X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fhardlink.c;h=5c2f20c4fc6720ef02d7fa522e7b24dfbcf59a0d;hp=a5d8b9fae05fae57f56dae0fab55262bcb9456ef;hb=63a5804943519281f206ca0fefacfe2b99fc9958;hpb=72cc9a31b518ca830fcc7a2ed74a20415d6e13ec diff --git a/src/hardlink.c b/src/hardlink.c index a5d8b9fa..5c2f20c4 100644 --- a/src/hardlink.c +++ b/src/hardlink.c @@ -1,149 +1,461 @@ +/* + * hardlink.c + * + * Code to deal with hard links in WIMs. Essentially, the WIM dentries are put + * into a hash table indexed by the inode ID field, then for each hard + * inode, a linked list is made to connect the dentries. + */ + +/* + * Copyright (C) 2012 Eric Biggers + * + * This file is part of wimlib, a library for working with WIM files. + * + * wimlib is free software; you can redistribute it and/or modify it under the + * terms of the GNU General Public License as published by the Free + * Software Foundation; either version 3 of the License, or (at your option) + * any later version. + * + * wimlib 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 General Public License for more + * details. + * + * You should have received a copy of the GNU General Public License + * along with wimlib; if not, see http://www.gnu.org/licenses/. + */ + #include "wimlib_internal.h" #include "dentry.h" #include "list.h" #include "lookup_table.h" -struct link_group { - u64 link_group_id; - struct link_group *next; - struct list_head dentry_list; -}; +/* NULL NULL + * ^ ^ + * dentry | | + * / \ ----------- ----------- + * | dentry<---| struct | | struct |---> dentry + * \ / |inode| |inode| + * dentry ------------ ------------ + * ^ ^ + * | | + * | | dentry + * ----------- ----------- / \ + * dentry<---| struct | | struct |---> dentry dentry + * / |inode| |inode| \ / + * dentry ------------ ------------ dentry + * ^ ^ + * | | + * ----------------- + * inode_table->array | idx 0 | idx 1 | + * ----------------- + */ + -struct link_group_table { - struct link_group **array; - u64 num_entries; - u64 capacity; -}; +int init_inode_table(struct inode_table *table, size_t capacity) +{ + 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; + INIT_HLIST_HEAD(&table->extra_inodes); + return 0; +} -#include -struct link_group_table *new_link_group_table(u64 capacity) +static size_t inode_link_count(const struct inode *inode) { - return (struct link_group_table*)new_lookup_table(capacity); + const struct list_head *cur; + size_t size = 0; + list_for_each(cur, &inode->dentry_list) + size++; + return size; } -/* Insert a dentry into the hard link group table based on its hard link group +/* + * Insert a dentry into the inode table based on its inode * ID. * - * If there is already a dentry in the table having the same hard link group ID, - * we link the dentries together in a circular list. + * If there is already a dentry in the table having the same inode ID, + * and the inode ID is not 0, the dentry is added to the circular + * linked list for that inode. * - * If the hard link group ID is 0, this is a no-op and the dentry is not - * inserted. + * If the inode ID is 0, this indicates a dentry that's in a hard link + * inode by itself (has a link count of 1). We can't insert it into the hash + * table itself because we don't know what inode numbers 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 inode + * numbers later. */ -int link_group_table_insert(struct dentry *dentry, struct link_group_table *table) +int inode_table_insert(struct dentry *dentry, void *__table) { - size_t pos; - struct link_group *group; + struct inode_table *table = __table; + struct inode *d_inode = dentry->inode; - if (dentry->hard_link == 0) { - INIT_LIST_HEAD(&dentry->link_group_list); - return 0; - } + if (d_inode->ino == 0) { + /* Single inode--- Add to the list of extra inodes (we can't put + * it in the table itself because all the singles have a link + * inode ID of 0) */ + hlist_add_head(&d_inode->hlist, &table->extra_inodes); + + wimlib_assert(d_inode->dentry_list.next == &dentry->inode_dentry_list); + wimlib_assert(d_inode->dentry_list.prev == &dentry->inode_dentry_list); + wimlib_assert(d_inode->link_count == 1); + } else { + /* Inode that may have multiple corresponding dentries (the code + * will work even if the inode actually contains only 1 dentry + * though) */ - /* Try adding to existing hard link group */ - pos = dentry->hard_link % table->capacity; - group = table->array[pos]; - while (group) { - if (group->link_group_id == dentry->hard_link) { - list_add(&dentry->link_group_list, &group->dentry_list); - return 0; + size_t pos; + struct inode *inode; + struct hlist_node *cur; + + /* Try adding to existing inode */ + pos = d_inode->ino % table->capacity; + hlist_for_each_entry(inode, cur, &table->array[pos], hlist) { + if (inode->ino == d_inode->ino) { + list_add(&dentry->inode_dentry_list, + &inode->dentry_list); + inode->link_count++; + return 0; + } } - group = group->next; - } - /* Add new hard link group to the table */ + /* Add new inode to the table */ + hlist_add_head(&d_inode->hlist, &table->array[pos]); - group = MALLOC(sizeof(struct link_group)); - if (!group) - return WIMLIB_ERR_NOMEM; - group->link_group_id = dentry->hard_link; - group->next = table->array[pos]; - INIT_LIST_HEAD(&group->dentry_list); - list_add(&dentry->link_group_list, &group->dentry_list); - table->array[pos] = group; - - /* XXX Make the table grow when too many entries have been inserted. */ - table->num_entries++; + wimlib_assert(d_inode->dentry_list.next == &dentry->inode_dentry_list); + wimlib_assert(d_inode->dentry_list.prev == &dentry->inode_dentry_list); + wimlib_assert(d_inode->link_count == 1); + + /* XXX Make the table grow when too many entries have been + * inserted. */ + table->num_entries++; + } return 0; } -/* Frees a link group table. */ -void free_link_group_table(struct link_group_table *table) +/* Assign the inode numbers to dentries in a inode table, and return the + * next available inode ID. */ +u64 assign_inode_numbers(struct hlist_head *inode_list) +{ + DEBUG("Assigning inode numbers"); + struct inode *inode; + struct hlist_node *cur; + u64 cur_ino = 1; + hlist_for_each_entry(inode, cur, inode_list, hlist) { + inode->ino = cur_ino; + cur_ino++; + } + return cur_ino; +} + + +static void print_inode_dentries(const struct inode *inode) +{ + struct dentry *dentry; + list_for_each_entry(dentry, &inode->dentry_list, inode_dentry_list) + printf("`%s'\n", dentry->full_path_utf8); +} + +static void inconsistent_inode(const struct inode *inode) +{ + ERROR("An inconsistent hard link group that we cannot correct has been " + "detected"); + ERROR("The dentries are located at the following paths:"); + print_inode_dentries(inode); +} + +static bool ref_inodes_consistent(const struct inode * restrict ref_inode_1, + const struct inode * restrict ref_inode_2) +{ + wimlib_assert(ref_inode_1 != ref_inode_2); + + if (ref_inode_1->num_ads != ref_inode_2->num_ads) + return false; + if (ref_inode_1->security_id != ref_inode_2->security_id + || ref_inode_1->attributes != ref_inode_2->attributes) + return false; + for (unsigned i = 0; i <= ref_inode_1->num_ads; i++) { + const u8 *ref_1_hash, *ref_2_hash; + 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_inode_1->ads_entries[i - 1], + ref_inode_2->ads_entries[i - 1])) + return false; + + } + return true; +} + +static bool inodes_consistent(const struct inode * restrict ref_inode, + const struct inode * restrict inode) +{ + wimlib_assert(ref_inode != inode); + + if (ref_inode->num_ads != inode->num_ads && + inode->num_ads != 0) + return false; + if (ref_inode->security_id != inode->security_id + || ref_inode->attributes != inode->attributes) + return false; + for (unsigned i = 0; i <= min(ref_inode->num_ads, inode->num_ads); i++) { + const u8 *ref_hash, *hash; + 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_inode->ads_entries[i - 1], + inode->ads_entries[i - 1])) + return false; + } + return true; +} + +#ifdef ENABLE_DEBUG +static void +print_dentry_list(const struct dentry *first_dentry) { - if (!table) - return; - if (table->array) { - for (u64 i = 0; i < table->capacity; i++) { - struct link_group *group = table->array[i]; - struct link_group *next; - while (group) { - next = group->next; - FREE(group); - group = next; + const struct dentry *dentry = first_dentry; + do { + printf("`%s'\n", dentry->full_path_utf8); + } while ((dentry = container_of(dentry->inode_dentry_list.next, + struct dentry, + inode_dentry_list)) != first_dentry); +} + +#endif + + +/* Fix up a "true" inode and check for inconsistencies */ +static int fix_true_inode(struct inode *inode) +{ + struct dentry *dentry; + struct dentry *ref_dentry = NULL; + struct inode *ref_inode; + u64 last_ctime = 0; + u64 last_mtime = 0; + u64 last_atime = 0; + bool found_short_name = false; + + inode_for_each_dentry(dentry, inode) { + if (!ref_dentry || ref_dentry->inode->num_ads == 0) + ref_dentry = dentry; + if (dentry->short_name_len) { + if (found_short_name) { + ERROR("Multiple short names in hard link " + "group!"); + inconsistent_inode(inode); + return WIMLIB_ERR_INVALID_DENTRY; + } else { + found_short_name = true; + } + } + if (dentry->inode->creation_time > last_ctime) + last_ctime = dentry->inode->creation_time; + if (dentry->inode->last_write_time > last_mtime) + last_mtime = dentry->inode->last_write_time; + if (dentry->inode->last_access_time > last_atime) + last_atime = dentry->inode->last_access_time; + } + + ref_inode = ref_dentry->inode; + ref_inode->link_count = 1; + + inode_for_each_dentry(dentry, inode) { + if (dentry != ref_dentry) { + if (!inodes_consistent(ref_inode, dentry->inode)) { + inconsistent_inode(dentry->inode); + return WIMLIB_ERR_INVALID_DENTRY; } + /* Free the unneeded `struct inode'. */ + free_inode(dentry->inode); + dentry->inode = ref_inode; + ref_inode->link_count++; } - FREE(table->array); } - FREE(table); + ref_inode->creation_time = last_ctime; + ref_inode->last_write_time = last_mtime; + ref_inode->last_access_time = last_atime; + wimlib_assert(inode_link_count(ref_inode) == ref_inode->link_count); + return 0; } -/* Assign the link group IDs to dentries in a link group table, and return the - * next available link group ID. */ -u64 assign_link_groups(struct link_group_table *table) +/* + * Fixes up a nominal inode. + * + * By a nominal inode we mean a group of two or more dentries that share + * the same hard link group ID. + * + * If dentries in the inode are found to be inconsistent, we may split the inode + * into several "true" inodes. @new_inodes points to a linked list of + * these split inodes, and if we create any, they will be added to this list. + * + * After splitting up each nominal inode into the "true" inodes we + * will canonicalize the link group by getting rid of all the superfluous + * `struct inodes'. There will be just one `struct inode' for each hard link + * group remaining. + */ +static int +fix_nominal_inode(struct inode *inode, struct hlist_head *inode_list) { - struct link_group *remaining_groups = NULL; - u64 id = 1; - for (u64 i = 0; i < table->capacity; i++) { - struct link_group *group = table->array[i]; - struct link_group *next_group; - struct dentry *dentry; - while (group) { - next_group = group->next; - if (list_is_singular(&group->dentry_list)) { - /* Hard link group of size 1. Change the hard - * link ID to 0 and discard the link_group */ - dentry = container_of(group->dentry_list.next, - struct dentry, - link_group_list); - dentry->hard_link = 0; - FREE(group); - } else { - /* Hard link group of size > 1. Assign the - * dentries in the group the next available hard - * link IDs and queue the group to be - * re-inserted into the table. */ - list_for_each_entry(dentry, &group->dentry_list, - link_group_list) - dentry->hard_link = id; - group->next = remaining_groups; - remaining_groups = group; - id++; + struct dentry *dentry, *ref_dentry; + struct hlist_node *cur, *tmp; + int ret; + size_t num_true_inodes; + + wimlib_assert(inode->link_count == inode_link_count(inode)); + + LIST_HEAD(dentries_with_data_streams); + LIST_HEAD(dentries_with_no_data_streams); + HLIST_HEAD(true_inodes); + + /* 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. */ + inode_for_each_dentry(dentry, inode) { + for (unsigned i = 0; i <= dentry->inode->num_ads; i++) { + const u8 *hash; + hash = inode_stream_hash(dentry->inode, i); + if (!is_zero_hash(hash)) { + list_add(&dentry->tmp_list, + &dentries_with_data_streams); + goto next_dentry; } - group = next_group; } + list_add(&dentry->tmp_list, + &dentries_with_no_data_streams); + next_dentry: + ; } - memset(table->array, 0, table->capacity * sizeof(table->array[0])); - table->num_entries = 0; - while (remaining_groups) { - struct link_group *group = remaining_groups; - size_t pos = group->link_group_id % table->capacity; - table->num_entries++; - group->next = table->array[pos]; - table->array[pos] = group; - remaining_groups = remaining_groups->next; + /* If there are no dentries with data streams, we require the nominal + * inode to be a true inode */ + if (list_empty(&dentries_with_data_streams)) { + DEBUG("No data streams"); + #ifdef ENABLE_DEBUG + { + if (inode->link_count > 1) { + DEBUG("Found link group of size %zu without " + "any data streams:", inode->link_count); + print_inode_dentries(inode); + DEBUG("We are going to interpret it as true " + "link group, provided that the dentries " + "are consistent."); + } + } + #endif + hlist_add_head(&inode->hlist, inode_list); + return fix_true_inode(inode); + } + + /* 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 + * 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, make a new one. */ + hlist_for_each_entry(inode, cur, &true_inodes, hlist) { + if (ref_inodes_consistent(inode, dentry->inode)) { + list_add(&dentry->inode_dentry_list, + &inode->dentry_list); + goto next_dentry_2; + } + } + num_true_inodes++; + INIT_LIST_HEAD(&dentry->inode->dentry_list); + list_add(&dentry->inode_dentry_list, &dentry->inode->dentry_list); + hlist_add_head(&dentry->inode->hlist, &true_inodes); +next_dentry_2: + ; } - return id; + + wimlib_assert(num_true_inodes != 0); + + /* If there were dentries with no data streams, we require there to only + * 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_inodes != 1) { + ERROR("Hard inode ambiguity detected!"); + ERROR("We split up inode 0x%"PRIx64" due to " + "inconsistencies,", inode->ino); + ERROR("but dentries with no stream information remained. " + "We don't know which true hard link"); + ERROR("inode to assign them to."); + return WIMLIB_ERR_INVALID_DENTRY; + } + inode = container_of(true_inodes.first, + struct inode, + 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->inode_dentry_list, &inode->dentry_list); + } + if (num_true_inodes != 1) { + #ifdef ENABLE_DEBUG + { + inode = container_of(true_inodes.first, + struct inode, + hlist); + + printf("Split nominal inode 0x%"PRIx64" into %zu " + "inodes:\n", + inode->ino, num_true_inodes); + puts("------------------------------------------------------------------------------"); + size_t i = 1; + hlist_for_each_entry(inode, cur, &true_inodes, hlist) { + printf("[Split inode %zu]\n", i++); + print_inode_dentries(inode); + putchar('\n'); + } + puts("------------------------------------------------------------------------------"); + } + #endif + } + + hlist_for_each_entry_safe(inode, cur, tmp, &true_inodes, hlist) { + hlist_add_head(&inode->hlist, inode_list); + ret = fix_true_inode(inode); + if (ret != 0) + return ret; + } + return 0; } -#if 0 -/* Load a dentry tree into the link group table */ -int load_link_groups(struct link_group_table *table, struct dentry *root) +/* + * Goes through each hard link group (dentries sharing the same hard link group + * ID field) that's been inserted into the inode table and shars the `struct + * inode's among members of each hard link group. + * + * In the process, the dentries belonging to each inode are checked for + * consistency. If they contain data features that indicate they cannot really + * correspond to the same inode, this should be an error, but in reality this + * case needs to be handled, so we split the dentries into different inodes. + */ +int fix_inodes(struct inode_table *table, struct hlist_head *inode_list) { - int ret = for_dentry_in_tree(dentry, link_group_table_insert, table); - if (ret == 0) - assign_link_groups(table); - return ret; + struct inode *inode; + struct hlist_node *cur, *tmp; + int ret; + INIT_HLIST_HEAD(inode_list); + for (u64 i = 0; i < table->capacity; i++) { + hlist_for_each_entry_safe(inode, cur, tmp, &table->array[i], hlist) { + ret = fix_nominal_inode(inode, inode_list); + if (ret != 0) + return ret; + } + } + hlist_for_each_safe(cur, tmp, &table->extra_inodes) + hlist_add_head(cur, inode_list); + return 0; } -#endif