dm-core.h

  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
 28struct dm_io;
 29
 30struct 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 */
 44struct dm_md_mempools {
 45	struct bio_set bs;
 46	struct bio_set io_bs;
 47};
 48
 49struct 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
166static inline sector_t dm_get_size(struct mapped_device *md)
167{
168	return get_capacity(md->disk);
169}
170
171static inline struct dm_stats *dm_get_stats(struct mapped_device *md)
172{
173	return &md->stats;
174}
175
176DECLARE_STATIC_KEY_FALSE(stats_enabled);
177DECLARE_STATIC_KEY_FALSE(swap_bios_enabled);
178DECLARE_STATIC_KEY_FALSE(zoned_enabled);
179
180static 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
189struct 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	struct rw_semaphore devices_lock;
219
220	/* events get handed up using this callback */
221	void (*event_fn)(void *data);
222	void *event_context;
223
224	struct dm_md_mempools *mempools;
225
226#ifdef CONFIG_BLK_INLINE_ENCRYPTION
227	struct blk_crypto_profile *crypto_profile;
228#endif
229};
230
231static inline struct dm_target *dm_table_get_target(struct dm_table *t,
232						    unsigned int index)
233{
234	BUG_ON(index >= t->num_targets);
235	return t->targets + index;
236}
237
238/*
239 * One of these is allocated per clone bio.
240 */
241#define DM_TIO_MAGIC 28714
242struct dm_target_io {
243	unsigned short magic;
244	blk_short_t flags;
245	unsigned int target_bio_nr;
246	struct dm_io *io;
247	struct dm_target *ti;
248	unsigned int *len_ptr;
249	sector_t old_sector;
250	struct bio clone;
251};
252#define DM_TARGET_IO_BIO_OFFSET (offsetof(struct dm_target_io, clone))
253#define DM_IO_BIO_OFFSET \
254	(offsetof(struct dm_target_io, clone) + offsetof(struct dm_io, tio))
255
256/*
257 * dm_target_io flags
258 */
259enum {
260	DM_TIO_INSIDE_DM_IO,
261	DM_TIO_IS_DUPLICATE_BIO
262};
263
264static inline bool dm_tio_flagged(struct dm_target_io *tio, unsigned int bit)
265{
266	return (tio->flags & (1U << bit)) != 0;
267}
268
269static inline void dm_tio_set_flag(struct dm_target_io *tio, unsigned int bit)
270{
271	tio->flags |= (1U << bit);
272}
273
274static inline bool dm_tio_is_normal(struct dm_target_io *tio)
275{
276	return (dm_tio_flagged(tio, DM_TIO_INSIDE_DM_IO) &&
277		!dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
278}
279
280/*
281 * One of these is allocated per original bio.
282 * It contains the first clone used for that original.
283 */
284#define DM_IO_MAGIC 19577
285struct dm_io {
286	unsigned short magic;
287	blk_short_t flags;
288	spinlock_t lock;
289	unsigned long start_time;
290	void *data;
291	struct dm_io *next;
292	struct dm_stats_aux stats_aux;
293	blk_status_t status;
294	bool requeue_flush_with_data;
295	atomic_t io_count;
296	struct mapped_device *md;
297
298	/* The three fields represent mapped part of original bio */
299	struct bio *orig_bio;
300	unsigned int sector_offset; /* offset to end of orig_bio */
301	unsigned int sectors;
302
303	/* last member of dm_target_io is 'struct bio' */
304	struct dm_target_io tio;
305};
306
307/*
308 * dm_io flags
309 */
310enum {
311	DM_IO_ACCOUNTED,
312	DM_IO_WAS_SPLIT,
313	DM_IO_BLK_STAT
314};
315
316static inline bool dm_io_flagged(struct dm_io *io, unsigned int bit)
317{
318	return (io->flags & (1U << bit)) != 0;
319}
320
321static inline void dm_io_set_flag(struct dm_io *io, unsigned int bit)
322{
323	io->flags |= (1U << bit);
324}
325
326void dm_io_rewind(struct dm_io *io, struct bio_set *bs);
327
328static inline struct completion *dm_get_completion_from_kobject(struct kobject *kobj)
329{
330	return &container_of(kobj, struct dm_kobject_holder, kobj)->completion;
331}
332
333unsigned int __dm_get_module_param(unsigned int *module_param, unsigned int def, unsigned int max);
334
335static inline bool dm_message_test_buffer_overflow(char *result, unsigned int maxlen)
336{
337	return !maxlen || strlen(result) + 1 >= maxlen;
338}
339
340extern atomic_t dm_global_event_nr;
341extern wait_queue_head_t dm_global_eventq;
342void dm_issue_global_event(void);
343
344#endif