xref: /spdk/lib/blob/blobstore.h (revision 66fc591ff7b1188e86b720feee0610a4b1f4fbe6)
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 	/* The disk state is being synchronized with the current
106 	 * blob state.
107 	 */
108 	SPDK_BLOB_STATE_SYNCING,
109 };
110 
111 TAILQ_HEAD(spdk_xattr_tailq, spdk_xattr);
112 
113 struct spdk_blob {
114 	struct spdk_blob_store *bs;
115 
116 	uint32_t	open_ref;
117 
118 	spdk_blob_id	id;
119 
120 	enum spdk_blob_state		state;
121 
122 	/* Two copies of the mutable data. One is a version
123 	 * that matches the last known data on disk (clean).
124 	 * The other (active) is the current data. Syncing
125 	 * a blob makes the clean match the active.
126 	 */
127 	struct spdk_blob_mut_data	clean;
128 	struct spdk_blob_mut_data	active;
129 
130 	bool		invalid;
131 	bool		data_ro;
132 	bool		md_ro;
133 
134 	uint64_t	invalid_flags;
135 	uint64_t	data_ro_flags;
136 	uint64_t	md_ro_flags;
137 
138 	struct spdk_bs_dev *back_bs_dev;
139 
140 	/* TODO: The xattrs are mutable, but we don't want to be
141 	 * copying them unecessarily. Figure this out.
142 	 */
143 	struct spdk_xattr_tailq xattrs;
144 	struct spdk_xattr_tailq xattrs_internal;
145 
146 	TAILQ_ENTRY(spdk_blob) link;
147 };
148 
149 struct spdk_blob_store {
150 	uint64_t			md_start; /* Offset from beginning of disk, in pages */
151 	uint32_t			md_len; /* Count, in pages */
152 
153 	struct spdk_io_channel		*md_channel;
154 	uint32_t			max_channel_ops;
155 
156 	struct spdk_thread		*md_thread;
157 
158 	struct spdk_bs_dev		*dev;
159 
160 	struct spdk_bit_array		*used_md_pages;
161 	struct spdk_bit_array		*used_clusters;
162 	struct spdk_bit_array		*used_blobids;
163 
164 	pthread_mutex_t			used_clusters_mutex;
165 
166 	uint32_t			cluster_sz;
167 	uint64_t			total_clusters;
168 	uint64_t			total_data_clusters;
169 	uint64_t			num_free_clusters;
170 	uint32_t			pages_per_cluster;
171 
172 	spdk_blob_id			super_blob;
173 	struct spdk_bs_type 		bstype;
174 
175 	struct spdk_bs_cpl		unload_cpl;
176 	int				unload_err;
177 
178 	TAILQ_HEAD(, spdk_blob) 	blobs;
179 };
180 
181 struct spdk_bs_channel {
182 	struct spdk_bs_request_set	*req_mem;
183 	TAILQ_HEAD(, spdk_bs_request_set) reqs;
184 
185 	struct spdk_blob_store		*bs;
186 
187 	struct spdk_bs_dev		*dev;
188 	struct spdk_io_channel		*dev_channel;
189 
190 	TAILQ_HEAD(, spdk_bs_request_set) need_cluster_alloc;
191 };
192 
193 /** operation type */
194 enum spdk_blob_op_type {
195 	SPDK_BLOB_WRITE,
196 	SPDK_BLOB_READ,
197 	SPDK_BLOB_UNMAP,
198 	SPDK_BLOB_WRITE_ZEROES,
199 	SPDK_BLOB_WRITEV,
200 	SPDK_BLOB_READV,
201 };
202 
203 /* On-Disk Data Structures
204  *
205  * The following data structures exist on disk.
206  */
207 #define SPDK_BS_INITIAL_VERSION 1
208 #define SPDK_BS_VERSION 3 /* current version */
209 
210 #pragma pack(push, 1)
211 
212 #define SPDK_MD_MASK_TYPE_USED_PAGES 0
213 #define SPDK_MD_MASK_TYPE_USED_CLUSTERS 1
214 #define SPDK_MD_MASK_TYPE_USED_BLOBIDS 2
215 
216 struct spdk_bs_md_mask {
217 	uint8_t		type;
218 	uint32_t	length; /* In bits */
219 	uint8_t		mask[0];
220 };
221 
222 #define SPDK_MD_DESCRIPTOR_TYPE_PADDING 0
223 #define SPDK_MD_DESCRIPTOR_TYPE_EXTENT 1
224 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR 2
225 #define SPDK_MD_DESCRIPTOR_TYPE_FLAGS 3
226 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL 4
227 
228 struct spdk_blob_md_descriptor_xattr {
229 	uint8_t		type;
230 	uint32_t	length;
231 
232 	uint16_t	name_length;
233 	uint16_t	value_length;
234 
235 	char		name[0];
236 	/* String name immediately followed by string value. */
237 };
238 
239 struct spdk_blob_md_descriptor_extent {
240 	uint8_t		type;
241 	uint32_t	length;
242 
243 	struct {
244 		uint32_t        cluster_idx;
245 		uint32_t        length; /* In units of clusters */
246 	} extents[0];
247 };
248 
249 #define SPDK_BLOB_THIN_PROV (1ULL << 0)
250 #define SPDK_BLOB_INTERNAL_XATTR (1ULL << 1)
251 #define SPDK_BLOB_INVALID_FLAGS_MASK	(SPDK_BLOB_THIN_PROV | SPDK_BLOB_INTERNAL_XATTR)
252 
253 #define SPDK_BLOB_READ_ONLY (1ULL << 0)
254 #define SPDK_BLOB_DATA_RO_FLAGS_MASK	SPDK_BLOB_READ_ONLY
255 #define SPDK_BLOB_MD_RO_FLAGS_MASK	0
256 
257 #define spdk_blob_is_thin_provisioned(blob) (blob->invalid_flags & SPDK_BLOB_THIN_PROV)
258 
259 struct spdk_blob_md_descriptor_flags {
260 	uint8_t		type;
261 	uint32_t	length;
262 
263 	/*
264 	 * If a flag in invalid_flags is set that the application is not aware of,
265 	 *  it will not allow the blob to be opened.
266 	 */
267 	uint64_t	invalid_flags;
268 
269 	/*
270 	 * If a flag in data_ro_flags is set that the application is not aware of,
271 	 *  allow the blob to be opened in data_read_only and md_read_only mode.
272 	 */
273 	uint64_t	data_ro_flags;
274 
275 	/*
276 	 * If a flag in md_ro_flags is set the the application is not aware of,
277 	 *  allow the blob to be opened in md_read_only mode.
278 	 */
279 	uint64_t	md_ro_flags;
280 };
281 
282 struct spdk_blob_md_descriptor {
283 	uint8_t		type;
284 	uint32_t	length;
285 };
286 
287 #define SPDK_INVALID_MD_PAGE UINT32_MAX
288 
289 struct spdk_blob_md_page {
290 	spdk_blob_id     id;
291 
292 	uint32_t        sequence_num;
293 	uint32_t	reserved0;
294 
295 	/* Descriptors here */
296 	uint8_t		descriptors[4072];
297 
298 	uint32_t	next;
299 	uint32_t	crc;
300 };
301 #define SPDK_BS_PAGE_SIZE 0x1000
302 SPDK_STATIC_ASSERT(SPDK_BS_PAGE_SIZE == sizeof(struct spdk_blob_md_page), "Invalid md page size");
303 
304 #define SPDK_BS_SUPER_BLOCK_SIG "SPDKBLOB"
305 
306 struct spdk_bs_super_block {
307 	uint8_t		signature[8];
308 	uint32_t        version;
309 	uint32_t        length;
310 	uint32_t	clean; /* If there was a clean shutdown, this is 1. */
311 	spdk_blob_id	super_blob;
312 
313 	uint32_t	cluster_size; /* In bytes */
314 
315 	uint32_t	used_page_mask_start; /* Offset from beginning of disk, in pages */
316 	uint32_t	used_page_mask_len; /* Count, in pages */
317 
318 	uint32_t	used_cluster_mask_start; /* Offset from beginning of disk, in pages */
319 	uint32_t	used_cluster_mask_len; /* Count, in pages */
320 
321 	uint32_t	md_start; /* Offset from beginning of disk, in pages */
322 	uint32_t	md_len; /* Count, in pages */
323 
324 	struct spdk_bs_type 	bstype; /* blobstore type */
325 
326 	uint32_t	used_blobid_mask_start; /* Offset from beginning of disk, in pages */
327 	uint32_t	used_blobid_mask_len; /* Count, in pages */
328 
329 	uint8_t		reserved[4012];
330 	uint32_t	crc;
331 };
332 SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_super_block) == 0x1000, "Invalid super block size");
333 
334 #pragma pack(pop)
335 
336 struct spdk_bs_dev *spdk_bs_create_zeroes_dev(void);
337 
338 /* Unit Conversions
339  *
340  * The blobstore works with several different units:
341  * - Byte: Self explanatory
342  * - LBA: The logical blocks on the backing storage device.
343  * - Page: The read/write units of blobs and metadata. This is
344  *         an offset into a blob in units of 4KiB.
345  * - Cluster Index: The disk is broken into a sequential list of
346  *		    clusters. This is the offset from the beginning.
347  *
348  * NOTE: These conversions all act on simple magnitudes, not with any sort
349  *        of knowledge about the blobs themselves. For instance, converting
350  *        a page to an lba with the conversion function below simply converts
351  *        a number of pages to an equivalent number of lbas, but that
352  *        lba certainly isn't the right lba that corresponds to a page offset
353  *        for a particular blob.
354  */
355 static inline uint64_t
356 _spdk_bs_byte_to_lba(struct spdk_blob_store *bs, uint64_t length)
357 {
358 	assert(length % bs->dev->blocklen == 0);
359 
360 	return length / bs->dev->blocklen;
361 }
362 
363 static inline uint64_t
364 _spdk_bs_dev_byte_to_lba(struct spdk_bs_dev *bs_dev, uint64_t length)
365 {
366 	assert(length % bs_dev->blocklen == 0);
367 
368 	return length / bs_dev->blocklen;
369 }
370 
371 static inline uint64_t
372 _spdk_bs_lba_to_byte(struct spdk_blob_store *bs, uint64_t lba)
373 {
374 	return lba * bs->dev->blocklen;
375 }
376 
377 static inline uint64_t
378 _spdk_bs_page_to_lba(struct spdk_blob_store *bs, uint64_t page)
379 {
380 	return page * SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
381 }
382 
383 static inline uint64_t
384 _spdk_bs_dev_page_to_lba(struct spdk_bs_dev *bs_dev, uint64_t page)
385 {
386 	return page * SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
387 }
388 
389 static inline uint32_t
390 _spdk_bs_lba_to_page(struct spdk_blob_store *bs, uint64_t lba)
391 {
392 	uint64_t	lbas_per_page;
393 
394 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
395 
396 	assert(lba % lbas_per_page == 0);
397 
398 	return lba / lbas_per_page;
399 }
400 
401 static inline uint64_t
402 _spdk_bs_dev_lba_to_page(struct spdk_bs_dev *bs_dev, uint64_t lba)
403 {
404 	uint64_t	lbas_per_page;
405 
406 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
407 
408 	assert(lba % lbas_per_page == 0);
409 
410 	return lba / lbas_per_page;
411 }
412 
413 static inline uint64_t
414 _spdk_bs_cluster_to_page(struct spdk_blob_store *bs, uint32_t cluster)
415 {
416 	return cluster * bs->pages_per_cluster;
417 }
418 
419 static inline uint32_t
420 _spdk_bs_page_to_cluster(struct spdk_blob_store *bs, uint64_t page)
421 {
422 	assert(page % bs->pages_per_cluster == 0);
423 
424 	return page / bs->pages_per_cluster;
425 }
426 
427 static inline uint64_t
428 _spdk_bs_cluster_to_lba(struct spdk_blob_store *bs, uint32_t cluster)
429 {
430 	return cluster * (bs->cluster_sz / bs->dev->blocklen);
431 }
432 
433 static inline uint32_t
434 _spdk_bs_lba_to_cluster(struct spdk_blob_store *bs, uint64_t lba)
435 {
436 	assert(lba % (bs->cluster_sz / bs->dev->blocklen) == 0);
437 
438 	return lba / (bs->cluster_sz / bs->dev->blocklen);
439 }
440 
441 static inline uint64_t
442 _spdk_bs_blob_lba_to_back_dev_lba(struct spdk_blob *blob, uint64_t lba)
443 {
444 	return lba * blob->bs->dev->blocklen / blob->back_bs_dev->blocklen;
445 }
446 
447 static inline uint64_t
448 _spdk_bs_blob_lba_from_back_dev_lba(struct spdk_blob *blob, uint64_t lba)
449 {
450 	return lba * blob->back_bs_dev->blocklen / blob->bs->dev->blocklen;
451 }
452 
453 /* End basic conversions */
454 
455 static inline uint32_t
456 _spdk_bs_blobid_to_page(spdk_blob_id id)
457 {
458 	return id & 0xFFFFFFFF;
459 }
460 
461 /* The blob id is a 64 bit number. The lower 32 bits are the page_idx. The upper
462  * 32 bits are not currently used. Stick a 1 there just to catch bugs where the
463  * code assumes blob id == page_idx.
464  */
465 static inline spdk_blob_id
466 _spdk_bs_page_to_blobid(uint32_t page_idx)
467 {
468 	return SPDK_BLOB_BLOBID_HIGH_BIT | page_idx;
469 }
470 
471 /* Given a page offset into a blob, look up the LBA for the
472  * start of that page.
473  */
474 static inline uint64_t
475 _spdk_bs_blob_page_to_lba(struct spdk_blob *blob, uint32_t page)
476 {
477 	uint64_t	lba;
478 	uint32_t	pages_per_cluster;
479 
480 	pages_per_cluster = blob->bs->pages_per_cluster;
481 
482 	assert(page < blob->active.num_clusters * pages_per_cluster);
483 
484 	lba = blob->active.clusters[page / pages_per_cluster];
485 	lba += _spdk_bs_page_to_lba(blob->bs, page % pages_per_cluster);
486 
487 	return lba;
488 }
489 
490 /* Given a page offset into a blob, look up the number of pages until the
491  * next cluster boundary.
492  */
493 static inline uint32_t
494 _spdk_bs_num_pages_to_cluster_boundary(struct spdk_blob *blob, uint32_t page)
495 {
496 	uint32_t	pages_per_cluster;
497 
498 	pages_per_cluster = blob->bs->pages_per_cluster;
499 
500 	return pages_per_cluster - (page % pages_per_cluster);
501 }
502 
503 /* Given a page offset into a blob, look up the number of pages into blob to beginning of current cluster */
504 static inline uint32_t
505 _spdk_bs_page_to_cluster_start(struct spdk_blob *blob, uint32_t page)
506 {
507 	uint32_t	pages_per_cluster;
508 
509 	pages_per_cluster = blob->bs->pages_per_cluster;
510 
511 	return page - (page % pages_per_cluster);
512 }
513 
514 /* Given a page offset into a blob, look up if it is from allocated cluster. */
515 static inline bool
516 _spdk_bs_page_is_allocated(struct spdk_blob *blob, uint32_t page)
517 {
518 	uint64_t	lba;
519 	uint32_t	pages_per_cluster;
520 
521 	pages_per_cluster = blob->bs->pages_per_cluster;
522 
523 	assert(page < blob->active.num_clusters * pages_per_cluster);
524 
525 	lba = blob->active.clusters[page / pages_per_cluster];
526 
527 	if (lba == 0) {
528 		assert(spdk_blob_is_thin_provisioned(blob));
529 		return false;
530 	} else {
531 		return true;
532 	}
533 }
534 
535 #endif
536