xref: /spdk/lib/blob/blobstore.h (revision da7673f71f15e27a7853ad18f3f07058d73e7274)
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 
152 struct spdk_blob_store {
153 	uint64_t			md_start; /* Offset from beginning of disk, in pages */
154 	uint32_t			md_len; /* Count, in pages */
155 
156 	struct spdk_io_channel		*md_channel;
157 	uint32_t			max_channel_ops;
158 
159 	struct spdk_thread		*md_thread;
160 
161 	struct spdk_bs_dev		*dev;
162 
163 	struct spdk_bit_array		*used_md_pages;
164 	struct spdk_bit_array		*used_clusters;
165 	struct spdk_bit_array		*used_blobids;
166 
167 	pthread_mutex_t			used_clusters_mutex;
168 
169 	uint32_t			cluster_sz;
170 	uint64_t			total_clusters;
171 	uint64_t			total_data_clusters;
172 	uint64_t			num_free_clusters;
173 	uint32_t			pages_per_cluster;
174 
175 	spdk_blob_id			super_blob;
176 	struct spdk_bs_type		bstype;
177 
178 	struct spdk_bs_cpl		unload_cpl;
179 	int				unload_err;
180 
181 	TAILQ_HEAD(, spdk_blob)		blobs;
182 	TAILQ_HEAD(, spdk_blob_list)	snapshots;
183 
184 	bool                            clean;
185 };
186 
187 struct spdk_bs_channel {
188 	struct spdk_bs_request_set	*req_mem;
189 	TAILQ_HEAD(, spdk_bs_request_set) reqs;
190 
191 	struct spdk_blob_store		*bs;
192 
193 	struct spdk_bs_dev		*dev;
194 	struct spdk_io_channel		*dev_channel;
195 
196 	TAILQ_HEAD(, spdk_bs_request_set) need_cluster_alloc;
197 };
198 
199 /** operation type */
200 enum spdk_blob_op_type {
201 	SPDK_BLOB_WRITE,
202 	SPDK_BLOB_READ,
203 	SPDK_BLOB_UNMAP,
204 	SPDK_BLOB_WRITE_ZEROES,
205 	SPDK_BLOB_WRITEV,
206 	SPDK_BLOB_READV,
207 };
208 
209 /* back bs_dev */
210 
211 #define BLOB_SNAPSHOT "SNAP"
212 #define SNAPSHOT_IN_PROGRESS "SNAPTMP"
213 
214 struct spdk_blob_bs_dev {
215 	struct spdk_bs_dev bs_dev;
216 	struct spdk_blob *blob;
217 };
218 
219 /* On-Disk Data Structures
220  *
221  * The following data structures exist on disk.
222  */
223 #define SPDK_BS_INITIAL_VERSION 1
224 #define SPDK_BS_VERSION 3 /* current version */
225 
226 #pragma pack(push, 1)
227 
228 #define SPDK_MD_MASK_TYPE_USED_PAGES 0
229 #define SPDK_MD_MASK_TYPE_USED_CLUSTERS 1
230 #define SPDK_MD_MASK_TYPE_USED_BLOBIDS 2
231 
232 struct spdk_bs_md_mask {
233 	uint8_t		type;
234 	uint32_t	length; /* In bits */
235 	uint8_t		mask[0];
236 };
237 
238 #define SPDK_MD_DESCRIPTOR_TYPE_PADDING 0
239 #define SPDK_MD_DESCRIPTOR_TYPE_EXTENT 1
240 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR 2
241 #define SPDK_MD_DESCRIPTOR_TYPE_FLAGS 3
242 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL 4
243 
244 struct spdk_blob_md_descriptor_xattr {
245 	uint8_t		type;
246 	uint32_t	length;
247 
248 	uint16_t	name_length;
249 	uint16_t	value_length;
250 
251 	char		name[0];
252 	/* String name immediately followed by string value. */
253 };
254 
255 struct spdk_blob_md_descriptor_extent {
256 	uint8_t		type;
257 	uint32_t	length;
258 
259 	struct {
260 		uint32_t        cluster_idx;
261 		uint32_t        length; /* In units of clusters */
262 	} extents[0];
263 };
264 
265 #define SPDK_BLOB_THIN_PROV (1ULL << 0)
266 #define SPDK_BLOB_INTERNAL_XATTR (1ULL << 1)
267 #define SPDK_BLOB_INVALID_FLAGS_MASK	(SPDK_BLOB_THIN_PROV | SPDK_BLOB_INTERNAL_XATTR)
268 
269 #define SPDK_BLOB_READ_ONLY (1ULL << 0)
270 #define SPDK_BLOB_DATA_RO_FLAGS_MASK	SPDK_BLOB_READ_ONLY
271 #define SPDK_BLOB_MD_RO_FLAGS_MASK	0
272 
273 struct spdk_blob_md_descriptor_flags {
274 	uint8_t		type;
275 	uint32_t	length;
276 
277 	/*
278 	 * If a flag in invalid_flags is set that the application is not aware of,
279 	 *  it will not allow the blob to be opened.
280 	 */
281 	uint64_t	invalid_flags;
282 
283 	/*
284 	 * If a flag in data_ro_flags is set that the application is not aware of,
285 	 *  allow the blob to be opened in data_read_only and md_read_only mode.
286 	 */
287 	uint64_t	data_ro_flags;
288 
289 	/*
290 	 * If a flag in md_ro_flags is set the the application is not aware of,
291 	 *  allow the blob to be opened in md_read_only mode.
292 	 */
293 	uint64_t	md_ro_flags;
294 };
295 
296 struct spdk_blob_md_descriptor {
297 	uint8_t		type;
298 	uint32_t	length;
299 };
300 
301 #define SPDK_INVALID_MD_PAGE UINT32_MAX
302 
303 struct spdk_blob_md_page {
304 	spdk_blob_id     id;
305 
306 	uint32_t        sequence_num;
307 	uint32_t	reserved0;
308 
309 	/* Descriptors here */
310 	uint8_t		descriptors[4072];
311 
312 	uint32_t	next;
313 	uint32_t	crc;
314 };
315 #define SPDK_BS_PAGE_SIZE 0x1000
316 SPDK_STATIC_ASSERT(SPDK_BS_PAGE_SIZE == sizeof(struct spdk_blob_md_page), "Invalid md page size");
317 
318 #define SPDK_BS_SUPER_BLOCK_SIG "SPDKBLOB"
319 
320 struct spdk_bs_super_block {
321 	uint8_t		signature[8];
322 	uint32_t        version;
323 	uint32_t        length;
324 	uint32_t	clean; /* If there was a clean shutdown, this is 1. */
325 	spdk_blob_id	super_blob;
326 
327 	uint32_t	cluster_size; /* In bytes */
328 
329 	uint32_t	used_page_mask_start; /* Offset from beginning of disk, in pages */
330 	uint32_t	used_page_mask_len; /* Count, in pages */
331 
332 	uint32_t	used_cluster_mask_start; /* Offset from beginning of disk, in pages */
333 	uint32_t	used_cluster_mask_len; /* Count, in pages */
334 
335 	uint32_t	md_start; /* Offset from beginning of disk, in pages */
336 	uint32_t	md_len; /* Count, in pages */
337 
338 	struct spdk_bs_type	bstype; /* blobstore type */
339 
340 	uint32_t	used_blobid_mask_start; /* Offset from beginning of disk, in pages */
341 	uint32_t	used_blobid_mask_len; /* Count, in pages */
342 
343 	uint8_t		reserved[4012];
344 	uint32_t	crc;
345 };
346 SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_super_block) == 0x1000, "Invalid super block size");
347 
348 #pragma pack(pop)
349 
350 struct spdk_bs_dev *spdk_bs_create_zeroes_dev(void);
351 struct spdk_bs_dev *spdk_bs_create_blob_bs_dev(struct spdk_blob *blob);
352 
353 /* Unit Conversions
354  *
355  * The blobstore works with several different units:
356  * - Byte: Self explanatory
357  * - LBA: The logical blocks on the backing storage device.
358  * - Page: The read/write units of blobs and metadata. This is
359  *         an offset into a blob in units of 4KiB.
360  * - Cluster Index: The disk is broken into a sequential list of
361  *		    clusters. This is the offset from the beginning.
362  *
363  * NOTE: These conversions all act on simple magnitudes, not with any sort
364  *        of knowledge about the blobs themselves. For instance, converting
365  *        a page to an lba with the conversion function below simply converts
366  *        a number of pages to an equivalent number of lbas, but that
367  *        lba certainly isn't the right lba that corresponds to a page offset
368  *        for a particular blob.
369  */
370 static inline uint64_t
371 _spdk_bs_byte_to_lba(struct spdk_blob_store *bs, uint64_t length)
372 {
373 	assert(length % bs->dev->blocklen == 0);
374 
375 	return length / bs->dev->blocklen;
376 }
377 
378 static inline uint64_t
379 _spdk_bs_dev_byte_to_lba(struct spdk_bs_dev *bs_dev, uint64_t length)
380 {
381 	assert(length % bs_dev->blocklen == 0);
382 
383 	return length / bs_dev->blocklen;
384 }
385 
386 static inline uint64_t
387 _spdk_bs_lba_to_byte(struct spdk_blob_store *bs, uint64_t lba)
388 {
389 	return lba * bs->dev->blocklen;
390 }
391 
392 static inline uint64_t
393 _spdk_bs_page_to_lba(struct spdk_blob_store *bs, uint64_t page)
394 {
395 	return page * SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
396 }
397 
398 static inline uint64_t
399 _spdk_bs_dev_page_to_lba(struct spdk_bs_dev *bs_dev, uint64_t page)
400 {
401 	return page * SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
402 }
403 
404 static inline uint32_t
405 _spdk_bs_lba_to_page(struct spdk_blob_store *bs, uint64_t lba)
406 {
407 	uint64_t	lbas_per_page;
408 
409 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
410 
411 	assert(lba % lbas_per_page == 0);
412 
413 	return lba / lbas_per_page;
414 }
415 
416 static inline uint64_t
417 _spdk_bs_dev_lba_to_page(struct spdk_bs_dev *bs_dev, uint64_t lba)
418 {
419 	uint64_t	lbas_per_page;
420 
421 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
422 
423 	assert(lba % lbas_per_page == 0);
424 
425 	return lba / lbas_per_page;
426 }
427 
428 static inline uint64_t
429 _spdk_bs_cluster_to_page(struct spdk_blob_store *bs, uint32_t cluster)
430 {
431 	return cluster * bs->pages_per_cluster;
432 }
433 
434 static inline uint32_t
435 _spdk_bs_page_to_cluster(struct spdk_blob_store *bs, uint64_t page)
436 {
437 	assert(page % bs->pages_per_cluster == 0);
438 
439 	return page / bs->pages_per_cluster;
440 }
441 
442 static inline uint64_t
443 _spdk_bs_cluster_to_lba(struct spdk_blob_store *bs, uint32_t cluster)
444 {
445 	return cluster * (bs->cluster_sz / bs->dev->blocklen);
446 }
447 
448 static inline uint32_t
449 _spdk_bs_lba_to_cluster(struct spdk_blob_store *bs, uint64_t lba)
450 {
451 	assert(lba % (bs->cluster_sz / bs->dev->blocklen) == 0);
452 
453 	return lba / (bs->cluster_sz / bs->dev->blocklen);
454 }
455 
456 static inline uint64_t
457 _spdk_bs_blob_lba_to_back_dev_lba(struct spdk_blob *blob, uint64_t lba)
458 {
459 	return lba * blob->bs->dev->blocklen / blob->back_bs_dev->blocklen;
460 }
461 
462 static inline uint64_t
463 _spdk_bs_blob_lba_from_back_dev_lba(struct spdk_blob *blob, uint64_t lba)
464 {
465 	return lba * blob->back_bs_dev->blocklen / blob->bs->dev->blocklen;
466 }
467 
468 /* End basic conversions */
469 
470 static inline uint32_t
471 _spdk_bs_blobid_to_page(spdk_blob_id id)
472 {
473 	return id & 0xFFFFFFFF;
474 }
475 
476 /* The blob id is a 64 bit number. The lower 32 bits are the page_idx. The upper
477  * 32 bits are not currently used. Stick a 1 there just to catch bugs where the
478  * code assumes blob id == page_idx.
479  */
480 static inline spdk_blob_id
481 _spdk_bs_page_to_blobid(uint32_t page_idx)
482 {
483 	return SPDK_BLOB_BLOBID_HIGH_BIT | page_idx;
484 }
485 
486 /* Given a page offset into a blob, look up the LBA for the
487  * start of that page.
488  */
489 static inline uint64_t
490 _spdk_bs_blob_page_to_lba(struct spdk_blob *blob, uint32_t page)
491 {
492 	uint64_t	lba;
493 	uint32_t	pages_per_cluster;
494 
495 	pages_per_cluster = blob->bs->pages_per_cluster;
496 
497 	assert(page < blob->active.num_clusters * pages_per_cluster);
498 
499 	lba = blob->active.clusters[page / pages_per_cluster];
500 	lba += _spdk_bs_page_to_lba(blob->bs, page % pages_per_cluster);
501 
502 	return lba;
503 }
504 
505 /* Given a page offset into a blob, look up the number of pages until the
506  * next cluster boundary.
507  */
508 static inline uint32_t
509 _spdk_bs_num_pages_to_cluster_boundary(struct spdk_blob *blob, uint32_t page)
510 {
511 	uint32_t	pages_per_cluster;
512 
513 	pages_per_cluster = blob->bs->pages_per_cluster;
514 
515 	return pages_per_cluster - (page % pages_per_cluster);
516 }
517 
518 /* Given a page offset into a blob, look up the number of pages into blob to beginning of current cluster */
519 static inline uint32_t
520 _spdk_bs_page_to_cluster_start(struct spdk_blob *blob, uint32_t page)
521 {
522 	uint32_t	pages_per_cluster;
523 
524 	pages_per_cluster = blob->bs->pages_per_cluster;
525 
526 	return page - (page % pages_per_cluster);
527 }
528 
529 /* Given a page offset into a blob, look up if it is from allocated cluster. */
530 static inline bool
531 _spdk_bs_page_is_allocated(struct spdk_blob *blob, uint32_t page)
532 {
533 	uint64_t	lba;
534 	uint32_t	pages_per_cluster;
535 
536 	pages_per_cluster = blob->bs->pages_per_cluster;
537 
538 	assert(page < blob->active.num_clusters * pages_per_cluster);
539 
540 	lba = blob->active.clusters[page / pages_per_cluster];
541 
542 	if (lba == 0) {
543 		assert(spdk_blob_is_thin_provisioned(blob));
544 		return false;
545 	} else {
546 		return true;
547 	}
548 }
549 
550 #endif
551