xref: /linux/fs/bcachefs/sb-members.c (revision 36df6f734a7ad69880c5262543165c47cb57169f)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "btree_cache.h"
5 #include "disk_groups.h"
6 #include "error.h"
7 #include "opts.h"
8 #include "recovery_passes.h"
9 #include "replicas.h"
10 #include "sb-members.h"
11 #include "super-io.h"
12 
bch2_dev_missing_bkey(struct bch_fs * c,struct bkey_s_c k,unsigned dev)13 int bch2_dev_missing_bkey(struct bch_fs *c, struct bkey_s_c k, unsigned dev)
14 {
15 	struct printbuf buf = PRINTBUF;
16 	bch2_log_msg_start(c, &buf);
17 
18 	prt_printf(&buf, "pointer to nonexistent device %u in key\n", dev);
19 	bch2_bkey_val_to_text(&buf, c, k);
20 
21 	bool print = bch2_count_fsck_err(c, ptr_to_invalid_device, &buf);
22 
23 	int ret = bch2_run_explicit_recovery_pass(c, &buf,
24 					BCH_RECOVERY_PASS_check_allocations, 0);
25 
26 	if (print)
27 		bch2_print_str(c, KERN_ERR, buf.buf);
28 	printbuf_exit(&buf);
29 	return ret;
30 }
31 
bch2_dev_missing_atomic(struct bch_fs * c,unsigned dev)32 void bch2_dev_missing_atomic(struct bch_fs *c, unsigned dev)
33 {
34 	if (dev != BCH_SB_MEMBER_INVALID)
35 		bch2_fs_inconsistent(c, "pointer to nonexistent device %u", dev);
36 }
37 
bch2_dev_bucket_missing(struct bch_dev * ca,u64 bucket)38 void bch2_dev_bucket_missing(struct bch_dev *ca, u64 bucket)
39 {
40 	bch2_fs_inconsistent(ca->fs,
41 		"pointer to nonexistent bucket %llu on device %s (valid range %u-%llu)",
42 		bucket, ca->name, ca->mi.first_bucket, ca->mi.nbuckets);
43 }
44 
45 #define x(t, n, ...) [n] = #t,
46 static const char * const bch2_iops_measurements[] = {
47 	BCH_IOPS_MEASUREMENTS()
48 	NULL
49 };
50 
51 char * const bch2_member_error_strs[] = {
52 	BCH_MEMBER_ERROR_TYPES()
53 	NULL
54 };
55 #undef x
56 
57 /* Code for bch_sb_field_members_v1: */
58 
bch2_members_v2_get_mut(struct bch_sb * sb,int i)59 struct bch_member *bch2_members_v2_get_mut(struct bch_sb *sb, int i)
60 {
61 	return __bch2_members_v2_get_mut(bch2_sb_field_get(sb, members_v2), i);
62 }
63 
members_v2_get(struct bch_sb_field_members_v2 * mi,int i)64 static struct bch_member members_v2_get(struct bch_sb_field_members_v2 *mi, int i)
65 {
66 	struct bch_member ret, *p = __bch2_members_v2_get_mut(mi, i);
67 	memset(&ret, 0, sizeof(ret));
68 	memcpy(&ret, p, min_t(size_t, le16_to_cpu(mi->member_bytes), sizeof(ret)));
69 	return ret;
70 }
71 
members_v1_get_mut(struct bch_sb_field_members_v1 * mi,int i)72 static struct bch_member *members_v1_get_mut(struct bch_sb_field_members_v1 *mi, int i)
73 {
74 	return (void *) mi->_members + (i * BCH_MEMBER_V1_BYTES);
75 }
76 
members_v1_get(struct bch_sb_field_members_v1 * mi,int i)77 static struct bch_member members_v1_get(struct bch_sb_field_members_v1 *mi, int i)
78 {
79 	struct bch_member ret, *p = members_v1_get_mut(mi, i);
80 	memset(&ret, 0, sizeof(ret));
81 	memcpy(&ret, p, min_t(size_t, BCH_MEMBER_V1_BYTES, sizeof(ret)));
82 	return ret;
83 }
84 
bch2_sb_member_get(struct bch_sb * sb,int i)85 struct bch_member bch2_sb_member_get(struct bch_sb *sb, int i)
86 {
87 	struct bch_sb_field_members_v2 *mi2 = bch2_sb_field_get(sb, members_v2);
88 	if (mi2)
89 		return members_v2_get(mi2, i);
90 	struct bch_sb_field_members_v1 *mi1 = bch2_sb_field_get(sb, members_v1);
91 	return members_v1_get(mi1, i);
92 }
93 
sb_members_v2_resize_entries(struct bch_fs * c)94 static int sb_members_v2_resize_entries(struct bch_fs *c)
95 {
96 	struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
97 
98 	if (le16_to_cpu(mi->member_bytes) < sizeof(struct bch_member)) {
99 		unsigned u64s = DIV_ROUND_UP((sizeof(*mi) + sizeof(mi->_members[0]) *
100 					      c->disk_sb.sb->nr_devices), 8);
101 
102 		mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
103 		if (!mi)
104 			return bch_err_throw(c, ENOSPC_sb_members_v2);
105 
106 		for (int i = c->disk_sb.sb->nr_devices - 1; i >= 0; --i) {
107 			void *dst = (void *) mi->_members + (i * sizeof(struct bch_member));
108 			memmove(dst, __bch2_members_v2_get_mut(mi, i), le16_to_cpu(mi->member_bytes));
109 			memset(dst + le16_to_cpu(mi->member_bytes),
110 			       0, (sizeof(struct bch_member) - le16_to_cpu(mi->member_bytes)));
111 		}
112 		mi->member_bytes = cpu_to_le16(sizeof(struct bch_member));
113 	}
114 	return 0;
115 }
116 
bch2_sb_members_v2_init(struct bch_fs * c)117 int bch2_sb_members_v2_init(struct bch_fs *c)
118 {
119 	struct bch_sb_field_members_v1 *mi1;
120 	struct bch_sb_field_members_v2 *mi2;
121 
122 	if (!bch2_sb_field_get(c->disk_sb.sb, members_v2)) {
123 		mi2 = bch2_sb_field_resize(&c->disk_sb, members_v2,
124 				DIV_ROUND_UP(sizeof(*mi2) +
125 					     sizeof(struct bch_member) * c->sb.nr_devices,
126 					     sizeof(u64)));
127 		mi1 = bch2_sb_field_get(c->disk_sb.sb, members_v1);
128 		memcpy(&mi2->_members[0], &mi1->_members[0],
129 		       BCH_MEMBER_V1_BYTES * c->sb.nr_devices);
130 		memset(&mi2->pad[0], 0, sizeof(mi2->pad));
131 		mi2->member_bytes = cpu_to_le16(BCH_MEMBER_V1_BYTES);
132 	}
133 
134 	return sb_members_v2_resize_entries(c);
135 }
136 
bch2_sb_members_cpy_v2_v1(struct bch_sb_handle * disk_sb)137 int bch2_sb_members_cpy_v2_v1(struct bch_sb_handle *disk_sb)
138 {
139 	struct bch_sb_field_members_v1 *mi1;
140 	struct bch_sb_field_members_v2 *mi2;
141 
142 	if (BCH_SB_VERSION_INCOMPAT(disk_sb->sb) > bcachefs_metadata_version_extent_flags) {
143 		bch2_sb_field_resize(disk_sb, members_v1, 0);
144 		return 0;
145 	}
146 
147 	mi1 = bch2_sb_field_resize(disk_sb, members_v1,
148 			DIV_ROUND_UP(sizeof(*mi1) + BCH_MEMBER_V1_BYTES *
149 				     disk_sb->sb->nr_devices, sizeof(u64)));
150 	if (!mi1)
151 		return -BCH_ERR_ENOSPC_sb_members;
152 
153 	mi2 = bch2_sb_field_get(disk_sb->sb, members_v2);
154 
155 	for (unsigned i = 0; i < disk_sb->sb->nr_devices; i++)
156 		memcpy(members_v1_get_mut(mi1, i), __bch2_members_v2_get_mut(mi2, i), BCH_MEMBER_V1_BYTES);
157 
158 	return 0;
159 }
160 
validate_member(struct printbuf * err,struct bch_member m,struct bch_sb * sb,int i)161 static int validate_member(struct printbuf *err,
162 			   struct bch_member m,
163 			   struct bch_sb *sb,
164 			   int i)
165 {
166 	if (le64_to_cpu(m.nbuckets) > BCH_MEMBER_NBUCKETS_MAX) {
167 		prt_printf(err, "device %u: too many buckets (got %llu, max %u)",
168 			   i, le64_to_cpu(m.nbuckets), BCH_MEMBER_NBUCKETS_MAX);
169 		return -BCH_ERR_invalid_sb_members;
170 	}
171 
172 	if (le64_to_cpu(m.nbuckets) -
173 	    le16_to_cpu(m.first_bucket) < BCH_MIN_NR_NBUCKETS) {
174 		prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
175 			   i, le64_to_cpu(m.nbuckets), BCH_MIN_NR_NBUCKETS);
176 		return -BCH_ERR_invalid_sb_members;
177 	}
178 
179 	if (le16_to_cpu(m.bucket_size) <
180 	    le16_to_cpu(sb->block_size)) {
181 		prt_printf(err, "device %u: bucket size %u smaller than block size %u",
182 			   i, le16_to_cpu(m.bucket_size), le16_to_cpu(sb->block_size));
183 		return -BCH_ERR_invalid_sb_members;
184 	}
185 
186 	if (le16_to_cpu(m.bucket_size) <
187 	    BCH_SB_BTREE_NODE_SIZE(sb)) {
188 		prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
189 			   i, le16_to_cpu(m.bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
190 		return -BCH_ERR_invalid_sb_members;
191 	}
192 
193 	if (m.btree_bitmap_shift >= BCH_MI_BTREE_BITMAP_SHIFT_MAX) {
194 		prt_printf(err, "device %u: invalid btree_bitmap_shift %u", i, m.btree_bitmap_shift);
195 		return -BCH_ERR_invalid_sb_members;
196 	}
197 
198 	if (BCH_MEMBER_FREESPACE_INITIALIZED(&m) &&
199 	    sb->features[0] & cpu_to_le64(BIT_ULL(BCH_FEATURE_no_alloc_info))) {
200 		prt_printf(err, "device %u: freespace initialized but fs has no alloc info", i);
201 		return -BCH_ERR_invalid_sb_members;
202 	}
203 
204 	return 0;
205 }
206 
member_to_text(struct printbuf * out,struct bch_member m,struct bch_sb_field_disk_groups * gi,struct bch_sb * sb,int i)207 static void member_to_text(struct printbuf *out,
208 			   struct bch_member m,
209 			   struct bch_sb_field_disk_groups *gi,
210 			   struct bch_sb *sb,
211 			   int i)
212 {
213 	unsigned data_have = bch2_sb_dev_has_data(sb, i);
214 	u64 bucket_size = le16_to_cpu(m.bucket_size);
215 	u64 device_size = le64_to_cpu(m.nbuckets) * bucket_size;
216 
217 	if (!bch2_member_alive(&m))
218 		return;
219 
220 	prt_printf(out, "Device:\t%u\n", i);
221 
222 	printbuf_indent_add(out, 2);
223 
224 	prt_printf(out, "Label:\t");
225 	if (BCH_MEMBER_GROUP(&m))
226 		bch2_disk_path_to_text_sb(out, sb,
227 				BCH_MEMBER_GROUP(&m) - 1);
228 	else
229 		prt_printf(out, "(none)");
230 	prt_newline(out);
231 
232 	prt_printf(out, "UUID:\t");
233 	pr_uuid(out, m.uuid.b);
234 	prt_newline(out);
235 
236 	prt_printf(out, "Size:\t");
237 	prt_units_u64(out, device_size << 9);
238 	prt_newline(out);
239 
240 	for (unsigned i = 0; i < BCH_MEMBER_ERROR_NR; i++)
241 		prt_printf(out, "%s errors:\t%llu\n", bch2_member_error_strs[i], le64_to_cpu(m.errors[i]));
242 
243 	for (unsigned i = 0; i < BCH_IOPS_NR; i++)
244 		prt_printf(out, "%s iops:\t%u\n", bch2_iops_measurements[i], le32_to_cpu(m.iops[i]));
245 
246 	prt_printf(out, "Bucket size:\t");
247 	prt_units_u64(out, bucket_size << 9);
248 	prt_newline(out);
249 
250 	prt_printf(out, "First bucket:\t%u\n", le16_to_cpu(m.first_bucket));
251 	prt_printf(out, "Buckets:\t%llu\n", le64_to_cpu(m.nbuckets));
252 
253 	prt_printf(out, "Last mount:\t");
254 	if (m.last_mount)
255 		bch2_prt_datetime(out, le64_to_cpu(m.last_mount));
256 	else
257 		prt_printf(out, "(never)");
258 	prt_newline(out);
259 
260 	prt_printf(out, "Last superblock write:\t%llu\n", le64_to_cpu(m.seq));
261 
262 	prt_printf(out, "State:\t%s\n",
263 		   BCH_MEMBER_STATE(&m) < BCH_MEMBER_STATE_NR
264 		   ? bch2_member_states[BCH_MEMBER_STATE(&m)]
265 		   : "unknown");
266 
267 	prt_printf(out, "Data allowed:\t");
268 	if (BCH_MEMBER_DATA_ALLOWED(&m))
269 		prt_bitflags(out, __bch2_data_types, BCH_MEMBER_DATA_ALLOWED(&m));
270 	else
271 		prt_printf(out, "(none)");
272 	prt_newline(out);
273 
274 	prt_printf(out, "Has data:\t");
275 	if (data_have)
276 		prt_bitflags(out, __bch2_data_types, data_have);
277 	else
278 		prt_printf(out, "(none)");
279 	prt_newline(out);
280 
281 	prt_printf(out, "Btree allocated bitmap blocksize:\t");
282 	if (m.btree_bitmap_shift < 64)
283 		prt_units_u64(out, 1ULL << m.btree_bitmap_shift);
284 	else
285 		prt_printf(out, "(invalid shift %u)", m.btree_bitmap_shift);
286 	prt_newline(out);
287 
288 	prt_printf(out, "Btree allocated bitmap:\t");
289 	bch2_prt_u64_base2_nbits(out, le64_to_cpu(m.btree_allocated_bitmap), 64);
290 	prt_newline(out);
291 
292 	prt_printf(out, "Durability:\t%llu\n", BCH_MEMBER_DURABILITY(&m) ? BCH_MEMBER_DURABILITY(&m) - 1 : 1);
293 
294 	prt_printf(out, "Discard:\t%llu\n", BCH_MEMBER_DISCARD(&m));
295 	prt_printf(out, "Freespace initialized:\t%llu\n", BCH_MEMBER_FREESPACE_INITIALIZED(&m));
296 	prt_printf(out, "Resize on mount:\t%llu\n", BCH_MEMBER_RESIZE_ON_MOUNT(&m));
297 
298 	printbuf_indent_sub(out, 2);
299 }
300 
bch2_sb_members_v1_validate(struct bch_sb * sb,struct bch_sb_field * f,enum bch_validate_flags flags,struct printbuf * err)301 static int bch2_sb_members_v1_validate(struct bch_sb *sb, struct bch_sb_field *f,
302 				enum bch_validate_flags flags, struct printbuf *err)
303 {
304 	struct bch_sb_field_members_v1 *mi = field_to_type(f, members_v1);
305 	unsigned i;
306 
307 	if ((void *) members_v1_get_mut(mi, sb->nr_devices) > vstruct_end(&mi->field)) {
308 		prt_printf(err, "too many devices for section size");
309 		return -BCH_ERR_invalid_sb_members;
310 	}
311 
312 	for (i = 0; i < sb->nr_devices; i++) {
313 		struct bch_member m = members_v1_get(mi, i);
314 
315 		int ret = validate_member(err, m, sb, i);
316 		if (ret)
317 			return ret;
318 	}
319 
320 	return 0;
321 }
322 
bch2_sb_members_v1_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)323 static void bch2_sb_members_v1_to_text(struct printbuf *out, struct bch_sb *sb,
324 				       struct bch_sb_field *f)
325 {
326 	struct bch_sb_field_members_v1 *mi = field_to_type(f, members_v1);
327 	struct bch_sb_field_disk_groups *gi = bch2_sb_field_get(sb, disk_groups);
328 
329 	if (vstruct_end(&mi->field) <= (void *) &mi->_members[0]) {
330 		prt_printf(out, "field ends before start of entries");
331 		return;
332 	}
333 
334 	unsigned nr = (vstruct_end(&mi->field) - (void *) &mi->_members[0]) / sizeof(mi->_members[0]);
335 	if (nr != sb->nr_devices)
336 		prt_printf(out, "nr_devices mismatch: have %i entries, should be %u", nr, sb->nr_devices);
337 
338 	for (unsigned i = 0; i < min(sb->nr_devices, nr); i++)
339 		member_to_text(out, members_v1_get(mi, i), gi, sb, i);
340 }
341 
342 const struct bch_sb_field_ops bch_sb_field_ops_members_v1 = {
343 	.validate	= bch2_sb_members_v1_validate,
344 	.to_text	= bch2_sb_members_v1_to_text,
345 };
346 
bch2_sb_members_v2_to_text(struct printbuf * out,struct bch_sb * sb,struct bch_sb_field * f)347 static void bch2_sb_members_v2_to_text(struct printbuf *out, struct bch_sb *sb,
348 				       struct bch_sb_field *f)
349 {
350 	struct bch_sb_field_members_v2 *mi = field_to_type(f, members_v2);
351 	struct bch_sb_field_disk_groups *gi = bch2_sb_field_get(sb, disk_groups);
352 
353 	if (vstruct_end(&mi->field) <= (void *) &mi->_members[0]) {
354 		prt_printf(out, "field ends before start of entries");
355 		return;
356 	}
357 
358 	if (!le16_to_cpu(mi->member_bytes)) {
359 		prt_printf(out, "member_bytes 0");
360 		return;
361 	}
362 
363 	unsigned nr = (vstruct_end(&mi->field) - (void *) &mi->_members[0]) / le16_to_cpu(mi->member_bytes);
364 	if (nr != sb->nr_devices)
365 		prt_printf(out, "nr_devices mismatch: have %i entries, should be %u", nr, sb->nr_devices);
366 
367 	/*
368 	 * We call to_text() on superblock sections that haven't passed
369 	 * validate, so we can't trust sb->nr_devices.
370 	 */
371 
372 	for (unsigned i = 0; i < min(sb->nr_devices, nr); i++)
373 		member_to_text(out, members_v2_get(mi, i), gi, sb, i);
374 }
375 
bch2_sb_members_v2_validate(struct bch_sb * sb,struct bch_sb_field * f,enum bch_validate_flags flags,struct printbuf * err)376 static int bch2_sb_members_v2_validate(struct bch_sb *sb, struct bch_sb_field *f,
377 				enum bch_validate_flags flags, struct printbuf *err)
378 {
379 	struct bch_sb_field_members_v2 *mi = field_to_type(f, members_v2);
380 	size_t mi_bytes = (void *) __bch2_members_v2_get_mut(mi, sb->nr_devices) -
381 		(void *) mi;
382 
383 	if (mi_bytes > vstruct_bytes(&mi->field)) {
384 		prt_printf(err, "section too small (%zu > %zu)",
385 			   mi_bytes, vstruct_bytes(&mi->field));
386 		return -BCH_ERR_invalid_sb_members;
387 	}
388 
389 	for (unsigned i = 0; i < sb->nr_devices; i++) {
390 		int ret = validate_member(err, members_v2_get(mi, i), sb, i);
391 		if (ret)
392 			return ret;
393 	}
394 
395 	return 0;
396 }
397 
398 const struct bch_sb_field_ops bch_sb_field_ops_members_v2 = {
399 	.validate	= bch2_sb_members_v2_validate,
400 	.to_text	= bch2_sb_members_v2_to_text,
401 };
402 
bch2_sb_members_from_cpu(struct bch_fs * c)403 void bch2_sb_members_from_cpu(struct bch_fs *c)
404 {
405 	struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
406 
407 	guard(rcu)();
408 	for_each_member_device_rcu(c, ca, NULL) {
409 		struct bch_member *m = __bch2_members_v2_get_mut(mi, ca->dev_idx);
410 
411 		for (unsigned e = 0; e < BCH_MEMBER_ERROR_NR; e++)
412 			m->errors[e] = cpu_to_le64(atomic64_read(&ca->errors[e]));
413 	}
414 }
415 
bch2_dev_io_errors_to_text(struct printbuf * out,struct bch_dev * ca)416 void bch2_dev_io_errors_to_text(struct printbuf *out, struct bch_dev *ca)
417 {
418 	struct bch_fs *c = ca->fs;
419 	struct bch_member m;
420 
421 	mutex_lock(&ca->fs->sb_lock);
422 	m = bch2_sb_member_get(c->disk_sb.sb, ca->dev_idx);
423 	mutex_unlock(&ca->fs->sb_lock);
424 
425 	printbuf_tabstop_push(out, 12);
426 
427 	prt_str(out, "IO errors since filesystem creation");
428 	prt_newline(out);
429 
430 	printbuf_indent_add(out, 2);
431 	for (unsigned i = 0; i < BCH_MEMBER_ERROR_NR; i++)
432 		prt_printf(out, "%s:\t%llu\n", bch2_member_error_strs[i], atomic64_read(&ca->errors[i]));
433 	printbuf_indent_sub(out, 2);
434 
435 	prt_str(out, "IO errors since ");
436 	bch2_pr_time_units(out, (ktime_get_real_seconds() - le64_to_cpu(m.errors_reset_time)) * NSEC_PER_SEC);
437 	prt_str(out, " ago");
438 	prt_newline(out);
439 
440 	printbuf_indent_add(out, 2);
441 	for (unsigned i = 0; i < BCH_MEMBER_ERROR_NR; i++)
442 		prt_printf(out, "%s:\t%llu\n", bch2_member_error_strs[i],
443 			   atomic64_read(&ca->errors[i]) - le64_to_cpu(m.errors_at_reset[i]));
444 	printbuf_indent_sub(out, 2);
445 }
446 
bch2_dev_errors_reset(struct bch_dev * ca)447 void bch2_dev_errors_reset(struct bch_dev *ca)
448 {
449 	struct bch_fs *c = ca->fs;
450 	struct bch_member *m;
451 
452 	mutex_lock(&c->sb_lock);
453 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
454 	for (unsigned i = 0; i < ARRAY_SIZE(m->errors_at_reset); i++)
455 		m->errors_at_reset[i] = cpu_to_le64(atomic64_read(&ca->errors[i]));
456 	m->errors_reset_time = cpu_to_le64(ktime_get_real_seconds());
457 
458 	bch2_write_super(c);
459 	mutex_unlock(&c->sb_lock);
460 }
461 
462 /*
463  * Per member "range has btree nodes" bitmap:
464  *
465  * This is so that if we ever have to run the btree node scan to repair we don't
466  * have to scan full devices:
467  */
468 
bch2_dev_btree_bitmap_marked(struct bch_fs * c,struct bkey_s_c k)469 bool bch2_dev_btree_bitmap_marked(struct bch_fs *c, struct bkey_s_c k)
470 {
471 	guard(rcu)();
472 	bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
473 		struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
474 		if (ca &&
475 		    !bch2_dev_btree_bitmap_marked_sectors(ca, ptr->offset, btree_sectors(c)))
476 			return false;
477 	}
478 	return true;
479 }
480 
__bch2_dev_btree_bitmap_mark(struct bch_sb_field_members_v2 * mi,unsigned dev,u64 start,unsigned sectors)481 static void __bch2_dev_btree_bitmap_mark(struct bch_sb_field_members_v2 *mi, unsigned dev,
482 				u64 start, unsigned sectors)
483 {
484 	struct bch_member *m = __bch2_members_v2_get_mut(mi, dev);
485 	u64 bitmap = le64_to_cpu(m->btree_allocated_bitmap);
486 
487 	u64 end = start + sectors;
488 
489 	int resize = ilog2(roundup_pow_of_two(end)) - (m->btree_bitmap_shift + 6);
490 	if (resize > 0) {
491 		u64 new_bitmap = 0;
492 
493 		for (unsigned i = 0; i < 64; i++)
494 			if (bitmap & BIT_ULL(i))
495 				new_bitmap |= BIT_ULL(i >> resize);
496 		bitmap = new_bitmap;
497 		m->btree_bitmap_shift += resize;
498 	}
499 
500 	BUG_ON(m->btree_bitmap_shift >= BCH_MI_BTREE_BITMAP_SHIFT_MAX);
501 	BUG_ON(end > 64ULL << m->btree_bitmap_shift);
502 
503 	for (unsigned bit = start >> m->btree_bitmap_shift;
504 	     (u64) bit << m->btree_bitmap_shift < end;
505 	     bit++)
506 		bitmap |= BIT_ULL(bit);
507 
508 	m->btree_allocated_bitmap = cpu_to_le64(bitmap);
509 }
510 
bch2_dev_btree_bitmap_mark(struct bch_fs * c,struct bkey_s_c k)511 void bch2_dev_btree_bitmap_mark(struct bch_fs *c, struct bkey_s_c k)
512 {
513 	lockdep_assert_held(&c->sb_lock);
514 
515 	struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
516 	bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
517 		if (!bch2_member_exists(c->disk_sb.sb, ptr->dev))
518 			continue;
519 
520 		__bch2_dev_btree_bitmap_mark(mi, ptr->dev, ptr->offset, btree_sectors(c));
521 	}
522 }
523 
bch2_sb_nr_devices(const struct bch_sb * sb)524 unsigned bch2_sb_nr_devices(const struct bch_sb *sb)
525 {
526 	unsigned nr = 0;
527 
528 	for (unsigned i = 0; i < sb->nr_devices; i++)
529 		nr += bch2_member_exists((struct bch_sb *) sb, i);
530 	return nr;
531 }
532 
bch2_sb_member_alloc(struct bch_fs * c)533 int bch2_sb_member_alloc(struct bch_fs *c)
534 {
535 	unsigned dev_idx = c->sb.nr_devices;
536 	struct bch_sb_field_members_v2 *mi;
537 	unsigned nr_devices;
538 	unsigned u64s;
539 	int best = -1;
540 	u64 best_last_mount = 0;
541 	unsigned nr_deleted = 0;
542 
543 	if (dev_idx < BCH_SB_MEMBERS_MAX)
544 		goto have_slot;
545 
546 	for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) {
547 		/* eventually BCH_SB_MEMBERS_MAX will be raised */
548 		if (dev_idx == BCH_SB_MEMBER_INVALID)
549 			continue;
550 
551 		struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
552 
553 		nr_deleted += uuid_equal(&m.uuid, &BCH_SB_MEMBER_DELETED_UUID);
554 
555 		if (!bch2_is_zero(&m.uuid, sizeof(m.uuid)))
556 			continue;
557 
558 		u64 last_mount = le64_to_cpu(m.last_mount);
559 		if (best < 0 || last_mount < best_last_mount) {
560 			best = dev_idx;
561 			best_last_mount = last_mount;
562 		}
563 	}
564 	if (best >= 0) {
565 		dev_idx = best;
566 		goto have_slot;
567 	}
568 
569 	if (nr_deleted)
570 		bch_err(c, "unable to allocate new member, but have %u deleted: run fsck",
571 			nr_deleted);
572 
573 	return -BCH_ERR_ENOSPC_sb_members;
574 have_slot:
575 	nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
576 
577 	mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
578 	u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
579 			    le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
580 
581 	mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
582 	if (!mi)
583 		return -BCH_ERR_ENOSPC_sb_members;
584 
585 	c->disk_sb.sb->nr_devices = nr_devices;
586 	return dev_idx;
587 }
588 
bch2_sb_members_clean_deleted(struct bch_fs * c)589 void bch2_sb_members_clean_deleted(struct bch_fs *c)
590 {
591 	mutex_lock(&c->sb_lock);
592 	bool write_sb = false;
593 
594 	for (unsigned i = 0; i < c->sb.nr_devices; i++) {
595 		struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, i);
596 
597 		if (uuid_equal(&m->uuid, &BCH_SB_MEMBER_DELETED_UUID)) {
598 			memset(&m->uuid, 0, sizeof(m->uuid));
599 			write_sb = true;
600 		}
601 	}
602 
603 	if (write_sb)
604 		bch2_write_super(c);
605 	mutex_unlock(&c->sb_lock);
606 }
607