block-rsv.c 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534
  1. // SPDX-License-Identifier: GPL-2.0
  2. #include "misc.h"
  3. #include "ctree.h"
  4. #include "block-rsv.h"
  5. #include "space-info.h"
  6. #include "transaction.h"
  7. #include "block-group.h"
  8. /*
  9. * HOW DO BLOCK RESERVES WORK
  10. *
  11. * Think of block_rsv's as buckets for logically grouped metadata
  12. * reservations. Each block_rsv has a ->size and a ->reserved. ->size is
  13. * how large we want our block rsv to be, ->reserved is how much space is
  14. * currently reserved for this block reserve.
  15. *
  16. * ->failfast exists for the truncate case, and is described below.
  17. *
  18. * NORMAL OPERATION
  19. *
  20. * -> Reserve
  21. * Entrance: btrfs_block_rsv_add, btrfs_block_rsv_refill
  22. *
  23. * We call into btrfs_reserve_metadata_bytes() with our bytes, which is
  24. * accounted for in space_info->bytes_may_use, and then add the bytes to
  25. * ->reserved, and ->size in the case of btrfs_block_rsv_add.
  26. *
  27. * ->size is an over-estimation of how much we may use for a particular
  28. * operation.
  29. *
  30. * -> Use
  31. * Entrance: btrfs_use_block_rsv
  32. *
  33. * When we do a btrfs_alloc_tree_block() we call into btrfs_use_block_rsv()
  34. * to determine the appropriate block_rsv to use, and then verify that
  35. * ->reserved has enough space for our tree block allocation. Once
  36. * successful we subtract fs_info->nodesize from ->reserved.
  37. *
  38. * -> Finish
  39. * Entrance: btrfs_block_rsv_release
  40. *
  41. * We are finished with our operation, subtract our individual reservation
  42. * from ->size, and then subtract ->size from ->reserved and free up the
  43. * excess if there is any.
  44. *
  45. * There is some logic here to refill the delayed refs rsv or the global rsv
  46. * as needed, otherwise the excess is subtracted from
  47. * space_info->bytes_may_use.
  48. *
  49. * TYPES OF BLOCK RESERVES
  50. *
  51. * BLOCK_RSV_TRANS, BLOCK_RSV_DELOPS, BLOCK_RSV_CHUNK
  52. * These behave normally, as described above, just within the confines of the
  53. * lifetime of their particular operation (transaction for the whole trans
  54. * handle lifetime, for example).
  55. *
  56. * BLOCK_RSV_GLOBAL
  57. * It is impossible to properly account for all the space that may be required
  58. * to make our extent tree updates. This block reserve acts as an overflow
  59. * buffer in case our delayed refs reserve does not reserve enough space to
  60. * update the extent tree.
  61. *
  62. * We can steal from this in some cases as well, notably on evict() or
  63. * truncate() in order to help users recover from ENOSPC conditions.
  64. *
  65. * BLOCK_RSV_DELALLOC
  66. * The individual item sizes are determined by the per-inode size
  67. * calculations, which are described with the delalloc code. This is pretty
  68. * straightforward, it's just the calculation of ->size encodes a lot of
  69. * different items, and thus it gets used when updating inodes, inserting file
  70. * extents, and inserting checksums.
  71. *
  72. * BLOCK_RSV_DELREFS
  73. * We keep a running tally of how many delayed refs we have on the system.
  74. * We assume each one of these delayed refs are going to use a full
  75. * reservation. We use the transaction items and pre-reserve space for every
  76. * operation, and use this reservation to refill any gap between ->size and
  77. * ->reserved that may exist.
  78. *
  79. * From there it's straightforward, removing a delayed ref means we remove its
  80. * count from ->size and free up reservations as necessary. Since this is
  81. * the most dynamic block reserve in the system, we will try to refill this
  82. * block reserve first with any excess returned by any other block reserve.
  83. *
  84. * BLOCK_RSV_EMPTY
  85. * This is the fallback block reserve to make us try to reserve space if we
  86. * don't have a specific bucket for this allocation. It is mostly used for
  87. * updating the device tree and such, since that is a separate pool we're
  88. * content to just reserve space from the space_info on demand.
  89. *
  90. * BLOCK_RSV_TEMP
  91. * This is used by things like truncate and iput. We will temporarily
  92. * allocate a block reserve, set it to some size, and then truncate bytes
  93. * until we have no space left. With ->failfast set we'll simply return
  94. * ENOSPC from btrfs_use_block_rsv() to signal that we need to unwind and try
  95. * to make a new reservation. This is because these operations are
  96. * unbounded, so we want to do as much work as we can, and then back off and
  97. * re-reserve.
  98. */
  99. static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
  100. struct btrfs_block_rsv *block_rsv,
  101. struct btrfs_block_rsv *dest, u64 num_bytes,
  102. u64 *qgroup_to_release_ret)
  103. {
  104. struct btrfs_space_info *space_info = block_rsv->space_info;
  105. u64 qgroup_to_release = 0;
  106. u64 ret;
  107. spin_lock(&block_rsv->lock);
  108. if (num_bytes == (u64)-1) {
  109. num_bytes = block_rsv->size;
  110. qgroup_to_release = block_rsv->qgroup_rsv_size;
  111. }
  112. block_rsv->size -= num_bytes;
  113. if (block_rsv->reserved >= block_rsv->size) {
  114. num_bytes = block_rsv->reserved - block_rsv->size;
  115. block_rsv->reserved = block_rsv->size;
  116. block_rsv->full = 1;
  117. } else {
  118. num_bytes = 0;
  119. }
  120. if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
  121. qgroup_to_release = block_rsv->qgroup_rsv_reserved -
  122. block_rsv->qgroup_rsv_size;
  123. block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
  124. } else {
  125. qgroup_to_release = 0;
  126. }
  127. spin_unlock(&block_rsv->lock);
  128. ret = num_bytes;
  129. if (num_bytes > 0) {
  130. if (dest) {
  131. spin_lock(&dest->lock);
  132. if (!dest->full) {
  133. u64 bytes_to_add;
  134. bytes_to_add = dest->size - dest->reserved;
  135. bytes_to_add = min(num_bytes, bytes_to_add);
  136. dest->reserved += bytes_to_add;
  137. if (dest->reserved >= dest->size)
  138. dest->full = 1;
  139. num_bytes -= bytes_to_add;
  140. }
  141. spin_unlock(&dest->lock);
  142. }
  143. if (num_bytes)
  144. btrfs_space_info_free_bytes_may_use(fs_info,
  145. space_info,
  146. num_bytes);
  147. }
  148. if (qgroup_to_release_ret)
  149. *qgroup_to_release_ret = qgroup_to_release;
  150. return ret;
  151. }
  152. int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
  153. struct btrfs_block_rsv *dst, u64 num_bytes,
  154. bool update_size)
  155. {
  156. int ret;
  157. ret = btrfs_block_rsv_use_bytes(src, num_bytes);
  158. if (ret)
  159. return ret;
  160. btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
  161. return 0;
  162. }
  163. void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
  164. {
  165. memset(rsv, 0, sizeof(*rsv));
  166. spin_lock_init(&rsv->lock);
  167. rsv->type = type;
  168. }
  169. void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
  170. struct btrfs_block_rsv *rsv,
  171. unsigned short type)
  172. {
  173. btrfs_init_block_rsv(rsv, type);
  174. rsv->space_info = btrfs_find_space_info(fs_info,
  175. BTRFS_BLOCK_GROUP_METADATA);
  176. }
  177. struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
  178. unsigned short type)
  179. {
  180. struct btrfs_block_rsv *block_rsv;
  181. block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
  182. if (!block_rsv)
  183. return NULL;
  184. btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
  185. return block_rsv;
  186. }
  187. void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
  188. struct btrfs_block_rsv *rsv)
  189. {
  190. if (!rsv)
  191. return;
  192. btrfs_block_rsv_release(fs_info, rsv, (u64)-1, NULL);
  193. kfree(rsv);
  194. }
  195. int btrfs_block_rsv_add(struct btrfs_root *root,
  196. struct btrfs_block_rsv *block_rsv, u64 num_bytes,
  197. enum btrfs_reserve_flush_enum flush)
  198. {
  199. int ret;
  200. if (num_bytes == 0)
  201. return 0;
  202. ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
  203. if (!ret)
  204. btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
  205. return ret;
  206. }
  207. int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
  208. {
  209. u64 num_bytes = 0;
  210. int ret = -ENOSPC;
  211. if (!block_rsv)
  212. return 0;
  213. spin_lock(&block_rsv->lock);
  214. num_bytes = div_factor(block_rsv->size, min_factor);
  215. if (block_rsv->reserved >= num_bytes)
  216. ret = 0;
  217. spin_unlock(&block_rsv->lock);
  218. return ret;
  219. }
  220. int btrfs_block_rsv_refill(struct btrfs_root *root,
  221. struct btrfs_block_rsv *block_rsv, u64 min_reserved,
  222. enum btrfs_reserve_flush_enum flush)
  223. {
  224. u64 num_bytes = 0;
  225. int ret = -ENOSPC;
  226. if (!block_rsv)
  227. return 0;
  228. spin_lock(&block_rsv->lock);
  229. num_bytes = min_reserved;
  230. if (block_rsv->reserved >= num_bytes)
  231. ret = 0;
  232. else
  233. num_bytes -= block_rsv->reserved;
  234. spin_unlock(&block_rsv->lock);
  235. if (!ret)
  236. return 0;
  237. ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
  238. if (!ret) {
  239. btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false);
  240. return 0;
  241. }
  242. return ret;
  243. }
  244. u64 btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
  245. struct btrfs_block_rsv *block_rsv, u64 num_bytes,
  246. u64 *qgroup_to_release)
  247. {
  248. struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
  249. struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
  250. struct btrfs_block_rsv *target = NULL;
  251. /*
  252. * If we are the delayed_rsv then push to the global rsv, otherwise dump
  253. * into the delayed rsv if it is not full.
  254. */
  255. if (block_rsv == delayed_rsv)
  256. target = global_rsv;
  257. else if (block_rsv != global_rsv && !delayed_rsv->full)
  258. target = delayed_rsv;
  259. if (target && block_rsv->space_info != target->space_info)
  260. target = NULL;
  261. return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
  262. qgroup_to_release);
  263. }
  264. int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes)
  265. {
  266. int ret = -ENOSPC;
  267. spin_lock(&block_rsv->lock);
  268. if (block_rsv->reserved >= num_bytes) {
  269. block_rsv->reserved -= num_bytes;
  270. if (block_rsv->reserved < block_rsv->size)
  271. block_rsv->full = 0;
  272. ret = 0;
  273. }
  274. spin_unlock(&block_rsv->lock);
  275. return ret;
  276. }
  277. void btrfs_block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
  278. u64 num_bytes, bool update_size)
  279. {
  280. spin_lock(&block_rsv->lock);
  281. block_rsv->reserved += num_bytes;
  282. if (update_size)
  283. block_rsv->size += num_bytes;
  284. else if (block_rsv->reserved >= block_rsv->size)
  285. block_rsv->full = 1;
  286. spin_unlock(&block_rsv->lock);
  287. }
  288. int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
  289. struct btrfs_block_rsv *dest, u64 num_bytes,
  290. int min_factor)
  291. {
  292. struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
  293. u64 min_bytes;
  294. if (global_rsv->space_info != dest->space_info)
  295. return -ENOSPC;
  296. spin_lock(&global_rsv->lock);
  297. min_bytes = div_factor(global_rsv->size, min_factor);
  298. if (global_rsv->reserved < min_bytes + num_bytes) {
  299. spin_unlock(&global_rsv->lock);
  300. return -ENOSPC;
  301. }
  302. global_rsv->reserved -= num_bytes;
  303. if (global_rsv->reserved < global_rsv->size)
  304. global_rsv->full = 0;
  305. spin_unlock(&global_rsv->lock);
  306. btrfs_block_rsv_add_bytes(dest, num_bytes, true);
  307. return 0;
  308. }
  309. void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
  310. {
  311. struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
  312. struct btrfs_space_info *sinfo = block_rsv->space_info;
  313. u64 num_bytes;
  314. unsigned min_items;
  315. /*
  316. * The global block rsv is based on the size of the extent tree, the
  317. * checksum tree and the root tree. If the fs is empty we want to set
  318. * it to a minimal amount for safety.
  319. */
  320. num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
  321. btrfs_root_used(&fs_info->csum_root->root_item) +
  322. btrfs_root_used(&fs_info->tree_root->root_item);
  323. /*
  324. * We at a minimum are going to modify the csum root, the tree root, and
  325. * the extent root.
  326. */
  327. min_items = 3;
  328. /*
  329. * But we also want to reserve enough space so we can do the fallback
  330. * global reserve for an unlink, which is an additional 5 items (see the
  331. * comment in __unlink_start_trans for what we're modifying.)
  332. *
  333. * But we also need space for the delayed ref updates from the unlink,
  334. * so its 10, 5 for the actual operation, and 5 for the delayed ref
  335. * updates.
  336. */
  337. min_items += 10;
  338. num_bytes = max_t(u64, num_bytes,
  339. btrfs_calc_insert_metadata_size(fs_info, min_items));
  340. spin_lock(&sinfo->lock);
  341. spin_lock(&block_rsv->lock);
  342. block_rsv->size = min_t(u64, num_bytes, SZ_512M);
  343. if (block_rsv->reserved < block_rsv->size) {
  344. num_bytes = block_rsv->size - block_rsv->reserved;
  345. btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
  346. num_bytes);
  347. block_rsv->reserved = block_rsv->size;
  348. } else if (block_rsv->reserved > block_rsv->size) {
  349. num_bytes = block_rsv->reserved - block_rsv->size;
  350. btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
  351. -num_bytes);
  352. block_rsv->reserved = block_rsv->size;
  353. btrfs_try_granting_tickets(fs_info, sinfo);
  354. }
  355. if (block_rsv->reserved == block_rsv->size)
  356. block_rsv->full = 1;
  357. else
  358. block_rsv->full = 0;
  359. if (block_rsv->size >= sinfo->total_bytes)
  360. sinfo->force_alloc = CHUNK_ALLOC_FORCE;
  361. spin_unlock(&block_rsv->lock);
  362. spin_unlock(&sinfo->lock);
  363. }
  364. void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
  365. {
  366. struct btrfs_space_info *space_info;
  367. space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
  368. fs_info->chunk_block_rsv.space_info = space_info;
  369. space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
  370. fs_info->global_block_rsv.space_info = space_info;
  371. fs_info->trans_block_rsv.space_info = space_info;
  372. fs_info->empty_block_rsv.space_info = space_info;
  373. fs_info->delayed_block_rsv.space_info = space_info;
  374. fs_info->delayed_refs_rsv.space_info = space_info;
  375. fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
  376. fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
  377. fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
  378. fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
  379. if (fs_info->quota_root)
  380. fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
  381. fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
  382. btrfs_update_global_block_rsv(fs_info);
  383. }
  384. void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
  385. {
  386. btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1,
  387. NULL);
  388. WARN_ON(fs_info->trans_block_rsv.size > 0);
  389. WARN_ON(fs_info->trans_block_rsv.reserved > 0);
  390. WARN_ON(fs_info->chunk_block_rsv.size > 0);
  391. WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
  392. WARN_ON(fs_info->delayed_block_rsv.size > 0);
  393. WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
  394. WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
  395. WARN_ON(fs_info->delayed_refs_rsv.size > 0);
  396. }
  397. static struct btrfs_block_rsv *get_block_rsv(
  398. const struct btrfs_trans_handle *trans,
  399. const struct btrfs_root *root)
  400. {
  401. struct btrfs_fs_info *fs_info = root->fs_info;
  402. struct btrfs_block_rsv *block_rsv = NULL;
  403. if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
  404. (root == fs_info->csum_root && trans->adding_csums) ||
  405. (root == fs_info->uuid_root))
  406. block_rsv = trans->block_rsv;
  407. if (!block_rsv)
  408. block_rsv = root->block_rsv;
  409. if (!block_rsv)
  410. block_rsv = &fs_info->empty_block_rsv;
  411. return block_rsv;
  412. }
  413. struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
  414. struct btrfs_root *root,
  415. u32 blocksize)
  416. {
  417. struct btrfs_fs_info *fs_info = root->fs_info;
  418. struct btrfs_block_rsv *block_rsv;
  419. struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
  420. int ret;
  421. bool global_updated = false;
  422. block_rsv = get_block_rsv(trans, root);
  423. if (unlikely(block_rsv->size == 0))
  424. goto try_reserve;
  425. again:
  426. ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
  427. if (!ret)
  428. return block_rsv;
  429. if (block_rsv->failfast)
  430. return ERR_PTR(ret);
  431. if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
  432. global_updated = true;
  433. btrfs_update_global_block_rsv(fs_info);
  434. goto again;
  435. }
  436. /*
  437. * The global reserve still exists to save us from ourselves, so don't
  438. * warn_on if we are short on our delayed refs reserve.
  439. */
  440. if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
  441. btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
  442. static DEFINE_RATELIMIT_STATE(_rs,
  443. DEFAULT_RATELIMIT_INTERVAL * 10,
  444. /*DEFAULT_RATELIMIT_BURST*/ 1);
  445. if (__ratelimit(&_rs))
  446. WARN(1, KERN_DEBUG
  447. "BTRFS: block rsv %d returned %d\n",
  448. block_rsv->type, ret);
  449. }
  450. try_reserve:
  451. ret = btrfs_reserve_metadata_bytes(root, block_rsv, blocksize,
  452. BTRFS_RESERVE_NO_FLUSH);
  453. if (!ret)
  454. return block_rsv;
  455. /*
  456. * If we couldn't reserve metadata bytes try and use some from
  457. * the global reserve if its space type is the same as the global
  458. * reservation.
  459. */
  460. if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
  461. block_rsv->space_info == global_rsv->space_info) {
  462. ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);
  463. if (!ret)
  464. return global_rsv;
  465. }
  466. return ERR_PTR(ret);
  467. }