X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fhardlink.c;h=3022cadfc47069340e5ae86c67fe1a06dd85bd10;hp=5e8ae1a94b9c7e654a5bd4b1354ec1c6514de310;hb=1770f3592215549707de3c5f2a1c4858e50b17ba;hpb=60b8f54df8fed44136bdc8ec615ee62703d87b69 diff --git a/src/hardlink.c b/src/hardlink.c index 5e8ae1a9..3022cadf 100644 --- a/src/hardlink.c +++ b/src/hardlink.c @@ -1,13 +1,11 @@ /* * hardlink.c * - * Code to deal with hard links in WIMs. Essentially, the WIM dentries are put - * into a hash table indexed by the hard link group ID field, then for each hard - * link group, a linked list is made to connect the dentries. + * Code to deal with hard links in WIMs. */ /* - * Copyright (C) 2012 Eric Biggers + * Copyright (C) 2012, 2013 Eric Biggers * * This file is part of wimlib, a library for working with WIM files. * @@ -32,147 +30,90 @@ /* NULL NULL * ^ ^ - * dentry | | - * / \ ----------- ----------- + * dentry | | + * / \ ----------- ----------- * | dentry<---| struct | | struct |---> dentry - * \ / |link_group| |link_group| + * \ / | wim_inode| | wim_inode| * dentry ------------ ------------ * ^ ^ * | | * | | dentry * ----------- ----------- / \ * dentry<---| struct | | struct |---> dentry dentry - * / |link_group| |link_group| \ / + * / | wim_inode| | wim_inode| \ / * dentry ------------ ------------ dentry * ^ ^ * | | * ----------------- - * link_group_table->array | idx 0 | idx 1 | + * wim_inode_table->array | idx 0 | idx 1 | * ----------------- */ -/* Hard link group; it's identified by its hard link group ID and points to a - * circularly linked list of dentries. */ -struct link_group { - u64 link_group_id; - - /* Pointer to use to make a singly-linked list of link groups. */ - struct link_group *next; - - /* This is a pointer to the circle and not part of the circle itself. - * This makes it easy to iterate through other dentries hard-linked to a - * given dentry without having to find the "head" of the list first. */ - struct list_head *dentry_list; -}; - -/* Hash table to find hard link groups, identified by their hard link group ID. - * */ -struct link_group_table { - /* Fields for the hash table */ - struct link_group **array; - u64 num_entries; - u64 capacity; - - /* - * Linked list of "extra" groups. These may be: - * - * - Hard link groups of size 1, which are all allowed to have 0 for - * their hard link group ID, meaning we cannot insert them into the - * hash table before calling assign_link_group_ids(). - * - * - Groups we create ourselves by splitting a nominal hard link group - * due to inconsistencies in the dentries. These groups will share a - * hard link group ID with some other group until - * assign_link_group_ids() is called. - */ - struct link_group *extra_groups; -}; - -/* Returns pointer to a new link group table having the specified capacity */ -struct link_group_table *new_link_group_table(size_t capacity) + +int +init_inode_table(struct wim_inode_table *table, size_t capacity) { - struct link_group_table *table; - struct link_group **array; - - table = MALLOC(sizeof(struct link_group_table)); - if (!table) - goto err; - array = CALLOC(capacity, sizeof(array[0])); - if (!array) { - FREE(table); - goto err; + table->array = CALLOC(capacity, sizeof(table->array[0])); + if (!table->array) { + ERROR("Cannot initalize inode table: out of memory"); + return WIMLIB_ERR_NOMEM; } table->num_entries = 0; table->capacity = capacity; - table->array = array; - table->extra_groups = NULL; - return table; -err: - ERROR("Failed to allocate memory for link group table with capacity %zu", - capacity); - return NULL; + INIT_LIST_HEAD(&table->extra_inodes); + return 0; } -/* - * Insert a dentry into the hard link group table based on its hard link group - * ID. - * - * If there is already a dentry in the table having the same hard link group ID, - * and the hard link group ID is not 0, the dentry is added to the circular - * linked list for that hard link group. - * - * If the hard link group ID is 0, this indicates a dentry that's in a hard link - * group by itself (has a link count of 1). We can't insert it into the hash - * table itself because we don't know what hard link group IDs are available to - * give it (this could be kept track of but would be more difficult). Instead - * we keep a linked list of the single dentries, and assign them hard link group - * IDs later. - */ -int link_group_table_insert(struct dentry *dentry, void *__table) +static inline size_t +inode_link_count(const struct wim_inode *inode) { - struct link_group_table *table = __table; - size_t pos; - struct link_group *group; - - if (dentry->link_group_id == 0) { - /* Single group--- Add to the list of extra groups (we can't put - * it in the table itself because all the singles have a link - * group ID of 0) */ - group = MALLOC(sizeof(struct link_group)); - if (!group) - return WIMLIB_ERR_NOMEM; - group->link_group_id = 0; - group->next = table->extra_groups; - table->extra_groups = group; - INIT_LIST_HEAD(&dentry->link_group_list); - group->dentry_list = &dentry->link_group_list; + const struct list_head *cur; + size_t size = 0; + list_for_each(cur, &inode->i_dentry) + size++; + return size; +} + +/* Insert a dentry into the inode table based on the inode number of the + * attached inode (which came from the hard link group ID field of the on-disk + * WIM dentry) */ +static int +inode_table_insert(struct wim_dentry *dentry, void *_table) +{ + struct wim_inode_table *table = _table; + struct wim_inode *d_inode = dentry->d_inode; + + if (d_inode->i_ino == 0) { + /* A dentry with a hard link group ID of 0 indicates that it's + * in a hard link group by itself. Add it to the list of extra + * inodes rather than inserting it into the hash lists. */ + list_add_tail(&d_inode->i_list, &table->extra_inodes); + + wimlib_assert(d_inode->i_dentry.next == &dentry->d_alias); + wimlib_assert(d_inode->i_dentry.prev == &dentry->d_alias); + wimlib_assert(d_inode->i_nlink == 1); } else { - /* Hard link group that may contain multiple dentries (the code - * will work even if the group actually contains only 1 dentry - * though) */ - - /* Try adding to existing hard link group */ - pos = dentry->link_group_id % table->capacity; - group = table->array[pos]; - while (group) { - if (group->link_group_id == dentry->link_group_id) { - list_add(&dentry->link_group_list, - group->dentry_list); + size_t pos; + struct wim_inode *inode; + struct hlist_node *cur; + + /* Try adding this dentry to an existing inode */ + pos = d_inode->i_ino % table->capacity; + hlist_for_each_entry(inode, cur, &table->array[pos], i_hlist) { + if (inode->i_ino == d_inode->i_ino) { + inode_add_dentry(dentry, inode); + inode->i_nlink++; return 0; } - group = group->next; } - /* Add new hard link group to the table */ + /* No inode in the table has the same number as this one, so add + * it to the table. */ + hlist_add_head(&d_inode->i_hlist, &table->array[pos]); - group = MALLOC(sizeof(struct link_group)); - if (!group) - return WIMLIB_ERR_NOMEM; - group->link_group_id = dentry->link_group_id; - group->next = table->array[pos]; - INIT_LIST_HEAD(&dentry->link_group_list); - group->dentry_list = &dentry->link_group_list; - table->array[pos] = group; + wimlib_assert(d_inode->i_dentry.next == &dentry->d_alias); + wimlib_assert(d_inode->i_dentry.prev == &dentry->d_alias); + wimlib_assert(d_inode->i_nlink == 1); /* XXX Make the table grow when too many entries have been * inserted. */ @@ -181,277 +122,232 @@ int link_group_table_insert(struct dentry *dentry, void *__table) return 0; } -static void free_link_group_list(struct link_group *group) +static struct wim_inode * +inode_table_get_inode(struct wim_inode_table *table, u64 ino, u64 devno) { - struct link_group *next_group; - while (group) { - next_group = group->next; - FREE(group); - group = next_group; + u64 hash = hash_u64(hash_u64(ino) + hash_u64(devno)); + size_t pos = hash % table->capacity; + struct wim_inode *inode; + struct hlist_node *cur; + + hlist_for_each_entry(inode, cur, &table->array[pos], i_hlist) { + if (inode->i_ino == ino && inode->i_devno == devno) { + DEBUG("Using existing inode {devno=%"PRIu64", ino=%"PRIu64"}", + devno, ino); + inode->i_nlink++; + return inode; + } } -} - -/* Frees a link group table. */ -void free_link_group_table(struct link_group_table *table) -{ - if (table) { - if (table->array) - for (size_t i = 0; i < table->capacity; i++) - free_link_group_list(table->array[i]); - free_link_group_list(table->extra_groups); - FREE(table); - } -} - -static u64 -assign_link_group_ids_to_list(struct link_group *group, u64 id, - struct link_group **extra_groups) -{ - struct dentry *dentry; - struct list_head *cur_head; - struct link_group *prev_group = NULL; - struct link_group *cur_group = group; - while (cur_group) { - cur_head = cur_group->dentry_list; - do { - dentry = container_of(cur_head, - struct dentry, - link_group_list); - dentry->link_group_id = id; - cur_head = cur_head->next; - } while (cur_head != cur_group->dentry_list); - cur_group->link_group_id = id; - id++; - prev_group = cur_group; - cur_group = cur_group->next; + inode = new_timeless_inode(); + if (inode) { + inode->i_ino = ino; + inode->i_devno = devno; + hlist_add_head(&inode->i_hlist, &table->array[pos]); + table->num_entries++; } - if (group && extra_groups) { - prev_group->next = *extra_groups; - *extra_groups = group; - } - return id; + return inode; } -/* Insert the link groups in the `extra_groups' list into the hash table */ -static void insert_extra_groups(struct link_group_table *table) +void +inode_ref_streams(struct wim_inode *inode) { - struct link_group *group, *next_group; - size_t pos; - - group = table->extra_groups; - while (group) { - next_group = group->next; - pos = group->link_group_id % table->capacity; - group->next = table->array[pos]; - table->array[pos] = group; - group = next_group; + for (unsigned i = 0; i <= inode->i_num_ads; i++) { + struct wim_lookup_table_entry *lte; + lte = inode_stream_lte_resolved(inode, i); + if (lte) + lte->refcnt++; } - table->extra_groups = NULL; } -/* Assign the link group IDs to dentries in a link group table, and return the - * next available link group ID. */ -u64 assign_link_group_ids(struct link_group_table *table) +/* Given a directory entry with the name @name for the file with the inode + * number @ino and device number @devno, create a new WIM dentry with an + * associated inode, where the inode is shared if an inode with the same @ino + * and @devno has already been created. On success, the new WIM dentry is + * written to *dentry_ret, and its inode has i_nlink > 1 if a previously + * existing inode was used. + */ +int +inode_table_new_dentry(struct wim_inode_table *table, const tchar *name, + u64 ino, u64 devno, struct wim_dentry **dentry_ret) { - DEBUG("Assigning link groups"); - struct link_group *extra_groups = table->extra_groups; + struct wim_dentry *dentry; + struct wim_inode *inode; + int ret; - /* Assign consecutive link group IDs to each link group in the hash - * table */ - u64 id = 1; - for (size_t i = 0; i < table->capacity; i++) { - id = assign_link_group_ids_to_list(table->array[i], id, - &table->extra_groups); - table->array[i] = NULL; - } + ret = new_dentry(name, &dentry); + if (ret) + return ret; - /* Assign link group IDs to the "extra" link groups and insert them into - * the hash table */ - id = assign_link_group_ids_to_list(extra_groups, id, NULL); - insert_extra_groups(table); - return id; + inode = inode_table_get_inode(table, ino, devno); + if (!inode) { + free_dentry(dentry); + return WIMLIB_ERR_NOMEM; + } + if (inode->i_nlink > 1) + inode_ref_streams(inode); + dentry->d_inode = inode; + inode_add_dentry(dentry, inode); + *dentry_ret = dentry; + return 0; } - - -static void inconsistent_link_group(const struct dentry *first_dentry) +#if defined(ENABLE_ERROR_MESSAGES) || defined(ENABLE_DEBUG) +static void +print_inode_dentries(const struct wim_inode *inode) { - const struct dentry *dentry = first_dentry; + struct wim_dentry *dentry; + inode_for_each_dentry(dentry, inode) + tfprintf(stderr, T("%"TS"\n"), dentry_full_path(dentry)); +} +#endif - ERROR("An inconsistent hard link group that we cannot correct has been " - "detected"); +static void +inconsistent_inode(const struct wim_inode *inode) +{ +#ifdef ENABLE_ERROR_MESSAGES + ERROR("An inconsistent hard link group that cannot be corrected has " + "been detected"); ERROR("The dentries are located at the following paths:"); - do { - ERROR("`%s'", dentry->full_path_utf8); - } while ((dentry = container_of(dentry->link_group_list.next, - const struct dentry, - link_group_list)) != first_dentry); + print_inode_dentries(inode); +#endif } -static bool ref_dentries_consistent(const struct dentry * restrict ref_dentry_1, - const struct dentry * restrict ref_dentry_2) +static bool +ref_inodes_consistent(const struct wim_inode * restrict ref_inode_1, + const struct wim_inode * restrict ref_inode_2) { - wimlib_assert(ref_dentry_1 != ref_dentry_2); + wimlib_assert(ref_inode_1 != ref_inode_2); - if (ref_dentry_1->num_ads != ref_dentry_2->num_ads) + if (ref_inode_1->i_num_ads != ref_inode_2->i_num_ads) return false; - if (ref_dentry_1->security_id != ref_dentry_2->security_id - || ref_dentry_1->attributes != ref_dentry_2->attributes) + if (ref_inode_1->i_security_id != ref_inode_2->i_security_id + || ref_inode_1->i_attributes != ref_inode_2->i_attributes) return false; - for (unsigned i = 0; i <= ref_dentry_1->num_ads; i++) { + for (unsigned i = 0; i <= ref_inode_1->i_num_ads; i++) { const u8 *ref_1_hash, *ref_2_hash; - ref_1_hash = dentry_stream_hash(ref_dentry_1, i); - ref_2_hash = dentry_stream_hash(ref_dentry_2, i); + ref_1_hash = inode_stream_hash(ref_inode_1, i); + ref_2_hash = inode_stream_hash(ref_inode_2, i); if (!hashes_equal(ref_1_hash, ref_2_hash)) return false; - if (i && !ads_entries_have_same_name(&ref_dentry_1->ads_entries[i - 1], - &ref_dentry_2->ads_entries[i - 1])) + if (i && !ads_entries_have_same_name(&ref_inode_1->i_ads_entries[i - 1], + &ref_inode_2->i_ads_entries[i - 1])) return false; } return true; } -static bool dentries_consistent(const struct dentry * restrict ref_dentry, - const struct dentry * restrict dentry) +static bool +inodes_consistent(const struct wim_inode * restrict ref_inode, + const struct wim_inode * restrict inode) { - wimlib_assert(ref_dentry != dentry); + wimlib_assert(ref_inode != inode); - if (ref_dentry->num_ads != dentry->num_ads && dentry->num_ads != 0) + if (ref_inode->i_num_ads != inode->i_num_ads && + inode->i_num_ads != 0) return false; - if (ref_dentry->security_id != dentry->security_id - || ref_dentry->attributes != dentry->attributes) + if (ref_inode->i_security_id != inode->i_security_id + || ref_inode->i_attributes != inode->i_attributes) return false; - for (unsigned i = 0; i <= min(ref_dentry->num_ads, dentry->num_ads); i++) { + for (unsigned i = 0; i <= min(ref_inode->i_num_ads, inode->i_num_ads); i++) { const u8 *ref_hash, *hash; - ref_hash = dentry_stream_hash(ref_dentry, i); - hash = dentry_stream_hash(dentry, i); + ref_hash = inode_stream_hash(ref_inode, i); + hash = inode_stream_hash(inode, i); if (!hashes_equal(ref_hash, hash) && !is_zero_hash(hash)) return false; - if (i && !ads_entries_have_same_name(&ref_dentry->ads_entries[i - 1], - &dentry->ads_entries[i - 1])) + if (i && !ads_entries_have_same_name(&ref_inode->i_ads_entries[i - 1], + &inode->i_ads_entries[i - 1])) return false; } return true; } -#ifdef ENABLE_DEBUG -static void -print_dentry_list(const struct dentry *first_dentry) -{ - const struct dentry *dentry = first_dentry; - do { - printf("`%s'\n", dentry->full_path_utf8); - } while ((dentry = container_of(dentry->link_group_list.next, - struct dentry, - link_group_list)) != first_dentry); -} -#endif - -/* Fix up a "true" link group and check for inconsistencies */ +/* Fix up a "true" inode and check for inconsistencies */ static int -fix_true_link_group(struct dentry *first_dentry) +fix_true_inode(struct wim_inode *inode, struct list_head *inode_list) { - struct dentry *dentry; - struct dentry *ref_dentry = NULL; + struct wim_dentry *dentry; + struct wim_dentry *ref_dentry = NULL; + struct wim_inode *ref_inode; u64 last_ctime = 0; u64 last_mtime = 0; u64 last_atime = 0; - dentry = first_dentry; - do { - if (!ref_dentry || ref_dentry->num_ads == 0) + inode_for_each_dentry(dentry, inode) { + if (!ref_dentry || dentry->d_inode->i_num_ads > ref_dentry->d_inode->i_num_ads) ref_dentry = dentry; - if (dentry->creation_time > last_ctime) - last_ctime = dentry->creation_time; - if (dentry->last_write_time > last_mtime) - last_mtime = dentry->last_write_time; - if (dentry->last_access_time > last_atime) - last_atime = dentry->last_access_time; - } while ((dentry = container_of(dentry->link_group_list.next, - struct dentry, - link_group_list)) != first_dentry); - - - ref_dentry->ads_entries_status = ADS_ENTRIES_OWNER; - dentry = first_dentry; - do { + if (dentry->d_inode->i_creation_time > last_ctime) + last_ctime = dentry->d_inode->i_creation_time; + if (dentry->d_inode->i_last_write_time > last_mtime) + last_mtime = dentry->d_inode->i_last_write_time; + if (dentry->d_inode->i_last_access_time > last_atime) + last_atime = dentry->d_inode->i_last_access_time; + } + + ref_inode = ref_dentry->d_inode; + ref_inode->i_nlink = 1; + list_add_tail(&ref_inode->i_list, inode_list); + + list_del(&inode->i_dentry); + list_add(&ref_inode->i_dentry, &ref_dentry->d_alias); + + inode_for_each_dentry(dentry, ref_inode) { if (dentry != ref_dentry) { - if (!dentries_consistent(ref_dentry, dentry)) { - inconsistent_link_group(first_dentry); + if (!inodes_consistent(ref_inode, dentry->d_inode)) { + inconsistent_inode(ref_inode); return WIMLIB_ERR_INVALID_DENTRY; } - copy_hash(dentry->hash, ref_dentry->hash); - dentry_free_ads_entries(dentry); - dentry->num_ads = ref_dentry->num_ads; - dentry->ads_entries = ref_dentry->ads_entries; - dentry->ads_entries_status = ADS_ENTRIES_USER; + /* Free the unneeded `struct wim_inode'. */ + dentry->d_inode->i_hlist.next = NULL; + dentry->d_inode->i_hlist.pprev = NULL; + free_inode(dentry->d_inode); + dentry->d_inode = ref_inode; + ref_inode->i_nlink++; } - dentry->creation_time = last_ctime; - dentry->last_write_time = last_mtime; - dentry->last_access_time = last_atime; - } while ((dentry = container_of(dentry->link_group_list.next, - struct dentry, - link_group_list)) != first_dentry); + } + ref_inode->i_creation_time = last_ctime; + ref_inode->i_last_write_time = last_mtime; + ref_inode->i_last_access_time = last_atime; + wimlib_assert(inode_link_count(ref_inode) == ref_inode->i_nlink); return 0; } -/* - * Fixes up a nominal link group. - * - * By a nominal link group we mean a group of two or more dentries that share - * the same hard link group ID. - * - * If dentries in the group are found to be inconsistent, we may split the group - * into several "true" hard link groups. @new_groups points to a linked list of - * these split groups, and if we create any, they will be added to this list. - * - * After splitting up each nominal link group into the "true" link groups we - * will canonicalize the link groups. To do this, we: - * - * - Assign all the dentries in the link group the most recent timestamp - * among all the corresponding timestamps in the link group, for each of - * the three categories of time stamps. +/* + * Fixes up a nominal inode. * - * - Make sure the dentry->hash field is valid in all the dentries, if - * possible (this field may be all zeroes, and in the context of a hard - * link group this must be interpreted as implicitly refering to the same - * stream as another dentry in the hard link group that does NOT have all - * zeroes for this field). + * By a nominal inode we mean a group of two or more dentries that share the + * same hard link group ID. * - * - Make sure dentry->num_ads is the same in all the dentries in the link - * group. In some cases, it's possible for it to be set to 0 when it - * actually must be interpreted as being the same as the number of - * alternate data streams in another dentry in the hard link group that has - * a nonzero number of alternate data streams. + * If dentries in the inode are found to be inconsistent, we may split the inode + * into several "true" inodes. * - * - Make sure only the dentry->ads_entries array is only allocated for one - * dentry in the hard link group. This dentry will have - * dentry->ads_entries_status set to ADS_ENTRIES_OWNER, while the others - * will have dentry->ads_entries_status set to ADS_ENTRIES_USER. + * After splitting up each nominal inode into the "true" inodes we will + * canonicalize the link group by getting rid of all the unnecessary `struct + * wim_inode's. There will be just one `struct wim_inode' for each hard link + * group remaining. */ static int -fix_nominal_link_group(struct link_group *group, - struct link_group **new_groups) +fix_nominal_inode(struct wim_inode *inode, struct list_head *inode_list, + bool *ino_changes_needed) { - struct dentry *tmp, *dentry, *ref_dentry; + struct wim_dentry *dentry; + struct hlist_node *cur, *tmp; int ret; - size_t num_true_link_groups; - struct list_head *head; + size_t num_true_inodes; + + wimlib_assert(inode->i_nlink == inode_link_count(inode)); LIST_HEAD(dentries_with_data_streams); LIST_HEAD(dentries_with_no_data_streams); - LIST_HEAD(true_link_groups); + HLIST_HEAD(true_inodes); - /* Create a list of dentries in the nominal hard link group that have at + /* Create a list of dentries in the nominal inode that have at * least one data stream with a non-zero hash, and another list that * contains the dentries that have a zero hash for all data streams. */ - dentry = container_of(group->dentry_list, struct dentry, - link_group_list); - do { - for (unsigned i = 0; i <= dentry->num_ads; i++) { + inode_for_each_dentry(dentry, inode) { + for (unsigned i = 0; i <= dentry->d_inode->i_num_ads; i++) { const u8 *hash; - hash = dentry_stream_hash(dentry, i); + hash = inode_stream_hash(dentry->d_inode, i); if (!is_zero_hash(hash)) { list_add(&dentry->tmp_list, &dentries_with_data_streams); @@ -461,150 +357,199 @@ fix_nominal_link_group(struct link_group *group, list_add(&dentry->tmp_list, &dentries_with_no_data_streams); next_dentry: - dentry = container_of(dentry->link_group_list.next, - struct dentry, - link_group_list); - } while (&dentry->link_group_list != group->dentry_list); + ; + } /* If there are no dentries with data streams, we require the nominal - * link group to be a true link group */ + * inode to be a true inode */ if (list_empty(&dentries_with_data_streams)) { #ifdef ENABLE_DEBUG - { - size_t size = dentry_link_group_size(dentry); - if (size > 1) { - DEBUG("Found link group of size %zu without " - "any data streams:", size); - print_dentry_list(dentry); - DEBUG("We are going to interpret it as true " - "link group, provided that the dentries " - "are consistent."); - } + if (inode->i_nlink > 1) { + DEBUG("Found link group of size %u without " + "any data streams:", inode->i_nlink); + print_inode_dentries(inode); + DEBUG("We are going to interpret it as true " + "link group, provided that the dentries " + "are consistent."); } #endif - return fix_true_link_group(container_of(group->dentry_list, - struct dentry, - link_group_list)); + return fix_true_inode(inode, inode_list); } /* One or more dentries had data streams specified. We check each of * these dentries for consistency with the others to form a set of true - * link groups. */ - num_true_link_groups = 0; - list_for_each_entry_safe(dentry, tmp, &dentries_with_data_streams, - tmp_list) - { - list_del(&dentry->tmp_list); - - /* Look for a true link group that is consistent with - * this dentry and add this dentry to it. Or, if none - * of the true link groups are consistent with this - * dentry, make a new one. */ - list_for_each_entry(ref_dentry, &true_link_groups, tmp_list) { - if (ref_dentries_consistent(ref_dentry, dentry)) { - list_add(&dentry->link_group_list, - &ref_dentry->link_group_list); + * inodes. */ + num_true_inodes = 0; + list_for_each_entry(dentry, &dentries_with_data_streams, tmp_list) { + /* Look for a true inode that is consistent with this dentry and + * add this dentry to it. Or, if none of the true inodes are + * consistent with this dentry, add a new one (if that happens, + * we have split the hard link group). */ + hlist_for_each_entry(inode, cur, &true_inodes, i_hlist) { + if (ref_inodes_consistent(inode, dentry->d_inode)) { + inode_add_dentry(dentry, inode); goto next_dentry_2; } } - num_true_link_groups++; - list_add(&dentry->tmp_list, &true_link_groups); - INIT_LIST_HEAD(&dentry->link_group_list); + num_true_inodes++; + INIT_LIST_HEAD(&dentry->d_inode->i_dentry); + inode_add_dentry(dentry, dentry->d_inode); + hlist_add_head(&dentry->d_inode->i_hlist, &true_inodes); next_dentry_2: ; } - wimlib_assert(num_true_link_groups != 0); + wimlib_assert(num_true_inodes != 0); /* If there were dentries with no data streams, we require there to only - * be one true link group so that we know which link group to assign the + * be one true inode so that we know which inode to assign the * streamless dentries to. */ if (!list_empty(&dentries_with_no_data_streams)) { - if (num_true_link_groups != 1) { - ERROR("Hard link group ambiguity detected!"); - ERROR("We split up hard link group 0x%"PRIx64" due to " - "inconsistencies,", group->link_group_id); + if (num_true_inodes != 1) { + ERROR("Hard inode ambiguity detected!"); + ERROR("We split up inode 0x%"PRIx64" due to " + "inconsistencies,", inode->i_ino); ERROR("but dentries with no stream information remained. " - "We don't know which true hard link"); - ERROR("group to assign them to."); + "We don't know which inode"); + ERROR("to assign them to."); return WIMLIB_ERR_INVALID_DENTRY; } - /* Assign the streamless dentries to the one and only true link - * group. */ - ref_dentry = container_of(true_link_groups.next, - struct dentry, - tmp_list); + inode = container_of(true_inodes.first, struct wim_inode, i_hlist); + /* Assign the streamless dentries to the one and only true + * inode. */ list_for_each_entry(dentry, &dentries_with_no_data_streams, tmp_list) - list_add(&dentry->link_group_list, &ref_dentry->link_group_list); + inode_add_dentry(dentry, inode); } - if (num_true_link_groups != 1) { - #ifdef ENABLE_DEBUG - { - printf("Split nominal link group 0x%"PRIx64" into %zu " - "link groups:\n", - group->link_group_id, num_true_link_groups); - puts("------------------------------------------------------------------------------"); - size_t i = 1; - list_for_each_entry(dentry, &true_link_groups, tmp_list) { - printf("[Split link group %zu]\n", i++); - print_dentry_list(dentry); - putchar('\n'); - } - puts("------------------------------------------------------------------------------"); + if (num_true_inodes != 1) { + #ifdef ENABLE_DEBUG + inode = container_of(true_inodes.first, struct wim_inode, i_hlist); + + tprintf(T("Split nominal inode 0x%"PRIx64" into %zu " + "inodes:\n"), inode->i_ino, num_true_inodes); + tputs(T("----------------------------------------------------" + "--------------------------")); + size_t i = 1; + hlist_for_each_entry(inode, cur, &true_inodes, i_hlist) { + tprintf(T("[Split inode %zu]\n"), i++); + print_inode_dentries(inode); + tputchar(T('\n')); } - #endif + tputs(T("----------------------------------------------------" + "--------------------------")); + #endif + *ino_changes_needed = true; } - list_for_each_entry(dentry, &true_link_groups, tmp_list) { - ret = fix_true_link_group(dentry); - if (ret != 0) + hlist_for_each_entry_safe(inode, cur, tmp, &true_inodes, i_hlist) { + ret = fix_true_inode(inode, inode_list); + if (ret) return ret; } + return 0; +} - /* Make new `struct link_group's for the new link groups */ - for (head = true_link_groups.next->next; - head != &true_link_groups; - head = head->next) - { - dentry = container_of(head, struct dentry, tmp_list); - group = MALLOC(sizeof(*group)); - if (!group) { - ERROR("Out of memory"); - return WIMLIB_ERR_NOMEM; +static int +fix_inodes(struct wim_inode_table *table, struct list_head *inode_list, + bool *ino_changes_needed) +{ + struct wim_inode *inode; + struct hlist_node *cur, *tmp; + int ret; + INIT_LIST_HEAD(inode_list); + for (u64 i = 0; i < table->capacity; i++) { + hlist_for_each_entry_safe(inode, cur, tmp, &table->array[i], i_hlist) { + ret = fix_nominal_inode(inode, inode_list, ino_changes_needed); + if (ret) + return ret; } - group->link_group_id = dentry->link_group_id; - group->dentry_list = &dentry->link_group_list; - group->next = *new_groups; - *new_groups = group; } + list_splice_tail(&table->extra_inodes, inode_list); return 0; } /* - * Goes through each link group and shares the ads_entries (Alternate Data - * Stream entries) field of each dentry among members of a hard link group. + * dentry_tree_fix_inodes(): * - * In the process, the dentries in each link group are checked for consistency. - * If they contain data features that indicate they cannot really be in the same - * hard link group, this should be an error, but in reality this case needs to - * be handled, so we split the dentries into different hard link groups. + * This function takes as input a tree of WIM dentries that initially has a + * different inode associated with each dentry. Sets of dentries that should + * share the same inode (a.k.a. hard link groups) are built using the i_ino + * field of each inode, then the link count and alias list for one inode in each + * set is set correctly and the unnecessary struct wim_inode's freed. The + * effect is to correctly associate exactly one struct wim_inode with each + * original inode, regardless of how many dentries are aliases for that inode. * - * One of the dentries in each hard link group group is arbitrarily assigned the - * role of "owner" of the memory pointed to by the @ads_entries field, - * (ADS_ENTRIES_OWNER), while the others are "users" (ADS_ENTRIES_USER) who are - * not allowed to free the memory. + * The special inode number of 0 indicates that the dentry is in a hard link + * group by itself, and therefore has a 'struct wim_inode' with i_nlink=1 to + * itself. + * + * This function also checks the dentries in each hard link group for + * consistency. In some WIMs, such as install.wim for some versions of Windows + * 7, dentries can share the same hard link group ID but not actually be hard + * linked to each other (based on conflicting information, such as file + * contents). This should be an error, but this case needs be handled. So, + * each "nominal" inode (the inode based on the inode numbers provided in the + * WIM) is examined for consistency and may be split into multiple "true" inodes + * that are maximally sized consistent sets of dentries. + * + * Return 0 on success; WIMLIB_ERR_NOMEM or WIMLIB_ERR_INVALID_DENTRY on + * failure. On success, the list of "true" inodes, linked by the i_hlist field, + * is returned in the hlist @inode_list. */ -int fix_link_groups(struct link_group_table *table) +int +dentry_tree_fix_inodes(struct wim_dentry *root, struct list_head *inode_list) { - for (u64 i = 0; i < table->capacity; i++) { - struct link_group *group = table->array[i]; - while (group) { - int ret; - ret = fix_nominal_link_group(group, &table->extra_groups); - if (ret != 0) - return ret; - group = group->next; + struct wim_inode_table inode_tab; + int ret; + bool ino_changes_needed; + + DEBUG("Inserting dentries into inode table"); + ret = init_inode_table(&inode_tab, 9001); + if (ret) + return ret; + + for_dentry_in_tree(root, inode_table_insert, &inode_tab); + + DEBUG("Cleaning up the hard link groups"); + ino_changes_needed = false; + ret = fix_inodes(&inode_tab, inode_list, &ino_changes_needed); + destroy_inode_table(&inode_tab); + + if (ret == 0 && ino_changes_needed) { + u64 cur_ino = 1; + struct wim_inode *inode; + + WARNING("Re-assigning inode numbers due to inode inconsistencies"); + list_for_each_entry(inode, inode_list, i_list) { + if (inode->i_nlink > 1) + inode->i_ino = cur_ino++; + else + inode->i_ino = 0; } } - return 0; + return ret; +} + +/* Assign consecutive inode numbers to the inodes in the inode table, and move + * the inodes to a single list @head. */ +void +inode_table_prepare_inode_list(struct wim_inode_table *table, + struct list_head *head) +{ + struct wim_inode *inode; + struct hlist_node *cur, *tmp; + u64 cur_ino = 1; + + INIT_LIST_HEAD(head); + for (size_t i = 0; i < table->capacity; i++) { + hlist_for_each_entry_safe(inode, cur, tmp, &table->array[i], i_hlist) + { + if (inode->i_nlink > 1) + inode->i_ino = cur_ino++; + else + inode->i_ino = 0; + list_add_tail(&inode->i_list, head); + } + INIT_HLIST_HEAD(&table->array[i]); + } + table->num_entries = 0; }