X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fntfs-3g_capture.c;h=ca8d1b2ffe4a335002569c0a6de780a52d13e9f2;hp=22b9e55422497ae85e5a45a375863d6b806482f0;hb=43f92ce30b2398d1823e34e39ff248de521d015c;hpb=5f3757249c17cb9e2826f645e0f8618534c80fcc diff --git a/src/ntfs-3g_capture.c b/src/ntfs-3g_capture.c index 22b9e554..ca8d1b2f 100644 --- a/src/ntfs-3g_capture.c +++ b/src/ntfs-3g_capture.c @@ -6,7 +6,7 @@ */ /* - * Copyright (C) 2012, 2013, 2014 Eric Biggers + * Copyright (C) 2012-2016 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 @@ -29,56 +29,98 @@ #ifdef WITH_NTFS_3G #include -#include - -#ifdef HAVE_ALLOCA_H -# include -#endif #include +#include /* for ENODATA, if needed */ +#include #include #include #include +#include "wimlib/alloca.h" #include "wimlib/assert.h" -#include "wimlib/capture.h" +#include "wimlib/blob_table.h" #include "wimlib/dentry.h" #include "wimlib/encoding.h" #include "wimlib/endianness.h" #include "wimlib/error.h" -#include "wimlib/lookup_table.h" #include "wimlib/ntfs_3g.h" +#include "wimlib/object_id.h" #include "wimlib/paths.h" +#include "wimlib/reparse.h" +#include "wimlib/scan.h" #include "wimlib/security.h" -static inline ntfschar * -attr_record_name(ATTR_RECORD *ar) +/* NTFS-3G 2013 renamed MS_RDONLY to NTFS_MNT_RDONLY. We can't check for the + * existence of NTFS_MNT_RDONLY at compilation time because it's an enum. We + * also can't check for MS_RDONLY being missing because it's also a system + * constant. So check if the NTFS-3G specific MS_IGNORE_HIBERFILE is defined; + * if yes, then we need to use the old MS_RDONLY. */ +#ifdef MS_IGNORE_HIBERFILE +# define NTFS_MNT_RDONLY MS_RDONLY +#endif + +/* A reference-counted NTFS volume than is automatically unmounted when the + * reference count reaches 0 */ +struct ntfs_volume_wrapper { + ntfs_volume *vol; + size_t refcnt; + bool dedup_warned; +}; + +/* Description of where data is located in an NTFS volume */ +struct ntfs_location { + struct ntfs_volume_wrapper *volume; + u64 mft_no; + ATTR_TYPES attr_type; + u32 attr_name_nchars; + ntfschar *attr_name; + u64 sort_key; +}; + +static struct ntfs_volume_wrapper * +get_ntfs_volume(struct ntfs_volume_wrapper *volume) +{ + volume->refcnt++; + return volume; +} + +static void +put_ntfs_volume(struct ntfs_volume_wrapper *volume) +{ + if (--volume->refcnt == 0) { + ntfs_umount(volume->vol, FALSE); + FREE(volume); + } +} + +static inline const ntfschar * +attr_record_name(const ATTR_RECORD *record) { - return (ntfschar*)((u8*)ar + le16_to_cpu(ar->name_offset)); + return (const ntfschar *) + ((const u8 *)record + le16_to_cpu(record->name_offset)); } static ntfs_attr * -open_ntfs_attr(ntfs_inode *ni, struct ntfs_location *loc) +open_ntfs_attr(ntfs_inode *ni, const struct ntfs_location *loc) { ntfs_attr *na; - na = ntfs_attr_open(ni, - loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA, - loc->stream_name, - loc->stream_name_nchars); + na = ntfs_attr_open(ni, loc->attr_type, loc->attr_name, + loc->attr_name_nchars); if (!na) { - ERROR_WITH_ERRNO("Failed to open attribute of \"%"TS"\" in " - "NTFS volume", loc->path); + ERROR_WITH_ERRNO("Failed to open attribute of NTFS inode %"PRIu64, + loc->mft_no); } return na; } int -read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t cb, void *cb_ctx) +read_ntfs_attribute_prefix(const struct blob_descriptor *blob, u64 size, + const struct consume_chunk_callback *cb) { - struct ntfs_location *loc = lte->ntfs_loc; - ntfs_volume *vol = loc->ntfs_vol; + const struct ntfs_location *loc = blob->ntfs_loc; + ntfs_volume *vol = loc->volume->vol; ntfs_inode *ni; ntfs_attr *na; s64 pos; @@ -86,9 +128,10 @@ read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte, u64 size, int ret; u8 buf[BUFFER_SIZE]; - ni = ntfs_pathname_to_inode(vol, NULL, loc->path); + ni = ntfs_inode_open(vol, loc->mft_no); if (!ni) { - ERROR_WITH_ERRNO("Can't find NTFS inode for \"%"TS"\"", loc->path); + ERROR_WITH_ERRNO("Failed to open NTFS inode %"PRIu64, + loc->mft_no); ret = WIMLIB_ERR_NTFS_3G; goto out; } @@ -99,18 +142,19 @@ read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte, u64 size, goto out_close_ntfs_inode; } - pos = (loc->is_reparse_point) ? 8 : 0; + pos = (loc->attr_type == AT_REPARSE_POINT) ? REPARSE_DATA_OFFSET : 0; bytes_remaining = size; while (bytes_remaining) { s64 to_read = min(bytes_remaining, sizeof(buf)); if (ntfs_attr_pread(na, pos, to_read, buf) != to_read) { - ERROR_WITH_ERRNO("Error reading \"%"TS"\"", loc->path); + ERROR_WITH_ERRNO("Error reading data from NTFS inode " + "%"PRIu64, loc->mft_no); ret = WIMLIB_ERR_NTFS_3G; goto out_close_ntfs_attr; } pos += to_read; bytes_remaining -= to_read; - ret = cb(buf, to_read, cb_ctx); + ret = consume_chunk(cb, buf, to_read); if (ret) goto out_close_ntfs_attr; } @@ -123,187 +167,359 @@ out: return ret; } +void +free_ntfs_location(struct ntfs_location *loc) +{ + put_ntfs_volume(loc->volume); + if (loc->attr_name != AT_UNNAMED) + FREE(loc->attr_name); + FREE(loc); +} + +struct ntfs_location * +clone_ntfs_location(const struct ntfs_location *loc) +{ + struct ntfs_location *new = memdup(loc, sizeof(*loc)); + if (!new) + goto err0; + if (loc->attr_name != AT_UNNAMED) { + new->attr_name = utf16le_dup(loc->attr_name); + if (!new->attr_name) + goto err1; + } + new->volume = get_ntfs_volume(loc->volume); + return new; + +err1: + FREE(new); +err0: + return NULL; +} + +int +cmp_ntfs_locations(const struct ntfs_location *loc1, + const struct ntfs_location *loc2) +{ + return cmp_u64(loc1->sort_key, loc2->sort_key); +} + +/* Read rptag and rpreserved from the NTFS inode and save them in the WIM inode. + */ static int -read_reparse_tag(ntfs_inode *ni, struct ntfs_location *loc, - u32 *reparse_tag_ret) +read_reparse_header(ntfs_inode *ni, struct wim_inode *inode) { - int ret; - le32 reparse_tag; + struct { + le32 rptag; + le16 rpdatalen; + le16 rpreserved; + } hdr; + s64 res; ntfs_attr *na; - na = open_ntfs_attr(ni, loc); - if (!na) { - ret = WIMLIB_ERR_NTFS_3G; - goto out; - } + na = ntfs_attr_open(ni, AT_REPARSE_POINT, AT_UNNAMED, 0); + if (!na) + return WIMLIB_ERR_NTFS_3G; + + res = ntfs_attr_pread(na, 0, sizeof(hdr), &hdr); + + ntfs_attr_close(na); + + if (res != sizeof(hdr)) + return WIMLIB_ERR_NTFS_3G; - if (ntfs_attr_pread(na, 0, sizeof(reparse_tag), - &reparse_tag) != sizeof(reparse_tag)) + inode->i_reparse_tag = le32_to_cpu(hdr.rptag); + inode->i_rp_reserved = le16_to_cpu(hdr.rpreserved); + return 0; +} + + +static void +warn_special_reparse_points(const struct wim_inode *inode, + const struct scan_params *params, + struct ntfs_volume_wrapper *volume) +{ + if (inode->i_reparse_tag == WIM_IO_REPARSE_TAG_DEDUP && + (params->add_flags & WIMLIB_ADD_FLAG_WINCONFIG) && + !volume->dedup_warned) { - ERROR_WITH_ERRNO("Error reading reparse data"); - ret = WIMLIB_ERR_NTFS_3G; - goto out_close_ntfs_attr; + WARNING( + "Filesystem includes files deduplicated with Windows'\n" +" Data Deduplication feature, which to properly restore\n" +" would require that the chunk store in \"System Volume Information\"\n" +" be included in the WIM image. By default \"System Volume Information\"\n" +" is excluded, so you may want to use a custom capture configuration\n" +" file which includes it."); + volume->dedup_warned = true; } - *reparse_tag_ret = le32_to_cpu(reparse_tag); - DEBUG("ReparseTag = %#x", *reparse_tag_ret); - ret = 0; -out_close_ntfs_attr: +} + +static int +attr_type_to_wimlib_stream_type(ATTR_TYPES type) +{ + switch (type) { + case AT_DATA: + return STREAM_TYPE_DATA; + case AT_REPARSE_POINT: + return STREAM_TYPE_REPARSE_POINT; + default: + wimlib_assert(0); + return STREAM_TYPE_UNKNOWN; + } +} + +/* When sorting blobs located in NTFS volumes for sequential reading, we sort + * first by starting LCN of the attribute if available, otherwise no sort order + * is defined. This usually results in better sequential access to the volume. + */ +static int +set_attr_sort_key(ntfs_inode *ni, struct ntfs_location *loc) +{ + ntfs_attr *na; + runlist_element *rl; + + na = open_ntfs_attr(ni, loc); + if (!na) + return WIMLIB_ERR_NTFS_3G; + + rl = ntfs_attr_find_vcn(na, 0); + if (rl && rl->lcn != LCN_HOLE) + loc->sort_key = rl->lcn; + else + loc->sort_key = 0; + ntfs_attr_close(na); -out: - return ret; + return 0; +} + +/* + * Add a new stream to the specified inode, with duplicate checking. + * + * This works around a problem where NTFS-3G can list multiple unnamed data + * streams for a single file. In this case we can only keep one. We'll prefer + * one that is nonempty. + */ +static int +add_stream(struct wim_inode *inode, const char *path, int stream_type, + const utf16lechar *stream_name, struct blob_descriptor **blob_p, + struct list_head *unhashed_blobs) +{ + struct blob_descriptor *blob = *blob_p; + struct wim_inode_stream *strm; + + strm = inode_get_stream(inode, stream_type, stream_name); + if (unlikely(strm)) { + /* Stream already existed. */ + if (!blob) + return 0; + if (stream_blob_resolved(strm)) { + WARNING("\"%s\" has multiple nonempty streams " + "with the same type and name! Only the first " + "will be saved.", path); + return 0; + } + inode_replace_stream_blob(inode, strm, blob, NULL); + } else { + strm = inode_add_stream(inode, stream_type, stream_name, blob); + if (unlikely(!strm)) + return WIMLIB_ERR_NOMEM; + } + prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs); + *blob_p = NULL; + return 0; } -/* Load the streams from a file or reparse point in the NTFS volume */ +/* Save information about an NTFS attribute (stream) to a WIM inode. */ static int -capture_ntfs_streams(struct wim_inode *inode, - ntfs_inode *ni, - char *path, - size_t path_len, - struct list_head *unhashed_streams, - ntfs_volume *vol, - ATTR_TYPES type) +scan_ntfs_attr(struct wim_inode *inode, + ntfs_inode *ni, + const char *path, + struct list_head *unhashed_blobs, + struct ntfs_volume_wrapper *volume, + ATTR_TYPES type, + const ATTR_RECORD *record) { - ntfs_attr_search_ctx *actx; - struct ntfs_location *ntfs_loc; + u64 data_size = ntfs_get_attribute_value_length(record); + const u32 name_nchars = record->name_length; + struct blob_descriptor *blob = NULL; + utf16lechar *stream_name = (utf16lechar *)NO_STREAM_NAME; int ret; - struct wim_lookup_table_entry *lte; - DEBUG("Capturing NTFS data streams from `%s'", path); + if (unlikely(name_nchars)) { + /* Named stream */ + stream_name = utf16le_dupz(attr_record_name(record), + name_nchars * sizeof(ntfschar)); + if (!stream_name) { + ret = WIMLIB_ERR_NOMEM; + goto out_cleanup; + } + } + + if (unlikely(type == AT_REPARSE_POINT)) { + if (data_size < REPARSE_DATA_OFFSET) { + ERROR("Reparse point attribute of \"%s\" " + "is too short!", path); + ret = WIMLIB_ERR_INVALID_REPARSE_DATA; + goto out_cleanup; + } + data_size -= REPARSE_DATA_OFFSET; + + ret = read_reparse_header(ni, inode); + if (ret) { + ERROR_WITH_ERRNO("Error reading reparse point header " + "of \"%s\"", path); + goto out_cleanup; + } + } + + /* If the stream is non-empty, set up a blob descriptor for it. */ + if (data_size != 0) { + blob = new_blob_descriptor(); + if (unlikely(!blob)) { + ret = WIMLIB_ERR_NOMEM; + goto out_cleanup; + } + + blob->ntfs_loc = MALLOC(sizeof(struct ntfs_location)); + if (unlikely(!blob->ntfs_loc)) { + ret = WIMLIB_ERR_NOMEM; + goto out_cleanup; + } + + blob->blob_location = BLOB_IN_NTFS_VOLUME; + blob->size = data_size; + blob->ntfs_loc->volume = get_ntfs_volume(volume); + blob->ntfs_loc->mft_no = ni->mft_no; + blob->ntfs_loc->attr_type = type; + + if (unlikely(name_nchars)) { + blob->ntfs_loc->attr_name_nchars = name_nchars; + blob->ntfs_loc->attr_name = utf16le_dup(stream_name); + if (!blob->ntfs_loc->attr_name) { + ret = WIMLIB_ERR_NOMEM; + goto out_cleanup; + } + } else { + blob->ntfs_loc->attr_name_nchars = 0; + blob->ntfs_loc->attr_name = AT_UNNAMED; + } + + ret = set_attr_sort_key(ni, blob->ntfs_loc); + if (ret) + goto out_cleanup; + } + + ret = add_stream(inode, path, attr_type_to_wimlib_stream_type(type), + stream_name, &blob, unhashed_blobs); +out_cleanup: + free_blob_descriptor(blob); + if (stream_name != NO_STREAM_NAME) + FREE(stream_name); + return ret; +} + +/* Scan attributes of the specified type from a file in the NTFS volume */ +static int +scan_ntfs_attrs_with_type(struct wim_inode *inode, + ntfs_inode *ni, + const char *path, + struct list_head *unhashed_blobs, + struct ntfs_volume_wrapper *volume, + ATTR_TYPES type) +{ + ntfs_attr_search_ctx *actx; + int ret; - /* Get context to search the streams of the NTFS file. */ actx = ntfs_attr_get_search_ctx(ni, NULL); if (!actx) { - ERROR_WITH_ERRNO("Cannot get NTFS attribute search " + ERROR_WITH_ERRNO("Failed to get NTFS attribute search " "context for \"%s\"", path); return WIMLIB_ERR_NTFS_3G; } - /* Capture each data stream or reparse data stream. */ while (!ntfs_attr_lookup(type, NULL, 0, CASE_SENSITIVE, 0, NULL, 0, actx)) { - u64 data_size = ntfs_get_attribute_value_length(actx->attr); - u64 name_length = actx->attr->name_length; - u32 stream_id; - - if (data_size == 0) { - /* Empty stream. No lookup table entry is needed. */ - lte = NULL; - ntfs_loc = NULL; - } else { - ntfs_loc = CALLOC(1, sizeof(*ntfs_loc)); - if (!ntfs_loc) { - ret = WIMLIB_ERR_NOMEM; - goto out_put_actx; - } - ntfs_loc->ntfs_vol = vol; - ntfs_loc->path = memdup(path, path_len + 1); - if (!ntfs_loc->path) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_ntfs_loc; - } - if (name_length) { - ntfs_loc->stream_name = memdup(attr_record_name(actx->attr), - name_length * 2); - if (!ntfs_loc->stream_name) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_ntfs_loc; - } - ntfs_loc->stream_name_nchars = name_length; - } - - lte = new_lookup_table_entry(); - if (!lte) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_ntfs_loc; - } - lte->resource_location = RESOURCE_IN_NTFS_VOLUME; - lte->ntfs_loc = ntfs_loc; - ntfs_loc = NULL; - if (type == AT_REPARSE_POINT) { - if (data_size < 8) { - ERROR("Invalid reparse data on \"%s\" " - "(only %u bytes)!", path, (unsigned)data_size); - ret = WIMLIB_ERR_NTFS_3G; - goto out_free_lte; - } - lte->ntfs_loc->is_reparse_point = true; - lte->size = data_size - 8; - ret = read_reparse_tag(ni, lte->ntfs_loc, - &inode->i_reparse_tag); - if (ret) - goto out_free_lte; - } else { - lte->ntfs_loc->is_reparse_point = false; - lte->size = data_size; - } - } - if (name_length == 0) { - /* Unnamed data stream. Put the reference to it in the - * dentry's inode. */ - if (inode->i_lte) { - if (lte) { - if (!(inode->i_attributes & - FILE_ATTRIBUTE_REPARSE_POINT)) - { - WARNING("Found two un-named " - "data streams for \"%s\" " - "(sizes = %"PRIu64", " - "%"PRIu64")", - path, - inode->i_lte->size, - lte->size); - } - free_lookup_table_entry(lte); - continue; - } - } else { - stream_id = 0; - inode->i_lte = lte; - } - } else { - /* Named data stream. Put the reference to it in the - * alternate data stream entries */ - struct wim_ads_entry *new_ads_entry; - - new_ads_entry = inode_add_ads_utf16le(inode, - attr_record_name(actx->attr), - name_length * 2); - if (!new_ads_entry) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_lte; - } - wimlib_assert(new_ads_entry->stream_name_nbytes == name_length * 2); - stream_id = new_ads_entry->stream_id; - new_ads_entry->lte = lte; - } - if (lte) { - add_unhashed_stream(lte, inode, - stream_id, unhashed_streams); - } + ret = scan_ntfs_attr(inode, + ni, + path, + unhashed_blobs, + volume, + type, + actx->attr); + if (ret) + goto out_put_actx; } - if (errno == ENOENT) { - ret = 0; - } else { + if (errno != ENOENT) { ERROR_WITH_ERRNO("Error listing NTFS attributes of \"%s\"", path); ret = WIMLIB_ERR_NTFS_3G; + goto out_put_actx; } - goto out_put_actx; -out_free_lte: - free_lookup_table_entry(lte); -out_free_ntfs_loc: - if (ntfs_loc) { - FREE(ntfs_loc->path); - FREE(ntfs_loc->stream_name); - FREE(ntfs_loc); - } + ret = 0; out_put_actx: ntfs_attr_put_search_ctx(actx); - if (ret == 0) - DEBUG("Successfully captured NTFS streams from \"%s\"", path); - else - ERROR("Failed to capture NTFS streams from \"%s\"", path); + return ret; +} + +static noinline_for_stack int +load_object_id(ntfs_inode *ni, struct wim_inode *inode) +{ + OBJECT_ID_ATTR attr; + int len; + + len = ntfs_get_ntfs_object_id(ni, (char *)&attr, sizeof(attr)); + if (likely(len == -ENODATA || len == 0)) + return 0; + if (len < 0) + return WIMLIB_ERR_NTFS_3G; + if (!inode_set_object_id(inode, &attr, len)) + return WIMLIB_ERR_NOMEM; + return 0; +} + +/* Load the security descriptor of an NTFS inode into the corresponding WIM + * inode and the WIM image's security descriptor set. */ +static noinline_for_stack int +get_security_descriptor(ntfs_inode *ni, struct wim_inode *inode, + ntfs_volume *vol, struct wim_sd_set *sd_set) +{ + struct SECURITY_CONTEXT scx = {.vol = vol}; + char _buf[4096]; + char *buf = _buf; + size_t avail_size = sizeof(_buf); + int ret; + +retry: + ret = ntfs_get_ntfs_acl(&scx, ni, buf, avail_size); + if (unlikely(ret < 0)) { + ret = WIMLIB_ERR_NTFS_3G; + goto out; + } + + if (unlikely(ret > avail_size)) { + if (unlikely(buf != _buf)) + FREE(buf); + buf = MALLOC(ret); + if (!buf) { + ret = WIMLIB_ERR_NOMEM; + goto out; + } + avail_size = ret; + goto retry; + } + + if (likely(ret > 0)) { + inode->i_security_id = sd_set_add_sd(sd_set, buf, ret); + if (unlikely(inode->i_security_id < 0)) { + ret = WIMLIB_ERR_NOMEM; + goto out; + } + } + + ret = 0; +out: + if (unlikely(buf != _buf)) + FREE(buf); return ret; } @@ -316,7 +532,7 @@ struct dos_name_node { struct avl_tree_node index_node; char dos_name[24]; int name_nbytes; - le64 ntfs_ino; + u64 ntfs_ino; }; #define DOS_NAME_NODE(avl_node) \ @@ -333,11 +549,10 @@ _avl_cmp_by_ntfs_ino(const struct avl_tree_node *n1, /* Inserts a new DOS name into the map */ static int insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name, - size_t name_nbytes, le64 ntfs_ino) + size_t name_nbytes, u64 ntfs_ino) { struct dos_name_node *new_node; - DEBUG("DOS name_len = %zu", name_nbytes); new_node = MALLOC(sizeof(struct dos_name_node)); if (!new_node) return WIMLIB_ERR_NOMEM; @@ -356,15 +571,12 @@ insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name, if (avl_tree_insert(&map->root, &new_node->index_node, _avl_cmp_by_ntfs_ino)) { - /* This should be impossible since an NTFS inode cannot - * have multiple DOS names, and we only should get each - * DOS name entry once from the ntfs_readdir() calls. */ - ERROR("NTFS inode %"PRIu64" has multiple DOS names", - le64_to_cpu(ntfs_ino)); + /* This should be impossible since an NTFS inode cannot have + * multiple DOS names, and we only should get each DOS name + * entry once from the ntfs_readdir() calls. */ + WARNING("NTFS inode %"PRIu64" has multiple DOS names", ntfs_ino); FREE(new_node); - return WIMLIB_ERR_NOMEM; } - DEBUG("Inserted DOS name for inode %"PRIu64, le64_to_cpu(ntfs_ino)); return 0; } @@ -376,7 +588,7 @@ lookup_dos_name(const struct dos_name_map *map, u64 ntfs_ino) struct dos_name_node dummy; struct avl_tree_node *res; - dummy.ntfs_ino = cpu_to_le64(ntfs_ino); + dummy.ntfs_ino = ntfs_ino; res = avl_tree_lookup_node(map->root, &dummy.index_node, _avl_cmp_by_ntfs_ino); @@ -390,16 +602,14 @@ set_dentry_dos_name(struct wim_dentry *dentry, const struct dos_name_map *map) { const struct dos_name_node *node; - if (dentry->is_win32_name) { + if (dentry->d_is_win32_name) { node = lookup_dos_name(map, dentry->d_inode->i_ino); if (node) { - dentry->short_name = utf16le_dupz(node->dos_name, - node->name_nbytes); - if (!dentry->short_name) + dentry->d_short_name = utf16le_dupz(node->dos_name, + node->name_nbytes); + if (!dentry->d_short_name) return WIMLIB_ERR_NOMEM; - dentry->short_name_nbytes = node->name_nbytes; - DEBUG("Assigned DOS name to ino %"PRIu64, - dentry->d_inode->i_ino); + dentry->d_short_name_nbytes = node->name_nbytes; } else { WARNING("NTFS inode %"PRIu64" has Win32 name with no " "corresponding DOS name", @@ -409,138 +619,152 @@ set_dentry_dos_name(struct wim_dentry *dentry, const struct dos_name_map *map) return 0; } -static void -free_dos_name_tree(struct avl_tree_node *node) { - if (node) { - free_dos_name_tree(node->left); - free_dos_name_tree(node->right); - FREE(DOS_NAME_NODE(node)); - } -} - static void destroy_dos_name_map(struct dos_name_map *map) { - free_dos_name_tree(map->root); + struct dos_name_node *node; + avl_tree_for_each_in_postorder(node, map->root, + struct dos_name_node, index_node) + FREE(node); } struct readdir_ctx { struct wim_dentry *parent; char *path; size_t path_len; - struct dos_name_map *dos_name_map; - ntfs_volume *vol; - struct capture_params *params; + struct dos_name_map dos_name_map; + struct ntfs_volume_wrapper *volume; + struct scan_params *params; int ret; }; static int -build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p, - ntfs_inode *ni, - char *path, - size_t path_len, - int name_type, - ntfs_volume *ntfs_vol, - struct capture_params *params); +ntfs_3g_build_dentry_tree_recursive(struct wim_dentry **root_p, + const MFT_REF mref, + char *path, + size_t path_len, + int name_type, + struct ntfs_volume_wrapper *volume, + struct scan_params *params); static int -wim_ntfs_capture_filldir(void *dirent, const ntfschar *name, - const int name_nchars, const int name_type, - const s64 pos, const MFT_REF mref, - const unsigned dt_type) +filldir(void *_ctx, const ntfschar *name, const int name_nchars, + const int name_type, const s64 pos, const MFT_REF mref, + const unsigned dt_type) { - struct readdir_ctx *ctx; - size_t mbs_name_nbytes; + struct readdir_ctx *ctx = _ctx; + const size_t name_nbytes = name_nchars * sizeof(ntfschar); char *mbs_name; + size_t mbs_name_nbytes; + size_t path_len; struct wim_dentry *child; int ret; - size_t path_len; - size_t name_nbytes = name_nchars * sizeof(ntfschar); - ctx = dirent; if (name_type & FILE_NAME_DOS) { /* If this is the entry for a DOS name, store it for later. */ - ret = insert_dos_name(ctx->dos_name_map, name, - name_nbytes, mref & MFT_REF_MASK_CPU); + ret = insert_dos_name(&ctx->dos_name_map, name, + name_nbytes, MREF(mref)); /* Return now if an error occurred or if this is just a DOS name * and not a Win32+DOS name. */ if (ret != 0 || name_type == FILE_NAME_DOS) goto out; } - ret = utf16le_to_tstr(name, name_nbytes, - &mbs_name, &mbs_name_nbytes); + + ret = utf16le_to_tstr(name, name_nbytes, &mbs_name, &mbs_name_nbytes); if (ret) goto out; - if (mbs_name[0] == '.' && - (mbs_name[1] == '\0' || - (mbs_name[1] == '.' && mbs_name[2] == '\0'))) { - /* . or .. entries - * - * note: name_type is POSIX for these, so DOS names will not - * have been inserted for them. */ - ret = 0; + if (should_ignore_filename(mbs_name, mbs_name_nbytes)) goto out_free_mbs_name; - } - /* Open the inode for this directory entry and recursively capture the - * directory tree rooted at it */ - ntfs_inode *ni = ntfs_inode_open(ctx->vol, mref); - if (!ni) { - /* XXX This used to be treated as an error, but NTFS-3g seemed - * to be unable to read some inodes on a Windows 8 image for - * some reason. */ - WARNING_WITH_ERRNO("Failed to open NTFS file \"%s/%s\"", - ctx->path, mbs_name); - ret = 0; - goto out_free_mbs_name; - } path_len = ctx->path_len; if (path_len != 1) ctx->path[path_len++] = '/'; memcpy(ctx->path + path_len, mbs_name, mbs_name_nbytes + 1); path_len += mbs_name_nbytes; child = NULL; - ret = build_dentry_tree_ntfs_recursive(&child, ni, ctx->path, - path_len, name_type, - ctx->vol, ctx->params); - path_len -= mbs_name_nbytes + 1; - if (child) - dentry_add_child(ctx->parent, child); - ntfs_inode_close(ni); + ret = ntfs_3g_build_dentry_tree_recursive(&child, mref, ctx->path, + path_len, name_type, + ctx->volume, ctx->params); + attach_scanned_tree(ctx->parent, child, ctx->params->blob_table); out_free_mbs_name: FREE(mbs_name); out: - ctx->path[ctx->path_len] = '\0'; ctx->ret = ret; return ret; } -/* Recursively build a WIM dentry tree corresponding to an NTFS volume. - * At the same time, update the WIM lookup table with lookup table entries for - * the NTFS streams, and build an array of security descriptors. - */ static int -build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret, - ntfs_inode *ni, - char *path, - size_t path_len, - int name_type, - ntfs_volume *vol, - struct capture_params *params) +ntfs_3g_recurse_directory(ntfs_inode *ni, char *path, size_t path_len, + struct wim_dentry *parent, + struct ntfs_volume_wrapper *volume, + struct scan_params *params) +{ + int ret; + s64 pos = 0; + struct readdir_ctx ctx = { + .parent = parent, + .path = path, + .path_len = path_len, + .dos_name_map = { .root = NULL }, + .volume = volume, + .params = params, + .ret = 0, + }; + ret = ntfs_readdir(ni, &pos, &ctx, filldir); + path[path_len] = '\0'; + if (unlikely(ret)) { + if (ctx.ret) { + /* wimlib error */ + ret = ctx.ret; + } else { + /* error from ntfs_readdir() itself */ + ERROR_WITH_ERRNO("Error reading directory \"%s\"", path); + ret = WIMLIB_ERR_NTFS_3G; + } + } else { + struct wim_dentry *child; + + ret = 0; + for_dentry_child(child, parent) { + ret = set_dentry_dos_name(child, &ctx.dos_name_map); + if (ret) + break; + } + } + destroy_dos_name_map(&ctx.dos_name_map); + return ret; +} + +static int +ntfs_3g_build_dentry_tree_recursive(struct wim_dentry **root_ret, + const MFT_REF mref, + char *path, + size_t path_len, + int name_type, + struct ntfs_volume_wrapper *volume, + struct scan_params *params) { u32 attributes; int ret; struct wim_dentry *root = NULL; struct wim_inode *inode = NULL; + ntfs_inode *ni = NULL; - ret = try_exclude(path, path_len, params); - if (ret < 0) /* Excluded? */ + ret = try_exclude(path, params); + if (unlikely(ret < 0)) /* Excluded? */ goto out_progress; - if (ret > 0) /* Error? */ + if (unlikely(ret > 0)) /* Error? */ goto out; + ni = ntfs_inode_open(volume->vol, mref); + if (!ni) { + ERROR_WITH_ERRNO("Failed to open NTFS file \"%s\"", path); + ret = WIMLIB_ERR_NTFS_3G; + goto out; + } + /* Get file attributes */ ret = ntfs_get_ntfs_attrib(ni, (char*)&attributes, sizeof(attributes)); if (ret != sizeof(attributes)) { @@ -549,17 +773,16 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret, goto out; } - if ((attributes & (FILE_ATTRIBUTE_DIRECTORY | - FILE_ATTRIBUTE_ENCRYPTED)) == FILE_ATTRIBUTE_ENCRYPTED) - { + if (unlikely(attributes & FILE_ATTRIBUTE_ENCRYPTED)) { if (params->add_flags & WIMLIB_ADD_FLAG_NO_UNSUPPORTED_EXCLUDE) { - ERROR("Can't archive unsupported encrypted file \"%s\"", path); + ERROR("Can't archive \"%s\" because NTFS-3G capture mode " + "does not support encrypted files and directories", path); ret = WIMLIB_ERR_UNSUPPORTED_FILE; goto out; } params->progress.scan.cur_path = path; - ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_UNSUPPORTED, NULL); + ret = do_scan_progress(params, WIMLIB_SCAN_DENTRY_UNSUPPORTED, NULL); goto out; } @@ -571,7 +794,7 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret, goto out; if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name (rather than POSIX) */ - root->is_win32_name = 1; + root->d_is_win32_name = 1; inode = root->d_inode; @@ -584,227 +807,143 @@ build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret, inode->i_last_write_time = le64_to_cpu(ni->last_data_change_time); inode->i_last_access_time = le64_to_cpu(ni->last_access_time); inode->i_attributes = attributes; - inode->i_resolved = 1; - /* Capture streams. */ - - if (attributes & FILE_ATTR_REPARSE_POINT) { - /* Capture reparse data stream. */ - ret = capture_ntfs_streams(inode, ni, path, path_len, - params->unhashed_streams, - vol, AT_REPARSE_POINT); + if (attributes & FILE_ATTRIBUTE_REPARSE_POINT) { + /* Scan the reparse point stream. */ + ret = scan_ntfs_attrs_with_type(inode, ni, path, + params->unhashed_blobs, + volume, AT_REPARSE_POINT); if (ret) goto out; + + warn_special_reparse_points(inode, params, volume); } - /* Capture data streams. - * - * Directories should not have an unnamed data stream, but they may have - * named data streams. - * - * Reparse points may have an unnamed data stream (which will be ignored - * in favor of the reparse data stream), and they also may have named - * data streams. + /* Load the object ID. */ + ret = load_object_id(ni, inode); + if (ret) { + ERROR_WITH_ERRNO("Error reading object ID of \"%s\"", path); + goto out; + } + + /* Scan the data streams. * - * Regular files can have an unnamed data stream as well as named data + * Note: directories should not have an unnamed data stream, but they + * may have named data streams. Nondirectories (including reparse + * points) can have an unnamed data stream as well as named data * streams. */ - ret = capture_ntfs_streams(inode, ni, path, path_len, - params->unhashed_streams, vol, AT_DATA); + ret = scan_ntfs_attrs_with_type(inode, ni, path, params->unhashed_blobs, + volume, AT_DATA); if (ret) goto out; - if (ni->mrec->flags & MFT_RECORD_IS_DIRECTORY) { - - /* Recurse to directory children */ - s64 pos = 0; - struct dos_name_map dos_name_map = { .root = NULL }; - struct readdir_ctx ctx = { - .parent = root, - .path = path, - .path_len = path_len, - .dos_name_map = &dos_name_map, - .vol = vol, - .params = params, - .ret = 0, - }; - ret = ntfs_readdir(ni, &pos, &ctx, wim_ntfs_capture_filldir); - if (ret) { - if (ctx.ret) { - /* wimlib error */ - ret = ctx.ret; - } else { - /* error from ntfs_readdir() itself */ - ERROR_WITH_ERRNO("Error reading directory \"%s\"", path); - ret = WIMLIB_ERR_NTFS_3G; - } - } else { - struct wim_dentry *child; - - ret = 0; - for_dentry_child(child, root) { - ret = set_dentry_dos_name(child, &dos_name_map); - if (ret) - break; - } - } - destroy_dos_name_map(&dos_name_map); - if (ret) - goto out; - } - path[path_len] = '\0'; - - /* Reparse-point fixups are a no-op because in NTFS-3g capture mode we + /* Reparse-point fixups are a no-op because in NTFS-3G capture mode we * only allow capturing an entire volume. */ if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX && inode_is_symlink(inode)) - inode->i_not_rpfixed = 0; + inode->i_rp_flags &= ~WIM_RP_FLAG_NOT_FIXED; if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) { - struct SECURITY_CONTEXT sec_ctx; - char _sd[4096]; - char *sd; - - /* Get security descriptor */ - memset(&sec_ctx, 0, sizeof(sec_ctx)); - sec_ctx.vol = vol; - - errno = 0; - sd = _sd; - ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, sizeof(_sd)); - if (ret > sizeof(_sd)) { - sd = alloca(ret); - ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, ret); - } - if (ret > 0) { - inode->i_security_id = sd_set_add_sd(params->sd_set, - sd, ret); - if (inode->i_security_id == -1) { - ERROR("Out of memory"); - ret = WIMLIB_ERR_NOMEM; - goto out; - } - DEBUG("Added security ID = %u for `%s'", - inode->i_security_id, path); - ret = 0; - } else if (ret < 0) { - ERROR_WITH_ERRNO("Failed to get security information from " - "`%s'", path); - ret = WIMLIB_ERR_NTFS_3G; - } else { - inode->i_security_id = -1; - DEBUG("No security ID for `%s'", path); + ret = get_security_descriptor(ni, inode, volume->vol, + params->sd_set); + if (ret) { + ERROR_WITH_ERRNO("Error reading security descriptor " + "of \"%s\"", path); + goto out; } } - if (ret) - goto out; + + if (inode_is_directory(inode)) { + ret = ntfs_3g_recurse_directory(ni, path, path_len, root, + volume, params); + if (ret) + goto out; + } out_progress: params->progress.scan.cur_path = path; if (root == NULL) - ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL); + ret = do_scan_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL); else - ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_OK, inode); + ret = do_scan_progress(params, WIMLIB_SCAN_DENTRY_OK, inode); out: + if (ni) + ntfs_inode_close(ni); if (unlikely(ret)) { - free_dentry_tree(root, params->lookup_table); + free_dentry_tree(root, params->blob_table); root = NULL; - ret = report_capture_error(params, ret, path); + ret = report_scan_error(params, ret, path); } *root_ret = root; return ret; } - -int -do_ntfs_umount(struct _ntfs_volume *vol) -{ - DEBUG("Unmounting NTFS volume"); - if (ntfs_umount(vol, FALSE)) - return WIMLIB_ERR_NTFS_3G; - else - return 0; -} - int -build_dentry_tree_ntfs(struct wim_dentry **root_p, - const char *device, - struct capture_params *params) +ntfs_3g_build_dentry_tree(struct wim_dentry **root_ret, + const char *device, struct scan_params *params) { + struct ntfs_volume_wrapper *volume; ntfs_volume *vol; - ntfs_inode *root_ni; + char *path; int ret; - DEBUG("Mounting NTFS volume `%s' read-only", device); + volume = CALLOC(1, sizeof(struct ntfs_volume_wrapper)); + if (!volume) + return WIMLIB_ERR_NOMEM; -/* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to - * NTFS_MNT_RDONLY. - * - * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because - * NTFS_MNT_RDONLY is an enumerated constant. Also, the NTFS-3g headers don't - * seem to contain any explicit version information. So we have to rely on a - * test done at configure time to detect whether NTFS_MNT_RDONLY should be used. - * */ -#ifdef HAVE_NTFS_MNT_RDONLY - /* NTFS-3g 2013 */ vol = ntfs_mount(device, NTFS_MNT_RDONLY); -#elif defined(MS_RDONLY) - /* NTFS-3g 2011, 2012 */ - vol = ntfs_mount(device, MS_RDONLY); -#else - #error "Can't find NTFS_MNT_RDONLY or MS_RDONLY flags" -#endif if (!vol) { - ERROR_WITH_ERRNO("Failed to mount NTFS volume `%s' read-only", + ERROR_WITH_ERRNO("Failed to mount NTFS volume \"%s\" read-only", device); + FREE(volume); return WIMLIB_ERR_NTFS_3G; } - ntfs_open_secure(vol); + + volume->vol = vol; + volume->refcnt = 1; + + /* Currently, libntfs-3g users that need to read security descriptors + * are required to call ntfs_open_secure() to open the volume's security + * descriptor index, "$Secure". This is only required to work on NTFS + * v3.0+, as older versions did not have a security descriptor index. */ + if (ntfs_open_secure(vol) && vol->major_ver >= 3) { + ERROR_WITH_ERRNO("Unable to open security descriptor index of " + "NTFS volume \"%s\"", device); + ret = WIMLIB_ERR_NTFS_3G; + goto out_put_ntfs_volume; + } /* We don't want to capture the special NTFS files such as $Bitmap. Not * to be confused with "hidden" or "system" files which are real files * that we do need to capture. */ NVolClearShowSysFiles(vol); - DEBUG("Opening root NTFS dentry"); - root_ni = ntfs_inode_open(vol, FILE_root); - if (!root_ni) { - ERROR_WITH_ERRNO("Failed to open root inode of NTFS volume " - "`%s'", device); - ret = WIMLIB_ERR_NTFS_3G; - goto out; - } - /* Currently we assume that all the paths fit into this length and there - * is no check for overflow. */ - char *path = MALLOC(32768); + * is no check for overflow. */ + path = MALLOC(32768); if (!path) { - ERROR("Could not allocate memory for NTFS pathname"); ret = WIMLIB_ERR_NOMEM; - goto out_cleanup; + goto out_close_secure; } path[0] = '/'; path[1] = '\0'; - ret = build_dentry_tree_ntfs_recursive(root_p, root_ni, path, 1, - FILE_NAME_POSIX, vol, params); -out_cleanup: + ret = ntfs_3g_build_dentry_tree_recursive(root_ret, FILE_root, path, 1, + FILE_NAME_POSIX, volume, + params); FREE(path); - ntfs_inode_close(root_ni); -out: - ntfs_index_ctx_put(vol->secure_xsii); - ntfs_index_ctx_put(vol->secure_xsdh); - ntfs_inode_close(vol->secure_ni); - - if (ret) { - if (do_ntfs_umount(vol)) { - ERROR_WITH_ERRNO("Failed to unmount NTFS volume `%s'", - device); - } - } else { - /* We need to leave the NTFS volume mounted so that we can read - * the NTFS files again when we are actually writing the WIM */ - *(ntfs_volume**)params->extra_arg = vol; +out_close_secure: + /* Undo the effects of ntfs_open_secure(). This is not yet done + * automatically by ntfs_umount(). But NULL out the inode to + * potentially be robust against future versions doing so. */ + if (vol->secure_ni) { + ntfs_index_ctx_put(vol->secure_xsii); + ntfs_index_ctx_put(vol->secure_xsdh); + ntfs_inode_close(vol->secure_ni); + vol->secure_ni = NULL; } +out_put_ntfs_volume: + put_ntfs_volume(volume); return ret; } #endif /* WITH_NTFS_3G */