1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * alloc.h
6 *
7 * Function prototypes
8 *
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 */
11
12 #ifndef OCFS2_ALLOC_H
13 #define OCFS2_ALLOC_H
14
15
16 /*
17 * For xattr tree leaf, we limit the leaf byte size to be 64K.
18 */
19 #define OCFS2_MAX_XATTR_TREE_LEAF_SIZE 65536
20
21 /*
22 * ocfs2_extent_tree and ocfs2_extent_tree_operations are used to abstract
23 * the b-tree operations in ocfs2. Now all the b-tree operations are not
24 * limited to ocfs2_dinode only. Any data which need to allocate clusters
25 * to store can use b-tree. And it only needs to implement its ocfs2_extent_tree
26 * and operation.
27 *
28 * ocfs2_extent_tree becomes the first-class object for extent tree
29 * manipulation. Callers of the alloc.c code need to fill it via one of
30 * the ocfs2_init_*_extent_tree() operations below.
31 *
32 * ocfs2_extent_tree contains info for the root of the b-tree, it must have a
33 * root ocfs2_extent_list and a root_bh so that they can be used in the b-tree
34 * functions. It needs the ocfs2_caching_info structure associated with
35 * I/O on the tree. With metadata ecc, we now call different journal_access
36 * functions for each type of metadata, so it must have the
37 * root_journal_access function.
38 * ocfs2_extent_tree_operations abstract the normal operations we do for
39 * the root of extent b-tree.
40 */
41 struct ocfs2_extent_tree_operations;
42 struct ocfs2_extent_tree {
43 const struct ocfs2_extent_tree_operations *et_ops;
44 struct buffer_head *et_root_bh;
45 struct ocfs2_extent_list *et_root_el;
46 struct ocfs2_caching_info *et_ci;
47 ocfs2_journal_access_func et_root_journal_access;
48 void *et_object;
49 unsigned int et_max_leaf_clusters;
50 struct ocfs2_cached_dealloc_ctxt *et_dealloc;
51 };
52
53 /*
54 * ocfs2_init_*_extent_tree() will fill an ocfs2_extent_tree from the
55 * specified object buffer.
56 */
57 void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
58 struct ocfs2_caching_info *ci,
59 struct buffer_head *bh);
60 void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
61 struct ocfs2_caching_info *ci,
62 struct buffer_head *bh);
63 struct ocfs2_xattr_value_buf;
64 void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
65 struct ocfs2_caching_info *ci,
66 struct ocfs2_xattr_value_buf *vb);
67 void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
68 struct ocfs2_caching_info *ci,
69 struct buffer_head *bh);
70 void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et,
71 struct ocfs2_caching_info *ci,
72 struct buffer_head *bh);
73
74 /*
75 * Read an extent block into *bh. If *bh is NULL, a bh will be
76 * allocated. This is a cached read. The extent block will be validated
77 * with ocfs2_validate_extent_block().
78 */
79 int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
80 struct buffer_head **bh);
81
82 struct ocfs2_alloc_context;
83 int ocfs2_insert_extent(handle_t *handle,
84 struct ocfs2_extent_tree *et,
85 u32 cpos,
86 u64 start_blk,
87 u32 new_clusters,
88 u8 flags,
89 struct ocfs2_alloc_context *meta_ac);
90
91 enum ocfs2_alloc_restarted {
92 RESTART_NONE = 0,
93 RESTART_TRANS,
94 RESTART_META
95 };
96 int ocfs2_add_clusters_in_btree(handle_t *handle,
97 struct ocfs2_extent_tree *et,
98 u32 *logical_offset,
99 u32 clusters_to_add,
100 int mark_unwritten,
101 struct ocfs2_alloc_context *data_ac,
102 struct ocfs2_alloc_context *meta_ac,
103 enum ocfs2_alloc_restarted *reason_ret);
104 struct ocfs2_cached_dealloc_ctxt;
105 struct ocfs2_path;
106 int ocfs2_split_extent(handle_t *handle,
107 struct ocfs2_extent_tree *et,
108 struct ocfs2_path *path,
109 int split_index,
110 struct ocfs2_extent_rec *split_rec,
111 struct ocfs2_alloc_context *meta_ac,
112 struct ocfs2_cached_dealloc_ctxt *dealloc);
113 int ocfs2_mark_extent_written(struct inode *inode,
114 struct ocfs2_extent_tree *et,
115 handle_t *handle, u32 cpos, u32 len, u32 phys,
116 struct ocfs2_alloc_context *meta_ac,
117 struct ocfs2_cached_dealloc_ctxt *dealloc);
118 int ocfs2_change_extent_flag(handle_t *handle,
119 struct ocfs2_extent_tree *et,
120 u32 cpos, u32 len, u32 phys,
121 struct ocfs2_alloc_context *meta_ac,
122 struct ocfs2_cached_dealloc_ctxt *dealloc,
123 int new_flags, int clear_flags);
124 int ocfs2_remove_extent(handle_t *handle, struct ocfs2_extent_tree *et,
125 u32 cpos, u32 len,
126 struct ocfs2_alloc_context *meta_ac,
127 struct ocfs2_cached_dealloc_ctxt *dealloc);
128 int ocfs2_remove_btree_range(struct inode *inode,
129 struct ocfs2_extent_tree *et,
130 u32 cpos, u32 phys_cpos, u32 len, int flags,
131 struct ocfs2_cached_dealloc_ctxt *dealloc,
132 u64 refcount_loc, bool refcount_tree_locked);
133
134 int ocfs2_num_free_extents(struct ocfs2_extent_tree *et);
135
136 /*
137 * how many new metadata chunks would an allocation need at maximum?
138 *
139 * Please note that the caller must make sure that root_el is the root
140 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
141 * the result may be wrong.
142 */
ocfs2_extend_meta_needed(struct ocfs2_extent_list * root_el)143 static inline int ocfs2_extend_meta_needed(struct ocfs2_extent_list *root_el)
144 {
145 /*
146 * Rather than do all the work of determining how much we need
147 * (involves a ton of reads and locks), just ask for the
148 * maximal limit. That's a tree depth shift. So, one block for
149 * level of the tree (current l_tree_depth), one block for the
150 * new tree_depth==0 extent_block, and one block at the new
151 * top-of-the tree.
152 */
153 return le16_to_cpu(root_el->l_tree_depth) + 2;
154 }
155
156 void ocfs2_dinode_new_extent_list(struct inode *inode, struct ocfs2_dinode *di);
157 void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di);
158 int ocfs2_convert_inline_data_to_extents(struct inode *inode,
159 struct buffer_head *di_bh);
160
161 int ocfs2_truncate_log_init(struct ocfs2_super *osb);
162 void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb);
163 void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
164 int cancel);
165 int ocfs2_flush_truncate_log(struct ocfs2_super *osb);
166 int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
167 int slot_num,
168 struct ocfs2_dinode **tl_copy);
169 int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
170 struct ocfs2_dinode *tl_copy);
171 int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb);
172 int ocfs2_truncate_log_append(struct ocfs2_super *osb,
173 handle_t *handle,
174 u64 start_blk,
175 unsigned int num_clusters);
176 int __ocfs2_flush_truncate_log(struct ocfs2_super *osb);
177 int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
178 unsigned int needed);
179
180 /*
181 * Process local structure which describes the block unlinks done
182 * during an operation. This is populated via
183 * ocfs2_cache_block_dealloc().
184 *
185 * ocfs2_run_deallocs() should be called after the potentially
186 * de-allocating routines. No journal handles should be open, and most
187 * locks should have been dropped.
188 */
189 struct ocfs2_cached_dealloc_ctxt {
190 struct ocfs2_per_slot_free_list *c_first_suballocator;
191 struct ocfs2_cached_block_free *c_global_allocator;
192 };
ocfs2_init_dealloc_ctxt(struct ocfs2_cached_dealloc_ctxt * c)193 static inline void ocfs2_init_dealloc_ctxt(struct ocfs2_cached_dealloc_ctxt *c)
194 {
195 c->c_first_suballocator = NULL;
196 c->c_global_allocator = NULL;
197 }
198 int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
199 u64 blkno, unsigned int bit);
200 int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
201 int type, int slot, u64 suballoc, u64 blkno,
202 unsigned int bit);
ocfs2_dealloc_has_cluster(struct ocfs2_cached_dealloc_ctxt * c)203 static inline int ocfs2_dealloc_has_cluster(struct ocfs2_cached_dealloc_ctxt *c)
204 {
205 return c->c_global_allocator != NULL;
206 }
207 int ocfs2_run_deallocs(struct ocfs2_super *osb,
208 struct ocfs2_cached_dealloc_ctxt *ctxt);
209
210 struct ocfs2_truncate_context {
211 struct ocfs2_cached_dealloc_ctxt tc_dealloc;
212 int tc_ext_alloc_locked; /* is it cluster locked? */
213 /* these get destroyed once it's passed to ocfs2_commit_truncate. */
214 struct buffer_head *tc_last_eb_bh;
215 };
216
217 int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
218 u64 range_start, u64 range_end);
219 int ocfs2_commit_truncate(struct ocfs2_super *osb,
220 struct inode *inode,
221 struct buffer_head *di_bh);
222 int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
223 unsigned int start, unsigned int end, int trunc);
224
225 int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
226 struct ocfs2_extent_list *root_el, u32 cpos,
227 struct buffer_head **leaf_bh);
228 int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster);
229
230 int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range);
231 /*
232 * Helper function to look at the # of clusters in an extent record.
233 */
ocfs2_rec_clusters(struct ocfs2_extent_list * el,struct ocfs2_extent_rec * rec)234 static inline unsigned int ocfs2_rec_clusters(struct ocfs2_extent_list *el,
235 struct ocfs2_extent_rec *rec)
236 {
237 /*
238 * Cluster count in extent records is slightly different
239 * between interior nodes and leaf nodes. This is to support
240 * unwritten extents which need a flags field in leaf node
241 * records, thus shrinking the available space for a clusters
242 * field.
243 */
244 if (el->l_tree_depth)
245 return le32_to_cpu(rec->e_int_clusters);
246 else
247 return le16_to_cpu(rec->e_leaf_clusters);
248 }
249
250 /*
251 * This is only valid for leaf nodes, which are the only ones that can
252 * have empty extents anyway.
253 */
ocfs2_is_empty_extent(struct ocfs2_extent_rec * rec)254 static inline int ocfs2_is_empty_extent(struct ocfs2_extent_rec *rec)
255 {
256 return !rec->e_leaf_clusters;
257 }
258
259 int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
260 struct page **pages, int *num);
261 void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
262 unsigned int from, unsigned int to,
263 struct page *page, int zero, u64 *phys);
264 /*
265 * Structures which describe a path through a btree, and functions to
266 * manipulate them.
267 *
268 * The idea here is to be as generic as possible with the tree
269 * manipulation code.
270 */
271 struct ocfs2_path_item {
272 struct buffer_head *bh;
273 struct ocfs2_extent_list *el;
274 };
275
276 #define OCFS2_MAX_PATH_DEPTH 5
277
278 struct ocfs2_path {
279 int p_tree_depth;
280 ocfs2_journal_access_func p_root_access;
281 struct ocfs2_path_item p_node[OCFS2_MAX_PATH_DEPTH];
282 };
283
284 #define path_root_bh(_path) ((_path)->p_node[0].bh)
285 #define path_root_el(_path) ((_path)->p_node[0].el)
286 #define path_root_access(_path)((_path)->p_root_access)
287 #define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh)
288 #define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el)
289 #define path_num_items(_path) ((_path)->p_tree_depth + 1)
290
291 void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root);
292 void ocfs2_free_path(struct ocfs2_path *path);
293 int ocfs2_find_path(struct ocfs2_caching_info *ci,
294 struct ocfs2_path *path,
295 u32 cpos);
296 struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path);
297 struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et);
298 int ocfs2_path_bh_journal_access(handle_t *handle,
299 struct ocfs2_caching_info *ci,
300 struct ocfs2_path *path,
301 int idx);
302 int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
303 handle_t *handle,
304 struct ocfs2_path *path);
305 int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
306 struct ocfs2_path *path, u32 *cpos);
307 int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
308 struct ocfs2_path *path, u32 *cpos);
309 int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
310 struct ocfs2_path *left,
311 struct ocfs2_path *right);
312 #endif /* OCFS2_ALLOC_H */
313