1  /* SPDX-License-Identifier: GPL-2.0 */
2  /*
3   * Code for manipulating bucket marks for garbage collection.
4   *
5   * Copyright 2014 Datera, Inc.
6   */
7  
8  #ifndef _BUCKETS_H
9  #define _BUCKETS_H
10  
11  #include "buckets_types.h"
12  #include "extents.h"
13  #include "sb-members.h"
14  
sector_to_bucket(const struct bch_dev * ca,sector_t s)15  static inline u64 sector_to_bucket(const struct bch_dev *ca, sector_t s)
16  {
17  	return div_u64(s, ca->mi.bucket_size);
18  }
19  
bucket_to_sector(const struct bch_dev * ca,size_t b)20  static inline sector_t bucket_to_sector(const struct bch_dev *ca, size_t b)
21  {
22  	return ((sector_t) b) * ca->mi.bucket_size;
23  }
24  
bucket_remainder(const struct bch_dev * ca,sector_t s)25  static inline sector_t bucket_remainder(const struct bch_dev *ca, sector_t s)
26  {
27  	u32 remainder;
28  
29  	div_u64_rem(s, ca->mi.bucket_size, &remainder);
30  	return remainder;
31  }
32  
sector_to_bucket_and_offset(const struct bch_dev * ca,sector_t s,u32 * offset)33  static inline u64 sector_to_bucket_and_offset(const struct bch_dev *ca, sector_t s, u32 *offset)
34  {
35  	return div_u64_rem(s, ca->mi.bucket_size, offset);
36  }
37  
38  #define for_each_bucket(_b, _buckets)				\
39  	for (_b = (_buckets)->b + (_buckets)->first_bucket;	\
40  	     _b < (_buckets)->b + (_buckets)->nbuckets; _b++)
41  
42  /*
43   * Ugly hack alert:
44   *
45   * We need to cram a spinlock in a single byte, because that's what we have left
46   * in struct bucket, and we care about the size of these - during fsck, we need
47   * in memory state for every single bucket on every device.
48   *
49   * We used to do
50   *   while (xchg(&b->lock, 1) cpu_relax();
51   * but, it turns out not all architectures support xchg on a single byte.
52   *
53   * So now we use bit_spin_lock(), with fun games since we can't burn a whole
54   * ulong for this - we just need to make sure the lock bit always ends up in the
55   * first byte.
56   */
57  
58  #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
59  #define BUCKET_LOCK_BITNR	0
60  #else
61  #define BUCKET_LOCK_BITNR	(BITS_PER_LONG - 1)
62  #endif
63  
64  union ulong_byte_assert {
65  	ulong	ulong;
66  	u8	byte;
67  };
68  
bucket_unlock(struct bucket * b)69  static inline void bucket_unlock(struct bucket *b)
70  {
71  	BUILD_BUG_ON(!((union ulong_byte_assert) { .ulong = 1UL << BUCKET_LOCK_BITNR }).byte);
72  
73  	clear_bit_unlock(BUCKET_LOCK_BITNR, (void *) &b->lock);
74  	wake_up_bit((void *) &b->lock, BUCKET_LOCK_BITNR);
75  }
76  
bucket_lock(struct bucket * b)77  static inline void bucket_lock(struct bucket *b)
78  {
79  	wait_on_bit_lock((void *) &b->lock, BUCKET_LOCK_BITNR,
80  			 TASK_UNINTERRUPTIBLE);
81  }
82  
gc_bucket(struct bch_dev * ca,size_t b)83  static inline struct bucket *gc_bucket(struct bch_dev *ca, size_t b)
84  {
85  	return genradix_ptr(&ca->buckets_gc, b);
86  }
87  
bucket_gens(struct bch_dev * ca)88  static inline struct bucket_gens *bucket_gens(struct bch_dev *ca)
89  {
90  	return rcu_dereference_check(ca->bucket_gens,
91  				     !ca->fs ||
92  				     percpu_rwsem_is_held(&ca->fs->mark_lock) ||
93  				     lockdep_is_held(&ca->fs->state_lock) ||
94  				     lockdep_is_held(&ca->bucket_lock));
95  }
96  
bucket_gen(struct bch_dev * ca,size_t b)97  static inline u8 *bucket_gen(struct bch_dev *ca, size_t b)
98  {
99  	struct bucket_gens *gens = bucket_gens(ca);
100  
101  	if (b - gens->first_bucket >= gens->nbuckets_minus_first)
102  		return NULL;
103  	return gens->b + b;
104  }
105  
bucket_gen_get_rcu(struct bch_dev * ca,size_t b)106  static inline int bucket_gen_get_rcu(struct bch_dev *ca, size_t b)
107  {
108  	u8 *gen = bucket_gen(ca, b);
109  	return gen ? *gen : -1;
110  }
111  
bucket_gen_get(struct bch_dev * ca,size_t b)112  static inline int bucket_gen_get(struct bch_dev *ca, size_t b)
113  {
114  	rcu_read_lock();
115  	int ret = bucket_gen_get_rcu(ca, b);
116  	rcu_read_unlock();
117  	return ret;
118  }
119  
PTR_BUCKET_NR(const struct bch_dev * ca,const struct bch_extent_ptr * ptr)120  static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
121  				   const struct bch_extent_ptr *ptr)
122  {
123  	return sector_to_bucket(ca, ptr->offset);
124  }
125  
PTR_BUCKET_POS(const struct bch_dev * ca,const struct bch_extent_ptr * ptr)126  static inline struct bpos PTR_BUCKET_POS(const struct bch_dev *ca,
127  					 const struct bch_extent_ptr *ptr)
128  {
129  	return POS(ptr->dev, PTR_BUCKET_NR(ca, ptr));
130  }
131  
PTR_BUCKET_POS_OFFSET(const struct bch_dev * ca,const struct bch_extent_ptr * ptr,u32 * bucket_offset)132  static inline struct bpos PTR_BUCKET_POS_OFFSET(const struct bch_dev *ca,
133  						const struct bch_extent_ptr *ptr,
134  						u32 *bucket_offset)
135  {
136  	return POS(ptr->dev, sector_to_bucket_and_offset(ca, ptr->offset, bucket_offset));
137  }
138  
PTR_GC_BUCKET(struct bch_dev * ca,const struct bch_extent_ptr * ptr)139  static inline struct bucket *PTR_GC_BUCKET(struct bch_dev *ca,
140  					   const struct bch_extent_ptr *ptr)
141  {
142  	return gc_bucket(ca, PTR_BUCKET_NR(ca, ptr));
143  }
144  
ptr_data_type(const struct bkey * k,const struct bch_extent_ptr * ptr)145  static inline enum bch_data_type ptr_data_type(const struct bkey *k,
146  					       const struct bch_extent_ptr *ptr)
147  {
148  	if (bkey_is_btree_ptr(k))
149  		return BCH_DATA_btree;
150  
151  	return ptr->cached ? BCH_DATA_cached : BCH_DATA_user;
152  }
153  
ptr_disk_sectors(s64 sectors,struct extent_ptr_decoded p)154  static inline s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p)
155  {
156  	EBUG_ON(sectors < 0);
157  
158  	return crc_is_compressed(p.crc)
159  		? DIV_ROUND_UP_ULL(sectors * p.crc.compressed_size,
160  				   p.crc.uncompressed_size)
161  		: sectors;
162  }
163  
gen_cmp(u8 a,u8 b)164  static inline int gen_cmp(u8 a, u8 b)
165  {
166  	return (s8) (a - b);
167  }
168  
gen_after(u8 a,u8 b)169  static inline int gen_after(u8 a, u8 b)
170  {
171  	int r = gen_cmp(a, b);
172  
173  	return r > 0 ? r : 0;
174  }
175  
dev_ptr_stale_rcu(struct bch_dev * ca,const struct bch_extent_ptr * ptr)176  static inline int dev_ptr_stale_rcu(struct bch_dev *ca, const struct bch_extent_ptr *ptr)
177  {
178  	int gen = bucket_gen_get_rcu(ca, PTR_BUCKET_NR(ca, ptr));
179  	return gen < 0 ? gen : gen_after(gen, ptr->gen);
180  }
181  
182  /**
183   * dev_ptr_stale() - check if a pointer points into a bucket that has been
184   * invalidated.
185   */
dev_ptr_stale(struct bch_dev * ca,const struct bch_extent_ptr * ptr)186  static inline int dev_ptr_stale(struct bch_dev *ca, const struct bch_extent_ptr *ptr)
187  {
188  	rcu_read_lock();
189  	int ret = dev_ptr_stale_rcu(ca, ptr);
190  	rcu_read_unlock();
191  	return ret;
192  }
193  
194  /* Device usage: */
195  
196  void bch2_dev_usage_read_fast(struct bch_dev *, struct bch_dev_usage *);
bch2_dev_usage_read(struct bch_dev * ca)197  static inline struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca)
198  {
199  	struct bch_dev_usage ret;
200  
201  	bch2_dev_usage_read_fast(ca, &ret);
202  	return ret;
203  }
204  
205  void bch2_dev_usage_to_text(struct printbuf *, struct bch_dev *, struct bch_dev_usage *);
206  
bch2_dev_buckets_reserved(struct bch_dev * ca,enum bch_watermark watermark)207  static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum bch_watermark watermark)
208  {
209  	s64 reserved = 0;
210  
211  	switch (watermark) {
212  	case BCH_WATERMARK_NR:
213  		BUG();
214  	case BCH_WATERMARK_stripe:
215  		reserved += ca->mi.nbuckets >> 6;
216  		fallthrough;
217  	case BCH_WATERMARK_normal:
218  		reserved += ca->mi.nbuckets >> 6;
219  		fallthrough;
220  	case BCH_WATERMARK_copygc:
221  		reserved += ca->nr_btree_reserve;
222  		fallthrough;
223  	case BCH_WATERMARK_btree:
224  		reserved += ca->nr_btree_reserve;
225  		fallthrough;
226  	case BCH_WATERMARK_btree_copygc:
227  	case BCH_WATERMARK_reclaim:
228  	case BCH_WATERMARK_interior_updates:
229  		break;
230  	}
231  
232  	return reserved;
233  }
234  
dev_buckets_free(struct bch_dev * ca,struct bch_dev_usage usage,enum bch_watermark watermark)235  static inline u64 dev_buckets_free(struct bch_dev *ca,
236  				   struct bch_dev_usage usage,
237  				   enum bch_watermark watermark)
238  {
239  	return max_t(s64, 0,
240  		     usage.d[BCH_DATA_free].buckets -
241  		     ca->nr_open_buckets -
242  		     bch2_dev_buckets_reserved(ca, watermark));
243  }
244  
__dev_buckets_available(struct bch_dev * ca,struct bch_dev_usage usage,enum bch_watermark watermark)245  static inline u64 __dev_buckets_available(struct bch_dev *ca,
246  					  struct bch_dev_usage usage,
247  					  enum bch_watermark watermark)
248  {
249  	return max_t(s64, 0,
250  		       usage.d[BCH_DATA_free].buckets
251  		     + usage.d[BCH_DATA_cached].buckets
252  		     + usage.d[BCH_DATA_need_gc_gens].buckets
253  		     + usage.d[BCH_DATA_need_discard].buckets
254  		     - ca->nr_open_buckets
255  		     - bch2_dev_buckets_reserved(ca, watermark));
256  }
257  
dev_buckets_available(struct bch_dev * ca,enum bch_watermark watermark)258  static inline u64 dev_buckets_available(struct bch_dev *ca,
259  					enum bch_watermark watermark)
260  {
261  	return __dev_buckets_available(ca, bch2_dev_usage_read(ca), watermark);
262  }
263  
264  /* Filesystem usage: */
265  
dev_usage_u64s(void)266  static inline unsigned dev_usage_u64s(void)
267  {
268  	return sizeof(struct bch_dev_usage) / sizeof(u64);
269  }
270  
271  struct bch_fs_usage_short
272  bch2_fs_usage_read_short(struct bch_fs *);
273  
274  int bch2_bucket_ref_update(struct btree_trans *, struct bch_dev *,
275  			   struct bkey_s_c, const struct bch_extent_ptr *,
276  			   s64, enum bch_data_type, u8, u8, u32 *);
277  
278  int bch2_check_fix_ptrs(struct btree_trans *,
279  			enum btree_id, unsigned, struct bkey_s_c,
280  			enum btree_iter_update_trigger_flags);
281  
282  int bch2_trigger_extent(struct btree_trans *, enum btree_id, unsigned,
283  			struct bkey_s_c, struct bkey_s,
284  			enum btree_iter_update_trigger_flags);
285  int bch2_trigger_reservation(struct btree_trans *, enum btree_id, unsigned,
286  			  struct bkey_s_c, struct bkey_s,
287  			  enum btree_iter_update_trigger_flags);
288  
289  #define trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)\
290  ({												\
291  	int ret = 0;										\
292  												\
293  	if (_old.k->type)									\
294  		ret = _fn(_trans, _btree_id, _level, _old, _flags & ~BTREE_TRIGGER_insert);	\
295  	if (!ret && _new.k->type)								\
296  		ret = _fn(_trans, _btree_id, _level, _new.s_c, _flags & ~BTREE_TRIGGER_overwrite);\
297  	ret;											\
298  })
299  
300  void bch2_trans_account_disk_usage_change(struct btree_trans *);
301  
302  int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *, u64,
303  				    enum bch_data_type, unsigned,
304  				    enum btree_iter_update_trigger_flags);
305  int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *,
306  				    enum btree_iter_update_trigger_flags);
307  int bch2_trans_mark_dev_sbs_flags(struct bch_fs *,
308  				    enum btree_iter_update_trigger_flags);
309  int bch2_trans_mark_dev_sbs(struct bch_fs *);
310  
is_superblock_bucket(struct bch_dev * ca,u64 b)311  static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b)
312  {
313  	struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
314  	u64 b_offset	= bucket_to_sector(ca, b);
315  	u64 b_end	= bucket_to_sector(ca, b + 1);
316  	unsigned i;
317  
318  	if (!b)
319  		return true;
320  
321  	for (i = 0; i < layout->nr_superblocks; i++) {
322  		u64 offset = le64_to_cpu(layout->sb_offset[i]);
323  		u64 end = offset + (1 << layout->sb_max_size_bits);
324  
325  		if (!(offset >= b_end || end <= b_offset))
326  			return true;
327  	}
328  
329  	return false;
330  }
331  
bch2_data_type_str(enum bch_data_type type)332  static inline const char *bch2_data_type_str(enum bch_data_type type)
333  {
334  	return type < BCH_DATA_NR
335  		? __bch2_data_types[type]
336  		: "(invalid data type)";
337  }
338  
339  /* disk reservations: */
340  
bch2_disk_reservation_put(struct bch_fs * c,struct disk_reservation * res)341  static inline void bch2_disk_reservation_put(struct bch_fs *c,
342  					     struct disk_reservation *res)
343  {
344  	if (res->sectors) {
345  		this_cpu_sub(*c->online_reserved, res->sectors);
346  		res->sectors = 0;
347  	}
348  }
349  
350  enum bch_reservation_flags {
351  	BCH_DISK_RESERVATION_NOFAIL	= 1 << 0,
352  	BCH_DISK_RESERVATION_PARTIAL	= 1 << 1,
353  };
354  
355  int __bch2_disk_reservation_add(struct bch_fs *, struct disk_reservation *,
356  				u64, enum bch_reservation_flags);
357  
bch2_disk_reservation_add(struct bch_fs * c,struct disk_reservation * res,u64 sectors,enum bch_reservation_flags flags)358  static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
359  					    u64 sectors, enum bch_reservation_flags flags)
360  {
361  #ifdef __KERNEL__
362  	u64 old, new;
363  
364  	old = this_cpu_read(c->pcpu->sectors_available);
365  	do {
366  		if (sectors > old)
367  			return __bch2_disk_reservation_add(c, res, sectors, flags);
368  
369  		new = old - sectors;
370  	} while (!this_cpu_try_cmpxchg(c->pcpu->sectors_available, &old, new));
371  
372  	this_cpu_add(*c->online_reserved, sectors);
373  	res->sectors			+= sectors;
374  	return 0;
375  #else
376  	return __bch2_disk_reservation_add(c, res, sectors, flags);
377  #endif
378  }
379  
380  static inline struct disk_reservation
bch2_disk_reservation_init(struct bch_fs * c,unsigned nr_replicas)381  bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas)
382  {
383  	return (struct disk_reservation) {
384  		.sectors	= 0,
385  #if 0
386  		/* not used yet: */
387  		.gen		= c->capacity_gen,
388  #endif
389  		.nr_replicas	= nr_replicas,
390  	};
391  }
392  
bch2_disk_reservation_get(struct bch_fs * c,struct disk_reservation * res,u64 sectors,unsigned nr_replicas,int flags)393  static inline int bch2_disk_reservation_get(struct bch_fs *c,
394  					    struct disk_reservation *res,
395  					    u64 sectors, unsigned nr_replicas,
396  					    int flags)
397  {
398  	*res = bch2_disk_reservation_init(c, nr_replicas);
399  
400  	return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags);
401  }
402  
403  #define RESERVE_FACTOR	6
404  
avail_factor(u64 r)405  static inline u64 avail_factor(u64 r)
406  {
407  	return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1);
408  }
409  
410  void bch2_buckets_nouse_free(struct bch_fs *);
411  int bch2_buckets_nouse_alloc(struct bch_fs *);
412  
413  int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64);
414  void bch2_dev_buckets_free(struct bch_dev *);
415  int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *);
416  
417  #endif /* _BUCKETS_H */
418