state->done = false; + state->ask_for_readonly_copy = false; + + status = db_ctdb_ltdb_parse( + ctx, key, db_ctdb_parse_record_parser_nonpersistent, state); + if (NT_STATUS_IS_OK(status) && state->done) { + if (state->empty_record) { + /* + * We know authoritatively, that this is an empty + * record. Since ctdb does not distinguish between empty + * and deleted records, this can be a record stored as + * empty or a not-yet-vacuumed tombstone record of a + * deleted record. Now Samba right now can live without + * empty records, so we can safely report this record + * as non-existing. + * + * See bugs 10008 and 12005. + */ + return NT_STATUS_NOT_FOUND; + } + return NT_STATUS_OK; + } + + return NT_STATUS_MORE_PROCESSING_REQUIRED; +} + +static NTSTATUS db_ctdb_parse_record(struct db_context *db, TDB_DATA key, + void (*parser)(TDB_DATA key, + TDB_DATA data, + void *private_data), + void *private_data) +{ + struct db_ctdb_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_ctdb_ctx); + struct db_ctdb_parse_record_state state; + NTSTATUS status; + int ret; + + state.parser = parser; + state.private_data = private_data; + state.my_vnn = get_my_vnn(); + state.empty_record = false; + + status = db_ctdb_try_parse_local_record(ctx, key, &state); + if (!NT_STATUS_EQUAL(status, NT_STATUS_MORE_PROCESSING_REQUIRED)) { + return status; + } + + ret = ctdbd_parse(messaging_ctdb_connection(), ctx->db_id, key, + state.ask_for_readonly_copy, parser, private_data); + if (ret != 0) { + if (ret == ENOENT) { + /* + * This maps to + * NT_STATUS_OBJECT_NAME_NOT_FOUND. Our upper + * layers expect NT_STATUS_NOT_FOUND for "no + * record around". We need to convert dbwrap + * to 0/errno away from NTSTATUS ... :-) + */ + return NT_STATUS_NOT_FOUND; + } + return map_nt_error_from_unix(ret); + } + return NT_STATUS_OK; +} + +static void db_ctdb_parse_record_done(struct tevent_req *subreq); + +static struct tevent_req *db_ctdb_parse_record_send( + TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + struct db_context *db, + TDB_DATA key, + void (*parser)(TDB_DATA key, + TDB_DATA data, + void *private_data), + void *private_data, + enum dbwrap_req_state *req_state) +{ + struct db_ctdb_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_ctdb_ctx); + struct tevent_req *req = NULL; + struct tevent_req *subreq = NULL; + struct db_ctdb_parse_record_state *state = NULL; + NTSTATUS status; + + req = tevent_req_create(mem_ctx, &state, + struct db_ctdb_parse_record_state); + if (req == NULL) { + *req_state = DBWRAP_REQ_ERROR; + return NULL; + + } + + *state = (struct db_ctdb_parse_record_state) { + .parser = parser, + .private_data = private_data, + .my_vnn = get_my_vnn(), + .empty_record = false, + }; + + status = db_ctdb_try_parse_local_record(ctx, key, state); + if (!NT_STATUS_EQUAL(status, NT_STATUS_MORE_PROCESSING_REQUIRED)) { + if (tevent_req_nterror(req, status)) { + *req_state = DBWRAP_REQ_ERROR; + return tevent_req_post(req, ev); + } + *req_state = DBWRAP_REQ_DONE; + tevent_req_done(req); + return tevent_req_post(req, ev); + } + + subreq = ctdbd_parse_send(state, + ev, + ctdb_async_ctx.async_conn, + ctx->db_id, + key, + state->ask_for_readonly_copy, + parser, + private_data, + req_state); + if (tevent_req_nomem(subreq, req)) { + *req_state = DBWRAP_REQ_ERROR; + return tevent_req_post(req, ev); + } + tevent_req_set_callback(subreq, db_ctdb_parse_record_done, req); + + return req; +} + +static void db_ctdb_parse_record_done(struct tevent_req *subreq) +{ + struct tevent_req *req = tevent_req_callback_data( + subreq, struct tevent_req); + int ret; + + ret = ctdbd_parse_recv(subreq); + TALLOC_FREE(subreq); + if (ret != 0) { + if (ret == ENOENT) { + /* + * This maps to NT_STATUS_OBJECT_NAME_NOT_FOUND. Our + * upper layers expect NT_STATUS_NOT_FOUND for "no + * record around". We need to convert dbwrap to 0/errno + * away from NTSTATUS ... :-) + */ + tevent_req_nterror(req, NT_STATUS_NOT_FOUND); + return; + } + tevent_req_nterror(req, map_nt_error_from_unix(ret)); + return; + } + + tevent_req_done(req); + return; +} + +static NTSTATUS db_ctdb_parse_record_recv(struct tevent_req *req) +{ + return tevent_req_simple_recv_ntstatus(req); +} + +struct traverse_state { + struct db_context *db; + int (*fn)(struct db_record *rec, void *private_data); + void *private_data; + int count; +}; + +static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data) +{ + struct traverse_state *state = (struct traverse_state *)private_data; + struct db_record *rec = NULL; + TALLOC_CTX *tmp_ctx = NULL; + + tmp_ctx = talloc_new(state->db); + if (tmp_ctx == NULL) { + DBG_ERR("talloc_new failed\n"); + return; + } + + /* we have to give them a locked record to prevent races */ + rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key); + if (rec != NULL && rec->value.dsize > 0) { + state->fn(rec, state->private_data); + state->count++; + } + talloc_free(tmp_ctx); +} + +static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf, + void *private_data) +{ + struct traverse_state *state = (struct traverse_state *)private_data; + struct db_record *rec; + TALLOC_CTX *tmp_ctx = talloc_new(state->db); + int ret = 0; + + /* + * Skip the __db_sequence_number__ key: + * This is used for persistent transactions internally. + */ + if (kbuf.dsize == strlen(CTDB_DB_SEQNUM_KEY) + 1 && + strcmp((const char*)kbuf.dptr, CTDB_DB_SEQNUM_KEY) == 0) + { + goto done; + } + + /* we have to give them a locked record to prevent races */ + rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf); + if (rec && rec->value.dsize > 0) { + ret = state->fn(rec, state->private_data); + } + +done: + talloc_free(tmp_ctx); + return ret; +} + +/* wrapper to use traverse_persistent_callback with dbwrap */ +static int traverse_persistent_callback_dbwrap(struct db_record *rec, void* data) +{ + return traverse_persistent_callback(NULL, rec->key, rec->value, data); +} + +static int db_ctdbd_traverse(uint32_t db_id, + void (*fn)(TDB_DATA key, TDB_DATA data, + void *private_data), + void *private_data) +{ + struct ctdbd_connection *conn; + int ret; + + become_root(); + ret = ctdbd_init_connection(talloc_tos(), lp_ctdbd_socket(), + lp_ctdb_timeout(), &conn); + unbecome_root(); + if (ret != 0) { + DBG_WARNING("ctdbd_init_connection failed: %s\n", + strerror(ret)); + return ret; + } + + ret = ctdbd_traverse(conn, db_id, fn, private_data); + TALLOC_FREE(conn); + + if (ret != 0) { + DBG_WARNING("ctdbd_traverse failed: %s\n", + strerror(ret)); + return ret; + } + + return 0; +} + + +static int db_ctdb_traverse(struct db_context *db, + int (*fn)(struct db_record *rec, + void *private_data), + void *private_data) +{ + int ret; + struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data, + struct db_ctdb_ctx); + struct traverse_state state; + + state = (struct traverse_state) { + .db = db, + .fn = fn, + .private_data = private_data, + }; + + if (db->persistent) { + struct tdb_context *ltdb = ctx->wtdb->tdb; + + /* for persistent databases we don't need to do a ctdb traverse, + we can do a faster local traverse */ + ret = tdb_traverse(ltdb, traverse_persistent_callback, &state); + if (ret < 0) { + return ret; + } + if (ctx->transaction && ctx->transaction->m_write) { + /* + * we now have to handle keys not yet + * present at transaction start + */ + struct db_context *newkeys = db_open_rbt(talloc_tos()); + struct ctdb_marshall_buffer *mbuf = ctx->transaction->m_write; + struct ctdb_rec_data_old *rec=NULL; + uint32_t i; + int count = 0; + NTSTATUS status; + + if (newkeys == NULL) { + return -1; + } + + for (i=0; icount; i++) { + TDB_DATA key; + rec = db_ctdb_marshall_loop_next_key( + mbuf, rec, &key); + SMB_ASSERT(rec != NULL); + + if (!tdb_exists(ltdb, key)) { + dbwrap_store(newkeys, key, tdb_null, 0); + } + } + status = dbwrap_traverse(newkeys, + traverse_persistent_callback_dbwrap, + &state, + &count); + talloc_free(newkeys); + if (!NT_STATUS_IS_OK(status)) { + return -1; + } + ret += count; + } + return ret; + } + + ret = db_ctdbd_traverse(ctx->db_id, traverse_callback, &state); + if (ret != 0) { + return -1; + } + return state.count; +} + +static NTSTATUS db_ctdb_storev_deny(struct db_record *rec, + const TDB_DATA *dbufs, int num_dbufs, int flag) +{ + return NT_STATUS_MEDIA_WRITE_PROTECTED; +} + +static NTSTATUS db_ctdb_delete_deny(struct db_record *rec) +{ + return NT_STATUS_MEDIA_WRITE_PROTECTED; +} + +static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data) +{ + struct traverse_state *state = (struct traverse_state *)private_data; + struct db_record rec; + + ZERO_STRUCT(rec); + rec.db = state->db; + rec.key = key; + rec.value = data; + rec.storev = db_ctdb_storev_deny; + rec.delete_rec = db_ctdb_delete_deny; + rec.private_data = NULL; + rec.value_valid = true; + state->fn(&rec, state->private_data); + state->count++; +} + +static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf, + void *private_data) +{ + struct traverse_state *state = (struct traverse_state *)private_data; + struct db_record rec; + + /* + * Skip the __db_sequence_number__ key: + * This is used for persistent transactions internally. + */ + if (kbuf.dsize == strlen(CTDB_DB_SEQNUM_KEY) + 1 && + strcmp((const char*)kbuf.dptr, CTDB_DB_SEQNUM_KEY) == 0) + { + return 0; + } + + ZERO_STRUCT(rec); + rec.db = state->db; + rec.key = kbuf; + rec.value = dbuf; + rec.value_valid = true; + rec.storev = db_ctdb_storev_deny; + rec.delete_rec = db_ctdb_delete_deny; + rec.private_data = NULL; + + if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) { + /* a deleted record */ + return 0; + } + rec.value.dsize -= sizeof(struct ctdb_ltdb_header); + rec.value.dptr += sizeof(struct ctdb_ltdb_header); + + state->count++; + return state->fn(&rec, state->private_data); +} + +static int db_ctdb_traverse_read(struct db_context *db, + int (*fn)(struct db_record *rec, + void *private_data), + void *private_data) +{ + int ret; + struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data, + struct db_ctdb_ctx); + struct traverse_state state; + + state = (struct traverse_state) { + .db = db, + .fn = fn, + .private_data = private_data, + }; + + if (db->persistent) { + /* for persistent databases we don't need to do a ctdb traverse, + we can do a faster local traverse */ + int nrecs; + + nrecs = tdb_traverse_read(ctx->wtdb->tdb, + traverse_persistent_callback_read, + &state); + if (nrecs == -1) { + return -1; + } + return state.count; + } + + ret = db_ctdbd_traverse(ctx->db_id, traverse_read_callback, &state); + if (ret != 0) { + return -1; + } + return state.count; +} + +static int db_ctdb_get_seqnum(struct db_context *db) +{ + struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data, + struct db_ctdb_ctx); + return tdb_get_seqnum(ctx->wtdb->tdb); +} + +static size_t db_ctdb_id(struct db_context *db, uint8_t *id, size_t idlen) +{ + struct db_ctdb_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_ctdb_ctx); + + if (idlen >= sizeof(ctx->db_id)) { + memcpy(id, &ctx->db_id, sizeof(ctx->db_id)); + } + + return sizeof(ctx->db_id); +} + +struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx, + struct messaging_context *msg_ctx, + const char *name, + int hash_size, int tdb_flags, + int open_flags, mode_t mode, + enum dbwrap_lock_order lock_order, + uint64_t dbwrap_flags) +{ + struct db_context *result; + struct db_ctdb_ctx *db_ctdb; + char *db_path; + struct loadparm_context *lp_ctx; + TDB_DATA data; + TDB_DATA outdata = {0}; + bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0; + int32_t cstatus; + int ret; + + if (!lp_clustering()) { + DEBUG(10, ("Clustering disabled -- no ctdb\n")); + return NULL; + } + + if (!(result = talloc_zero(mem_ctx, struct db_context))) { + DEBUG(0, ("talloc failed\n")); + TALLOC_FREE(result); + return NULL; + } + + if (!(db_ctdb = talloc(result, struct db_ctdb_ctx))) { + DEBUG(0, ("talloc failed\n")); + TALLOC_FREE(result); + return NULL; + } + + result->name = talloc_strdup(result, name); + if (result->name == NULL) { + DEBUG(0, ("talloc failed\n")); + TALLOC_FREE(result); + return NULL; + } + + db_ctdb->transaction = NULL; + db_ctdb->db = result; + + ret = ctdbd_db_attach(messaging_ctdb_connection(), name, + &db_ctdb->db_id, persistent); + if (ret != 0) { + DEBUG(0, ("ctdbd_db_attach failed for %s: %s\n", name, + strerror(ret))); + TALLOC_FREE(result); + return NULL; + } + + if (tdb_flags & TDB_SEQNUM) { + data.dptr = (uint8_t *)&db_ctdb->db_id; + data.dsize = sizeof(db_ctdb->db_id); + + ret = ctdbd_control_local(messaging_ctdb_connection(), + CTDB_CONTROL_ENABLE_SEQNUM, + 0, 0, data, + NULL, NULL, &cstatus); + if ((ret != 0) || cstatus != 0) { + DBG_ERR("ctdb_control for enable seqnum " + "failed: %s\n", strerror(ret)); + TALLOC_FREE(result); + return NULL; + } + } + + db_path = ctdbd_dbpath(messaging_ctdb_connection(), db_ctdb, + db_ctdb->db_id); + if (db_path == NULL) { + DBG_ERR("ctdbd_dbpath failed\n"); + TALLOC_FREE(result); + return NULL; + } + + result->persistent = persistent; + result->lock_order = lock_order; + + data.dptr = (uint8_t *)&db_ctdb->db_id; + data.dsize = sizeof(db_ctdb->db_id); + + ret = ctdbd_control_local(messaging_ctdb_connection(), + CTDB_CONTROL_DB_OPEN_FLAGS, + 0, 0, data, NULL, &outdata, &cstatus); + if (ret != 0) { + DBG_ERR(" ctdb control for db_open_flags " + "failed: %s\n", strerror(ret)); + TALLOC_FREE(result); + return NULL; + } + + if (cstatus != 0 || outdata.dsize != sizeof(int)) { + DBG_ERR("ctdb_control for db_open_flags failed\n"); + TALLOC_FREE(outdata.dptr); + TALLOC_FREE(result); + return NULL; + } + + tdb_flags = *(int *)outdata.dptr; + TALLOC_FREE(outdata.dptr); + + if (!result->persistent) { + ret = ctdb_async_ctx_init(NULL, messaging_tevent_context(msg_ctx)); + if (ret != 0) { + DBG_ERR("ctdb_async_ctx_init failed: %s\n", strerror(ret)); + TALLOC_FREE(result); + return NULL; + } + } + + if (!result->persistent && + (dbwrap_flags & DBWRAP_FLAG_OPTIMIZE_READONLY_ACCESS)) + { + TDB_DATA indata; + + indata = make_tdb_data((uint8_t *)&db_ctdb->db_id, + sizeof(db_ctdb->db_id)); + + ret = ctdbd_control_local( + messaging_ctdb_connection(), + CTDB_CONTROL_SET_DB_READONLY, 0, 0, + indata, NULL, NULL, &cstatus); + if ((ret != 0) || (cstatus != 0)) { + DEBUG(1, ("CTDB_CONTROL_SET_DB_READONLY failed: " + "%s, %"PRIi32"\n", strerror(ret), cstatus)); + TALLOC_FREE(result); + return NULL; + } + } + + lp_ctx = loadparm_init_s3(db_path, loadparm_s3_helpers()); + + if (hash_size == 0) { + hash_size = lpcfg_tdb_hash_size(lp_ctx, db_path); + } + + db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, + lpcfg_tdb_flags(lp_ctx, tdb_flags), + O_RDWR, 0); + talloc_unlink(db_path, lp_ctx); + if (db_ctdb->wtdb == NULL) { + DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno))); + TALLOC_FREE(result); + return NULL; + } + talloc_free(db_path); + + /* honor permissions if user has specified O_CREAT */ + if (open_flags & O_CREAT) { + int fd; + fd = tdb_fd(db_ctdb->wtdb->tdb); + ret = fchmod(fd, mode); + if (ret == -1) { + DBG_WARNING("fchmod failed: %s\n", + strerror(errno)); + TALLOC_FREE(result); + return NULL; + } + } + + if (result->persistent) { + db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb, msg_ctx); + if (db_ctdb->lock_ctx == NULL) { + DEBUG(0, ("g_lock_ctx_init failed\n")); + TALLOC_FREE(result); + return NULL; + } + } + + db_ctdb->warn_unlock_msecs = lp_parm_int(-1, "ctdb", + "unlock_warn_threshold", 5); + db_ctdb->warn_migrate_attempts = lp_parm_int(-1, "ctdb", + "migrate_attempts", 10); + db_ctdb->warn_migrate_msecs = lp_parm_int(-1, "ctdb", + "migrate_duration", 5000); + db_ctdb->warn_locktime_msecs = lp_ctdb_locktime_warn_threshold(); + + result->private_data = (void *)db_ctdb; + result->fetch_locked = db_ctdb_fetch_locked; + result->parse_record = db_ctdb_parse_record; + result->parse_record_send = db_ctdb_parse_record_send; + result->parse_record_recv = db_ctdb_parse_record_recv; + result->traverse = db_ctdb_traverse; + result->traverse_read = db_ctdb_traverse_read; + result->get_seqnum = db_ctdb_get_seqnum; + result->transaction_start = db_ctdb_transaction_start; + result->transaction_commit = db_ctdb_transaction_commit; + result->transaction_cancel = db_ctdb_transaction_cancel; + result->id = db_ctdb_id; + + DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n", + name, db_ctdb->db_id)); + + return result; +} diff --git a/source3/lib/dbwrap/dbwrap_ctdb.h b/source3/lib/dbwrap/dbwrap_ctdb.h new file mode 100644 index 0000000..0b82479 --- /dev/null +++ b/source3/lib/dbwrap/dbwrap_ctdb.h @@ -0,0 +1,40 @@ +/* + Unix SMB/CIFS implementation. + Database interface wrapper around ctdbd + Copyright (C) Volker Lendecke 2007-2009 + Copyright (C) Michael Adam 2009 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#ifndef __DBWRAP_CTDB_H__ +#define __DBWRAP_CTDB_H__ + +#include + +#include "dbwrap/dbwrap_private.h" + +struct db_context; +struct ctdbd_connection; + +struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx, + struct messaging_context *msg_ctx, + const char *name, + int hash_size, int tdb_flags, + int open_flags, mode_t mode, + enum dbwrap_lock_order lock_order, + uint64_t dbwrap_flags); +int ctdb_async_ctx_reinit(TALLOC_CTX *mem_ctx, struct tevent_context *ev); + +#endif /* __DBWRAP_CTDB_H__ */ diff --git a/source3/lib/dbwrap/dbwrap_open.c b/source3/lib/dbwrap/dbwrap_open.c new file mode 100644 index 0000000..52c8a94 --- /dev/null +++ b/source3/lib/dbwrap/dbwrap_open.c @@ -0,0 +1,197 @@ +/* + Unix SMB/CIFS implementation. + Database interface wrapper + + Copyright (C) Volker Lendecke 2005-2007 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#include "includes.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap/dbwrap_private.h" +#include "dbwrap/dbwrap_open.h" +#include "dbwrap/dbwrap_tdb.h" +#include "dbwrap/dbwrap_ctdb.h" +#include "lib/param/param.h" +#include "lib/cluster_support.h" +#include "lib/messages_ctdb.h" +#include "util_tdb.h" +#include "ctdbd_conn.h" +#include "global_contexts.h" + +bool db_is_local(const char *name) +{ + const char *sockname = lp_ctdbd_socket(); + + if (lp_clustering() && socket_exist(sockname)) { + const char *partname; + /* ctdb only wants the file part of the name */ + partname = strrchr(name, '/'); + if (partname) { + partname++; + } else { + partname = name; + } + /* allow ctdb for individual databases to be disabled */ + if (lp_parm_bool(-1, "ctdb", partname, True)) { + return false; + } + } + + return true; +} + +/** + * open a database + */ +struct db_context *db_open(TALLOC_CTX *mem_ctx, + const char *name, + int hash_size, int tdb_flags, + int open_flags, mode_t mode, + enum dbwrap_lock_order lock_order, + uint64_t dbwrap_flags) +{ + struct db_context *result = NULL; + const char *base; + struct loadparm_context *lp_ctx = NULL; + + if ((lock_order != DBWRAP_LOCK_ORDER_NONE) && + !DBWRAP_LOCK_ORDER_VALID(lock_order)) { + errno = EINVAL; + return NULL; + } + + base = strrchr_m(name, '/'); + if (base != NULL) { + base++; + } else { + base = name; + } + + if (tdb_flags & TDB_CLEAR_IF_FIRST) { + bool try_readonly = false; + + if (dbwrap_flags & DBWRAP_FLAG_OPTIMIZE_READONLY_ACCESS) { + try_readonly = true; + } + + try_readonly = lp_parm_bool(-1, "dbwrap_optimize_readonly", "*", try_readonly); + try_readonly = lp_parm_bool(-1, "dbwrap_optimize_readonly", base, try_readonly); + + if (try_readonly) { + dbwrap_flags |= DBWRAP_FLAG_OPTIMIZE_READONLY_ACCESS; + } else { + dbwrap_flags &= ~DBWRAP_FLAG_OPTIMIZE_READONLY_ACCESS; + } + } + + if (tdb_flags & TDB_CLEAR_IF_FIRST) { + bool try_mutex = true; + bool require_mutex = false; + + try_mutex = lp_parm_bool(-1, "dbwrap_tdb_mutexes", "*", try_mutex); + try_mutex = lp_parm_bool(-1, "dbwrap_tdb_mutexes", base, try_mutex); + + if (!lp_use_mmap()) { + /* + * Mutexes require mmap. "use mmap = no" can + * be a debugging tool, so let it override the + * mutex parameters + */ + try_mutex = false; + } + + if (try_mutex && tdb_runtime_check_for_robust_mutexes()) { + tdb_flags |= TDB_MUTEX_LOCKING; + } + + require_mutex = lp_parm_bool(-1, "dbwrap_tdb_require_mutexes", + "*", require_mutex); + require_mutex = lp_parm_bool(-1, "dbwrap_tdb_require_mutexes", + base, require_mutex); + + if (require_mutex) { + tdb_flags |= TDB_MUTEX_LOCKING; + } + } + + if (lp_clustering()) { + const char *sockname; + + sockname = lp_ctdbd_socket(); + if (!socket_exist(sockname)) { + DBG_WARNING("ctdb socket does %s not exist - " + "is ctdb not running?\n", + sockname); + return NULL; + } + + /* allow ctdb for individual databases to be disabled */ + if (lp_parm_bool(-1, "ctdb", base, true)) { + struct messaging_context *msg_ctx; + struct ctdbd_connection *conn; + + /* + * Initialize messaging before getting the ctdb + * connection, as the ctdb connection requires messaging + * to be initialized. + */ + msg_ctx = global_messaging_context(); + if (msg_ctx == NULL) { + DBG_ERR("Failed to initialize messaging\n"); + return NULL; + } + + conn = messaging_ctdb_connection(); + if (conn == NULL) { + DBG_WARNING("No ctdb connection\n"); + errno = EIO; + return NULL; + } + + result = db_open_ctdb(mem_ctx, msg_ctx, base, + hash_size, + tdb_flags, open_flags, mode, + lock_order, dbwrap_flags); + if (result == NULL) { + DBG_ERR("failed to attach to ctdb %s\n", base); + if (errno == 0) { + errno = EIO; + } + return NULL; + } + + return result; + } + } + + lp_ctx = loadparm_init_s3(mem_ctx, loadparm_s3_helpers()); + + if (hash_size == 0) { + hash_size = lpcfg_tdb_hash_size(lp_ctx, name); + } + tdb_flags = lpcfg_tdb_flags(lp_ctx, tdb_flags); + + result = dbwrap_local_open(mem_ctx, + name, + hash_size, + tdb_flags, + open_flags, + mode, + lock_order, + dbwrap_flags); + talloc_unlink(mem_ctx, lp_ctx); + return result; +} diff --git a/source3/lib/dbwrap/dbwrap_open.h b/source3/lib/dbwrap/dbwrap_open.h new file mode 100644 index 0000000..d14794e --- /dev/null +++ b/source3/lib/dbwrap/dbwrap_open.h @@ -0,0 +1,45 @@ +/* + Unix SMB/CIFS implementation. + Database interface wrapper around tdb + + Copyright (C) Volker Lendecke 2005-2007 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#ifndef __DBWRAP_OPEN_H__ +#define __DBWRAP_OPEN_H__ + +struct db_context; + +/** + * Convenience function to check whether a tdb database + * is local or clustered (ctdb) in a clustered environment. + */ +bool db_is_local(const char *name); + +/** + * Convenience function that will determine whether to + * open a tdb database via the tdb backend or via the ctdb + * backend, based on lp_clustering() and a db-specific + * settings. + */ +struct db_context *db_open(TALLOC_CTX *mem_ctx, + const char *name, + int hash_size, int tdb_flags, + int open_flags, mode_t mode, + enum dbwrap_lock_order lock_order, + uint64_t dbwrap_flags); + +#endif /* __DBWRAP_OPEN_H__ */ diff --git a/source3/lib/dbwrap/dbwrap_watch.c b/source3/lib/dbwrap/dbwrap_watch.c new file mode 100644 index 0000000..df93119 --- /dev/null +++ b/source3/lib/dbwrap/dbwrap_watch.c @@ -0,0 +1,1285 @@ +/* + Unix SMB/CIFS implementation. + Watch dbwrap record changes + Copyright (C) Volker Lendecke 2012 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#include "includes.h" +#include "system/filesys.h" +#include "lib/util/server_id.h" +#include "dbwrap/dbwrap.h" +#include "dbwrap_watch.h" +#include "dbwrap_open.h" +#include "lib/util/util_tdb.h" +#include "lib/util/tevent_ntstatus.h" +#include "serverid.h" +#include "server_id_watch.h" +#include "lib/dbwrap/dbwrap_private.h" + +struct dbwrap_watcher { + /* + * Process watching this record + */ + struct server_id pid; + /* + * Individual instance inside the waiter, incremented each + * time a watcher is created + */ + uint64_t instance; +}; + +#define DBWRAP_WATCHER_BUF_LENGTH (SERVER_ID_BUF_LENGTH + sizeof(uint64_t)) +#define DBWRAP_MAX_WATCHERS (INT32_MAX/DBWRAP_WATCHER_BUF_LENGTH) + +/* + * Watched records contain a header of: + * + * [uint32] num_records + * 0 [DBWRAP_WATCHER_BUF_LENGTH] \ + * 1 [DBWRAP_WATCHER_BUF_LENGTH] | + * .. |- Array of watchers + * (num_records-1)[DBWRAP_WATCHER_BUF_LENGTH] / + * + * [Remainder of record....] + * + * If this header is absent then this is a + * fresh record of length zero (no watchers). + */ + +static bool dbwrap_watch_rec_parse( + TDB_DATA data, + uint8_t **pwatchers, + size_t *pnum_watchers, + TDB_DATA *pdata) +{ + size_t num_watchers; + + if (data.dsize == 0) { + /* Fresh record */ + if (pwatchers != NULL) { + *pwatchers = NULL; + } + if (pnum_watchers != NULL) { + *pnum_watchers = 0; + } + if (pdata != NULL) { + *pdata = (TDB_DATA) { .dptr = NULL }; + } + return true; + } + + if (data.dsize < sizeof(uint32_t)) { + /* Invalid record */ + return false; + } + + num_watchers = IVAL(data.dptr, 0); + + data.dptr += sizeof(uint32_t); + data.dsize -= sizeof(uint32_t); + + if (num_watchers > data.dsize/DBWRAP_WATCHER_BUF_LENGTH) { + /* Invalid record */ + return false; + } + + if (pwatchers != NULL) { + *pwatchers = data.dptr; + } + if (pnum_watchers != NULL) { + *pnum_watchers = num_watchers; + } + if (pdata != NULL) { + size_t watchers_len = num_watchers * DBWRAP_WATCHER_BUF_LENGTH; + *pdata = (TDB_DATA) { + .dptr = data.dptr + watchers_len, + .dsize = data.dsize - watchers_len + }; + } + + return true; +} + +static void dbwrap_watcher_get(struct dbwrap_watcher *w, + const uint8_t buf[DBWRAP_WATCHER_BUF_LENGTH]) +{ + server_id_get(&w->pid, buf); + w->instance = BVAL(buf, SERVER_ID_BUF_LENGTH); +} + +static void dbwrap_watcher_put(uint8_t buf[DBWRAP_WATCHER_BUF_LENGTH], + const struct dbwrap_watcher *w) +{ + server_id_put(buf, w->pid); + SBVAL(buf, SERVER_ID_BUF_LENGTH, w->instance); +} + +static void dbwrap_watch_log_invalid_record( + struct db_context *db, TDB_DATA key, TDB_DATA value) +{ + DBG_ERR("Found invalid record in %s\n", dbwrap_name(db)); + dump_data(1, key.dptr, key.dsize); + dump_data(1, value.dptr, value.dsize); +} + +struct db_watched_ctx { + struct db_context *backend; + struct messaging_context *msg; +}; + +struct db_watched_record { + struct db_record *rec; + struct server_id self; + struct { + struct db_record *rec; + TDB_DATA initial_value; + bool initial_valid; + } backend; + bool force_fini_store; + struct dbwrap_watcher added; + bool removed_first; + struct { + /* + * The is the number of watcher records + * parsed from backend.initial_value + */ + size_t count; + /* + * This is the pointer to + * the optentially first watcher record + * parsed from backend.initial_value + * + * The pointer actually points to memory + * in backend.initial_value. + * + * Note it might be NULL, if count is 0. + */ + uint8_t *first; + /* + * This remembers if we already + * notified the watchers. + * + * As we only need to do that once during: + * do_locked + * or: + * between rec = fetch_locked + * and + * TALLOC_FREE(rec) + */ + bool alerted; + } watchers; + struct { + struct dbwrap_watcher watcher; + } wakeup; +}; + +static struct db_watched_record *db_record_get_watched_record(struct db_record *rec) +{ + /* + * we can't use wrec = talloc_get_type_abort() here! + * because wrec is likely a stack variable in + * dbwrap_watched_do_locked_fn() + * + * In order to have a least some protection + * we verify the cross reference pointers + * between rec and wrec + */ + struct db_watched_record *wrec = + (struct db_watched_record *)rec->private_data; + SMB_ASSERT(wrec->rec == rec); + return wrec; +} + +static NTSTATUS dbwrap_watched_record_storev( + struct db_watched_record *wrec, + const TDB_DATA *dbufs, int num_dbufs, int flags); +static NTSTATUS dbwrap_watched_storev(struct db_record *rec, + const TDB_DATA *dbufs, int num_dbufs, + int flags); +static NTSTATUS dbwrap_watched_delete(struct db_record *rec); +static void dbwrap_watched_trigger_wakeup(struct messaging_context *msg_ctx, + struct dbwrap_watcher *watcher); +static int db_watched_record_destructor(struct db_watched_record *wrec); + +static void db_watched_record_init(struct db_context *db, + struct messaging_context *msg_ctx, + struct db_record *rec, + struct db_watched_record *wrec, + struct db_record *backend_rec, + TDB_DATA backend_value) +{ + bool ok; + + *rec = (struct db_record) { + .db = db, + .key = dbwrap_record_get_key(backend_rec), + .storev = dbwrap_watched_storev, + .delete_rec = dbwrap_watched_delete, + .private_data = wrec, + }; + + *wrec = (struct db_watched_record) { + .rec = rec, + .self = messaging_server_id(msg_ctx), + .backend = { + .rec = backend_rec, + .initial_value = backend_value, + .initial_valid = true, + }, + }; + + ok = dbwrap_watch_rec_parse(backend_value, + &wrec->watchers.first, + &wrec->watchers.count, + &rec->value); + if (!ok) { + dbwrap_watch_log_invalid_record(rec->db, rec->key, backend_value); + /* wipe invalid data */ + rec->value = (TDB_DATA) { .dptr = NULL, .dsize = 0 }; + } +} + +static struct db_record *dbwrap_watched_fetch_locked( + struct db_context *db, TALLOC_CTX *mem_ctx, TDB_DATA key) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + struct db_record *rec = NULL; + struct db_watched_record *wrec = NULL; + struct db_record *backend_rec = NULL; + TDB_DATA backend_value = { .dptr = NULL, }; + + rec = talloc_zero(mem_ctx, struct db_record); + if (rec == NULL) { + return NULL; + } + wrec = talloc_zero(rec, struct db_watched_record); + if (wrec == NULL) { + TALLOC_FREE(rec); + return NULL; + } + + backend_rec = dbwrap_fetch_locked(ctx->backend, wrec, key); + if (backend_rec == NULL) { + TALLOC_FREE(rec); + return NULL; + } + backend_value = dbwrap_record_get_value(backend_rec); + + db_watched_record_init(db, ctx->msg, + rec, wrec, + backend_rec, backend_value); + rec->value_valid = true; + talloc_set_destructor(wrec, db_watched_record_destructor); + + return rec; +} + +struct db_watched_record_fini_state { + struct db_watched_record *wrec; + TALLOC_CTX *frame; + TDB_DATA dbufs[2]; + int num_dbufs; + bool ok; +}; + +static void db_watched_record_fini_fetcher(TDB_DATA key, + TDB_DATA backend_value, + void *private_data) +{ + struct db_watched_record_fini_state *state = + (struct db_watched_record_fini_state *)private_data; + struct db_watched_record *wrec = state->wrec; + struct db_record *rec = wrec->rec; + TDB_DATA value = {}; + bool ok; + size_t copy_size; + + /* + * We're within dbwrap_parse_record() + * and backend_value directly points into + * the mmap'ed tdb, so we need to copy the + * parts we require. + */ + + ok = dbwrap_watch_rec_parse(backend_value, NULL, NULL, &value); + if (!ok) { + struct db_context *db = dbwrap_record_get_db(rec); + + dbwrap_watch_log_invalid_record(db, key, backend_value); + + /* wipe invalid data */ + value = (TDB_DATA) { .dptr = NULL, .dsize = 0 }; + } + + copy_size = MIN(rec->value.dsize, value.dsize); + if (copy_size != 0) { + /* + * First reuse the buffer we already had + * as much as we can. + */ + memcpy(rec->value.dptr, value.dptr, copy_size); + state->dbufs[state->num_dbufs++] = rec->value; + value.dsize -= copy_size; + value.dptr += copy_size; + } + + if (value.dsize != 0) { + uint8_t *p = NULL; + + /* + * There's still new data left + * allocate it on callers stackframe + */ + p = talloc_memdup(state->frame, value.dptr, value.dsize); + if (p == NULL) { + DBG_WARNING("failed to allocate %zu bytes\n", + value.dsize); + return; + } + + state->dbufs[state->num_dbufs++] = (TDB_DATA) { + .dptr = p, .dsize = value.dsize, + }; + } + + state->ok = true; +} + +static void db_watched_record_fini(struct db_watched_record *wrec) +{ + struct db_watched_record_fini_state state = { .wrec = wrec, }; + struct db_context *backend = dbwrap_record_get_db(wrec->backend.rec); + struct db_record *rec = wrec->rec; + TDB_DATA key = dbwrap_record_get_key(wrec->backend.rec); + NTSTATUS status; + + if (!wrec->force_fini_store) { + return; + } + + if (wrec->backend.initial_valid) { + if (rec->value.dsize != 0) { + state.dbufs[state.num_dbufs++] = rec->value; + } + } else { + /* + * We need to fetch the current + * value from the backend again, + * which may need to allocate memory + * on the provided stackframe. + */ + + state.frame = talloc_stackframe(); + + status = dbwrap_parse_record(backend, key, + db_watched_record_fini_fetcher, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_WARNING("dbwrap_parse_record failed: %s\n", + nt_errstr(status)); + TALLOC_FREE(state.frame); + return; + } + if (!state.ok) { + TALLOC_FREE(state.frame); + return; + } + } + + /* + * We don't want to wake up others just because + * we added ourself as new watcher. But if we + * removed outself from the first position + * we need to alert the next one. + */ + if (!wrec->removed_first) { + dbwrap_watched_watch_skip_alerting(rec); + } + + status = dbwrap_watched_record_storev(wrec, state.dbufs, state.num_dbufs, 0); + TALLOC_FREE(state.frame); + if (!NT_STATUS_IS_OK(status)) { + DBG_WARNING("dbwrap_watched_record_storev failed: %s\n", + nt_errstr(status)); + return; + } + + return; +} + +static int db_watched_record_destructor(struct db_watched_record *wrec) +{ + struct db_record *rec = wrec->rec; + struct db_watched_ctx *ctx = talloc_get_type_abort( + rec->db->private_data, struct db_watched_ctx); + + db_watched_record_fini(wrec); + TALLOC_FREE(wrec->backend.rec); + dbwrap_watched_trigger_wakeup(ctx->msg, &wrec->wakeup.watcher); + return 0; +} + +struct dbwrap_watched_do_locked_state { + struct db_context *db; + struct messaging_context *msg_ctx; + struct db_watched_record *wrec; + struct db_record *rec; + void (*fn)(struct db_record *rec, + TDB_DATA value, + void *private_data); + void *private_data; +}; + +static void dbwrap_watched_do_locked_fn( + struct db_record *backend_rec, + TDB_DATA backend_value, + void *private_data) +{ + struct dbwrap_watched_do_locked_state *state = + (struct dbwrap_watched_do_locked_state *)private_data; + + db_watched_record_init(state->db, state->msg_ctx, + state->rec, state->wrec, + backend_rec, backend_value); + + state->fn(state->rec, state->rec->value, state->private_data); + + db_watched_record_fini(state->wrec); +} + +static NTSTATUS dbwrap_watched_do_locked(struct db_context *db, TDB_DATA key, + void (*fn)(struct db_record *rec, + TDB_DATA value, + void *private_data), + void *private_data) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + struct db_watched_record wrec; + struct db_record rec; + struct dbwrap_watched_do_locked_state state = { + .db = db, .msg_ctx = ctx->msg, + .rec = &rec, .wrec = &wrec, + .fn = fn, .private_data = private_data, + }; + NTSTATUS status; + + status = dbwrap_do_locked( + ctx->backend, key, dbwrap_watched_do_locked_fn, &state); + if (!NT_STATUS_IS_OK(status)) { + DBG_DEBUG("dbwrap_do_locked returned %s\n", nt_errstr(status)); + return status; + } + + DBG_DEBUG("dbwrap_watched_do_locked_fn returned\n"); + + dbwrap_watched_trigger_wakeup(state.msg_ctx, &wrec.wakeup.watcher); + + return NT_STATUS_OK; +} + +static void dbwrap_watched_record_prepare_wakeup( + struct db_watched_record *wrec) +{ + /* + * Wakeup only needs to happen once (if at all) + */ + if (wrec->watchers.alerted) { + /* already done */ + return; + } + wrec->watchers.alerted = true; + + if (wrec->watchers.count == 0) { + DBG_DEBUG("No watchers\n"); + return; + } + + while (wrec->watchers.count != 0) { + struct server_id_buf tmp; + bool exists; + + dbwrap_watcher_get(&wrec->wakeup.watcher, wrec->watchers.first); + exists = serverid_exists(&wrec->wakeup.watcher.pid); + if (!exists) { + DBG_DEBUG("Discard non-existing waiter %s:%"PRIu64"\n", + server_id_str_buf(wrec->wakeup.watcher.pid, &tmp), + wrec->wakeup.watcher.instance); + wrec->watchers.first += DBWRAP_WATCHER_BUF_LENGTH; + wrec->watchers.count -= 1; + continue; + } + + /* + * We will only wakeup the first waiter, via + * dbwrap_watched_trigger_wakeup(), but keep + * all (including the first one) in the list that + * will be flushed back to the backend record + * again. Waiters are removing their entries + * via dbwrap_watched_watch_remove_instance() + * when they no longer want to monitor the record. + */ + DBG_DEBUG("Will alert first waiter %s:%"PRIu64"\n", + server_id_str_buf(wrec->wakeup.watcher.pid, &tmp), + wrec->wakeup.watcher.instance); + break; + } +} + +static void dbwrap_watched_trigger_wakeup(struct messaging_context *msg_ctx, + struct dbwrap_watcher *watcher) +{ + struct server_id_buf tmp; + uint8_t instance_buf[8]; + NTSTATUS status; + + if (watcher->instance == 0) { + DBG_DEBUG("No one to wakeup\n"); + return; + } + + DBG_DEBUG("Alerting %s:%"PRIu64"\n", + server_id_str_buf(watcher->pid, &tmp), + watcher->instance); + + SBVAL(instance_buf, 0, watcher->instance); + + status = messaging_send_buf( + msg_ctx, + watcher->pid, + MSG_DBWRAP_MODIFIED, + instance_buf, + sizeof(instance_buf)); + if (!NT_STATUS_IS_OK(status)) { + DBG_WARNING("messaging_send_buf to %s failed: %s - ignoring...\n", + server_id_str_buf(watcher->pid, &tmp), + nt_errstr(status)); + } +} + +static NTSTATUS dbwrap_watched_record_storev( + struct db_watched_record *wrec, + const TDB_DATA *dbufs, int num_dbufs, int flags) +{ + uint8_t num_watchers_buf[4] = { 0 }; + uint8_t add_buf[DBWRAP_WATCHER_BUF_LENGTH]; + size_t num_store_watchers; + TDB_DATA my_dbufs[num_dbufs+3]; + int num_my_dbufs = 0; + NTSTATUS status; + size_t add_count = 0; + + dbwrap_watched_record_prepare_wakeup(wrec); + + wrec->backend.initial_valid = false; + wrec->force_fini_store = false; + + if (wrec->added.pid.pid != 0) { + dbwrap_watcher_put(add_buf, &wrec->added); + add_count = 1; + } + + num_store_watchers = wrec->watchers.count + add_count; + if (num_store_watchers == 0 && num_dbufs == 0) { + status = dbwrap_record_delete(wrec->backend.rec); + return status; + } + if (num_store_watchers >= DBWRAP_MAX_WATCHERS) { + DBG_WARNING("Can't handle %zu watchers\n", + num_store_watchers); + return NT_STATUS_INSUFFICIENT_RESOURCES; + } + + SIVAL(num_watchers_buf, 0, num_store_watchers); + + my_dbufs[num_my_dbufs++] = (TDB_DATA) { + .dptr = num_watchers_buf, .dsize = sizeof(num_watchers_buf), + }; + if (wrec->watchers.count != 0) { + my_dbufs[num_my_dbufs++] = (TDB_DATA) { + .dptr = wrec->watchers.first, .dsize = wrec->watchers.count * DBWRAP_WATCHER_BUF_LENGTH, + }; + } + if (add_count != 0) { + my_dbufs[num_my_dbufs++] = (TDB_DATA) { + .dptr = add_buf, + .dsize = sizeof(add_buf), + }; + } + if (num_dbufs != 0) { + memcpy(my_dbufs+num_my_dbufs, dbufs, num_dbufs * sizeof(*dbufs)); + num_my_dbufs += num_dbufs; + } + + SMB_ASSERT(num_my_dbufs <= ARRAY_SIZE(my_dbufs)); + + status = dbwrap_record_storev( + wrec->backend.rec, my_dbufs, num_my_dbufs, flags); + return status; +} + +static NTSTATUS dbwrap_watched_storev(struct db_record *rec, + const TDB_DATA *dbufs, int num_dbufs, + int flags) +{ + struct db_watched_record *wrec = db_record_get_watched_record(rec); + + return dbwrap_watched_record_storev(wrec, dbufs, num_dbufs, flags); +} + +static NTSTATUS dbwrap_watched_delete(struct db_record *rec) +{ + struct db_watched_record *wrec = db_record_get_watched_record(rec); + + /* + * dbwrap_watched_record_storev() will figure out + * if the record should be deleted or if there are still + * watchers to be stored. + */ + return dbwrap_watched_record_storev(wrec, NULL, 0, 0); +} + +struct dbwrap_watched_traverse_state { + int (*fn)(struct db_record *rec, void *private_data); + void *private_data; +}; + +static int dbwrap_watched_traverse_fn(struct db_record *rec, + void *private_data) +{ + struct dbwrap_watched_traverse_state *state = private_data; + struct db_record prec = *rec; + bool ok; + + ok = dbwrap_watch_rec_parse(rec->value, NULL, NULL, &prec.value); + if (!ok) { + return 0; + } + if (prec.value.dsize == 0) { + return 0; + } + prec.value_valid = true; + + return state->fn(&prec, state->private_data); +} + +static int dbwrap_watched_traverse(struct db_context *db, + int (*fn)(struct db_record *rec, + void *private_data), + void *private_data) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + struct dbwrap_watched_traverse_state state = { + .fn = fn, .private_data = private_data }; + NTSTATUS status; + int ret; + + status = dbwrap_traverse( + ctx->backend, dbwrap_watched_traverse_fn, &state, &ret); + if (!NT_STATUS_IS_OK(status)) { + return -1; + } + return ret; +} + +static int dbwrap_watched_traverse_read(struct db_context *db, + int (*fn)(struct db_record *rec, + void *private_data), + void *private_data) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + struct dbwrap_watched_traverse_state state = { + .fn = fn, .private_data = private_data }; + NTSTATUS status; + int ret; + + status = dbwrap_traverse_read( + ctx->backend, dbwrap_watched_traverse_fn, &state, &ret); + if (!NT_STATUS_IS_OK(status)) { + return -1; + } + return ret; +} + +static int dbwrap_watched_get_seqnum(struct db_context *db) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + return dbwrap_get_seqnum(ctx->backend); +} + +static int dbwrap_watched_transaction_start(struct db_context *db) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + return dbwrap_transaction_start(ctx->backend); +} + +static int dbwrap_watched_transaction_commit(struct db_context *db) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + return dbwrap_transaction_commit(ctx->backend); +} + +static int dbwrap_watched_transaction_cancel(struct db_context *db) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + return dbwrap_transaction_cancel(ctx->backend); +} + +struct dbwrap_watched_parse_record_state { + struct db_context *db; + void (*parser)(TDB_DATA key, TDB_DATA data, void *private_data); + void *private_data; + bool ok; +}; + +static void dbwrap_watched_parse_record_parser(TDB_DATA key, TDB_DATA data, + void *private_data) +{ + struct dbwrap_watched_parse_record_state *state = private_data; + TDB_DATA userdata; + + state->ok = dbwrap_watch_rec_parse(data, NULL, NULL, &userdata); + if (!state->ok) { + dbwrap_watch_log_invalid_record(state->db, key, data); + return; + } + + state->parser(key, userdata, state->private_data); +} + +static NTSTATUS dbwrap_watched_parse_record( + struct db_context *db, TDB_DATA key, + void (*parser)(TDB_DATA key, TDB_DATA data, void *private_data), + void *private_data) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + struct dbwrap_watched_parse_record_state state = { + .db = db, + .parser = parser, + .private_data = private_data, + }; + NTSTATUS status; + + status = dbwrap_parse_record( + ctx->backend, key, dbwrap_watched_parse_record_parser, &state); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + if (!state.ok) { + return NT_STATUS_NOT_FOUND; + } + return NT_STATUS_OK; +} + +static void dbwrap_watched_parse_record_done(struct tevent_req *subreq); + +static struct tevent_req *dbwrap_watched_parse_record_send( + TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + struct db_context *db, + TDB_DATA key, + void (*parser)(TDB_DATA key, TDB_DATA data, void *private_data), + void *private_data, + enum dbwrap_req_state *req_state) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + struct tevent_req *req = NULL; + struct tevent_req *subreq = NULL; + struct dbwrap_watched_parse_record_state *state = NULL; + + req = tevent_req_create(mem_ctx, &state, + struct dbwrap_watched_parse_record_state); + if (req == NULL) { + *req_state = DBWRAP_REQ_ERROR; + return NULL; + } + + *state = (struct dbwrap_watched_parse_record_state) { + .parser = parser, + .private_data = private_data, + .ok = true, + }; + + subreq = dbwrap_parse_record_send(state, + ev, + ctx->backend, + key, + dbwrap_watched_parse_record_parser, + state, + req_state); + if (tevent_req_nomem(subreq, req)) { + *req_state = DBWRAP_REQ_ERROR; + return tevent_req_post(req, ev); + } + + tevent_req_set_callback(subreq, dbwrap_watched_parse_record_done, req); + return req; +} + +static void dbwrap_watched_parse_record_done(struct tevent_req *subreq) +{ + struct tevent_req *req = tevent_req_callback_data( + subreq, struct tevent_req); + struct dbwrap_watched_parse_record_state *state = tevent_req_data( + req, struct dbwrap_watched_parse_record_state); + NTSTATUS status; + + status = dbwrap_parse_record_recv(subreq); + TALLOC_FREE(subreq); + if (tevent_req_nterror(req, status)) { + return; + } + + if (!state->ok) { + tevent_req_nterror(req, NT_STATUS_NOT_FOUND); + return; + } + + tevent_req_done(req); + return; +} + +static NTSTATUS dbwrap_watched_parse_record_recv(struct tevent_req *req) +{ + NTSTATUS status; + + if (tevent_req_is_nterror(req, &status)) { + tevent_req_received(req); + return status; + } + + tevent_req_received(req); + return NT_STATUS_OK; +} + +static int dbwrap_watched_exists(struct db_context *db, TDB_DATA key) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + + return dbwrap_exists(ctx->backend, key); +} + +static size_t dbwrap_watched_id(struct db_context *db, uint8_t *id, + size_t idlen) +{ + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + + return dbwrap_db_id(ctx->backend, id, idlen); +} + +struct db_context *db_open_watched(TALLOC_CTX *mem_ctx, + struct db_context **backend, + struct messaging_context *msg) +{ + struct db_context *db; + struct db_watched_ctx *ctx; + + db = talloc_zero(mem_ctx, struct db_context); + if (db == NULL) { + return NULL; + } + ctx = talloc_zero(db, struct db_watched_ctx); + if (ctx == NULL) { + TALLOC_FREE(db); + return NULL; + } + db->private_data = ctx; + + ctx->msg = msg; + + ctx->backend = talloc_move(ctx, backend); + db->lock_order = ctx->backend->lock_order; + ctx->backend->lock_order = DBWRAP_LOCK_ORDER_NONE; + + db->fetch_locked = dbwrap_watched_fetch_locked; + db->do_locked = dbwrap_watched_do_locked; + db->traverse = dbwrap_watched_traverse; + db->traverse_read = dbwrap_watched_traverse_read; + db->get_seqnum = dbwrap_watched_get_seqnum; + db->transaction_start = dbwrap_watched_transaction_start; + db->transaction_commit = dbwrap_watched_transaction_commit; + db->transaction_cancel = dbwrap_watched_transaction_cancel; + db->parse_record = dbwrap_watched_parse_record; + db->parse_record_send = dbwrap_watched_parse_record_send; + db->parse_record_recv = dbwrap_watched_parse_record_recv; + db->exists = dbwrap_watched_exists; + db->id = dbwrap_watched_id; + db->name = dbwrap_name(ctx->backend); + + return db; +} + +uint64_t dbwrap_watched_watch_add_instance(struct db_record *rec) +{ + struct db_watched_record *wrec = db_record_get_watched_record(rec); + static uint64_t global_instance = 1; + + SMB_ASSERT(wrec->added.instance == 0); + + wrec->added = (struct dbwrap_watcher) { + .pid = wrec->self, + .instance = global_instance++, + }; + + wrec->force_fini_store = true; + + return wrec->added.instance; +} + +void dbwrap_watched_watch_remove_instance(struct db_record *rec, uint64_t instance) +{ + struct db_watched_record *wrec = db_record_get_watched_record(rec); + struct dbwrap_watcher clear_watcher = { + .pid = wrec->self, + .instance = instance, + }; + size_t i; + struct server_id_buf buf; + + if (instance == 0) { + return; + } + + if (wrec->added.instance == instance) { + SMB_ASSERT(server_id_equal(&wrec->added.pid, &wrec->self)); + DBG_DEBUG("Watcher %s:%"PRIu64" reverted from adding\n", + server_id_str_buf(clear_watcher.pid, &buf), + clear_watcher.instance); + ZERO_STRUCT(wrec->added); + } + + for (i=0; i < wrec->watchers.count; i++) { + struct dbwrap_watcher watcher; + size_t off = i*DBWRAP_WATCHER_BUF_LENGTH; + size_t next_off; + size_t full_len; + size_t move_len; + + dbwrap_watcher_get(&watcher, wrec->watchers.first + off); + + if (clear_watcher.instance != watcher.instance) { + continue; + } + if (!server_id_equal(&clear_watcher.pid, &watcher.pid)) { + continue; + } + + wrec->force_fini_store = true; + + if (i == 0) { + DBG_DEBUG("Watcher %s:%"PRIu64" removed from first position of %zu\n", + server_id_str_buf(clear_watcher.pid, &buf), + clear_watcher.instance, + wrec->watchers.count); + wrec->watchers.first += DBWRAP_WATCHER_BUF_LENGTH; + wrec->watchers.count -= 1; + wrec->removed_first = true; + return; + } + if (i == (wrec->watchers.count-1)) { + DBG_DEBUG("Watcher %s:%"PRIu64" removed from last position of %zu\n", + server_id_str_buf(clear_watcher.pid, &buf), + clear_watcher.instance, + wrec->watchers.count); + wrec->watchers.count -= 1; + return; + } + + DBG_DEBUG("Watcher %s:%"PRIu64" cleared at position %zu from %zu\n", + server_id_str_buf(clear_watcher.pid, &buf), + clear_watcher.instance, i+1, + wrec->watchers.count); + + next_off = off + DBWRAP_WATCHER_BUF_LENGTH; + full_len = wrec->watchers.count * DBWRAP_WATCHER_BUF_LENGTH; + move_len = full_len - next_off; + memmove(wrec->watchers.first + off, + wrec->watchers.first + next_off, + move_len); + wrec->watchers.count -= 1; + return; + } + + DBG_DEBUG("Watcher %s:%"PRIu64" not found in %zu watchers\n", + server_id_str_buf(clear_watcher.pid, &buf), + clear_watcher.instance, + wrec->watchers.count); + return; +} + +void dbwrap_watched_watch_skip_alerting(struct db_record *rec) +{ + struct db_watched_record *wrec = db_record_get_watched_record(rec); + + wrec->wakeup.watcher = (struct dbwrap_watcher) { .instance = 0, }; + wrec->watchers.alerted = true; +} + +void dbwrap_watched_watch_reset_alerting(struct db_record *rec) +{ + struct db_watched_record *wrec = db_record_get_watched_record(rec); + + wrec->wakeup.watcher = (struct dbwrap_watcher) { .instance = 0, }; + wrec->watchers.alerted = false; +} + +void dbwrap_watched_watch_force_alerting(struct db_record *rec) +{ + struct db_watched_record *wrec = db_record_get_watched_record(rec); + + dbwrap_watched_record_prepare_wakeup(wrec); +} + +struct dbwrap_watched_watch_state { + struct db_context *db; + TDB_DATA key; + struct dbwrap_watcher watcher; + struct server_id blocker; + bool blockerdead; +}; + +static bool dbwrap_watched_msg_filter(struct messaging_rec *rec, + void *private_data); +static void dbwrap_watched_watch_done(struct tevent_req *subreq); +static void dbwrap_watched_watch_blocker_died(struct tevent_req *subreq); +static int dbwrap_watched_watch_state_destructor( + struct dbwrap_watched_watch_state *state); + +struct tevent_req *dbwrap_watched_watch_send(TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + struct db_record *rec, + uint64_t resumed_instance, + struct server_id blocker) +{ + struct db_context *db = dbwrap_record_get_db(rec); + struct db_watched_ctx *ctx = talloc_get_type_abort( + db->private_data, struct db_watched_ctx); + struct db_watched_record *wrec = db_record_get_watched_record(rec); + struct tevent_req *req, *subreq; + struct dbwrap_watched_watch_state *state; + uint64_t instance; + + req = tevent_req_create(mem_ctx, &state, + struct dbwrap_watched_watch_state); + if (req == NULL) { + return NULL; + } + state->db = db; + state->blocker = blocker; + + if (ctx->msg == NULL) { + tevent_req_nterror(req, NT_STATUS_NOT_SUPPORTED); + return tevent_req_post(req, ev); + } + + if (resumed_instance == 0 && wrec->added.instance == 0) { + /* + * Adding a new instance + */ + instance = dbwrap_watched_watch_add_instance(rec); + } else if (resumed_instance != 0 && wrec->added.instance == 0) { + /* + * Resuming an existing instance that was + * already present before do_locked started + */ + instance = resumed_instance; + } else if (resumed_instance == wrec->added.instance) { + /* + * The caller used dbwrap_watched_watch_add_instance() + * already during this do_locked() invocation. + */ + instance = resumed_instance; + } else { + tevent_req_nterror(req, NT_STATUS_REQUEST_NOT_ACCEPTED); + return tevent_req_post(req, ev); + } + + state->watcher = (struct dbwrap_watcher) { + .pid = messaging_server_id(ctx->msg), + .instance = instance, + }; + + state->key = tdb_data_talloc_copy(state, rec->key); + if (tevent_req_nomem(state->key.dptr, req)) { + return tevent_req_post(req, ev); + } + + subreq = messaging_filtered_read_send( + state, ev, ctx->msg, dbwrap_watched_msg_filter, state); + if (tevent_req_nomem(subreq, req)) { + return tevent_req_post(req, ev); + } + tevent_req_set_callback(subreq, dbwrap_watched_watch_done, req); + + talloc_set_destructor(state, dbwrap_watched_watch_state_destructor); + + if (blocker.pid != 0) { + subreq = server_id_watch_send(state, ev, blocker); + if (tevent_req_nomem(subreq, req)) { + return tevent_req_post(req, ev); + } + tevent_req_set_callback( + subreq, dbwrap_watched_watch_blocker_died, req); + } + + return req; +} + +static void dbwrap_watched_watch_blocker_died(struct tevent_req *subreq) +{ + struct tevent_req *req = tevent_req_callback_data( + subreq, struct tevent_req); + struct dbwrap_watched_watch_state *state = tevent_req_data( + req, struct dbwrap_watched_watch_state); + int ret; + + ret = server_id_watch_recv(subreq, NULL); + TALLOC_FREE(subreq); + if (ret != 0) { + tevent_req_nterror(req, map_nt_error_from_unix(ret)); + return; + } + state->blockerdead = true; + tevent_req_done(req); +} + +static void dbwrap_watched_watch_state_destructor_fn( + struct db_record *rec, + TDB_DATA value, + void *private_data) +{ + struct dbwrap_watched_watch_state *state = talloc_get_type_abort( + private_data, struct dbwrap_watched_watch_state); + + /* + * Here we just remove ourself from the in memory + * watchers array and let db_watched_record_fini() + * call dbwrap_watched_record_storev() to do the magic + * of writing back the modified in memory copy. + */ + dbwrap_watched_watch_remove_instance(rec, state->watcher.instance); + return; +} + +static int dbwrap_watched_watch_state_destructor( + struct dbwrap_watched_watch_state *state) +{ + NTSTATUS status; + + status = dbwrap_do_locked( + state->db, + state->key, + dbwrap_watched_watch_state_destructor_fn, + state); + if (!NT_STATUS_IS_OK(status)) { + DBG_WARNING("dbwrap_do_locked failed: %s\n", + nt_errstr(status)); + } + return 0; +} + +static bool dbwrap_watched_msg_filter(struct messaging_rec *rec, + void *private_data) +{ + struct dbwrap_watched_watch_state *state = talloc_get_type_abort( + private_data, struct dbwrap_watched_watch_state); + uint64_t instance; + + if (rec->msg_type != MSG_DBWRAP_MODIFIED) { + return false; + } + if (rec->num_fds != 0) { + return false; + } + + if (rec->buf.length != sizeof(instance)) { + DBG_DEBUG("Got size %zu, expected %zu\n", + rec->buf.length, + sizeof(instance)); + return false; + } + + instance = BVAL(rec->buf.data, 0); + + if (instance != state->watcher.instance) { + DBG_DEBUG("Got instance %"PRIu64", expected %"PRIu64"\n", + instance, + state->watcher.instance); + return false; + } + + return true; +} + +static void dbwrap_watched_watch_done(struct tevent_req *subreq) +{ + struct tevent_req *req = tevent_req_callback_data( + subreq, struct tevent_req); + struct dbwrap_watched_watch_state *state = tevent_req_data( + req, struct dbwrap_watched_watch_state); + struct messaging_rec *rec; + int ret; + + ret = messaging_filtered_read_recv(subreq, state, &rec); + TALLOC_FREE(subreq); + if (ret != 0) { + tevent_req_nterror(req, map_nt_error_from_unix(ret)); + return; + } + tevent_req_done(req); +} + +NTSTATUS dbwrap_watched_watch_recv(struct tevent_req *req, + uint64_t *pkeep_instance, + bool *blockerdead, + struct server_id *blocker) +{ + struct dbwrap_watched_watch_state *state = tevent_req_data( + req, struct dbwrap_watched_watch_state); + NTSTATUS status; + + if (tevent_req_is_nterror(req, &status)) { + tevent_req_received(req); + return status; + } + if (pkeep_instance != NULL) { + *pkeep_instance = state->watcher.instance; + /* + * No need to remove ourselves anymore, + * the caller will take care of removing itself. + */ + talloc_set_destructor(state, NULL); + } + if (blockerdead != NULL) { + *blockerdead = state->blockerdead; + } + if (blocker != NULL) { + *blocker = state->blocker; + } + tevent_req_received(req); + return NT_STATUS_OK; +} + diff --git a/source3/lib/dbwrap/dbwrap_watch.h b/source3/lib/dbwrap/dbwrap_watch.h new file mode 100644 index 0000000..1b93e13 --- /dev/null +++ b/source3/lib/dbwrap/dbwrap_watch.h @@ -0,0 +1,45 @@ +/* + Unix SMB/CIFS implementation. + Watch dbwrap record changes + Copyright (C) Volker Lendecke 2012 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#ifndef __DBWRAP_WATCH_H__ +#define __DBWRAP_WATCH_H__ + +#include +#include "dbwrap/dbwrap.h" +#include "messages.h" + +struct db_context *db_open_watched(TALLOC_CTX *mem_ctx, + struct db_context **backend, + struct messaging_context *msg); +uint64_t dbwrap_watched_watch_add_instance(struct db_record *rec); +void dbwrap_watched_watch_remove_instance(struct db_record *rec, uint64_t instance); +void dbwrap_watched_watch_skip_alerting(struct db_record *rec); +void dbwrap_watched_watch_reset_alerting(struct db_record *rec); +void dbwrap_watched_watch_force_alerting(struct db_record *rec); +struct tevent_req *dbwrap_watched_watch_send(TALLOC_CTX *mem_ctx, + struct tevent_context *ev, + struct db_record *rec, + uint64_t resume_instance, + struct server_id blocker); +NTSTATUS dbwrap_watched_watch_recv(struct tevent_req *req, + uint64_t *pkeep_instance, + bool *blockerdead, + struct server_id *blocker); + +#endif /* __DBWRAP_WATCH_H__ */ -- cgit v1.2.3