summaryrefslogtreecommitdiff
path: root/lib/sqfs/block_processor/backend.c
diff options
context:
space:
mode:
authorDavid Oberhollenzer <david.oberhollenzer@sigma-star.at>2021-03-22 11:30:15 +0100
committerDavid Oberhollenzer <david.oberhollenzer@sigma-star.at>2021-03-22 15:26:47 +0100
commit5aa0f30173ecf3b6538b9136cb4783fc19266288 (patch)
treee163429bb7ecf5c3cd343532e0fd60b5f1819d39 /lib/sqfs/block_processor/backend.c
parentc7056c1853b5defd5b933e651bf58dc94b4d3f8b (diff)
Cleanup the block processor file structure
A cleaner separation between common code, frontend code and backend code is made. The "is this byte blob zero" function is moved out to libutil (with test case and everything) with a more optimized implementation. Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
Diffstat (limited to 'lib/sqfs/block_processor/backend.c')
-rw-r--r--lib/sqfs/block_processor/backend.c288
1 files changed, 288 insertions, 0 deletions
diff --git a/lib/sqfs/block_processor/backend.c b/lib/sqfs/block_processor/backend.c
new file mode 100644
index 0000000..ff142c9
--- /dev/null
+++ b/lib/sqfs/block_processor/backend.c
@@ -0,0 +1,288 @@
+/* SPDX-License-Identifier: LGPL-3.0-or-later */
+/*
+ * backend.c
+ *
+ * Copyright (C) 2019 David Oberhollenzer <goliath@infraroot.at>
+ */
+#define SQFS_BUILDING_DLL
+#include "internal.h"
+
+static int set_block_size(sqfs_inode_generic_t **inode,
+ sqfs_u32 index, sqfs_u32 size)
+{
+ size_t min_size = (index + 1) * sizeof(sqfs_u32);
+ size_t avail = (*inode)->payload_bytes_available;
+ sqfs_inode_generic_t *new;
+ size_t newsz;
+
+ if (avail < min_size) {
+ newsz = avail ? avail : (sizeof(sqfs_u32) * 4);
+ while (newsz < min_size)
+ newsz *= 2;
+
+ if (SZ_ADD_OV(newsz, sizeof(**inode), &newsz))
+ return SQFS_ERROR_OVERFLOW;
+
+ if (sizeof(size_t) > sizeof(sqfs_u32)) {
+ if ((newsz - sizeof(**inode)) > 0x0FFFFFFFFUL)
+ return SQFS_ERROR_OVERFLOW;
+ }
+
+ new = realloc((*inode), newsz);
+ if (new == NULL)
+ return SQFS_ERROR_ALLOC;
+
+ (*inode) = new;
+ (*inode)->payload_bytes_available = newsz - sizeof(**inode);
+ }
+
+ (*inode)->extra[index] = size;
+
+ if (min_size >= (*inode)->payload_bytes_used)
+ (*inode)->payload_bytes_used = min_size;
+
+ return 0;
+}
+
+static void release_old_block(sqfs_block_processor_t *proc, sqfs_block_t *blk)
+{
+ blk->next = proc->free_list;
+ proc->free_list = blk;
+}
+
+static int process_completed_block(sqfs_block_processor_t *proc, sqfs_block_t *blk)
+{
+ sqfs_u64 location;
+ sqfs_u32 size;
+ int err;
+
+ err = proc->wr->write_data_block(proc->wr, blk->user, blk->size,
+ blk->checksum,
+ blk->flags & ~BLK_FLAG_INTERNAL,
+ blk->data, &location);
+ if (err)
+ goto out;
+
+ proc->stats.output_bytes_generated += blk->size;
+
+ if (blk->flags & SQFS_BLK_IS_SPARSE) {
+ if (blk->inode != NULL) {
+ sqfs_inode_make_extended(*(blk->inode));
+ (*(blk->inode))->data.file_ext.sparse += blk->size;
+
+ err = set_block_size(blk->inode, blk->index, 0);
+ if (err)
+ goto out;
+ }
+ proc->stats.sparse_block_count += 1;
+ } else if (blk->size != 0) {
+ size = blk->size;
+ if (!(blk->flags & SQFS_BLK_IS_COMPRESSED))
+ size |= 1 << 24;
+
+ if (blk->flags & SQFS_BLK_FRAGMENT_BLOCK) {
+ if (proc->frag_tbl != NULL) {
+ err = sqfs_frag_table_set(proc->frag_tbl,
+ blk->index, location,
+ size);
+ if (err)
+ goto out;
+ }
+ proc->stats.frag_block_count += 1;
+ } else {
+ if (blk->inode != NULL) {
+ err = set_block_size(blk->inode, blk->index,
+ size);
+ if (err)
+ goto out;
+ }
+ proc->stats.data_block_count += 1;
+ }
+ }
+
+ if (blk->flags & SQFS_BLK_LAST_BLOCK && blk->inode != NULL)
+ sqfs_inode_set_file_block_start(*(blk->inode), location);
+out:
+ release_old_block(proc, blk);
+ return err;
+}
+
+static int process_completed_fragment(sqfs_block_processor_t *proc,
+ sqfs_block_t *frag)
+{
+ chunk_info_t *chunk = NULL, search;
+ struct hash_entry *entry;
+ sqfs_u32 index, offset;
+ int err;
+
+ if (frag->flags & SQFS_BLK_IS_SPARSE) {
+ if (frag->inode != NULL) {
+ sqfs_inode_make_extended(*(frag->inode));
+ set_block_size(frag->inode, frag->index, 0);
+ (*(frag->inode))->data.file_ext.sparse += frag->size;
+ }
+ proc->stats.sparse_block_count += 1;
+ release_old_block(proc, frag);
+ return 0;
+ }
+
+ proc->stats.total_frag_count += 1;
+
+ if (!(frag->flags & SQFS_BLK_DONT_DEDUPLICATE)) {
+ search.hash = frag->checksum;
+ search.size = frag->size;
+
+ entry = hash_table_search_pre_hashed(proc->frag_ht,
+ search.hash, &search);
+
+ if (entry != NULL) {
+ if (frag->inode != NULL) {
+ chunk = entry->data;
+ sqfs_inode_set_frag_location(*(frag->inode),
+ chunk->index,
+ chunk->offset);
+ }
+ release_old_block(proc, frag);
+ return 0;
+ }
+ }
+
+ if (proc->frag_block != NULL) {
+ size_t size = proc->frag_block->size + frag->size;
+
+ if (size > proc->max_block_size) {
+ proc->frag_block->io_seq_num = proc->io_seq_num++;
+
+ err = enqueue_block(proc, proc->frag_block);
+ proc->frag_block = NULL;
+
+ if (err)
+ goto fail;
+ }
+ }
+
+ if (proc->frag_block == NULL) {
+ if (proc->frag_tbl == NULL) {
+ index = 0;
+ } else {
+ err = sqfs_frag_table_append(proc->frag_tbl,
+ 0, 0, &index);
+ if (err)
+ goto fail;
+ }
+
+ offset = 0;
+ proc->frag_block = frag;
+ proc->frag_block->index = index;
+ proc->frag_block->flags &= SQFS_BLK_DONT_COMPRESS;
+ proc->frag_block->flags |= SQFS_BLK_FRAGMENT_BLOCK;
+ } else {
+ index = proc->frag_block->index;
+ offset = proc->frag_block->size;
+
+ memcpy(proc->frag_block->data + proc->frag_block->size,
+ frag->data, frag->size);
+
+ proc->frag_block->size += frag->size;
+ proc->frag_block->flags |=
+ (frag->flags & SQFS_BLK_DONT_COMPRESS);
+ }
+
+ if (proc->frag_tbl != NULL) {
+ err = SQFS_ERROR_ALLOC;
+ chunk = calloc(1, sizeof(*chunk));
+ if (chunk == NULL)
+ goto fail;
+
+ chunk->index = index;
+ chunk->offset = offset;
+ chunk->size = frag->size;
+ chunk->hash = frag->checksum;
+
+ entry = hash_table_insert_pre_hashed(proc->frag_ht, chunk->hash,
+ chunk, chunk);
+
+ if (entry == NULL)
+ goto fail;
+ }
+
+ if (frag->inode != NULL)
+ sqfs_inode_set_frag_location(*(frag->inode), index, offset);
+
+ if (frag != proc->frag_block)
+ release_old_block(proc, frag);
+
+ proc->stats.actual_frag_count += 1;
+ return 0;
+fail:
+ free(chunk);
+ free(frag);
+ return err;
+}
+
+static void store_io_block(sqfs_block_processor_t *proc, sqfs_block_t *blk)
+{
+ sqfs_block_t *prev = NULL, *it = proc->io_queue;
+
+ while (it != NULL && (it->io_seq_num < blk->io_seq_num)) {
+ prev = it;
+ it = it->next;
+ }
+
+ if (prev == NULL) {
+ proc->io_queue = blk;
+ } else {
+ prev->next = blk;
+ }
+
+ blk->next = it;
+ proc->backlog += 1;
+}
+
+int dequeue_block(sqfs_block_processor_t *proc)
+{
+ size_t backlog_old = proc->backlog;
+ sqfs_block_t *blk;
+ int status;
+
+ do {
+ while (proc->io_queue != NULL) {
+ if (proc->io_queue->io_seq_num != proc->io_deq_seq_num)
+ break;
+
+ blk = proc->io_queue;
+ proc->io_queue = blk->next;
+ proc->io_deq_seq_num += 1;
+ proc->backlog -= 1;
+
+ status = process_completed_block(proc, blk);
+ if (status != 0)
+ return status;
+ }
+
+ if (proc->backlog < backlog_old)
+ break;
+
+ blk = proc->pool->dequeue(proc->pool);
+
+ if (blk == NULL) {
+ status = proc->pool->get_status(proc->pool);
+ return status ? status : SQFS_ERROR_INTERNAL;
+ }
+
+ proc->backlog -= 1;
+
+ if (blk->flags & SQFS_BLK_IS_FRAGMENT) {
+ status = process_completed_fragment(proc, blk);
+ if (status != 0)
+ return status;
+ } else {
+ if (!(blk->flags & SQFS_BLK_FRAGMENT_BLOCK))
+ blk->io_seq_num = proc->io_seq_num++;
+
+ store_io_block(proc, blk);
+ }
+ } while (proc->backlog >= backlog_old);
+
+ return 0;
+}