[linux-block.git] / fs / btrfs / raid56.h

/* SPDX-License-Identifier: GPL-2.0 */
/*
 * Copyright (C) 2012 Fusion-io  All rights reserved.
 * Copyright (C) 2012 Intel Corp. All rights reserved.
 */

#ifndef BTRFS_RAID56_H
#define BTRFS_RAID56_H

#include <linux/workqueue.h>
#include "volumes.h"

enum btrfs_rbio_ops {
	BTRFS_RBIO_WRITE,
	BTRFS_RBIO_READ_REBUILD,
	BTRFS_RBIO_PARITY_SCRUB,
	BTRFS_RBIO_REBUILD_MISSING,
};

struct btrfs_raid_bio {
	struct btrfs_io_context *bioc;

	/*
	 * While we're doing RMW on a stripe we put it into a hash table so we
	 * can lock the stripe and merge more rbios into it.
	 */
	struct list_head hash_list;

	/* LRU list for the stripe cache */
	struct list_head stripe_cache;

	/* For scheduling work in the helper threads */
	struct work_struct work;

	/*
	 * bio_list and bio_list_lock are used to add more bios into the stripe
	 * in hopes of avoiding the full RMW
	 */
	struct bio_list bio_list;
	spinlock_t bio_list_lock;

	/*
	 * Also protected by the bio_list_lock, the plug list is used by the
	 * plugging code to collect partial bios while plugged.  The stripe
	 * locking code also uses it to hand off the stripe lock to the next
	 * pending IO.
	 */
	struct list_head plug_list;

	/* Flags that tell us if it is safe to merge with this bio. */
	unsigned long flags;

	/*
	 * Set if we're doing a parity rebuild for a read from higher up, which
	 * is handled differently from a parity rebuild as part of RMW.
	 */
	enum btrfs_rbio_ops operation;

	/* How many pages there are for the full stripe including P/Q */
	u16 nr_pages;

	/* How many sectors there are for the full stripe including P/Q */
	u16 nr_sectors;

	/* Number of data stripes (no p/q) */
	u8 nr_data;

	/* Number of all stripes (including P/Q) */
	u8 real_stripes;

	/* How many pages there are for each stripe */
	u8 stripe_npages;

	/* How many sectors there are for each stripe */
	u8 stripe_nsectors;

	/* Stripe number that we're scrubbing  */
	u8 scrubp;

	/*
	 * Size of all the bios in the bio_list.  This helps us decide if the
	 * rbio maps to a full stripe or not.
	 */
	int bio_list_bytes;

	refcount_t refs;

	atomic_t stripes_pending;

	wait_queue_head_t io_wait;

	/* Bitmap to record which horizontal stripe has data */
	unsigned long dbitmap;

	/* Allocated with stripe_nsectors-many bits for finish_*() calls */
	unsigned long finish_pbitmap;

	/*
	 * These are two arrays of pointers.  We allocate the rbio big enough
	 * to hold them both and setup their locations when the rbio is
	 * allocated.
	 */

	/*
	 * Pointers to pages that we allocated for reading/writing stripes
	 * directly from the disk (including P/Q).
	 */
	struct page **stripe_pages;

	/* Pointers to the sectors in the bio_list, for faster lookup */
	struct sector_ptr *bio_sectors;

	/*
	 * For subpage support, we need to map each sector to above
	 * stripe_pages.
	 */
	struct sector_ptr *stripe_sectors;

	/* Allocated with real_stripes-many pointers for finish_*() calls */
	void **finish_pointers;

	/*
	 * The bitmap recording where IO errors happened.
	 * Each bit is corresponding to one sector in either bio_sectors[] or
	 * stripe_sectors[] array.
	 *
	 * The reason we don't use another bit in sector_ptr is, we have two
	 * arrays of sectors, and a lot of IO can use sectors in both arrays.
	 * Thus making it much harder to iterate.
	 */
	unsigned long *error_bitmap;

	/*
	 * Checksum buffer if the rbio is for data.  The buffer should cover
	 * all data sectors (excluding P/Q sectors).
	 */
	u8 *csum_buf;

	/*
	 * Each bit represents if the corresponding sector has data csum found.
	 * Should only cover data sectors (excluding P/Q sectors).
	 */
	unsigned long *csum_bitmap;
};

/*
 * For trace event usage only. Records useful debug info for each bio submitted
 * by RAID56 to each physical device.
 *
 * No matter signed or not, (-1) is always the one indicating we can not grab
 * the proper stripe number.
 */
struct raid56_bio_trace_info {
	u64 devid;

	/* The offset inside the stripe. (<= STRIPE_LEN) */
	u32 offset;

	/*
	 * Stripe number.
	 * 0 is the first data stripe, and nr_data for P stripe,
	 * nr_data + 1 for Q stripe.
	 * >= real_stripes for
	 */
	u8 stripe_nr;
};

static inline int nr_data_stripes(const struct map_lookup *map)
{
	return map->num_stripes - btrfs_nr_parity_stripes(map->type);
}

static inline int nr_bioc_data_stripes(const struct btrfs_io_context *bioc)
{
	return bioc->num_stripes - btrfs_nr_parity_stripes(bioc->map_type);
}

#define RAID5_P_STRIPE ((u64)-2)
#define RAID6_Q_STRIPE ((u64)-1)

#define is_parity_stripe(x) (((x) == RAID5_P_STRIPE) ||		\
			     ((x) == RAID6_Q_STRIPE))

struct btrfs_device;

void raid56_parity_recover(struct bio *bio, struct btrfs_io_context *bioc,
			   int mirror_num);
void raid56_parity_write(struct bio *bio, struct btrfs_io_context *bioc);

struct btrfs_raid_bio *raid56_parity_alloc_scrub_rbio(struct bio *bio,
				struct btrfs_io_context *bioc,
				struct btrfs_device *scrub_dev,
				unsigned long *dbitmap, int stripe_nsectors);
void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio);

void raid56_parity_cache_data_pages(struct btrfs_raid_bio *rbio,
				    struct page **data_pages, u64 data_logical);

int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info);
void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info);

#endif
Commit	Line	Data
9888c340	1	/* SPDX-License-Identifier: GPL-2.0 */
53b381b3 DW	2	/*
	3	* Copyright (C) 2012 Fusion-io All rights reserved.
	4	* Copyright (C) 2012 Intel Corp. All rights reserved.
53b381b3 DW	5	*/
53b381b3 DW	6
9888c340 DS	7	#ifndef BTRFS_RAID56_H
	8	#define BTRFS_RAID56_H
	9
b8bea09a QW	10	#include <linux/workqueue.h>
	11	#include "volumes.h"
	12
	13	enum btrfs_rbio_ops {
	14	BTRFS_RBIO_WRITE,
	15	BTRFS_RBIO_READ_REBUILD,
	16	BTRFS_RBIO_PARITY_SCRUB,
	17	BTRFS_RBIO_REBUILD_MISSING,
	18	};
	19
	20	struct btrfs_raid_bio {
	21	struct btrfs_io_context *bioc;
	22
	23	/*
	24	* While we're doing RMW on a stripe we put it into a hash table so we
	25	* can lock the stripe and merge more rbios into it.
	26	*/
	27	struct list_head hash_list;
	28
	29	/* LRU list for the stripe cache */
	30	struct list_head stripe_cache;
	31
	32	/* For scheduling work in the helper threads */
	33	struct work_struct work;
	34
	35	/*
	36	* bio_list and bio_list_lock are used to add more bios into the stripe
	37	* in hopes of avoiding the full RMW
	38	*/
	39	struct bio_list bio_list;
	40	spinlock_t bio_list_lock;
	41
	42	/*
	43	* Also protected by the bio_list_lock, the plug list is used by the
	44	* plugging code to collect partial bios while plugged. The stripe
	45	* locking code also uses it to hand off the stripe lock to the next
	46	* pending IO.
	47	*/
	48	struct list_head plug_list;
	49
	50	/* Flags that tell us if it is safe to merge with this bio. */
	51	unsigned long flags;
	52
	53	/*
	54	* Set if we're doing a parity rebuild for a read from higher up, which
	55	* is handled differently from a parity rebuild as part of RMW.
	56	*/
	57	enum btrfs_rbio_ops operation;
	58
b8bea09a QW	59	/* How many pages there are for the full stripe including P/Q */
	60	u16 nr_pages;
	61
	62	/* How many sectors there are for the full stripe including P/Q */
	63	u16 nr_sectors;
	64
	65	/* Number of data stripes (no p/q) */
	66	u8 nr_data;
	67
67da05b3	68	/* Number of all stripes (including P/Q) */
b8bea09a QW	69	u8 real_stripes;
	70
	71	/* How many pages there are for each stripe */
	72	u8 stripe_npages;
	73
	74	/* How many sectors there are for each stripe */
	75	u8 stripe_nsectors;
	76
b8bea09a QW	77	/* Stripe number that we're scrubbing */
	78	u8 scrubp;
	79
	80	/*
	81	* Size of all the bios in the bio_list. This helps us decide if the
	82	* rbio maps to a full stripe or not.
	83	*/
	84	int bio_list_bytes;
	85
b8bea09a QW	86	refcount_t refs;
	87
	88	atomic_t stripes_pending;
	89
d817ce35 QW	90	wait_queue_head_t io_wait;
d817ce35 QW	91
b8bea09a QW	92	/* Bitmap to record which horizontal stripe has data */
	93	unsigned long dbitmap;
	94
	95	/* Allocated with stripe_nsectors-many bits for finish_() calls /
	96	unsigned long finish_pbitmap;
	97
	98	/*
	99	* These are two arrays of pointers. We allocate the rbio big enough
	100	* to hold them both and setup their locations when the rbio is
	101	* allocated.
	102	*/
	103
	104	/*
	105	* Pointers to pages that we allocated for reading/writing stripes
	106	* directly from the disk (including P/Q).
	107	*/
	108	struct page **stripe_pages;
	109
	110	/* Pointers to the sectors in the bio_list, for faster lookup */
	111	struct sector_ptr *bio_sectors;
	112
	113	/*
	114	* For subpage support, we need to map each sector to above
	115	* stripe_pages.
	116	*/
	117	struct sector_ptr *stripe_sectors;
	118
	119	/* Allocated with real_stripes-many pointers for finish_() calls /
	120	void **finish_pointers;
2942a50d QW	121
	122	/*
	123	* The bitmap recording where IO errors happened.
	124	* Each bit is corresponding to one sector in either bio_sectors[] or
	125	* stripe_sectors[] array.
	126	*
	127	* The reason we don't use another bit in sector_ptr is, we have two
	128	* arrays of sectors, and a lot of IO can use sectors in both arrays.
	129	* Thus making it much harder to iterate.
	130	*/
	131	unsigned long *error_bitmap;
c5a41562 QW	132
	133	/*
	134	* Checksum buffer if the rbio is for data. The buffer should cover
67da05b3	135	* all data sectors (excluding P/Q sectors).
c5a41562 QW	136	*/
	137	u8 *csum_buf;
	138
	139	/*
	140	* Each bit represents if the corresponding sector has data csum found.
	141	* Should only cover data sectors (excluding P/Q sectors).
	142	*/
	143	unsigned long *csum_bitmap;
b8bea09a QW	144	};
	145
	146	/*
	147	* For trace event usage only. Records useful debug info for each bio submitted
	148	* by RAID56 to each physical device.
	149	*
	150	* No matter signed or not, (-1) is always the one indicating we can not grab
	151	* the proper stripe number.
	152	*/
	153	struct raid56_bio_trace_info {
	154	u64 devid;
	155
	156	/* The offset inside the stripe. (<= STRIPE_LEN) */
	157	u32 offset;
	158
	159	/*
	160	* Stripe number.
	161	* 0 is the first data stripe, and nr_data for P stripe,
	162	* nr_data + 1 for Q stripe.
	163	* >= real_stripes for
	164	*/
	165	u8 stripe_nr;
	166	};
	167
72ad8131	168	static inline int nr_data_stripes(const struct map_lookup *map)
53b381b3	169	{
0b30f719	170	return map->num_stripes - btrfs_nr_parity_stripes(map->type);
53b381b3	171	}
b8bea09a	172
4886ff7b QW	173	static inline int nr_bioc_data_stripes(const struct btrfs_io_context *bioc)
	174	{
	175	return bioc->num_stripes - btrfs_nr_parity_stripes(bioc->map_type);
	176	}
	177
53b381b3 DW	178	#define RAID5_P_STRIPE ((u64)-2)
	179	#define RAID6_Q_STRIPE ((u64)-1)
	180
	181	#define is_parity_stripe(x) (((x) == RAID5_P_STRIPE) \|\| \
	182	((x) == RAID6_Q_STRIPE))
	183
5a6ac9ea MX	184	struct btrfs_device;
5a6ac9ea MX	185
6065fd95	186	void raid56_parity_recover(struct bio bio, struct btrfs_io_context bioc,
f1c29379	187	int mirror_num);
31683f4a	188	void raid56_parity_write(struct bio bio, struct btrfs_io_context bioc);
53b381b3	189
6a258d72	190	struct btrfs_raid_bio raid56_parity_alloc_scrub_rbio(struct bio bio,
ff18a4af	191	struct btrfs_io_context *bioc,
6a258d72 QW	192	struct btrfs_device *scrub_dev,
6a258d72 QW	193	unsigned long *dbitmap, int stripe_nsectors);
5a6ac9ea MX	194	void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio);
5a6ac9ea MX	195
94ead93e QW	196	void raid56_parity_cache_data_pages(struct btrfs_raid_bio *rbio,
	197	struct page **data_pages, u64 data_logical);
	198
53b381b3 DW	199	int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info);
53b381b3 DW	200	void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info);
9888c340	201
53b381b3	202	#endif