xref: /openbmc/linux/fs/btrfs/block-rsv.c (revision 005f3e67)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "misc.h"
4 #include "ctree.h"
5 #include "block-rsv.h"
6 #include "space-info.h"
7 #include "transaction.h"
8 
9 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
10 				    struct btrfs_block_rsv *block_rsv,
11 				    struct btrfs_block_rsv *dest, u64 num_bytes,
12 				    u64 *qgroup_to_release_ret)
13 {
14 	struct btrfs_space_info *space_info = block_rsv->space_info;
15 	u64 qgroup_to_release = 0;
16 	u64 ret;
17 
18 	spin_lock(&block_rsv->lock);
19 	if (num_bytes == (u64)-1) {
20 		num_bytes = block_rsv->size;
21 		qgroup_to_release = block_rsv->qgroup_rsv_size;
22 	}
23 	block_rsv->size -= num_bytes;
24 	if (block_rsv->reserved >= block_rsv->size) {
25 		num_bytes = block_rsv->reserved - block_rsv->size;
26 		block_rsv->reserved = block_rsv->size;
27 		block_rsv->full = 1;
28 	} else {
29 		num_bytes = 0;
30 	}
31 	if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
32 		qgroup_to_release = block_rsv->qgroup_rsv_reserved -
33 				    block_rsv->qgroup_rsv_size;
34 		block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
35 	} else {
36 		qgroup_to_release = 0;
37 	}
38 	spin_unlock(&block_rsv->lock);
39 
40 	ret = num_bytes;
41 	if (num_bytes > 0) {
42 		if (dest) {
43 			spin_lock(&dest->lock);
44 			if (!dest->full) {
45 				u64 bytes_to_add;
46 
47 				bytes_to_add = dest->size - dest->reserved;
48 				bytes_to_add = min(num_bytes, bytes_to_add);
49 				dest->reserved += bytes_to_add;
50 				if (dest->reserved >= dest->size)
51 					dest->full = 1;
52 				num_bytes -= bytes_to_add;
53 			}
54 			spin_unlock(&dest->lock);
55 		}
56 		if (num_bytes)
57 			btrfs_space_info_free_bytes_may_use(fs_info,
58 							    space_info,
59 							    num_bytes);
60 	}
61 	if (qgroup_to_release_ret)
62 		*qgroup_to_release_ret = qgroup_to_release;
63 	return ret;
64 }
65 
66 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
67 			    struct btrfs_block_rsv *dst, u64 num_bytes,
68 			    bool update_size)
69 {
70 	int ret;
71 
72 	ret = btrfs_block_rsv_use_bytes(src, num_bytes);
73 	if (ret)
74 		return ret;
75 
76 	btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
77 	return 0;
78 }
79 
80 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
81 {
82 	memset(rsv, 0, sizeof(*rsv));
83 	spin_lock_init(&rsv->lock);
84 	rsv->type = type;
85 }
86 
87 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
88 				   struct btrfs_block_rsv *rsv,
89 				   unsigned short type)
90 {
91 	btrfs_init_block_rsv(rsv, type);
92 	rsv->space_info = btrfs_find_space_info(fs_info,
93 					    BTRFS_BLOCK_GROUP_METADATA);
94 }
95 
96 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
97 					      unsigned short type)
98 {
99 	struct btrfs_block_rsv *block_rsv;
100 
101 	block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
102 	if (!block_rsv)
103 		return NULL;
104 
105 	btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
106 	return block_rsv;
107 }
108 
109 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
110 			  struct btrfs_block_rsv *rsv)
111 {
112 	if (!rsv)
113 		return;
114 	btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
115 	kfree(rsv);
116 }
117 
118 int btrfs_block_rsv_add(struct btrfs_root *root,
119 			struct btrfs_block_rsv *block_rsv, u64 num_bytes,
120 			enum btrfs_reserve_flush_enum flush)
121 {
122 	int ret;
123 
124 	if (num_bytes == 0)
125 		return 0;
126 
127 	ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
128 	if (!ret)
129 		btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
130 
131 	return ret;
132 }
133 
134 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
135 {
136 	u64 num_bytes = 0;
137 	int ret = -ENOSPC;
138 
139 	if (!block_rsv)
140 		return 0;
141 
142 	spin_lock(&block_rsv->lock);
143 	num_bytes = div_factor(block_rsv->size, min_factor);
144 	if (block_rsv->reserved >= num_bytes)
145 		ret = 0;
146 	spin_unlock(&block_rsv->lock);
147 
148 	return ret;
149 }
150 
151 int btrfs_block_rsv_refill(struct btrfs_root *root,
152 			   struct btrfs_block_rsv *block_rsv, u64 min_reserved,
153 			   enum btrfs_reserve_flush_enum flush)
154 {
155 	u64 num_bytes = 0;
156 	int ret = -ENOSPC;
157 
158 	if (!block_rsv)
159 		return 0;
160 
161 	spin_lock(&block_rsv->lock);
162 	num_bytes = min_reserved;
163 	if (block_rsv->reserved >= num_bytes)
164 		ret = 0;
165 	else
166 		num_bytes -= block_rsv->reserved;
167 	spin_unlock(&block_rsv->lock);
168 
169 	if (!ret)
170 		return 0;
171 
172 	ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
173 	if (!ret) {
174 		btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false);
175 		return 0;
176 	}
177 
178 	return ret;
179 }
180 
181 u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
182 			      struct btrfs_block_rsv *block_rsv,
183 			      u64 num_bytes, u64 *qgroup_to_release)
184 {
185 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
186 	struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
187 	struct btrfs_block_rsv *target = NULL;
188 
189 	/*
190 	 * If we are the delayed_rsv then push to the global rsv, otherwise dump
191 	 * into the delayed rsv if it is not full.
192 	 */
193 	if (block_rsv == delayed_rsv)
194 		target = global_rsv;
195 	else if (block_rsv != global_rsv && !delayed_rsv->full)
196 		target = delayed_rsv;
197 
198 	if (target && block_rsv->space_info != target->space_info)
199 		target = NULL;
200 
201 	return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
202 				       qgroup_to_release);
203 }
204 
205 int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes)
206 {
207 	int ret = -ENOSPC;
208 
209 	spin_lock(&block_rsv->lock);
210 	if (block_rsv->reserved >= num_bytes) {
211 		block_rsv->reserved -= num_bytes;
212 		if (block_rsv->reserved < block_rsv->size)
213 			block_rsv->full = 0;
214 		ret = 0;
215 	}
216 	spin_unlock(&block_rsv->lock);
217 	return ret;
218 }
219 
220 void btrfs_block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
221 			       u64 num_bytes, bool update_size)
222 {
223 	spin_lock(&block_rsv->lock);
224 	block_rsv->reserved += num_bytes;
225 	if (update_size)
226 		block_rsv->size += num_bytes;
227 	else if (block_rsv->reserved >= block_rsv->size)
228 		block_rsv->full = 1;
229 	spin_unlock(&block_rsv->lock);
230 }
231 
232 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
233 			     struct btrfs_block_rsv *dest, u64 num_bytes,
234 			     int min_factor)
235 {
236 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
237 	u64 min_bytes;
238 
239 	if (global_rsv->space_info != dest->space_info)
240 		return -ENOSPC;
241 
242 	spin_lock(&global_rsv->lock);
243 	min_bytes = div_factor(global_rsv->size, min_factor);
244 	if (global_rsv->reserved < min_bytes + num_bytes) {
245 		spin_unlock(&global_rsv->lock);
246 		return -ENOSPC;
247 	}
248 	global_rsv->reserved -= num_bytes;
249 	if (global_rsv->reserved < global_rsv->size)
250 		global_rsv->full = 0;
251 	spin_unlock(&global_rsv->lock);
252 
253 	btrfs_block_rsv_add_bytes(dest, num_bytes, true);
254 	return 0;
255 }
256 
257 void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
258 {
259 	struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
260 	struct btrfs_space_info *sinfo = block_rsv->space_info;
261 	u64 num_bytes;
262 	unsigned min_items;
263 
264 	/*
265 	 * The global block rsv is based on the size of the extent tree, the
266 	 * checksum tree and the root tree.  If the fs is empty we want to set
267 	 * it to a minimal amount for safety.
268 	 */
269 	num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
270 		btrfs_root_used(&fs_info->csum_root->root_item) +
271 		btrfs_root_used(&fs_info->tree_root->root_item);
272 
273 	/*
274 	 * We at a minimum are going to modify the csum root, the tree root, and
275 	 * the extent root.
276 	 */
277 	min_items = 3;
278 
279 	/*
280 	 * But we also want to reserve enough space so we can do the fallback
281 	 * global reserve for an unlink, which is an additional 5 items (see the
282 	 * comment in __unlink_start_trans for what we're modifying.)
283 	 *
284 	 * But we also need space for the delayed ref updates from the unlink,
285 	 * so its 10, 5 for the actual operation, and 5 for the delayed ref
286 	 * updates.
287 	 */
288 	min_items += 10;
289 
290 	num_bytes = max_t(u64, num_bytes,
291 			  btrfs_calc_insert_metadata_size(fs_info, min_items));
292 
293 	spin_lock(&sinfo->lock);
294 	spin_lock(&block_rsv->lock);
295 
296 	block_rsv->size = min_t(u64, num_bytes, SZ_512M);
297 
298 	if (block_rsv->reserved < block_rsv->size) {
299 		num_bytes = block_rsv->size - block_rsv->reserved;
300 		block_rsv->reserved += num_bytes;
301 		btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
302 						      num_bytes);
303 	} else if (block_rsv->reserved > block_rsv->size) {
304 		num_bytes = block_rsv->reserved - block_rsv->size;
305 		btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
306 						      -num_bytes);
307 		block_rsv->reserved = block_rsv->size;
308 		btrfs_try_granting_tickets(fs_info, sinfo);
309 	}
310 
311 	if (block_rsv->reserved == block_rsv->size)
312 		block_rsv->full = 1;
313 	else
314 		block_rsv->full = 0;
315 
316 	spin_unlock(&block_rsv->lock);
317 	spin_unlock(&sinfo->lock);
318 }
319 
320 void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
321 {
322 	struct btrfs_space_info *space_info;
323 
324 	space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
325 	fs_info->chunk_block_rsv.space_info = space_info;
326 
327 	space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
328 	fs_info->global_block_rsv.space_info = space_info;
329 	fs_info->trans_block_rsv.space_info = space_info;
330 	fs_info->empty_block_rsv.space_info = space_info;
331 	fs_info->delayed_block_rsv.space_info = space_info;
332 	fs_info->delayed_refs_rsv.space_info = space_info;
333 
334 	fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
335 	fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
336 	fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
337 	fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
338 	if (fs_info->quota_root)
339 		fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
340 	fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
341 
342 	btrfs_update_global_block_rsv(fs_info);
343 }
344 
345 void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
346 {
347 	btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1);
348 	WARN_ON(fs_info->trans_block_rsv.size > 0);
349 	WARN_ON(fs_info->trans_block_rsv.reserved > 0);
350 	WARN_ON(fs_info->chunk_block_rsv.size > 0);
351 	WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
352 	WARN_ON(fs_info->delayed_block_rsv.size > 0);
353 	WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
354 	WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
355 	WARN_ON(fs_info->delayed_refs_rsv.size > 0);
356 }
357 
358 static struct btrfs_block_rsv *get_block_rsv(
359 					const struct btrfs_trans_handle *trans,
360 					const struct btrfs_root *root)
361 {
362 	struct btrfs_fs_info *fs_info = root->fs_info;
363 	struct btrfs_block_rsv *block_rsv = NULL;
364 
365 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
366 	    (root == fs_info->csum_root && trans->adding_csums) ||
367 	    (root == fs_info->uuid_root))
368 		block_rsv = trans->block_rsv;
369 
370 	if (!block_rsv)
371 		block_rsv = root->block_rsv;
372 
373 	if (!block_rsv)
374 		block_rsv = &fs_info->empty_block_rsv;
375 
376 	return block_rsv;
377 }
378 
379 struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
380 					    struct btrfs_root *root,
381 					    u32 blocksize)
382 {
383 	struct btrfs_fs_info *fs_info = root->fs_info;
384 	struct btrfs_block_rsv *block_rsv;
385 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
386 	int ret;
387 	bool global_updated = false;
388 
389 	block_rsv = get_block_rsv(trans, root);
390 
391 	if (unlikely(block_rsv->size == 0))
392 		goto try_reserve;
393 again:
394 	ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
395 	if (!ret)
396 		return block_rsv;
397 
398 	if (block_rsv->failfast)
399 		return ERR_PTR(ret);
400 
401 	if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
402 		global_updated = true;
403 		btrfs_update_global_block_rsv(fs_info);
404 		goto again;
405 	}
406 
407 	/*
408 	 * The global reserve still exists to save us from ourselves, so don't
409 	 * warn_on if we are short on our delayed refs reserve.
410 	 */
411 	if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
412 	    btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
413 		static DEFINE_RATELIMIT_STATE(_rs,
414 				DEFAULT_RATELIMIT_INTERVAL * 10,
415 				/*DEFAULT_RATELIMIT_BURST*/ 1);
416 		if (__ratelimit(&_rs))
417 			WARN(1, KERN_DEBUG
418 				"BTRFS: block rsv returned %d\n", ret);
419 	}
420 try_reserve:
421 	ret = btrfs_reserve_metadata_bytes(root, block_rsv, blocksize,
422 					   BTRFS_RESERVE_NO_FLUSH);
423 	if (!ret)
424 		return block_rsv;
425 	/*
426 	 * If we couldn't reserve metadata bytes try and use some from
427 	 * the global reserve if its space type is the same as the global
428 	 * reservation.
429 	 */
430 	if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
431 	    block_rsv->space_info == global_rsv->space_info) {
432 		ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);
433 		if (!ret)
434 			return global_rsv;
435 	}
436 	return ERR_PTR(ret);
437 }
438