diff options
Diffstat (limited to 'reftable/writer.c')
| -rw-r--r-- | reftable/writer.c | 461 |
1 files changed, 291 insertions, 170 deletions
diff --git a/reftable/writer.c b/reftable/writer.c index 2e322a5683..be0fae6cb2 100644 --- a/reftable/writer.c +++ b/reftable/writer.c @@ -49,8 +49,14 @@ static int padded_write(struct reftable_writer *w, uint8_t *data, size_t len, { int n = 0; if (w->pending_padding > 0) { - uint8_t *zeroed = reftable_calloc(w->pending_padding); - int n = w->write(w->write_arg, zeroed, w->pending_padding); + uint8_t *zeroed; + int n; + + zeroed = reftable_calloc(w->pending_padding, sizeof(*zeroed)); + if (!zeroed) + return -1; + + n = w->write(w->write_arg, zeroed, w->pending_padding); if (n < 0) return n; @@ -102,43 +108,61 @@ static int writer_write_header(struct reftable_writer *w, uint8_t *dest) return header_size(writer_version(w)); } -static void writer_reinit_block_writer(struct reftable_writer *w, uint8_t typ) +static int writer_reinit_block_writer(struct reftable_writer *w, uint8_t typ) { - int block_start = 0; - if (w->next == 0) { + int block_start = 0, ret; + + if (w->next == 0) block_start = header_size(writer_version(w)); - } - strbuf_release(&w->last_key); - block_writer_init(&w->block_writer_data, typ, w->block, - w->opts.block_size, block_start, - hash_size(w->opts.hash_id)); + reftable_buf_reset(&w->last_key); + ret = block_writer_init(&w->block_writer_data, typ, w->block, + w->opts.block_size, block_start, + hash_size(w->opts.hash_id)); + if (ret < 0) + return ret; + w->block_writer = &w->block_writer_data; w->block_writer->restart_interval = w->opts.restart_interval; -} -static struct strbuf reftable_empty_strbuf = STRBUF_INIT; + return 0; +} -struct reftable_writer * -reftable_new_writer(ssize_t (*writer_func)(void *, const void *, size_t), - void *writer_arg, struct reftable_write_options *opts) +int reftable_writer_new(struct reftable_writer **out, + ssize_t (*writer_func)(void *, const void *, size_t), + int (*flush_func)(void *), + void *writer_arg, const struct reftable_write_options *_opts) { - struct reftable_writer *wp = - reftable_calloc(sizeof(struct reftable_writer)); - strbuf_init(&wp->block_writer_data.last_key, 0); - options_set_defaults(opts); - if (opts->block_size >= (1 << 24)) { - /* TODO - error return? */ - abort(); + struct reftable_write_options opts = {0}; + struct reftable_writer *wp; + + wp = reftable_calloc(1, sizeof(*wp)); + if (!wp) + return REFTABLE_OUT_OF_MEMORY_ERROR; + + if (_opts) + opts = *_opts; + options_set_defaults(&opts); + if (opts.block_size >= (1 << 24)) + BUG("configured block size exceeds 16MB"); + + reftable_buf_init(&wp->block_writer_data.last_key); + reftable_buf_init(&wp->last_key); + reftable_buf_init(&wp->scratch); + REFTABLE_CALLOC_ARRAY(wp->block, opts.block_size); + if (!wp->block) { + reftable_free(wp); + return REFTABLE_OUT_OF_MEMORY_ERROR; } - wp->last_key = reftable_empty_strbuf; - wp->block = reftable_calloc(opts->block_size); wp->write = writer_func; wp->write_arg = writer_arg; - wp->opts = *opts; + wp->opts = opts; + wp->flush = flush_func; writer_reinit_block_writer(wp, BLOCK_TYPE_REF); - return wp; + *out = wp; + + return 0; } void reftable_writer_set_limits(struct reftable_writer *w, uint64_t min, @@ -148,16 +172,27 @@ void reftable_writer_set_limits(struct reftable_writer *w, uint64_t min, w->max_update_index = max; } +static void writer_release(struct reftable_writer *w) +{ + if (w) { + reftable_free(w->block); + w->block = NULL; + block_writer_release(&w->block_writer_data); + w->block_writer = NULL; + writer_clear_index(w); + reftable_buf_release(&w->last_key); + reftable_buf_release(&w->scratch); + } +} + void reftable_writer_free(struct reftable_writer *w) { - if (!w) - return; - reftable_free(w->block); + writer_release(w); reftable_free(w); } struct obj_index_tree_node { - struct strbuf hash; + struct reftable_buf hash; uint64_t *offsets; size_t offset_len; size_t offset_cap; @@ -165,90 +200,119 @@ struct obj_index_tree_node { #define OBJ_INDEX_TREE_NODE_INIT \ { \ - .hash = STRBUF_INIT \ + .hash = REFTABLE_BUF_INIT \ } static int obj_index_tree_node_compare(const void *a, const void *b) { - return strbuf_cmp(&((const struct obj_index_tree_node *)a)->hash, + return reftable_buf_cmp(&((const struct obj_index_tree_node *)a)->hash, &((const struct obj_index_tree_node *)b)->hash); } -static void writer_index_hash(struct reftable_writer *w, struct strbuf *hash) +static int writer_index_hash(struct reftable_writer *w, struct reftable_buf *hash) { uint64_t off = w->next; - struct obj_index_tree_node want = { .hash = *hash }; + struct obj_index_tree_node *key; + struct tree_node *node; - struct tree_node *node = tree_search(&want, &w->obj_index_tree, - &obj_index_tree_node_compare, 0); - struct obj_index_tree_node *key = NULL; + node = tree_search(w->obj_index_tree, &want, &obj_index_tree_node_compare); if (!node) { struct obj_index_tree_node empty = OBJ_INDEX_TREE_NODE_INIT; - key = reftable_malloc(sizeof(struct obj_index_tree_node)); + int err; + + key = reftable_malloc(sizeof(*key)); + if (!key) + return REFTABLE_OUT_OF_MEMORY_ERROR; + *key = empty; - strbuf_reset(&key->hash); - strbuf_addbuf(&key->hash, hash); - tree_search((void *)key, &w->obj_index_tree, - &obj_index_tree_node_compare, 1); + reftable_buf_reset(&key->hash); + err = reftable_buf_add(&key->hash, hash->buf, hash->len); + if (err < 0) + return err; + tree_insert(&w->obj_index_tree, key, + &obj_index_tree_node_compare); } else { key = node->key; } - if (key->offset_len > 0 && key->offsets[key->offset_len - 1] == off) { - return; - } - - if (key->offset_len == key->offset_cap) { - key->offset_cap = 2 * key->offset_cap + 1; - key->offsets = reftable_realloc( - key->offsets, sizeof(uint64_t) * key->offset_cap); - } + if (key->offset_len > 0 && key->offsets[key->offset_len - 1] == off) + return 0; + REFTABLE_ALLOC_GROW(key->offsets, key->offset_len + 1, key->offset_cap); + if (!key->offsets) + return REFTABLE_OUT_OF_MEMORY_ERROR; key->offsets[key->offset_len++] = off; + + return 0; } static int writer_add_record(struct reftable_writer *w, struct reftable_record *rec) { - struct strbuf key = STRBUF_INIT; - int err = -1; - reftable_record_key(rec, &key); - if (strbuf_cmp(&w->last_key, &key) >= 0) { + int err; + + err = reftable_record_key(rec, &w->scratch); + if (err < 0) + goto done; + + if (reftable_buf_cmp(&w->last_key, &w->scratch) >= 0) { err = REFTABLE_API_ERROR; goto done; } - strbuf_reset(&w->last_key); - strbuf_addbuf(&w->last_key, &key); + reftable_buf_reset(&w->last_key); + err = reftable_buf_add(&w->last_key, w->scratch.buf, w->scratch.len); + if (err < 0) + goto done; + if (!w->block_writer) { - writer_reinit_block_writer(w, reftable_record_type(rec)); + err = writer_reinit_block_writer(w, reftable_record_type(rec)); + if (err < 0) + goto done; } - assert(block_writer_type(w->block_writer) == reftable_record_type(rec)); + if (block_writer_type(w->block_writer) != reftable_record_type(rec)) + BUG("record of type %d added to writer of type %d", + reftable_record_type(rec), block_writer_type(w->block_writer)); - if (block_writer_add(w->block_writer, rec) == 0) { + /* + * Try to add the record to the writer. If this succeeds then we're + * done. Otherwise the block writer may have hit the block size limit + * and needs to be flushed. + */ + if (!block_writer_add(w->block_writer, rec)) { err = 0; goto done; } + /* + * The current block is full, so we need to flush and reinitialize the + * writer to start writing the next block. + */ err = writer_flush_block(w); - if (err < 0) { + if (err < 0) + goto done; + err = writer_reinit_block_writer(w, reftable_record_type(rec)); + if (err < 0) goto done; - } - writer_reinit_block_writer(w, reftable_record_type(rec)); + /* + * Try to add the record to the writer again. If this still fails then + * the record does not fit into the block size. + * + * TODO: it would be great to have `block_writer_add()` return proper + * error codes so that we don't have to second-guess the failure + * mode here. + */ err = block_writer_add(w->block_writer, rec); - if (err == -1) { - /* we are writing into memory, so an error can only mean it - * doesn't fit. */ + if (err) { err = REFTABLE_ENTRY_TOO_BIG_ERROR; goto done; } done: - strbuf_release(&key); return err; } @@ -261,11 +325,10 @@ int reftable_writer_add_ref(struct reftable_writer *w, .ref = *ref }, }; - int err = 0; + int err; - if (!ref->refname) - return REFTABLE_API_ERROR; - if (ref->update_index < w->min_update_index || + if (!ref->refname || + ref->update_index < w->min_update_index || ref->update_index > w->max_update_index) return REFTABLE_API_ERROR; @@ -273,24 +336,36 @@ int reftable_writer_add_ref(struct reftable_writer *w, err = writer_add_record(w, &rec); if (err < 0) - return err; + goto out; if (!w->opts.skip_index_objects && reftable_ref_record_val1(ref)) { - struct strbuf h = STRBUF_INIT; - strbuf_add(&h, (char *)reftable_ref_record_val1(ref), - hash_size(w->opts.hash_id)); - writer_index_hash(w, &h); - strbuf_release(&h); + reftable_buf_reset(&w->scratch); + err = reftable_buf_add(&w->scratch, (char *)reftable_ref_record_val1(ref), + hash_size(w->opts.hash_id)); + if (err < 0) + goto out; + + err = writer_index_hash(w, &w->scratch); + if (err < 0) + goto out; } if (!w->opts.skip_index_objects && reftable_ref_record_val2(ref)) { - struct strbuf h = STRBUF_INIT; - strbuf_add(&h, reftable_ref_record_val2(ref), - hash_size(w->opts.hash_id)); - writer_index_hash(w, &h); - strbuf_release(&h); + reftable_buf_reset(&w->scratch); + err = reftable_buf_add(&w->scratch, reftable_ref_record_val2(ref), + hash_size(w->opts.hash_id)); + if (err < 0) + goto out; + + err = writer_index_hash(w, &w->scratch); + if (err < 0) + goto out; } - return 0; + + err = 0; + +out: + return err; } int reftable_writer_add_refs(struct reftable_writer *w, @@ -330,7 +405,7 @@ int reftable_writer_add_log(struct reftable_writer *w, struct reftable_log_record *log) { char *input_log_message = NULL; - struct strbuf cleaned_message = STRBUF_INIT; + struct reftable_buf cleaned_message = REFTABLE_BUF_INIT; int err = 0; if (log->value_type == REFTABLE_LOG_DELETION) @@ -341,24 +416,34 @@ int reftable_writer_add_log(struct reftable_writer *w, input_log_message = log->value.update.message; if (!w->opts.exact_log_message && log->value.update.message) { - strbuf_addstr(&cleaned_message, log->value.update.message); + err = reftable_buf_addstr(&cleaned_message, log->value.update.message); + if (err < 0) + goto done; + while (cleaned_message.len && - cleaned_message.buf[cleaned_message.len - 1] == '\n') - strbuf_setlen(&cleaned_message, - cleaned_message.len - 1); + cleaned_message.buf[cleaned_message.len - 1] == '\n') { + err = reftable_buf_setlen(&cleaned_message, + cleaned_message.len - 1); + if (err < 0) + goto done; + } if (strchr(cleaned_message.buf, '\n')) { /* multiple lines not allowed. */ err = REFTABLE_API_ERROR; goto done; } - strbuf_addstr(&cleaned_message, "\n"); + + err = reftable_buf_addstr(&cleaned_message, "\n"); + if (err < 0) + goto done; + log->value.update.message = cleaned_message.buf; } err = reftable_writer_add_log_verbatim(w, log); log->value.update.message = input_log_message; done: - strbuf_release(&cleaned_message); + reftable_buf_release(&cleaned_message); return err; } @@ -377,24 +462,45 @@ int reftable_writer_add_logs(struct reftable_writer *w, static int writer_finish_section(struct reftable_writer *w) { + struct reftable_block_stats *bstats = NULL; uint8_t typ = block_writer_type(w->block_writer); uint64_t index_start = 0; int max_level = 0; - int threshold = w->opts.unpadded ? 1 : 3; + size_t threshold = w->opts.unpadded ? 1 : 3; int before_blocks = w->stats.idx_stats.blocks; - int err = writer_flush_block(w); - int i = 0; - struct reftable_block_stats *bstats = NULL; + int err; + + err = writer_flush_block(w); if (err < 0) return err; + /* + * When the section we are about to index has a lot of blocks then the + * index itself may span across multiple blocks, as well. This would + * require a linear scan over index blocks only to find the desired + * indexed block, which is inefficient. Instead, we write a multi-level + * index where index records of level N+1 will refer to index blocks of + * level N. This isn't constant time, either, but at least logarithmic. + * + * This loop handles writing this multi-level index. Note that we write + * the lowest-level index pointing to the indexed blocks first. We then + * continue writing additional index levels until the current level has + * less blocks than the threshold so that the highest level will be at + * the end of the index section. + * + * Readers are thus required to start reading the index section from + * its end, which is why we set `index_start` to the beginning of the + * last index section. + */ while (w->index_len > threshold) { struct reftable_index_record *idx = NULL; - int idx_len = 0; + size_t i, idx_len; max_level++; index_start = w->next; - writer_reinit_block_writer(w, BLOCK_TYPE_INDEX); + err = writer_reinit_block_writer(w, BLOCK_TYPE_INDEX); + if (err < 0) + return err; idx = w->index; idx_len = w->index_len; @@ -409,48 +515,41 @@ static int writer_finish_section(struct reftable_writer *w) .idx = idx[i], }, }; - if (block_writer_add(w->block_writer, &rec) == 0) { - continue; - } - err = writer_flush_block(w); + err = writer_add_record(w, &rec); if (err < 0) return err; + } - writer_reinit_block_writer(w, BLOCK_TYPE_INDEX); + err = writer_flush_block(w); + if (err < 0) + return err; - err = block_writer_add(w->block_writer, &rec); - if (err != 0) { - /* write into fresh block should always succeed - */ - abort(); - } - } - for (i = 0; i < idx_len; i++) { - strbuf_release(&idx[i].last_key); - } + for (i = 0; i < idx_len; i++) + reftable_buf_release(&idx[i].last_key); reftable_free(idx); } + /* + * The index may still contain a number of index blocks lower than the + * threshold. Clear it so that these entries don't leak into the next + * index section. + */ writer_clear_index(w); - err = writer_flush_block(w); - if (err < 0) - return err; - bstats = writer_reftable_block_stats(w, typ); bstats->index_blocks = w->stats.idx_stats.blocks - before_blocks; bstats->index_offset = index_start; bstats->max_index_level = max_level; /* Reinit lastKey, as the next section can start with any key. */ - w->last_key.len = 0; + reftable_buf_reset(&w->last_key); return 0; } struct common_prefix_arg { - struct strbuf *last; + struct reftable_buf *last; int max; }; @@ -495,7 +594,10 @@ static void write_object_record(void *void_arg, void *key) if (arg->err < 0) goto done; - writer_reinit_block_writer(arg->w, BLOCK_TYPE_OBJ); + arg->err = writer_reinit_block_writer(arg->w, BLOCK_TYPE_OBJ); + if (arg->err < 0) + goto done; + arg->err = block_writer_add(arg->w->block_writer, &rec); if (arg->err == 0) goto done; @@ -509,12 +611,12 @@ static void write_object_record(void *void_arg, void *key) done:; } -static void object_record_free(void *void_arg, void *key) +static void object_record_free(void *void_arg UNUSED, void *key) { struct obj_index_tree_node *entry = key; - FREE_AND_NULL(entry->offsets); - strbuf_release(&entry->hash); + REFTABLE_FREE_AND_NULL(entry->offsets); + reftable_buf_release(&entry->hash); reftable_free(entry); } @@ -524,16 +626,18 @@ static int writer_dump_object_index(struct reftable_writer *w) struct common_prefix_arg common = { .max = 1, /* obj_id_len should be >= 2. */ }; - if (w->obj_index_tree) { + int err; + + if (w->obj_index_tree) infix_walk(w->obj_index_tree, &update_common, &common); - } w->stats.object_id_len = common.max + 1; - writer_reinit_block_writer(w, BLOCK_TYPE_OBJ); + err = writer_reinit_block_writer(w, BLOCK_TYPE_OBJ); + if (err < 0) + return err; - if (w->obj_index_tree) { + if (w->obj_index_tree) infix_walk(w->obj_index_tree, &write_object_record, &closure); - } if (closure.err < 0) return closure.err; @@ -603,6 +707,12 @@ int reftable_writer_close(struct reftable_writer *w) put_be32(p, crc32(0, footer, p - footer)); p += 4; + err = w->flush(w->write_arg); + if (err < 0) { + err = REFTABLE_IO_ERROR; + goto done; + } + err = padded_write(w, footer, footer_size(writer_version(w)), 0); if (err < 0) goto done; @@ -613,82 +723,93 @@ int reftable_writer_close(struct reftable_writer *w) } done: - /* free up memory. */ - block_writer_release(&w->block_writer_data); - writer_clear_index(w); - strbuf_release(&w->last_key); + writer_release(w); return err; } static void writer_clear_index(struct reftable_writer *w) { - int i = 0; - for (i = 0; i < w->index_len; i++) { - strbuf_release(&w->index[i].last_key); - } - - FREE_AND_NULL(w->index); + for (size_t i = 0; w->index && i < w->index_len; i++) + reftable_buf_release(&w->index[i].last_key); + REFTABLE_FREE_AND_NULL(w->index); w->index_len = 0; w->index_cap = 0; } -static const int debug = 0; - static int writer_flush_nonempty_block(struct reftable_writer *w) { + struct reftable_index_record index_record = { + .last_key = REFTABLE_BUF_INIT, + }; uint8_t typ = block_writer_type(w->block_writer); - struct reftable_block_stats *bstats = - writer_reftable_block_stats(w, typ); - uint64_t block_typ_off = (bstats->blocks == 0) ? w->next : 0; - int raw_bytes = block_writer_finish(w->block_writer); - int padding = 0; - int err = 0; - struct reftable_index_record ir = { .last_key = STRBUF_INIT }; + struct reftable_block_stats *bstats; + int raw_bytes, padding = 0, err; + uint64_t block_typ_off; + + /* + * Finish the current block. This will cause the block writer to emit + * restart points and potentially compress records in case we are + * writing a log block. + * + * Note that this is still happening in memory. + */ + raw_bytes = block_writer_finish(w->block_writer); if (raw_bytes < 0) return raw_bytes; - if (!w->opts.unpadded && typ != BLOCK_TYPE_LOG) { + /* + * By default, all records except for log records are padded to the + * block size. + */ + if (!w->opts.unpadded && typ != BLOCK_TYPE_LOG) padding = w->opts.block_size - raw_bytes; - } - if (block_typ_off > 0) { + bstats = writer_reftable_block_stats(w, typ); + block_typ_off = (bstats->blocks == 0) ? w->next : 0; + if (block_typ_off > 0) bstats->offset = block_typ_off; - } - bstats->entries += w->block_writer->entries; bstats->restarts += w->block_writer->restart_len; bstats->blocks++; w->stats.blocks++; - if (debug) { - fprintf(stderr, "block %c off %" PRIu64 " sz %d (%d)\n", typ, - w->next, raw_bytes, - get_be24(w->block + w->block_writer->header_off + 1)); - } - - if (w->next == 0) { + /* + * If this is the first block we're writing to the table then we need + * to also write the reftable header. + */ + if (!w->next) writer_write_header(w, w->block); - } err = padded_write(w, w->block, raw_bytes, padding); if (err < 0) return err; - if (w->index_cap == w->index_len) { - w->index_cap = 2 * w->index_cap + 1; - w->index = reftable_realloc( - w->index, - sizeof(struct reftable_index_record) * w->index_cap); - } - - ir.offset = w->next; - strbuf_reset(&ir.last_key); - strbuf_addbuf(&ir.last_key, &w->block_writer->last_key); - w->index[w->index_len] = ir; - + /* + * Add an index record for every block that we're writing. If we end up + * having more than a threshold of index records we will end up writing + * an index section in `writer_finish_section()`. Each index record + * contains the last record key of the block it is indexing as well as + * the offset of that block. + * + * Note that this also applies when flushing index blocks, in which + * case we will end up with a multi-level index. + */ + REFTABLE_ALLOC_GROW(w->index, w->index_len + 1, w->index_cap); + if (!w->index) + return REFTABLE_OUT_OF_MEMORY_ERROR; + + index_record.offset = w->next; + reftable_buf_reset(&index_record.last_key); + err = reftable_buf_add(&index_record.last_key, w->block_writer->last_key.buf, + w->block_writer->last_key.len); + if (err < 0) + return err; + w->index[w->index_len] = index_record; w->index_len++; + w->next += padding + raw_bytes; w->block_writer = NULL; + return 0; } |
