X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fhardlink.c;h=93030d8f7e10e6b89874a089c28da575c63e850d;hp=26c9e05d5e50b343c10944c74175ad048316e387;hb=266d03613339dbe9a433c9849b6b4c47e0090dc8;hpb=eb066ffe5f37eae975c285cbbfc377247b0d60ff diff --git a/src/hardlink.c b/src/hardlink.c index 26c9e05d..93030d8f 100644 --- a/src/hardlink.c +++ b/src/hardlink.c @@ -32,27 +32,53 @@ /* NULL NULL * ^ ^ - * dentry | | - * / \ ----------- ----------- + * dentry | | + * / \ ----------- ----------- * | dentry<---| struct | | struct |---> dentry - * \ / |inode| |inode| + * \ / | inode | | inode | * dentry ------------ ------------ * ^ ^ * | | * | | dentry * ----------- ----------- / \ * dentry<---| struct | | struct |---> dentry dentry - * / |inode| |inode| \ / + * / | inode | | inode | \ / * dentry ------------ ------------ dentry * ^ ^ * | | * ----------------- - * inode_table->array | idx 0 | idx 1 | + * inode_table->array | idx 0 | idx 1 | * ----------------- */ +/* Hash table to find inodes, identified by their inode ID. + * */ +struct inode_table { + /* Fields for the hash table */ + struct hlist_head *array; + u64 num_entries; + u64 capacity; + + /* + * Linked list of "extra" inodes. These may be: + * + * - inodes with link count 1, which are all allowed to have 0 for their + * inode number, meaning we cannot insert them into the hash table + * before calling assign_inode_numbers(). + * + * - Groups we create ourselves by splitting a nominal inode due to + * inconsistencies in the dentries. These inodes will share a inode + * ID with some other inode until assign_inode_numbers() is called. + */ + struct hlist_head extra_inodes; +}; + +static inline void destroy_inode_table(struct inode_table *table) +{ + FREE(table->array); +} -int init_inode_table(struct inode_table *table, size_t capacity) +static int init_inode_table(struct inode_table *table, size_t capacity) { table->array = CALLOC(capacity, sizeof(table->array[0])); if (!table->array) { @@ -65,8 +91,7 @@ int init_inode_table(struct inode_table *table, size_t capacity) return 0; } - -static size_t inode_link_count(const struct inode *inode) +static inline size_t inode_link_count(const struct inode *inode) { const struct list_head *cur; size_t size = 0; @@ -75,13 +100,7 @@ static size_t inode_link_count(const struct inode *inode) return size; } -static struct dentry *inode_first_dentry(struct inode *inode) -{ - return container_of(inode->dentry_list.next, struct dentry, - inode_dentry_list); -} - -/* +/* * Insert a dentry into the inode table based on its inode * ID. * @@ -96,10 +115,10 @@ static struct dentry *inode_first_dentry(struct inode *inode) * we keep a linked list of the single dentries, and assign them inode * numbers later. */ -int inode_table_insert(struct dentry *dentry, void *__table) +static int inode_table_insert(struct dentry *dentry, void *__table) { struct inode_table *table = __table; - struct inode *d_inode = dentry->inode; + struct inode *d_inode = dentry->d_inode; if (d_inode->ino == 0) { /* Single inode--- Add to the list of extra inodes (we can't put @@ -123,8 +142,7 @@ int inode_table_insert(struct dentry *dentry, void *__table) 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_add_dentry(dentry, inode); inode->link_count++; return 0; } @@ -148,6 +166,7 @@ int inode_table_insert(struct dentry *dentry, void *__table) * 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; @@ -159,20 +178,21 @@ u64 assign_inode_numbers(struct hlist_head *inode_list) } -static void -print_inode_dentries(const struct inode *inode) +static void print_inode_dentries(const struct inode *inode) { struct dentry *dentry; - list_for_each_entry(dentry, &inode->dentry_list, inode_dentry_list) + inode_for_each_dentry(dentry, inode) 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("An inconsistent hard link group that cannot be corrected has " + "been detected"); ERROR("The dentries are located at the following paths:"); +#ifdef ENABLE_ERROR_MESSAGES print_inode_dentries(inode); +#endif } static bool ref_inodes_consistent(const struct inode * restrict ref_inode_1, @@ -191,8 +211,8 @@ static bool ref_inodes_consistent(const struct inode * restrict ref_inode_1, 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])) + if (i && !ads_entries_have_same_name(&ref_inode_1->ads_entries[i - 1], + &ref_inode_2->ads_entries[i - 1])) return false; } @@ -216,30 +236,15 @@ static bool inodes_consistent(const struct inode * restrict ref_inode, 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])) + 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) -{ - 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) +static int fix_true_inode(struct inode *inode, struct hlist_head *inode_list) { struct dentry *dentry; struct dentry *ref_dentry = NULL; @@ -247,41 +252,36 @@ static int fix_true_inode(struct inode *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) + if (!ref_dentry || dentry->d_inode->num_ads > ref_dentry->d_inode->num_ads) 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; + if (dentry->d_inode->creation_time > last_ctime) + last_ctime = dentry->d_inode->creation_time; + if (dentry->d_inode->last_write_time > last_mtime) + last_mtime = dentry->d_inode->last_write_time; + if (dentry->d_inode->last_access_time > last_atime) + last_atime = dentry->d_inode->last_access_time; } - ref_inode = ref_dentry->inode; + ref_inode = ref_dentry->d_inode; ref_inode->link_count = 1; + hlist_add_head(&ref_inode->hlist, inode_list); - inode_for_each_dentry(dentry, inode) { + list_del(&inode->dentry_list); + list_add(&ref_inode->dentry_list, &ref_dentry->inode_dentry_list); + + inode_for_each_dentry(dentry, ref_inode) { if (dentry != ref_dentry) { - if (!inodes_consistent(ref_inode, dentry->inode)) { - inconsistent_inode(dentry->inode); + if (!inodes_consistent(ref_inode, dentry->d_inode)) { + inconsistent_inode(ref_inode); return WIMLIB_ERR_INVALID_DENTRY; } /* Free the unneeded `struct inode'. */ - free_inode(dentry->inode); - dentry->inode = ref_inode; + dentry->d_inode->hlist.next = NULL; + dentry->d_inode->hlist.pprev = NULL; + free_inode(dentry->d_inode); + dentry->d_inode = ref_inode; ref_inode->link_count++; } } @@ -292,25 +292,24 @@ static int fix_true_inode(struct inode *inode) return 0; } -/* +/* * 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. + * into several "true" inodes. * - * 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. + * After splitting up each nominal inode into the "true" inodes we will + * canonicalize the link group by getting rid of all the unnecessary `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) +static int fix_nominal_inode(struct inode *inode, + struct hlist_head *inode_list) { - struct dentry *dentry, *ref_dentry; + struct dentry *dentry; struct hlist_node *cur, *tmp; int ret; size_t num_true_inodes; @@ -325,9 +324,9 @@ fix_nominal_inode(struct inode *inode, struct hlist_head *inode_list) * 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++) { + for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) { const u8 *hash; - hash = inode_stream_hash(dentry->inode, i); + hash = inode_stream_hash(dentry->d_inode, i); if (!is_zero_hash(hash)) { list_add(&dentry->tmp_list, &dentries_with_data_streams); @@ -343,21 +342,17 @@ fix_nominal_inode(struct inode *inode, struct hlist_head *inode_list) /* 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."); - } + if (inode->link_count > 1) { + DEBUG("Found link group of size %u 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); + return fix_true_inode(inode, inode_list); } /* One or more dentries had data streams specified. We check each of @@ -365,21 +360,20 @@ fix_nominal_inode(struct inode *inode, struct hlist_head *inode_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, make a new one. */ + /* 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, hlist) { - if (ref_inodes_consistent(inode, dentry->inode)) { - list_add(&dentry->inode_dentry_list, - &inode->dentry_list); + if (ref_inodes_consistent(inode, dentry->d_inode)) { + inode_add_dentry(dentry, inode); 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); + INIT_LIST_HEAD(&dentry->d_inode->dentry_list); + inode_add_dentry(dentry, dentry->d_inode); + hlist_add_head(&dentry->d_inode->hlist, &true_inodes); next_dentry_2: ; } @@ -395,43 +389,36 @@ next_dentry_2: 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."); + "We don't know which inode"); + ERROR("to assign them to."); return WIMLIB_ERR_INVALID_DENTRY; } - inode = container_of(true_inodes.first, - struct inode, - hlist); + 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); + inode_add_dentry(dentry, inode); } + #ifdef ENABLE_DEBUG 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("------------------------------------------------------------------------------"); + 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'); } - #endif + 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); + ret = fix_true_inode(inode, inode_list); if (ret != 0) return ret; } @@ -439,15 +426,21 @@ next_dentry_2: } /* - * Goes through each inode and shares the inodes among members of a hard - * inode. + * Goes through each hard link group (dentries sharing the same hard link group + * ID field) that's been inserted into the inode table and shares 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. * - * In the process, the dentries in each inode are checked for consistency. - * If they contain data features that indicate they cannot really be in 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. + * After this function returns, the inodes are no longer in the inode table, and + * the inode table should be destroyed. A list of the inodes, including all + * split inodes as well as the inodes that were good before, is returned in the + * list @inode_list. */ -int fix_inodes(struct inode_table *table, struct hlist_head *inode_list) +static int fix_inodes(struct inode_table *table, struct hlist_head *inode_list) { struct inode *inode; struct hlist_node *cur, *tmp; @@ -464,3 +457,21 @@ int fix_inodes(struct inode_table *table, struct hlist_head *inode_list) hlist_add_head(cur, inode_list); return 0; } + +int dentry_tree_fix_inodes(struct dentry *root, struct hlist_head *inode_list) +{ + struct inode_table inode_tab; + int ret; + + DEBUG("Inserting dentries into inode table"); + ret = init_inode_table(&inode_tab, 9001); + if (ret != 0) + return ret; + + for_dentry_in_tree(root, inode_table_insert, &inode_tab); + + DEBUG("Cleaning up the hard link groups"); + ret = fix_inodes(&inode_tab, inode_list); + destroy_inode_table(&inode_tab); + return ret; +}