xref: /openbmc/linux/fs/btrfs/raid56.h (revision 9470114d)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Copyright (C) 2012 Fusion-io  All rights reserved.
4  * Copyright (C) 2012 Intel Corp. All rights reserved.
5  */
6 
7 #ifndef BTRFS_RAID56_H
8 #define BTRFS_RAID56_H
9 
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 };
18 
19 struct btrfs_raid_bio {
20 	struct btrfs_io_context *bioc;
21 
22 	/*
23 	 * While we're doing RMW on a stripe we put it into a hash table so we
24 	 * can lock the stripe and merge more rbios into it.
25 	 */
26 	struct list_head hash_list;
27 
28 	/* LRU list for the stripe cache */
29 	struct list_head stripe_cache;
30 
31 	/* For scheduling work in the helper threads */
32 	struct work_struct work;
33 
34 	/*
35 	 * bio_list and bio_list_lock are used to add more bios into the stripe
36 	 * in hopes of avoiding the full RMW
37 	 */
38 	struct bio_list bio_list;
39 	spinlock_t bio_list_lock;
40 
41 	/*
42 	 * Also protected by the bio_list_lock, the plug list is used by the
43 	 * plugging code to collect partial bios while plugged.  The stripe
44 	 * locking code also uses it to hand off the stripe lock to the next
45 	 * pending IO.
46 	 */
47 	struct list_head plug_list;
48 
49 	/* Flags that tell us if it is safe to merge with this bio. */
50 	unsigned long flags;
51 
52 	/*
53 	 * Set if we're doing a parity rebuild for a read from higher up, which
54 	 * is handled differently from a parity rebuild as part of RMW.
55 	 */
56 	enum btrfs_rbio_ops operation;
57 
58 	/* How many pages there are for the full stripe including P/Q */
59 	u16 nr_pages;
60 
61 	/* How many sectors there are for the full stripe including P/Q */
62 	u16 nr_sectors;
63 
64 	/* Number of data stripes (no p/q) */
65 	u8 nr_data;
66 
67 	/* Number of all stripes (including P/Q) */
68 	u8 real_stripes;
69 
70 	/* How many pages there are for each stripe */
71 	u8 stripe_npages;
72 
73 	/* How many sectors there are for each stripe */
74 	u8 stripe_nsectors;
75 
76 	/* Stripe number that we're scrubbing  */
77 	u8 scrubp;
78 
79 	/*
80 	 * Size of all the bios in the bio_list.  This helps us decide if the
81 	 * rbio maps to a full stripe or not.
82 	 */
83 	int bio_list_bytes;
84 
85 	refcount_t refs;
86 
87 	atomic_t stripes_pending;
88 
89 	wait_queue_head_t io_wait;
90 
91 	/* Bitmap to record which horizontal stripe has data */
92 	unsigned long dbitmap;
93 
94 	/* Allocated with stripe_nsectors-many bits for finish_*() calls */
95 	unsigned long finish_pbitmap;
96 
97 	/*
98 	 * These are two arrays of pointers.  We allocate the rbio big enough
99 	 * to hold them both and setup their locations when the rbio is
100 	 * allocated.
101 	 */
102 
103 	/*
104 	 * Pointers to pages that we allocated for reading/writing stripes
105 	 * directly from the disk (including P/Q).
106 	 */
107 	struct page **stripe_pages;
108 
109 	/* Pointers to the sectors in the bio_list, for faster lookup */
110 	struct sector_ptr *bio_sectors;
111 
112 	/*
113 	 * For subpage support, we need to map each sector to above
114 	 * stripe_pages.
115 	 */
116 	struct sector_ptr *stripe_sectors;
117 
118 	/* Allocated with real_stripes-many pointers for finish_*() calls */
119 	void **finish_pointers;
120 
121 	/*
122 	 * The bitmap recording where IO errors happened.
123 	 * Each bit is corresponding to one sector in either bio_sectors[] or
124 	 * stripe_sectors[] array.
125 	 *
126 	 * The reason we don't use another bit in sector_ptr is, we have two
127 	 * arrays of sectors, and a lot of IO can use sectors in both arrays.
128 	 * Thus making it much harder to iterate.
129 	 */
130 	unsigned long *error_bitmap;
131 
132 	/*
133 	 * Checksum buffer if the rbio is for data.  The buffer should cover
134 	 * all data sectors (excluding P/Q sectors).
135 	 */
136 	u8 *csum_buf;
137 
138 	/*
139 	 * Each bit represents if the corresponding sector has data csum found.
140 	 * Should only cover data sectors (excluding P/Q sectors).
141 	 */
142 	unsigned long *csum_bitmap;
143 };
144 
145 /*
146  * For trace event usage only. Records useful debug info for each bio submitted
147  * by RAID56 to each physical device.
148  *
149  * No matter signed or not, (-1) is always the one indicating we can not grab
150  * the proper stripe number.
151  */
152 struct raid56_bio_trace_info {
153 	u64 devid;
154 
155 	/* The offset inside the stripe. (<= STRIPE_LEN) */
156 	u32 offset;
157 
158 	/*
159 	 * Stripe number.
160 	 * 0 is the first data stripe, and nr_data for P stripe,
161 	 * nr_data + 1 for Q stripe.
162 	 * >= real_stripes for
163 	 */
164 	u8 stripe_nr;
165 };
166 
167 static inline int nr_data_stripes(const struct map_lookup *map)
168 {
169 	return map->num_stripes - btrfs_nr_parity_stripes(map->type);
170 }
171 
172 static inline int nr_bioc_data_stripes(const struct btrfs_io_context *bioc)
173 {
174 	return bioc->num_stripes - btrfs_nr_parity_stripes(bioc->map_type);
175 }
176 
177 #define RAID5_P_STRIPE ((u64)-2)
178 #define RAID6_Q_STRIPE ((u64)-1)
179 
180 #define is_parity_stripe(x) (((x) == RAID5_P_STRIPE) ||		\
181 			     ((x) == RAID6_Q_STRIPE))
182 
183 struct btrfs_device;
184 
185 void raid56_parity_recover(struct bio *bio, struct btrfs_io_context *bioc,
186 			   int mirror_num);
187 void raid56_parity_write(struct bio *bio, struct btrfs_io_context *bioc);
188 
189 struct btrfs_raid_bio *raid56_parity_alloc_scrub_rbio(struct bio *bio,
190 				struct btrfs_io_context *bioc,
191 				struct btrfs_device *scrub_dev,
192 				unsigned long *dbitmap, int stripe_nsectors);
193 void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio);
194 
195 void raid56_parity_cache_data_pages(struct btrfs_raid_bio *rbio,
196 				    struct page **data_pages, u64 data_logical);
197 
198 int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info);
199 void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info);
200 
201 #endif
202