1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3 * Internal header file _only_ for device mapper core
4 *
5 * Copyright (C) 2016 Red Hat, Inc. All rights reserved.
6 *
7 * This file is released under the LGPL.
8 */
9
10 #ifndef DM_CORE_INTERNAL_H
11 #define DM_CORE_INTERNAL_H
12
13 #include <linux/kthread.h>
14 #include <linux/ktime.h>
15 #include <linux/blk-mq.h>
16 #include <linux/blk-crypto-profile.h>
17 #include <linux/jump_label.h>
18
19 #include <trace/events/block.h>
20
21 #include "dm.h"
22 #include "dm-ima.h"
23
24 #define DM_RESERVED_MAX_IOS 1024
25 #define DM_MAX_TARGETS 1048576
26 #define DM_MAX_TARGET_PARAMS 1024
27
28 struct dm_io;
29
30 struct dm_kobject_holder {
31 struct kobject kobj;
32 struct completion completion;
33 };
34
35 /*
36 * DM core internal structures used directly by dm.c, dm-rq.c and dm-table.c.
37 * DM targets must _not_ deference a mapped_device or dm_table to directly
38 * access their members!
39 */
40
41 /*
42 * For mempools pre-allocation at the table loading time.
43 */
44 struct dm_md_mempools {
45 struct bio_set bs;
46 struct bio_set io_bs;
47 };
48
49 struct mapped_device {
50 struct mutex suspend_lock;
51
52 struct mutex table_devices_lock;
53 struct list_head table_devices;
54
55 /*
56 * The current mapping (struct dm_table *).
57 * Use dm_get_live_table{_fast} or take suspend_lock for
58 * dereference.
59 */
60 void __rcu *map;
61
62 unsigned long flags;
63
64 /* Protect queue and type against concurrent access. */
65 struct mutex type_lock;
66 enum dm_queue_mode type;
67
68 int numa_node_id;
69 struct request_queue *queue;
70
71 atomic_t holders;
72 atomic_t open_count;
73
74 struct dm_target *immutable_target;
75 struct target_type *immutable_target_type;
76
77 char name[16];
78 struct gendisk *disk;
79 struct dax_device *dax_dev;
80
81 wait_queue_head_t wait;
82 unsigned long __percpu *pending_io;
83
84 /* forced geometry settings */
85 struct hd_geometry geometry;
86
87 /*
88 * Processing queue (flush)
89 */
90 struct workqueue_struct *wq;
91
92 /*
93 * A list of ios that arrived while we were suspended.
94 */
95 struct work_struct work;
96 spinlock_t deferred_lock;
97 struct bio_list deferred;
98
99 /*
100 * requeue work context is needed for cloning one new bio
101 * to represent the dm_io to be requeued, since each
102 * dm_io may point to the original bio from FS.
103 */
104 struct work_struct requeue_work;
105 struct dm_io *requeue_list;
106
107 void *interface_ptr;
108
109 /*
110 * Event handling.
111 */
112 wait_queue_head_t eventq;
113 atomic_t event_nr;
114 atomic_t uevent_seq;
115 struct list_head uevent_list;
116 spinlock_t uevent_lock; /* Protect access to uevent_list */
117
118 /* for blk-mq request-based DM support */
119 bool init_tio_pdu:1;
120 struct blk_mq_tag_set *tag_set;
121
122 struct dm_stats stats;
123
124 /* the number of internal suspends */
125 unsigned int internal_suspend_count;
126
127 int swap_bios;
128 struct semaphore swap_bios_semaphore;
129 struct mutex swap_bios_lock;
130
131 /*
132 * io objects are allocated from here.
133 */
134 struct dm_md_mempools *mempools;
135
136 /* kobject and completion */
137 struct dm_kobject_holder kobj_holder;
138
139 struct srcu_struct io_barrier;
140
141 #ifdef CONFIG_BLK_DEV_ZONED
142 void *zone_revalidate_map;
143 struct task_struct *revalidate_map_task;
144 #endif
145
146 #ifdef CONFIG_IMA
147 struct dm_ima_measurements ima;
148 #endif
149 };
150
151 /*
152 * Bits for the flags field of struct mapped_device.
153 */
154 #define DMF_BLOCK_IO_FOR_SUSPEND 0
155 #define DMF_SUSPENDED 1
156 #define DMF_FROZEN 2
157 #define DMF_FREEING 3
158 #define DMF_DELETING 4
159 #define DMF_NOFLUSH_SUSPENDING 5
160 #define DMF_DEFERRED_REMOVE 6
161 #define DMF_SUSPENDED_INTERNALLY 7
162 #define DMF_POST_SUSPENDING 8
163 #define DMF_EMULATE_ZONE_APPEND 9
164 #define DMF_QUEUE_STOPPED 10
165
dm_get_size(struct mapped_device * md)166 static inline sector_t dm_get_size(struct mapped_device *md)
167 {
168 return get_capacity(md->disk);
169 }
170
dm_get_stats(struct mapped_device * md)171 static inline struct dm_stats *dm_get_stats(struct mapped_device *md)
172 {
173 return &md->stats;
174 }
175
176 DECLARE_STATIC_KEY_FALSE(stats_enabled);
177 DECLARE_STATIC_KEY_FALSE(swap_bios_enabled);
178 DECLARE_STATIC_KEY_FALSE(zoned_enabled);
179
dm_emulate_zone_append(struct mapped_device * md)180 static inline bool dm_emulate_zone_append(struct mapped_device *md)
181 {
182 if (blk_queue_is_zoned(md->queue))
183 return test_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
184 return false;
185 }
186
187 #define DM_TABLE_MAX_DEPTH 16
188
189 struct dm_table {
190 struct mapped_device *md;
191 enum dm_queue_mode type;
192
193 /* btree table */
194 unsigned int depth;
195 unsigned int counts[DM_TABLE_MAX_DEPTH]; /* in nodes */
196 sector_t *index[DM_TABLE_MAX_DEPTH];
197
198 unsigned int num_targets;
199 unsigned int num_allocated;
200 sector_t *highs;
201 struct dm_target *targets;
202
203 struct target_type *immutable_target_type;
204
205 bool integrity_supported:1;
206 bool singleton:1;
207 /* set if all the targets in the table have "flush_bypasses_map" set */
208 bool flush_bypasses_map:1;
209
210 /*
211 * Indicates the rw permissions for the new logical device. This
212 * should be a combination of BLK_OPEN_READ and BLK_OPEN_WRITE.
213 */
214 blk_mode_t mode;
215
216 /* a list of devices used by this table */
217 struct list_head devices;
218
219 /* events get handed up using this callback */
220 void (*event_fn)(void *data);
221 void *event_context;
222
223 struct dm_md_mempools *mempools;
224
225 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
226 struct blk_crypto_profile *crypto_profile;
227 #endif
228 };
229
dm_table_get_target(struct dm_table * t,unsigned int index)230 static inline struct dm_target *dm_table_get_target(struct dm_table *t,
231 unsigned int index)
232 {
233 BUG_ON(index >= t->num_targets);
234 return t->targets + index;
235 }
236
237 /*
238 * One of these is allocated per clone bio.
239 */
240 #define DM_TIO_MAGIC 28714
241 struct dm_target_io {
242 unsigned short magic;
243 blk_short_t flags;
244 unsigned int target_bio_nr;
245 struct dm_io *io;
246 struct dm_target *ti;
247 unsigned int *len_ptr;
248 sector_t old_sector;
249 struct bio clone;
250 };
251 #define DM_TARGET_IO_BIO_OFFSET (offsetof(struct dm_target_io, clone))
252 #define DM_IO_BIO_OFFSET \
253 (offsetof(struct dm_target_io, clone) + offsetof(struct dm_io, tio))
254
255 /*
256 * dm_target_io flags
257 */
258 enum {
259 DM_TIO_INSIDE_DM_IO,
260 DM_TIO_IS_DUPLICATE_BIO
261 };
262
dm_tio_flagged(struct dm_target_io * tio,unsigned int bit)263 static inline bool dm_tio_flagged(struct dm_target_io *tio, unsigned int bit)
264 {
265 return (tio->flags & (1U << bit)) != 0;
266 }
267
dm_tio_set_flag(struct dm_target_io * tio,unsigned int bit)268 static inline void dm_tio_set_flag(struct dm_target_io *tio, unsigned int bit)
269 {
270 tio->flags |= (1U << bit);
271 }
272
dm_tio_is_normal(struct dm_target_io * tio)273 static inline bool dm_tio_is_normal(struct dm_target_io *tio)
274 {
275 return (dm_tio_flagged(tio, DM_TIO_INSIDE_DM_IO) &&
276 !dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
277 }
278
279 /*
280 * One of these is allocated per original bio.
281 * It contains the first clone used for that original.
282 */
283 #define DM_IO_MAGIC 19577
284 struct dm_io {
285 unsigned short magic;
286 blk_short_t flags;
287 spinlock_t lock;
288 unsigned long start_time;
289 void *data;
290 struct dm_io *next;
291 struct dm_stats_aux stats_aux;
292 blk_status_t status;
293 bool requeue_flush_with_data;
294 atomic_t io_count;
295 struct mapped_device *md;
296
297 /* The three fields represent mapped part of original bio */
298 struct bio *orig_bio;
299 unsigned int sector_offset; /* offset to end of orig_bio */
300 unsigned int sectors;
301
302 /* last member of dm_target_io is 'struct bio' */
303 struct dm_target_io tio;
304 };
305
306 /*
307 * dm_io flags
308 */
309 enum {
310 DM_IO_ACCOUNTED,
311 DM_IO_WAS_SPLIT,
312 DM_IO_BLK_STAT
313 };
314
dm_io_flagged(struct dm_io * io,unsigned int bit)315 static inline bool dm_io_flagged(struct dm_io *io, unsigned int bit)
316 {
317 return (io->flags & (1U << bit)) != 0;
318 }
319
dm_io_set_flag(struct dm_io * io,unsigned int bit)320 static inline void dm_io_set_flag(struct dm_io *io, unsigned int bit)
321 {
322 io->flags |= (1U << bit);
323 }
324
325 void dm_io_rewind(struct dm_io *io, struct bio_set *bs);
326
dm_get_completion_from_kobject(struct kobject * kobj)327 static inline struct completion *dm_get_completion_from_kobject(struct kobject *kobj)
328 {
329 return &container_of(kobj, struct dm_kobject_holder, kobj)->completion;
330 }
331
332 unsigned int __dm_get_module_param(unsigned int *module_param, unsigned int def, unsigned int max);
333
dm_message_test_buffer_overflow(char * result,unsigned int maxlen)334 static inline bool dm_message_test_buffer_overflow(char *result, unsigned int maxlen)
335 {
336 return !maxlen || strlen(result) + 1 >= maxlen;
337 }
338
339 extern atomic_t dm_global_event_nr;
340 extern wait_queue_head_t dm_global_eventq;
341 void dm_issue_global_event(void);
342
343 #endif
344