[wimlib] / src / resource.c

/*
 * resource.c
 *
 * Code for reading blobs and resources, including compressed WIM resources.
 */

/*
 * Copyright (C) 2012, 2013, 2015 Eric Biggers
 *
 * This file is free software; you can redistribute it and/or modify it under
 * the terms of the GNU Lesser General Public License as published by the Free
 * Software Foundation; either version 3 of the License, or (at your option) any
 * later version.
 *
 * This file 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 Lesser General Public License for more
 * details.
 *
 * You should have received a copy of the GNU Lesser General Public License
 * along with this file; if not, see http://www.gnu.org/licenses/.
 */

#ifdef HAVE_CONFIG_H
#  include "config.h"
#endif

#include <errno.h>
#include <fcntl.h>
#include <unistd.h>

#include "wimlib/alloca.h"
#include "wimlib/assert.h"
#include "wimlib/bitops.h"
#include "wimlib/blob_table.h"
#include "wimlib/endianness.h"
#include "wimlib/error.h"
#include "wimlib/file_io.h"
#include "wimlib/ntfs_3g.h"
#include "wimlib/resource.h"
#include "wimlib/sha1.h"
#include "wimlib/wim.h"
#include "wimlib/win32.h"

/*
 *                         Compressed WIM resources
 *
 * A compressed resource in a WIM consists of a sequence of chunks.  Each chunk
 * decompresses to the same size except possibly for the last, which
 * decompresses to the remaining size.  Chunks that did not compress to less
 * than their original size are stored uncompressed.
 *
 * We support three variations on this resource format, independently of the
 * compression type and chunk size which can vary as well:
 *
 * - Original resource format: immediately before the compressed chunks, the
 *   "chunk table" provides the offset, in bytes relative to the end of the
 *   chunk table, of the start of each compressed chunk, except for the first
 *   chunk which is omitted as it always has an offset of 0.  Chunk table
 *   entries are 32-bit for resources <= 4 GiB uncompressed and 64-bit for
 *   resources > 4 GiB uncompressed.
 *
 * - Solid resource format (distinguished by the use of WIM_RESHDR_FLAG_SOLID
 *   instead of WIM_RESHDR_FLAG_COMPRESSED): similar to the original format, but
 *   the resource begins with a 16-byte header which specifies the uncompressed
 *   size of the resource, the compression type, and the chunk size.  (In the
 *   original format, these values were instead determined from outside the
 *   resource itself, from the blob table and the WIM file header.) In addition,
 *   in this format the entries in the chunk table contain compressed chunk
 *   sizes rather than offsets.  As a consequence of this, the chunk table
 *   entries are always 32-bit and there is an entry for chunk 0.
 *
 * - Pipable resource format (wimlib extension; all resources in a pipable WIM
 *   have this format): similar to the original format, but the chunk table is
 *   at the end of the resource rather than the beginning, and each compressed
 *   chunk is prefixed with its compressed size as a 32-bit integer.  This
 *   format allows a resource to be written without rewinding.
 */


struct data_range {
        u64 offset;
        u64 size;
};

/*
 * Read data from a compressed WIM resource.
 *
 * @rdesc
 *      Description of the compressed WIM resource to read from.
 * @ranges
 *      Nonoverlapping, nonempty ranges of the uncompressed resource data to
 *      read, sorted by increasing offset.
 * @num_ranges
 *      Number of ranges in @ranges; must be at least 1.
 * @cbs
 *      Structure which provides the consume_chunk() callback to feed the data
 *      being read.  Each call provides the next chunk of the requested data,
 *      uncompressed.  Each chunk will be nonempty and will not cross range
 *      boundaries but otherwise will be of unspecified size.
 *
 * Possible return values:
 *
 *      WIMLIB_ERR_SUCCESS (0)
 *      WIMLIB_ERR_READ                   (errno set)
 *      WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0)
 *      WIMLIB_ERR_NOMEM                  (errno set to ENOMEM)
 *      WIMLIB_ERR_DECOMPRESSION          (errno set to EINVAL)
 *      WIMLIB_ERR_INVALID_CHUNK_SIZE     (errno set to EINVAL)
 *
 *      or other error code returned by the cbs->consume_chunk() function.
 */
static int
read_compressed_wim_resource(const struct wim_resource_descriptor * const rdesc,
                             const struct data_range * const ranges,
                             const size_t num_ranges,
                             const struct read_blob_callbacks *cbs)
{
        int ret;
        u64 *chunk_offsets = NULL;
        u8 *ubuf = NULL;
        void *cbuf = NULL;
        bool chunk_offsets_malloced = false;
        bool ubuf_malloced = false;
        bool cbuf_malloced = false;
        struct wimlib_decompressor *decompressor = NULL;

        /* Sanity checks  */
        wimlib_assert(num_ranges != 0);
        for (size_t i = 0; i < num_ranges; i++) {
                wimlib_assert(ranges[i].offset + ranges[i].size > ranges[i].offset &&
                              ranges[i].offset + ranges[i].size <= rdesc->uncompressed_size);
        }
        for (size_t i = 0; i < num_ranges - 1; i++)
                wimlib_assert(ranges[i].offset + ranges[i].size <= ranges[i + 1].offset);

        /* Get the offsets of the first and last bytes of the read.  */
        const u64 first_offset = ranges[0].offset;
        const u64 last_offset = ranges[num_ranges - 1].offset + ranges[num_ranges - 1].size - 1;

        /* Get the file descriptor for the WIM.  */
        struct filedes * const in_fd = &rdesc->wim->in_fd;

        /* Determine if we're reading a pipable resource from a pipe or not.  */
        const bool is_pipe_read = (rdesc->is_pipable && !filedes_is_seekable(in_fd));

        /* Determine if the chunk table is in an alternate format.  */
        const bool alt_chunk_table = (rdesc->flags & WIM_RESHDR_FLAG_SOLID)
                                        && !is_pipe_read;

        /* Get the maximum size of uncompressed chunks in this resource, which
         * we require be a power of 2.  */
        u64 cur_read_offset = rdesc->offset_in_wim;
        int ctype = rdesc->compression_type;
        u32 chunk_size = rdesc->chunk_size;
        if (alt_chunk_table) {
                /* Alternate chunk table format.  Its header specifies the chunk
                 * size and compression format.  Note: it could be read here;
                 * however, the relevant data was already loaded into @rdesc by
                 * read_blob_table().  */
                cur_read_offset += sizeof(struct alt_chunk_table_header_disk);
        }

        if (unlikely(!is_power_of_2(chunk_size))) {
                ERROR("Invalid compressed resource: "
                      "expected power-of-2 chunk size (got %"PRIu32")",
                      chunk_size);
                ret = WIMLIB_ERR_INVALID_CHUNK_SIZE;
                errno = EINVAL;
                goto out_cleanup;
        }

        /* Get valid decompressor.  */
        if (likely(ctype == rdesc->wim->decompressor_ctype &&
                   chunk_size == rdesc->wim->decompressor_max_block_size))
        {
                /* Cached decompressor.  */
                decompressor = rdesc->wim->decompressor;
                rdesc->wim->decompressor_ctype = WIMLIB_COMPRESSION_TYPE_NONE;
                rdesc->wim->decompressor = NULL;
        } else {
                ret = wimlib_create_decompressor(ctype, chunk_size,
                                                 &decompressor);
                if (unlikely(ret)) {
                        if (ret != WIMLIB_ERR_NOMEM)
                                errno = EINVAL;
                        goto out_cleanup;
                }
        }

        const u32 chunk_order = fls32(chunk_size);

        /* Calculate the total number of chunks the resource is divided into.  */
        const u64 num_chunks = (rdesc->uncompressed_size + chunk_size - 1) >> chunk_order;

        /* Calculate the 0-based indices of the first and last chunks containing
         * data that needs to be passed to the callback.  */
        const u64 first_needed_chunk = first_offset >> chunk_order;
        const u64 last_needed_chunk = last_offset >> chunk_order;

        /* Calculate the 0-based index of the first chunk that actually needs to
         * be read.  This is normally first_needed_chunk, but for pipe reads we
         * must always start from the 0th chunk.  */
        const u64 read_start_chunk = (is_pipe_read ? 0 : first_needed_chunk);

        /* Calculate the number of chunk offsets that are needed for the chunks
         * being read.  */
        const u64 num_needed_chunk_offsets =
                last_needed_chunk - read_start_chunk + 1 +
                (last_needed_chunk < num_chunks - 1);

        /* Calculate the number of entries in the chunk table.  Normally, it's
         * one less than the number of chunks, since the first chunk has no
         * entry.  But in the alternate chunk table format, the chunk entries
         * contain chunk sizes, not offsets, and there is one per chunk.  */
        const u64 num_chunk_entries = (alt_chunk_table ? num_chunks : num_chunks - 1);

        /* Set the size of each chunk table entry based on the resource's
         * uncompressed size.  */
        const u64 chunk_entry_size = get_chunk_entry_size(rdesc->uncompressed_size,
                                                          alt_chunk_table);

        /* Calculate the size of the chunk table in bytes.  */
        const u64 chunk_table_size = num_chunk_entries * chunk_entry_size;

        /* Calculate the size of the chunk table in bytes, including the header
         * in the case of the alternate chunk table format.  */
        const u64 chunk_table_full_size =
                (alt_chunk_table) ? chunk_table_size + sizeof(struct alt_chunk_table_header_disk)
                                  : chunk_table_size;

        if (!is_pipe_read) {
                /* Read the needed chunk table entries into memory and use them
                 * to initialize the chunk_offsets array.  */

                u64 first_chunk_entry_to_read;
                u64 last_chunk_entry_to_read;

                if (alt_chunk_table) {
                        /* The alternate chunk table contains chunk sizes, not
                         * offsets, so we always must read all preceding entries
                         * in order to determine offsets.  */
                        first_chunk_entry_to_read = 0;
                        last_chunk_entry_to_read = last_needed_chunk;
                } else {
                        /* Here we must account for the fact that the first
                         * chunk has no explicit chunk table entry.  */

                        if (read_start_chunk == 0)
                                first_chunk_entry_to_read = 0;
                        else
                                first_chunk_entry_to_read = read_start_chunk - 1;

                        if (last_needed_chunk == 0)
                                last_chunk_entry_to_read = 0;
                        else
                                last_chunk_entry_to_read = last_needed_chunk - 1;

                        if (last_needed_chunk < num_chunks - 1)
                                last_chunk_entry_to_read++;
                }

                const u64 num_chunk_entries_to_read =
                        last_chunk_entry_to_read - first_chunk_entry_to_read + 1;

                const u64 chunk_offsets_alloc_size =
                        max(num_chunk_entries_to_read,
                            num_needed_chunk_offsets) * sizeof(chunk_offsets[0]);

                if (unlikely((size_t)chunk_offsets_alloc_size != chunk_offsets_alloc_size)) {
                        errno = ENOMEM;
                        goto oom;
                }

                if (likely(chunk_offsets_alloc_size <= STACK_MAX)) {
                        chunk_offsets = alloca(chunk_offsets_alloc_size);
                } else {
                        chunk_offsets = MALLOC(chunk_offsets_alloc_size);
                        if (unlikely(!chunk_offsets))
                                goto oom;
                        chunk_offsets_malloced = true;
                }

                const size_t chunk_table_size_to_read =
                        num_chunk_entries_to_read * chunk_entry_size;

                const u64 file_offset_of_needed_chunk_entries =
                        cur_read_offset
                        + (first_chunk_entry_to_read * chunk_entry_size)
                        + (rdesc->is_pipable ? (rdesc->size_in_wim - chunk_table_size) : 0);

                void * const chunk_table_data =
                        (u8*)chunk_offsets +
                        chunk_offsets_alloc_size -
                        chunk_table_size_to_read;

                ret = full_pread(in_fd, chunk_table_data, chunk_table_size_to_read,
                                 file_offset_of_needed_chunk_entries);
                if (unlikely(ret))
                        goto read_error;

                /* Now fill in chunk_offsets from the entries we have read in
                 * chunk_tab_data.  We break aliasing rules here to avoid having
                 * to allocate yet another array.  */
                typedef le64 _may_alias_attribute aliased_le64_t;
                typedef le32 _may_alias_attribute aliased_le32_t;
                u64 * chunk_offsets_p = chunk_offsets;

                if (alt_chunk_table) {
                        u64 cur_offset = 0;
                        aliased_le32_t *raw_entries = chunk_table_data;

                        for (size_t i = 0; i < num_chunk_entries_to_read; i++) {
                                u32 entry = le32_to_cpu(raw_entries[i]);
                                if (i >= read_start_chunk)
                                        *chunk_offsets_p++ = cur_offset;
                                cur_offset += entry;
                        }
                        if (last_needed_chunk < num_chunks - 1)
                                *chunk_offsets_p = cur_offset;
                } else {
                        if (read_start_chunk == 0)
                                *chunk_offsets_p++ = 0;

                        if (chunk_entry_size == 4) {
                                aliased_le32_t *raw_entries = chunk_table_data;
                                for (size_t i = 0; i < num_chunk_entries_to_read; i++)
                                        *chunk_offsets_p++ = le32_to_cpu(raw_entries[i]);
                        } else {
                                aliased_le64_t *raw_entries = chunk_table_data;
                                for (size_t i = 0; i < num_chunk_entries_to_read; i++)
                                        *chunk_offsets_p++ = le64_to_cpu(raw_entries[i]);
                        }
                }

                /* Set offset to beginning of first chunk to read.  */
                cur_read_offset += chunk_offsets[0];
                if (rdesc->is_pipable)
                        cur_read_offset += read_start_chunk * sizeof(struct pwm_chunk_hdr);
                else
                        cur_read_offset += chunk_table_size;
        }

        /* Allocate buffer for holding the uncompressed data of each chunk.  */
        if (chunk_size <= STACK_MAX) {
                ubuf = alloca(chunk_size);
        } else {
                ubuf = MALLOC(chunk_size);
                if (unlikely(!ubuf))
                        goto oom;
                ubuf_malloced = true;
        }

        /* Allocate a temporary buffer for reading compressed chunks, each of
         * which can be at most @chunk_size - 1 bytes.  This excludes compressed
         * chunks that are a full @chunk_size bytes, which are actually stored
         * uncompressed.  */
        if (chunk_size - 1 <= STACK_MAX) {
                cbuf = alloca(chunk_size - 1);
        } else {
                cbuf = MALLOC(chunk_size - 1);
                if (unlikely(!cbuf))
                        goto oom;
                cbuf_malloced = true;
        }

        /* Set current data range.  */
        const struct data_range *cur_range = ranges;
        const struct data_range * const end_range = &ranges[num_ranges];
        u64 cur_range_pos = cur_range->offset;
        u64 cur_range_end = cur_range->offset + cur_range->size;

        /* Read and process each needed chunk.  */
        for (u64 i = read_start_chunk; i <= last_needed_chunk; i++) {

                /* Calculate uncompressed size of next chunk.  */
                u32 chunk_usize;
                if ((i == num_chunks - 1) && (rdesc->uncompressed_size & (chunk_size - 1)))
                        chunk_usize = (rdesc->uncompressed_size & (chunk_size - 1));
                else
                        chunk_usize = chunk_size;

                /* Calculate compressed size of next chunk.  */
                u32 chunk_csize;
                if (is_pipe_read) {
                        struct pwm_chunk_hdr chunk_hdr;

                        ret = full_pread(in_fd, &chunk_hdr,
                                         sizeof(chunk_hdr), cur_read_offset);
                        if (unlikely(ret))
                                goto read_error;
                        chunk_csize = le32_to_cpu(chunk_hdr.compressed_size);
                } else {
                        if (i == num_chunks - 1) {
                                chunk_csize = rdesc->size_in_wim -
                                              chunk_table_full_size -
                                              chunk_offsets[i - read_start_chunk];
                                if (rdesc->is_pipable)
                                        chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr);
                        } else {
                                chunk_csize = chunk_offsets[i + 1 - read_start_chunk] -
                                              chunk_offsets[i - read_start_chunk];
                        }
                }
                if (unlikely(chunk_csize == 0 || chunk_csize > chunk_usize)) {
                        ERROR("Invalid chunk size in compressed resource!");
                        errno = EINVAL;
                        ret = WIMLIB_ERR_DECOMPRESSION;
                        goto out_cleanup;
                }
                if (rdesc->is_pipable)
                        cur_read_offset += sizeof(struct pwm_chunk_hdr);

                /* Offsets in the uncompressed resource at which this chunk
                 * starts and ends.  */
                const u64 chunk_start_offset = i << chunk_order;
                const u64 chunk_end_offset = chunk_start_offset + chunk_usize;

                if (chunk_end_offset <= cur_range_pos) {

                        /* The next range does not require data in this chunk,
                         * so skip it.  */
                        cur_read_offset += chunk_csize;
                        if (is_pipe_read) {
                                u8 dummy;

                                ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
                                if (unlikely(ret))
                                        goto read_error;
                        }
                } else {

                        /* Read the chunk and feed data to the callback
                         * function.  */
                        u8 *read_buf;

                        if (chunk_csize == chunk_usize)
                                read_buf = ubuf;
                        else
                                read_buf = cbuf;

                        ret = full_pread(in_fd,
                                         read_buf,
                                         chunk_csize,
                                         cur_read_offset);
                        if (unlikely(ret))
                                goto read_error;

                        if (read_buf == cbuf) {
                                ret = wimlib_decompress(cbuf,
                                                        chunk_csize,
                                                        ubuf,
                                                        chunk_usize,
                                                        decompressor);
                                if (unlikely(ret)) {
                                        ERROR("Failed to decompress data!");
                                        ret = WIMLIB_ERR_DECOMPRESSION;
                                        errno = EINVAL;
                                        goto out_cleanup;
                                }
                        }
                        cur_read_offset += chunk_csize;

                        /* At least one range requires data in this chunk.  */
                        do {
                                size_t start, end, size;

                                /* Calculate how many bytes of data should be
                                 * sent to the callback function, taking into
                                 * account that data sent to the callback
                                 * function must not overlap range boundaries.
                                 */
                                start = cur_range_pos - chunk_start_offset;
                                end = min(cur_range_end, chunk_end_offset) - chunk_start_offset;
                                size = end - start;

                                ret = call_consume_chunk(&ubuf[start], size, cbs);
                                if (unlikely(ret))
                                        goto out_cleanup;

                                cur_range_pos += size;
                                if (cur_range_pos == cur_range_end) {
                                        /* Advance to next range.  */
                                        if (++cur_range == end_range) {
                                                cur_range_pos = ~0ULL;
                                        } else {
                                                cur_range_pos = cur_range->offset;
                                                cur_range_end = cur_range->offset + cur_range->size;
                                        }
                                }
                        } while (cur_range_pos < chunk_end_offset);
                }
        }

        if (is_pipe_read &&
            last_offset == rdesc->uncompressed_size - 1 &&
            chunk_table_size)
        {
                u8 dummy;
                /* If reading a pipable resource from a pipe and the full data
                 * was requested, skip the chunk table at the end so that the
                 * file descriptor is fully clear of the resource after this
                 * returns.  */
                cur_read_offset += chunk_table_size;
                ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
                if (unlikely(ret))
                        goto read_error;
        }
        ret = 0;

out_cleanup:
        if (decompressor) {
                wimlib_free_decompressor(rdesc->wim->decompressor);
                rdesc->wim->decompressor = decompressor;
                rdesc->wim->decompressor_ctype = ctype;
                rdesc->wim->decompressor_max_block_size = chunk_size;
        }
        if (chunk_offsets_malloced)
                FREE(chunk_offsets);
        if (ubuf_malloced)
                FREE(ubuf);
        if (cbuf_malloced)
                FREE(cbuf);
        return ret;

oom:
        ERROR("Out of memory while reading compressed WIM resource");
        ret = WIMLIB_ERR_NOMEM;
        goto out_cleanup;

read_error:
        ERROR_WITH_ERRNO("Error reading data from WIM file");
        goto out_cleanup;
}

/* Read raw data from a file descriptor at the specified offset, feeding the
 * data in nonempty chunks into the cbs->consume_chunk() function.  */
static int
read_raw_file_data(struct filedes *in_fd, u64 offset, u64 size,
                   const struct read_blob_callbacks *cbs)
{
        u8 buf[BUFFER_SIZE];
        size_t bytes_to_read;
        int ret;

        while (size) {
                bytes_to_read = min(sizeof(buf), size);
                ret = full_pread(in_fd, buf, bytes_to_read, offset);
                if (unlikely(ret)) {
                        ERROR_WITH_ERRNO("Read error");
                        return ret;
                }
                ret = call_consume_chunk(buf, bytes_to_read, cbs);
                if (unlikely(ret))
                        return ret;
                size -= bytes_to_read;
                offset += bytes_to_read;
        }
        return 0;
}

/* A consume_chunk() implementation that simply concatenates all chunks into an
 * in-memory buffer.  */
static int
bufferer_cb(const void *chunk, size_t size, void *_ctx)
{
        void **buf_p = _ctx;

        *buf_p = mempcpy(*buf_p, chunk, size);
        return 0;
}

/*
 * Read @size bytes at @offset in the WIM resource described by @rdesc and feed
 * the data into the @cbs->consume_chunk callback function.
 *
 * @offset and @size are assumed to have already been validated against the
 * resource's uncompressed size.
 *
 * Returns 0 on success; or the first nonzero value returned by the callback
 * function; or a nonzero wimlib error code with errno set as well.
 */
static int
read_partial_wim_resource(const struct wim_resource_descriptor *rdesc,
                          const u64 offset, const u64 size,
                          const struct read_blob_callbacks *cbs)
{
        if (rdesc->flags & (WIM_RESHDR_FLAG_COMPRESSED |
                            WIM_RESHDR_FLAG_SOLID))
        {
                /* Compressed resource  */
                if (unlikely(!size))
                        return 0;
                struct data_range range = {
                        .offset = offset,
                        .size = size,
                };
                return read_compressed_wim_resource(rdesc, &range, 1, cbs);
        }

        /* Uncompressed resource  */
        return read_raw_file_data(&rdesc->wim->in_fd,
                                  rdesc->offset_in_wim + offset,
                                  size, cbs);
}

/* Read the specified range of uncompressed data from the specified blob, which
 * must be located in a WIM file, into the specified buffer.  */
int
read_partial_wim_blob_into_buf(const struct blob_descriptor *blob,
                               u64 offset, size_t size, void *buf)
{
        struct read_blob_callbacks cbs = {
                .consume_chunk  = bufferer_cb,
                .ctx            = &buf,
        };
        return read_partial_wim_resource(blob->rdesc,
                                         blob->offset_in_res + offset,
                                         size,
                                         &cbs);
}

/* Skip over the data of the specified WIM resource.  */
int
skip_wim_resource(const struct wim_resource_descriptor *rdesc)
{
        struct read_blob_callbacks cbs = {
        };
        return read_partial_wim_resource(rdesc, 0,
                                         rdesc->uncompressed_size, &cbs);
}

static int
read_wim_blob_prefix(const struct blob_descriptor *blob, u64 size,
                     const struct read_blob_callbacks *cbs)
{
        return read_partial_wim_resource(blob->rdesc, blob->offset_in_res,
                                         size, cbs);
}

/* This function handles reading blob data that is located in an external file,
 * such as a file that has been added to the WIM image through execution of a
 * wimlib_add_command.
 *
 * This assumes the file can be accessed using the standard POSIX open(),
 * read(), and close().  On Windows this will not necessarily be the case (since
 * the file may need FILE_FLAG_BACKUP_SEMANTICS to be opened, or the file may be
 * encrypted), so Windows uses its own code for its equivalent case.  */
static int
read_file_on_disk_prefix(const struct blob_descriptor *blob, u64 size,
                         const struct read_blob_callbacks *cbs)
{
        int ret;
        int raw_fd;
        struct filedes fd;

        raw_fd = topen(blob->file_on_disk, O_BINARY | O_RDONLY);
        if (unlikely(raw_fd < 0)) {
                ERROR_WITH_ERRNO("Can't open \"%"TS"\"", blob->file_on_disk);
                return WIMLIB_ERR_OPEN;
        }
        filedes_init(&fd, raw_fd);
        ret = read_raw_file_data(&fd, 0, size, cbs);
        filedes_close(&fd);
        return ret;
}

#ifdef WITH_FUSE
static int
read_staging_file_prefix(const struct blob_descriptor *blob, u64 size,
                         const struct read_blob_callbacks *cbs)
{
        int raw_fd;
        struct filedes fd;
        int ret;

        raw_fd = openat(blob->staging_dir_fd, blob->staging_file_name,
                        O_RDONLY | O_NOFOLLOW);
        if (unlikely(raw_fd < 0)) {
                ERROR_WITH_ERRNO("Can't open staging file \"%s\"",
                                 blob->staging_file_name);
                return WIMLIB_ERR_OPEN;
        }
        filedes_init(&fd, raw_fd);
        ret = read_raw_file_data(&fd, 0, size, cbs);
        filedes_close(&fd);
        return ret;
}
#endif

/* This function handles the trivial case of reading blob data that is, in fact,
 * already located in an in-memory buffer.  */
static int
read_buffer_prefix(const struct blob_descriptor *blob,
                   u64 size, const struct read_blob_callbacks *cbs)
{
        if (unlikely(!size))
                return 0;
        return call_consume_chunk(blob->attached_buffer, size, cbs);
}

typedef int (*read_blob_prefix_handler_t)(const struct blob_descriptor *blob,
                                          u64 size,
                                          const struct read_blob_callbacks *cbs);

/*
 * Read the first @size bytes from a generic "blob", which may be located in any
 * one of several locations, such as in a WIM resource (possibly compressed), in
 * an external file, or directly in an in-memory buffer.  The blob data will be
 * fed to the cbs->consume_chunk() callback function in chunks that are nonempty
 * but otherwise are of unspecified size.
 *
 * Returns 0 on success; nonzero on error.  A nonzero value will be returned if
 * the blob data cannot be successfully read (for a number of different reasons,
 * depending on the blob location), or if cbs->consume_chunk() returned nonzero
 * in which case that error code will be returned.
 */
static int
read_blob_prefix(const struct blob_descriptor *blob, u64 size,
                 const struct read_blob_callbacks *cbs)
{
        static const read_blob_prefix_handler_t handlers[] = {
                [BLOB_IN_WIM] = read_wim_blob_prefix,
                [BLOB_IN_FILE_ON_DISK] = read_file_on_disk_prefix,
                [BLOB_IN_ATTACHED_BUFFER] = read_buffer_prefix,
        #ifdef WITH_FUSE
                [BLOB_IN_STAGING_FILE] = read_staging_file_prefix,
        #endif
        #ifdef WITH_NTFS_3G
                [BLOB_IN_NTFS_VOLUME] = read_ntfs_attribute_prefix,
        #endif
        #ifdef __WIN32__
                [BLOB_IN_WINNT_FILE_ON_DISK] = read_winnt_stream_prefix,
                [BLOB_WIN32_ENCRYPTED] = read_win32_encrypted_file_prefix,
        #endif
        };
        wimlib_assert(blob->blob_location < ARRAY_LEN(handlers)
                      && handlers[blob->blob_location] != NULL);
        wimlib_assert(size <= blob->size);
        return handlers[blob->blob_location](blob, size, cbs);
}

/* Read the full data of the specified blob, passing the data into the specified
 * callbacks (all of which are optional).  */
int
read_blob_with_cbs(struct blob_descriptor *blob,
                   const struct read_blob_callbacks *cbs)
{
        int ret;

        ret = call_begin_blob(blob, cbs);
        if (unlikely(ret))
                return ret;

        ret = read_blob_prefix(blob, blob->size, cbs);

        return call_end_blob(blob, ret, cbs);
}

/* Read the full uncompressed data of the specified blob into the specified
 * buffer, which must have space for at least blob->size bytes.  The SHA-1
 * message digest is *not* checked.  */
int
read_blob_into_buf(const struct blob_descriptor *blob, void *buf)
{
        struct read_blob_callbacks cbs = {
                .consume_chunk  = bufferer_cb,
                .ctx            = &buf,
        };
        return read_blob_prefix(blob, blob->size, &cbs);
}

/* Retrieve the full uncompressed data of the specified blob.  A buffer large
 * enough hold the data is allocated and returned in @buf_ret.  The SHA-1
 * message digest is *not* checked.  */
int
read_blob_into_alloc_buf(const struct blob_descriptor *blob, void **buf_ret)
{
        int ret;
        void *buf;

        if (unlikely((size_t)blob->size != blob->size)) {
                ERROR("Can't read %"PRIu64" byte blob into memory", blob->size);
                return WIMLIB_ERR_NOMEM;
        }

        buf = MALLOC(blob->size);
        if (unlikely(!buf))
                return WIMLIB_ERR_NOMEM;

        ret = read_blob_into_buf(blob, buf);
        if (unlikely(ret)) {
                FREE(buf);
                return ret;
        }

        *buf_ret = buf;
        return 0;
}

/* Retrieve the full uncompressed data of a WIM resource specified as a raw
 * `wim_reshdr' and the corresponding WIM file.  A buffer large enough hold the
 * data is allocated and returned in @buf_ret.  */
int
wim_reshdr_to_data(const struct wim_reshdr *reshdr, WIMStruct *wim,
                   void **buf_ret)
{
        struct wim_resource_descriptor rdesc;
        struct blob_descriptor blob;

        wim_reshdr_to_desc(reshdr, wim, &rdesc);
        blob_set_is_located_in_nonsolid_wim_resource(&blob, &rdesc);

        return read_blob_into_alloc_buf(&blob, buf_ret);
}

/* Calculate the SHA-1 message digest of the uncompressed data of the specified
 * WIM resource.  */
int
wim_reshdr_to_hash(const struct wim_reshdr *reshdr, WIMStruct *wim,
                   u8 hash[SHA1_HASH_SIZE])
{
        struct wim_resource_descriptor rdesc;
        struct blob_descriptor blob;
        int ret;

        wim_reshdr_to_desc(reshdr, wim, &rdesc);
        blob_set_is_located_in_nonsolid_wim_resource(&blob, &rdesc);
        blob.unhashed = 1;

        ret = sha1_blob(&blob);
        if (unlikely(ret))
                return ret;

        copy_hash(hash, blob.hash);
        return 0;
}

struct blobifier_context {
        struct read_blob_callbacks cbs;
        struct blob_descriptor *cur_blob;
        struct blob_descriptor *next_blob;
        u64 cur_blob_offset;
        struct blob_descriptor *final_blob;
        size_t list_head_offset;
};

static struct blob_descriptor *
next_blob(struct blob_descriptor *blob, size_t list_head_offset)
{
        struct list_head *cur;

        cur = (struct list_head*)((u8*)blob + list_head_offset);

        return (struct blob_descriptor*)((u8*)cur->next - list_head_offset);
}

/* A consume_chunk() implementation that translates raw resource data into
 * blobs, calling the begin_blob, consume_chunk, and end_blob callbacks as
 * appropriate.  */
static int
blobifier_cb(const void *chunk, size_t size, void *_ctx)
{
        struct blobifier_context *ctx = _ctx;
        int ret;

        wimlib_assert(ctx->cur_blob != NULL);
        wimlib_assert(size <= ctx->cur_blob->size - ctx->cur_blob_offset);

        if (ctx->cur_blob_offset == 0) {
                /* Starting a new blob.  */
                ret = call_begin_blob(ctx->cur_blob, &ctx->cbs);
                if (ret)
                        return ret;
        }

        ctx->cur_blob_offset += size;

        ret = call_consume_chunk(chunk, size, &ctx->cbs);
        if (ret)
                return ret;

        if (ctx->cur_blob_offset == ctx->cur_blob->size) {
                /* Finished reading all the data for a blob.  */

                ctx->cur_blob_offset = 0;

                ret = call_end_blob(ctx->cur_blob, 0, &ctx->cbs);
                if (ret)
                        return ret;

                /* Advance to next blob.  */
                ctx->cur_blob = ctx->next_blob;
                if (ctx->cur_blob != NULL) {
                        if (ctx->cur_blob != ctx->final_blob)
                                ctx->next_blob = next_blob(ctx->cur_blob,
                                                           ctx->list_head_offset);
                        else
                                ctx->next_blob = NULL;
                }
        }
        return 0;
}

struct hasher_context {
        SHA_CTX sha_ctx;
        int flags;
        struct read_blob_callbacks cbs;
};

/* Callback for starting to read a blob while calculating its SHA-1 message
 * digest.  */
static int
hasher_begin_blob(struct blob_descriptor *blob, void *_ctx)
{
        struct hasher_context *ctx = _ctx;

        sha1_init(&ctx->sha_ctx);

        return call_begin_blob(blob, &ctx->cbs);
}

/* A consume_chunk() implementation that continues calculating the SHA-1 message
 * digest of the blob being read, then optionally passes the data on to another
 * consume_chunk() implementation.  This allows checking the SHA-1 message
 * digest of a blob being extracted, for example.  */
static int
hasher_consume_chunk(const void *chunk, size_t size, void *_ctx)
{
        struct hasher_context *ctx = _ctx;

        sha1_update(&ctx->sha_ctx, chunk, size);

        return call_consume_chunk(chunk, size, &ctx->cbs);
}

/* Callback for finishing reading a blob while calculating its SHA-1 message
 * digest.  */
static int
hasher_end_blob(struct blob_descriptor *blob, int status, void *_ctx)
{
        struct hasher_context *ctx = _ctx;
        u8 hash[SHA1_HASH_SIZE];
        int ret;

        if (unlikely(status)) {
                /* Error occurred; the full blob may not have been read.  */
                ret = status;
                goto out_next_cb;
        }

        /* Retrieve the final SHA-1 message digest.  */
        sha1_final(hash, &ctx->sha_ctx);

        /* Set the SHA-1 message digest of the blob, or compare the calculated
         * value with stored value.  */
        if (blob->unhashed) {
                if (ctx->flags & COMPUTE_MISSING_BLOB_HASHES)
                        copy_hash(blob->hash, hash);
        } else if ((ctx->flags & VERIFY_BLOB_HASHES) &&
                   unlikely(!hashes_equal(hash, blob->hash)))
        {
                if (wimlib_print_errors) {
                        tchar expected_hashstr[SHA1_HASH_SIZE * 2 + 1];
                        tchar actual_hashstr[SHA1_HASH_SIZE * 2 + 1];
                        sprint_hash(blob->hash, expected_hashstr);
                        sprint_hash(hash, actual_hashstr);
                        ERROR("The data is corrupted!\n"
                              "        (Expected SHA-1=%"TS", got SHA-1=%"TS")",
                              expected_hashstr, actual_hashstr);
                }
                ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
                goto out_next_cb;
        }
        ret = 0;
out_next_cb:
        return call_end_blob(blob, ret, &ctx->cbs);
}

/* Read the full data of the specified blob, passing the data into the specified
 * callbacks (all of which are optional) and either checking or computing the
 * SHA-1 message digest of the blob.  */
int
read_blob_with_sha1(struct blob_descriptor *blob,
                    const struct read_blob_callbacks *cbs)
{
        struct hasher_context hasher_ctx = {
                .flags = VERIFY_BLOB_HASHES | COMPUTE_MISSING_BLOB_HASHES,
                .cbs = *cbs,
        };
        struct read_blob_callbacks hasher_cbs = {
                .begin_blob     = hasher_begin_blob,
                .consume_chunk  = hasher_consume_chunk,
                .end_blob       = hasher_end_blob,
                .ctx            = &hasher_ctx,
        };
        return read_blob_with_cbs(blob, &hasher_cbs);
}

static int
read_blobs_in_solid_resource(struct blob_descriptor *first_blob,
                             struct blob_descriptor *last_blob,
                             size_t blob_count,
                             size_t list_head_offset,
                             const struct read_blob_callbacks *sink_cbs)
{
        struct data_range *ranges;
        bool ranges_malloced;
        struct blob_descriptor *cur_blob;
        size_t i;
        int ret;
        u64 ranges_alloc_size;

        /* Setup data ranges array (one range per blob to read); this way
         * read_compressed_wim_resource() does not need to be aware of blobs.
         */

        ranges_alloc_size = (u64)blob_count * sizeof(ranges[0]);

        if (unlikely((size_t)ranges_alloc_size != ranges_alloc_size))
                goto oom;

        if (ranges_alloc_size <= STACK_MAX) {
                ranges = alloca(ranges_alloc_size);
                ranges_malloced = false;
        } else {
                ranges = MALLOC(ranges_alloc_size);
                if (unlikely(!ranges))
                        goto oom;
                ranges_malloced = true;
        }

        for (i = 0, cur_blob = first_blob;
             i < blob_count;
             i++, cur_blob = next_blob(cur_blob, list_head_offset))
        {
                ranges[i].offset = cur_blob->offset_in_res;
                ranges[i].size = cur_blob->size;
        }

        struct blobifier_context blobifier_ctx = {
                .cbs                    = *sink_cbs,
                .cur_blob               = first_blob,
                .next_blob              = next_blob(first_blob, list_head_offset),
                .cur_blob_offset        = 0,
                .final_blob             = last_blob,
                .list_head_offset       = list_head_offset,
        };
        struct read_blob_callbacks cbs = {
                .consume_chunk  = blobifier_cb,
                .ctx            = &blobifier_ctx,
        };

        ret = read_compressed_wim_resource(first_blob->rdesc, ranges,
                                           blob_count, &cbs);

        if (ranges_malloced)
                FREE(ranges);

        if (unlikely(ret && blobifier_ctx.cur_blob_offset != 0)) {
                ret = call_end_blob(blobifier_ctx.cur_blob, ret,
                                    &blobifier_ctx.cbs);
        }
        return ret;

oom:
        ERROR("Too many blobs in one resource!");
        return WIMLIB_ERR_NOMEM;
}

/*
 * Read a list of blobs, each of which may be in any supported location (e.g.
 * in a WIM or in an external file).  This function optimizes the case where
 * multiple blobs are combined into a single solid compressed WIM resource by
 * reading the blobs in sequential order, only decompressing the solid resource
 * one time.
 *
 * @blob_list
 *      List of blobs to read.
 * @list_head_offset
 *      Offset of the `struct list_head' within each `struct blob_descriptor'
 *      that makes up the @blob_list.
 * @cbs
 *      Callback functions to accept the blob data.
 * @flags
 *      Bitwise OR of zero or more of the following flags:
 *
 *      VERIFY_BLOB_HASHES:
 *              For all blobs being read that have already had SHA-1 message
 *              digests computed, calculate the SHA-1 message digest of the read
 *              data and compare it with the previously computed value.  If they
 *              do not match, return WIMLIB_ERR_INVALID_RESOURCE_HASH.
 *
 *      COMPUTE_MISSING_BLOB_HASHES
 *              For all blobs being read that have not yet had their SHA-1
 *              message digests computed, calculate and save their SHA-1 message
 *              digests.
 *
 *      BLOB_LIST_ALREADY_SORTED
 *              @blob_list is already sorted in sequential order for reading.
 *
 * The callback functions are allowed to delete the current blob from the list
 * if necessary.
 *
 * Returns 0 on success; a nonzero error code on failure.  Failure can occur due
 * to an error reading the data or due to an error status being returned by any
 * of the callback functions.
 */
int
read_blob_list(struct list_head *blob_list, size_t list_head_offset,
               const struct read_blob_callbacks *cbs, int flags)
{
        int ret;
        struct list_head *cur, *next;
        struct blob_descriptor *blob;
        struct hasher_context *hasher_ctx;
        struct read_blob_callbacks *sink_cbs;

        if (!(flags & BLOB_LIST_ALREADY_SORTED)) {
                ret = sort_blob_list_by_sequential_order(blob_list,
                                                         list_head_offset);
                if (ret)
                        return ret;
        }

        if (flags & (VERIFY_BLOB_HASHES | COMPUTE_MISSING_BLOB_HASHES)) {
                hasher_ctx = alloca(sizeof(*hasher_ctx));
                *hasher_ctx = (struct hasher_context) {
                        .flags  = flags,
                        .cbs    = *cbs,
                };
                sink_cbs = alloca(sizeof(*sink_cbs));
                *sink_cbs = (struct read_blob_callbacks) {
                        .begin_blob     = hasher_begin_blob,
                        .consume_chunk  = hasher_consume_chunk,
                        .end_blob       = hasher_end_blob,
                        .ctx            = hasher_ctx,
                };
        } else {
                sink_cbs = (struct read_blob_callbacks *)cbs;
        }

        for (cur = blob_list->next, next = cur->next;
             cur != blob_list;
             cur = next, next = cur->next)
        {
                blob = (struct blob_descriptor*)((u8*)cur - list_head_offset);

                if (blob->blob_location == BLOB_IN_WIM &&
                    blob->size != blob->rdesc->uncompressed_size)
                {
                        struct blob_descriptor *blob_next, *blob_last;
                        struct list_head *next2;
                        size_t blob_count;

                        /* The next blob is a proper sub-sequence of a WIM
                         * resource.  See if there are other blobs in the same
                         * resource that need to be read.  Since
                         * sort_blob_list_by_sequential_order() sorted the blobs
                         * by offset in the WIM, this can be determined by
                         * simply scanning forward in the list.  */

                        blob_last = blob;
                        blob_count = 1;
                        for (next2 = next;
                             next2 != blob_list
                             && (blob_next = (struct blob_descriptor*)
                                                ((u8*)next2 - list_head_offset),
                                 blob_next->blob_location == BLOB_IN_WIM
                                 && blob_next->rdesc == blob->rdesc);
                             next2 = next2->next)
                        {
                                blob_last = blob_next;
                                blob_count++;
                        }
                        if (blob_count > 1) {
                                /* Reading multiple blobs combined into a single
                                 * WIM resource.  They are in the blob list,
                                 * sorted by offset; @blob specifies the first
                                 * blob in the resource that needs to be read
                                 * and @blob_last specifies the last blob in the
                                 * resource that needs to be read.  */
                                next = next2;
                                ret = read_blobs_in_solid_resource(blob, blob_last,
                                                                   blob_count,
                                                                   list_head_offset,
                                                                   sink_cbs);
                                if (ret)
                                        return ret;
                                continue;
                        }
                }

                ret = read_blob_with_cbs(blob, sink_cbs);
                if (unlikely(ret && ret != BEGIN_BLOB_STATUS_SKIP_BLOB))
                        return ret;
        }
        return 0;
}

static int
extract_chunk_to_fd(const void *chunk, size_t size, void *_fd)
{
        struct filedes *fd = _fd;
        int ret = full_write(fd, chunk, size);
        if (unlikely(ret))
                ERROR_WITH_ERRNO("Error writing to file descriptor");
        return ret;
}

/* Extract the first @size bytes of the specified blob to the specified file
 * descriptor.  This does *not* check the SHA-1 message digest.  */
int
extract_blob_prefix_to_fd(struct blob_descriptor *blob, u64 size,
                          struct filedes *fd)
{
        struct read_blob_callbacks cbs = {
                .consume_chunk  = extract_chunk_to_fd,
                .ctx            = fd,
        };
        return read_blob_prefix(blob, size, &cbs);
}

/* Extract the full uncompressed contents of the specified blob to the specified
 * file descriptor.  This checks the SHA-1 message digest.  */
int
extract_blob_to_fd(struct blob_descriptor *blob, struct filedes *fd)
{
        struct read_blob_callbacks cbs = {
                .consume_chunk  = extract_chunk_to_fd,
                .ctx            = fd,
        };
        return read_blob_with_sha1(blob, &cbs);
}

/* Calculate the SHA-1 message digest of a blob and store it in @blob->hash.  */
int
sha1_blob(struct blob_descriptor *blob)
{
        struct read_blob_callbacks cbs = {
        };
        return read_blob_with_sha1(blob, &cbs);
}

/*
 * Convert a short WIM resource header to a stand-alone WIM resource descriptor.
 *
 * Note: for solid resources some fields still need to be overridden.
 */
void
wim_reshdr_to_desc(const struct wim_reshdr *reshdr, WIMStruct *wim,
                   struct wim_resource_descriptor *rdesc)
{
        rdesc->wim = wim;
        rdesc->offset_in_wim = reshdr->offset_in_wim;
        rdesc->size_in_wim = reshdr->size_in_wim;
        rdesc->uncompressed_size = reshdr->uncompressed_size;
        INIT_LIST_HEAD(&rdesc->blob_list);
        rdesc->flags = reshdr->flags;
        rdesc->is_pipable = wim_is_pipable(wim);
        if (rdesc->flags & WIM_RESHDR_FLAG_COMPRESSED) {
                rdesc->compression_type = wim->compression_type;
                rdesc->chunk_size = wim->chunk_size;
        } else {
                rdesc->compression_type = WIMLIB_COMPRESSION_TYPE_NONE;
                rdesc->chunk_size = 0;
        }
}

/* Import a WIM resource header from the on-disk format.  */
void
get_wim_reshdr(const struct wim_reshdr_disk *disk_reshdr,
               struct wim_reshdr *reshdr)
{
        reshdr->offset_in_wim = le64_to_cpu(disk_reshdr->offset_in_wim);
        reshdr->size_in_wim = (((u64)disk_reshdr->size_in_wim[0] <<  0) |
                               ((u64)disk_reshdr->size_in_wim[1] <<  8) |
                               ((u64)disk_reshdr->size_in_wim[2] << 16) |
                               ((u64)disk_reshdr->size_in_wim[3] << 24) |
                               ((u64)disk_reshdr->size_in_wim[4] << 32) |
                               ((u64)disk_reshdr->size_in_wim[5] << 40) |
                               ((u64)disk_reshdr->size_in_wim[6] << 48));
        reshdr->uncompressed_size = le64_to_cpu(disk_reshdr->uncompressed_size);
        reshdr->flags = disk_reshdr->flags;
}

/* Export a WIM resource header to the on-disk format.  */
void
put_wim_reshdr(const struct wim_reshdr *reshdr,
               struct wim_reshdr_disk *disk_reshdr)
{
        disk_reshdr->size_in_wim[0] = reshdr->size_in_wim  >>  0;
        disk_reshdr->size_in_wim[1] = reshdr->size_in_wim  >>  8;
        disk_reshdr->size_in_wim[2] = reshdr->size_in_wim  >> 16;
        disk_reshdr->size_in_wim[3] = reshdr->size_in_wim  >> 24;
        disk_reshdr->size_in_wim[4] = reshdr->size_in_wim  >> 32;
        disk_reshdr->size_in_wim[5] = reshdr->size_in_wim  >> 40;
        disk_reshdr->size_in_wim[6] = reshdr->size_in_wim  >> 48;
        disk_reshdr->flags = reshdr->flags;
        disk_reshdr->offset_in_wim = cpu_to_le64(reshdr->offset_in_wim);
        disk_reshdr->uncompressed_size = cpu_to_le64(reshdr->uncompressed_size);
}