xref: /spdk/lib/blob/blobstore.h (revision 0d2745c94b03b159020b6812c6caddb4922e4449)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #ifndef SPDK_BLOBSTORE_H
35 #define SPDK_BLOBSTORE_H
36 
37 #include "spdk/assert.h"
38 #include "spdk/blob.h"
39 #include "spdk/queue.h"
40 #include "spdk/util.h"
41 
42 #include "request.h"
43 
44 /* In Memory Data Structures
45  *
46  * The following data structures exist only in memory.
47  */
48 
49 #define SPDK_BLOB_OPTS_CLUSTER_SZ (1024 * 1024)
50 #define SPDK_BLOB_OPTS_NUM_MD_PAGES UINT32_MAX
51 #define SPDK_BLOB_OPTS_MAX_MD_OPS 32
52 #define SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS 512
53 #define SPDK_BLOB_BLOBID_HIGH_BIT (1ULL << 32)
54 
55 struct spdk_xattr {
56 	/* TODO: reorder for best packing */
57 	uint32_t	index;
58 	char		*name;
59 	void		*value;
60 	uint16_t	value_len;
61 	TAILQ_ENTRY(spdk_xattr)	link;
62 };
63 
64 /* The mutable part of the blob data that is sync'd to
65  * disk. The data in here is both mutable and persistent.
66  */
67 struct spdk_blob_mut_data {
68 	/* Number of data clusters in the blob */
69 	uint64_t	num_clusters;
70 
71 	/* Array LBAs that are the beginning of a cluster, in
72 	 * the order they appear in the blob.
73 	 */
74 	uint64_t	*clusters;
75 
76 	/* The size of the clusters array. This is greater than or
77 	 * equal to 'num_clusters'.
78 	 */
79 	size_t		cluster_array_size;
80 
81 	/* Number of metadata pages */
82 	uint32_t	num_pages;
83 
84 	/* Array of page offsets into the metadata region, in
85 	 * the order of the metadata page sequence.
86 	 */
87 	uint32_t	*pages;
88 };
89 
90 enum spdk_blob_state {
91 	/* The blob in-memory version does not match the on-disk
92 	 * version.
93 	 */
94 	SPDK_BLOB_STATE_DIRTY,
95 
96 	/* The blob in memory version of the blob matches the on disk
97 	 * version.
98 	 */
99 	SPDK_BLOB_STATE_CLEAN,
100 
101 	/* The in-memory state being synchronized with the on-disk
102 	 * blob state. */
103 	SPDK_BLOB_STATE_LOADING,
104 };
105 
106 TAILQ_HEAD(spdk_xattr_tailq, spdk_xattr);
107 
108 struct spdk_blob_list {
109 	spdk_blob_id id;
110 	size_t clone_count;
111 	TAILQ_HEAD(, spdk_blob_list) clones;
112 	TAILQ_ENTRY(spdk_blob_list) link;
113 };
114 
115 struct spdk_blob {
116 	struct spdk_blob_store *bs;
117 
118 	uint32_t	open_ref;
119 
120 	spdk_blob_id	id;
121 	spdk_blob_id	parent_id;
122 
123 	enum spdk_blob_state		state;
124 
125 	/* Two copies of the mutable data. One is a version
126 	 * that matches the last known data on disk (clean).
127 	 * The other (active) is the current data. Syncing
128 	 * a blob makes the clean match the active.
129 	 */
130 	struct spdk_blob_mut_data	clean;
131 	struct spdk_blob_mut_data	active;
132 
133 	bool		invalid;
134 	bool		data_ro;
135 	bool		md_ro;
136 
137 	uint64_t	invalid_flags;
138 	uint64_t	data_ro_flags;
139 	uint64_t	md_ro_flags;
140 
141 	struct spdk_bs_dev *back_bs_dev;
142 
143 	/* TODO: The xattrs are mutable, but we don't want to be
144 	 * copying them unecessarily. Figure this out.
145 	 */
146 	struct spdk_xattr_tailq xattrs;
147 	struct spdk_xattr_tailq xattrs_internal;
148 
149 	TAILQ_ENTRY(spdk_blob) link;
150 
151 	uint32_t frozen_refcnt;
152 	bool resize_in_progress;
153 };
154 
155 struct spdk_blob_store {
156 	uint64_t			md_start; /* Offset from beginning of disk, in pages */
157 	uint32_t			md_len; /* Count, in pages */
158 
159 	struct spdk_io_channel		*md_channel;
160 	uint32_t			max_channel_ops;
161 
162 	struct spdk_thread		*md_thread;
163 
164 	struct spdk_bs_dev		*dev;
165 
166 	struct spdk_bit_array		*used_md_pages;
167 	struct spdk_bit_array		*used_clusters;
168 	struct spdk_bit_array		*used_blobids;
169 
170 	pthread_mutex_t			used_clusters_mutex;
171 
172 	uint32_t			cluster_sz;
173 	uint64_t			total_clusters;
174 	uint64_t			total_data_clusters;
175 	uint64_t			num_free_clusters;
176 	uint64_t			pages_per_cluster;
177 
178 	spdk_blob_id			super_blob;
179 	struct spdk_bs_type		bstype;
180 
181 	struct spdk_bs_cpl		unload_cpl;
182 	int				unload_err;
183 
184 	TAILQ_HEAD(, spdk_blob)		blobs;
185 	TAILQ_HEAD(, spdk_blob_list)	snapshots;
186 
187 	bool                            clean;
188 };
189 
190 struct spdk_bs_channel {
191 	struct spdk_bs_request_set	*req_mem;
192 	TAILQ_HEAD(, spdk_bs_request_set) reqs;
193 
194 	struct spdk_blob_store		*bs;
195 
196 	struct spdk_bs_dev		*dev;
197 	struct spdk_io_channel		*dev_channel;
198 
199 	TAILQ_HEAD(, spdk_bs_request_set) need_cluster_alloc;
200 	TAILQ_HEAD(, spdk_bs_request_set) queued_io;
201 };
202 
203 /** operation type */
204 enum spdk_blob_op_type {
205 	SPDK_BLOB_WRITE,
206 	SPDK_BLOB_READ,
207 	SPDK_BLOB_UNMAP,
208 	SPDK_BLOB_WRITE_ZEROES,
209 	SPDK_BLOB_WRITEV,
210 	SPDK_BLOB_READV,
211 };
212 
213 /* back bs_dev */
214 
215 #define BLOB_SNAPSHOT "SNAP"
216 #define SNAPSHOT_IN_PROGRESS "SNAPTMP"
217 
218 struct spdk_blob_bs_dev {
219 	struct spdk_bs_dev bs_dev;
220 	struct spdk_blob *blob;
221 };
222 
223 /* On-Disk Data Structures
224  *
225  * The following data structures exist on disk.
226  */
227 #define SPDK_BS_INITIAL_VERSION 1
228 #define SPDK_BS_VERSION 3 /* current version */
229 
230 #pragma pack(push, 1)
231 
232 #define SPDK_MD_MASK_TYPE_USED_PAGES 0
233 #define SPDK_MD_MASK_TYPE_USED_CLUSTERS 1
234 #define SPDK_MD_MASK_TYPE_USED_BLOBIDS 2
235 
236 struct spdk_bs_md_mask {
237 	uint8_t		type;
238 	uint32_t	length; /* In bits */
239 	uint8_t		mask[0];
240 };
241 
242 #define SPDK_MD_DESCRIPTOR_TYPE_PADDING 0
243 #define SPDK_MD_DESCRIPTOR_TYPE_EXTENT 1
244 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR 2
245 #define SPDK_MD_DESCRIPTOR_TYPE_FLAGS 3
246 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL 4
247 
248 struct spdk_blob_md_descriptor_xattr {
249 	uint8_t		type;
250 	uint32_t	length;
251 
252 	uint16_t	name_length;
253 	uint16_t	value_length;
254 
255 	char		name[0];
256 	/* String name immediately followed by string value. */
257 };
258 
259 struct spdk_blob_md_descriptor_extent {
260 	uint8_t		type;
261 	uint32_t	length;
262 
263 	struct {
264 		uint32_t        cluster_idx;
265 		uint32_t        length; /* In units of clusters */
266 	} extents[0];
267 };
268 
269 #define SPDK_BLOB_THIN_PROV (1ULL << 0)
270 #define SPDK_BLOB_INTERNAL_XATTR (1ULL << 1)
271 #define SPDK_BLOB_INVALID_FLAGS_MASK	(SPDK_BLOB_THIN_PROV | SPDK_BLOB_INTERNAL_XATTR)
272 
273 #define SPDK_BLOB_READ_ONLY (1ULL << 0)
274 #define SPDK_BLOB_DATA_RO_FLAGS_MASK	SPDK_BLOB_READ_ONLY
275 #define SPDK_BLOB_MD_RO_FLAGS_MASK	0
276 
277 struct spdk_blob_md_descriptor_flags {
278 	uint8_t		type;
279 	uint32_t	length;
280 
281 	/*
282 	 * If a flag in invalid_flags is set that the application is not aware of,
283 	 *  it will not allow the blob to be opened.
284 	 */
285 	uint64_t	invalid_flags;
286 
287 	/*
288 	 * If a flag in data_ro_flags is set that the application is not aware of,
289 	 *  allow the blob to be opened in data_read_only and md_read_only mode.
290 	 */
291 	uint64_t	data_ro_flags;
292 
293 	/*
294 	 * If a flag in md_ro_flags is set the the application is not aware of,
295 	 *  allow the blob to be opened in md_read_only mode.
296 	 */
297 	uint64_t	md_ro_flags;
298 };
299 
300 struct spdk_blob_md_descriptor {
301 	uint8_t		type;
302 	uint32_t	length;
303 };
304 
305 #define SPDK_INVALID_MD_PAGE UINT32_MAX
306 
307 struct spdk_blob_md_page {
308 	spdk_blob_id     id;
309 
310 	uint32_t        sequence_num;
311 	uint32_t	reserved0;
312 
313 	/* Descriptors here */
314 	uint8_t		descriptors[4072];
315 
316 	uint32_t	next;
317 	uint32_t	crc;
318 };
319 #define SPDK_BS_PAGE_SIZE 0x1000
320 SPDK_STATIC_ASSERT(SPDK_BS_PAGE_SIZE == sizeof(struct spdk_blob_md_page), "Invalid md page size");
321 
322 #define SPDK_BS_SUPER_BLOCK_SIG "SPDKBLOB"
323 
324 struct spdk_bs_super_block {
325 	uint8_t		signature[8];
326 	uint32_t        version;
327 	uint32_t        length;
328 	uint32_t	clean; /* If there was a clean shutdown, this is 1. */
329 	spdk_blob_id	super_blob;
330 
331 	uint32_t	cluster_size; /* In bytes */
332 
333 	uint32_t	used_page_mask_start; /* Offset from beginning of disk, in pages */
334 	uint32_t	used_page_mask_len; /* Count, in pages */
335 
336 	uint32_t	used_cluster_mask_start; /* Offset from beginning of disk, in pages */
337 	uint32_t	used_cluster_mask_len; /* Count, in pages */
338 
339 	uint32_t	md_start; /* Offset from beginning of disk, in pages */
340 	uint32_t	md_len; /* Count, in pages */
341 
342 	struct spdk_bs_type	bstype; /* blobstore type */
343 
344 	uint32_t	used_blobid_mask_start; /* Offset from beginning of disk, in pages */
345 	uint32_t	used_blobid_mask_len; /* Count, in pages */
346 
347 	uint64_t        size; /* size of blobstore in bytes */
348 
349 	uint8_t         reserved[4004];
350 	uint32_t	crc;
351 };
352 SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_super_block) == 0x1000, "Invalid super block size");
353 
354 #pragma pack(pop)
355 
356 struct spdk_bs_dev *spdk_bs_create_zeroes_dev(void);
357 struct spdk_bs_dev *spdk_bs_create_blob_bs_dev(struct spdk_blob *blob);
358 
359 /* Unit Conversions
360  *
361  * The blobstore works with several different units:
362  * - Byte: Self explanatory
363  * - LBA: The logical blocks on the backing storage device.
364  * - Page: The read/write units of blobs and metadata. This is
365  *         an offset into a blob in units of 4KiB.
366  * - Cluster Index: The disk is broken into a sequential list of
367  *		    clusters. This is the offset from the beginning.
368  *
369  * NOTE: These conversions all act on simple magnitudes, not with any sort
370  *        of knowledge about the blobs themselves. For instance, converting
371  *        a page to an lba with the conversion function below simply converts
372  *        a number of pages to an equivalent number of lbas, but that
373  *        lba certainly isn't the right lba that corresponds to a page offset
374  *        for a particular blob.
375  */
376 static inline uint64_t
377 _spdk_bs_byte_to_lba(struct spdk_blob_store *bs, uint64_t length)
378 {
379 	assert(length % bs->dev->blocklen == 0);
380 
381 	return length / bs->dev->blocklen;
382 }
383 
384 static inline uint64_t
385 _spdk_bs_dev_byte_to_lba(struct spdk_bs_dev *bs_dev, uint64_t length)
386 {
387 	assert(length % bs_dev->blocklen == 0);
388 
389 	return length / bs_dev->blocklen;
390 }
391 
392 static inline uint64_t
393 _spdk_bs_lba_to_byte(struct spdk_blob_store *bs, uint64_t lba)
394 {
395 	return lba * bs->dev->blocklen;
396 }
397 
398 static inline uint64_t
399 _spdk_bs_page_to_lba(struct spdk_blob_store *bs, uint64_t page)
400 {
401 	return page * SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
402 }
403 
404 static inline uint64_t
405 _spdk_bs_dev_page_to_lba(struct spdk_bs_dev *bs_dev, uint64_t page)
406 {
407 	return page * SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
408 }
409 
410 static inline uint64_t
411 _spdk_bs_lba_to_page(struct spdk_blob_store *bs, uint64_t lba)
412 {
413 	uint64_t	lbas_per_page;
414 
415 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
416 
417 	assert(lba % lbas_per_page == 0);
418 
419 	return lba / lbas_per_page;
420 }
421 
422 static inline uint64_t
423 _spdk_bs_dev_lba_to_page(struct spdk_bs_dev *bs_dev, uint64_t lba)
424 {
425 	uint64_t	lbas_per_page;
426 
427 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
428 
429 	assert(lba % lbas_per_page == 0);
430 
431 	return lba / lbas_per_page;
432 }
433 
434 static inline uint64_t
435 _spdk_bs_cluster_to_page(struct spdk_blob_store *bs, uint32_t cluster)
436 {
437 	return (uint64_t)cluster * bs->pages_per_cluster;
438 }
439 
440 static inline uint32_t
441 _spdk_bs_page_to_cluster(struct spdk_blob_store *bs, uint64_t page)
442 {
443 	assert(page % bs->pages_per_cluster == 0);
444 
445 	return page / bs->pages_per_cluster;
446 }
447 
448 static inline uint64_t
449 _spdk_bs_cluster_to_lba(struct spdk_blob_store *bs, uint32_t cluster)
450 {
451 	return (uint64_t)cluster * (bs->cluster_sz / bs->dev->blocklen);
452 }
453 
454 static inline uint32_t
455 _spdk_bs_lba_to_cluster(struct spdk_blob_store *bs, uint64_t lba)
456 {
457 	assert(lba % (bs->cluster_sz / bs->dev->blocklen) == 0);
458 
459 	return lba / (bs->cluster_sz / bs->dev->blocklen);
460 }
461 
462 static inline uint64_t
463 _spdk_bs_blob_lba_to_back_dev_lba(struct spdk_blob *blob, uint64_t lba)
464 {
465 	return lba * blob->bs->dev->blocklen / blob->back_bs_dev->blocklen;
466 }
467 
468 static inline uint64_t
469 _spdk_bs_blob_lba_from_back_dev_lba(struct spdk_blob *blob, uint64_t lba)
470 {
471 	return lba * blob->back_bs_dev->blocklen / blob->bs->dev->blocklen;
472 }
473 
474 /* End basic conversions */
475 
476 static inline uint64_t
477 _spdk_bs_blobid_to_page(spdk_blob_id id)
478 {
479 	return id & 0xFFFFFFFF;
480 }
481 
482 /* The blob id is a 64 bit number. The lower 32 bits are the page_idx. The upper
483  * 32 bits are not currently used. Stick a 1 there just to catch bugs where the
484  * code assumes blob id == page_idx.
485  */
486 static inline spdk_blob_id
487 _spdk_bs_page_to_blobid(uint64_t page_idx)
488 {
489 	if (page_idx > UINT32_MAX) {
490 		return SPDK_BLOBID_INVALID;
491 	}
492 	return SPDK_BLOB_BLOBID_HIGH_BIT | page_idx;
493 }
494 
495 /* Given a page offset into a blob, look up the LBA for the
496  * start of that page.
497  */
498 static inline uint64_t
499 _spdk_bs_blob_page_to_lba(struct spdk_blob *blob, uint64_t page)
500 {
501 	uint64_t	lba;
502 	uint64_t	pages_per_cluster;
503 
504 	pages_per_cluster = blob->bs->pages_per_cluster;
505 
506 	assert(page < blob->active.num_clusters * pages_per_cluster);
507 
508 	lba = blob->active.clusters[page / pages_per_cluster];
509 	lba += _spdk_bs_page_to_lba(blob->bs, page % pages_per_cluster);
510 
511 	return lba;
512 }
513 
514 /* Given a page offset into a blob, look up the number of pages until the
515  * next cluster boundary.
516  */
517 static inline uint32_t
518 _spdk_bs_num_pages_to_cluster_boundary(struct spdk_blob *blob, uint64_t page)
519 {
520 	uint64_t	pages_per_cluster;
521 
522 	pages_per_cluster = blob->bs->pages_per_cluster;
523 
524 	return pages_per_cluster - (page % pages_per_cluster);
525 }
526 
527 /* Given a page offset into a blob, look up the number of pages into blob to beginning of current cluster */
528 static inline uint32_t
529 _spdk_bs_page_to_cluster_start(struct spdk_blob *blob, uint64_t page)
530 {
531 	uint64_t	pages_per_cluster;
532 
533 	pages_per_cluster = blob->bs->pages_per_cluster;
534 
535 	return page - (page % pages_per_cluster);
536 }
537 
538 /* Given a page offset into a blob, look up if it is from allocated cluster. */
539 static inline bool
540 _spdk_bs_page_is_allocated(struct spdk_blob *blob, uint64_t page)
541 {
542 	uint64_t	lba;
543 	uint64_t	pages_per_cluster;
544 
545 	pages_per_cluster = blob->bs->pages_per_cluster;
546 
547 	assert(page < blob->active.num_clusters * pages_per_cluster);
548 
549 	lba = blob->active.clusters[page / pages_per_cluster];
550 
551 	if (lba == 0) {
552 		assert(spdk_blob_is_thin_provisioned(blob));
553 		return false;
554 	} else {
555 		return true;
556 	}
557 }
558 
559 #endif
560