xref: /spdk/lib/blob/blobstore.h (revision c7852cf98d4df6ce3e2b1f7660db634876c42e0e)
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_data {
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_data) link;
147 };
148 
149 #define __blob_to_data(x)	((struct spdk_blob_data *)(x))
150 #define __data_to_blob(x)	((struct spdk_blob *)(x))
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_data) 	blobs;
182 };
183 
184 struct spdk_bs_channel {
185 	struct spdk_bs_request_set	*req_mem;
186 	TAILQ_HEAD(, spdk_bs_request_set) reqs;
187 
188 	struct spdk_blob_store		*bs;
189 
190 	struct spdk_bs_dev		*dev;
191 	struct spdk_io_channel		*dev_channel;
192 
193 	TAILQ_HEAD(, spdk_bs_request_set) need_cluster_alloc;
194 };
195 
196 /** operation type */
197 enum spdk_blob_op_type {
198 	SPDK_BLOB_WRITE,
199 	SPDK_BLOB_READ,
200 	SPDK_BLOB_UNMAP,
201 	SPDK_BLOB_WRITE_ZEROES,
202 	SPDK_BLOB_WRITEV,
203 	SPDK_BLOB_READV,
204 };
205 
206 /* On-Disk Data Structures
207  *
208  * The following data structures exist on disk.
209  */
210 #define SPDK_BS_INITIAL_VERSION 1
211 #define SPDK_BS_VERSION 3 /* current version */
212 
213 #pragma pack(push, 1)
214 
215 #define SPDK_MD_MASK_TYPE_USED_PAGES 0
216 #define SPDK_MD_MASK_TYPE_USED_CLUSTERS 1
217 #define SPDK_MD_MASK_TYPE_USED_BLOBIDS 2
218 
219 struct spdk_bs_md_mask {
220 	uint8_t		type;
221 	uint32_t	length; /* In bits */
222 	uint8_t		mask[0];
223 };
224 
225 #define SPDK_MD_DESCRIPTOR_TYPE_PADDING 0
226 #define SPDK_MD_DESCRIPTOR_TYPE_EXTENT 1
227 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR 2
228 #define SPDK_MD_DESCRIPTOR_TYPE_FLAGS 3
229 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL 4
230 
231 struct spdk_blob_md_descriptor_xattr {
232 	uint8_t		type;
233 	uint32_t	length;
234 
235 	uint16_t	name_length;
236 	uint16_t	value_length;
237 
238 	char		name[0];
239 	/* String name immediately followed by string value. */
240 };
241 
242 struct spdk_blob_md_descriptor_extent {
243 	uint8_t		type;
244 	uint32_t	length;
245 
246 	struct {
247 		uint32_t        cluster_idx;
248 		uint32_t        length; /* In units of clusters */
249 	} extents[0];
250 };
251 
252 #define SPDK_BLOB_THIN_PROV (1ULL << 0)
253 #define SPDK_BLOB_INTERNAL_XATTR (1ULL << 1)
254 #define SPDK_BLOB_INVALID_FLAGS_MASK	(SPDK_BLOB_THIN_PROV | SPDK_BLOB_INTERNAL_XATTR)
255 
256 #define SPDK_BLOB_READ_ONLY (1ULL << 0)
257 #define SPDK_BLOB_DATA_RO_FLAGS_MASK	SPDK_BLOB_READ_ONLY
258 #define SPDK_BLOB_MD_RO_FLAGS_MASK	0
259 
260 #define spdk_blob_is_thin_provisioned(blob) (blob->invalid_flags & SPDK_BLOB_THIN_PROV)
261 
262 struct spdk_blob_md_descriptor_flags {
263 	uint8_t		type;
264 	uint32_t	length;
265 
266 	/*
267 	 * If a flag in invalid_flags is set that the application is not aware of,
268 	 *  it will not allow the blob to be opened.
269 	 */
270 	uint64_t	invalid_flags;
271 
272 	/*
273 	 * If a flag in data_ro_flags is set that the application is not aware of,
274 	 *  allow the blob to be opened in data_read_only and md_read_only mode.
275 	 */
276 	uint64_t	data_ro_flags;
277 
278 	/*
279 	 * If a flag in md_ro_flags is set the the application is not aware of,
280 	 *  allow the blob to be opened in md_read_only mode.
281 	 */
282 	uint64_t	md_ro_flags;
283 };
284 
285 struct spdk_blob_md_descriptor {
286 	uint8_t		type;
287 	uint32_t	length;
288 };
289 
290 #define SPDK_INVALID_MD_PAGE UINT32_MAX
291 
292 struct spdk_blob_md_page {
293 	spdk_blob_id     id;
294 
295 	uint32_t        sequence_num;
296 	uint32_t	reserved0;
297 
298 	/* Descriptors here */
299 	uint8_t		descriptors[4072];
300 
301 	uint32_t	next;
302 	uint32_t	crc;
303 };
304 #define SPDK_BS_PAGE_SIZE 0x1000
305 SPDK_STATIC_ASSERT(SPDK_BS_PAGE_SIZE == sizeof(struct spdk_blob_md_page), "Invalid md page size");
306 
307 #define SPDK_BS_SUPER_BLOCK_SIG "SPDKBLOB"
308 
309 struct spdk_bs_super_block {
310 	uint8_t		signature[8];
311 	uint32_t        version;
312 	uint32_t        length;
313 	uint32_t	clean; /* If there was a clean shutdown, this is 1. */
314 	spdk_blob_id	super_blob;
315 
316 	uint32_t	cluster_size; /* In bytes */
317 
318 	uint32_t	used_page_mask_start; /* Offset from beginning of disk, in pages */
319 	uint32_t	used_page_mask_len; /* Count, in pages */
320 
321 	uint32_t	used_cluster_mask_start; /* Offset from beginning of disk, in pages */
322 	uint32_t	used_cluster_mask_len; /* Count, in pages */
323 
324 	uint32_t	md_start; /* Offset from beginning of disk, in pages */
325 	uint32_t	md_len; /* Count, in pages */
326 
327 	struct spdk_bs_type 	bstype; /* blobstore type */
328 
329 	uint32_t	used_blobid_mask_start; /* Offset from beginning of disk, in pages */
330 	uint32_t	used_blobid_mask_len; /* Count, in pages */
331 
332 	uint8_t		reserved[4012];
333 	uint32_t	crc;
334 };
335 SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_super_block) == 0x1000, "Invalid super block size");
336 
337 #pragma pack(pop)
338 
339 struct spdk_bs_dev *spdk_bs_create_zeroes_dev(void);
340 
341 /* Unit Conversions
342  *
343  * The blobstore works with several different units:
344  * - Byte: Self explanatory
345  * - LBA: The logical blocks on the backing storage device.
346  * - Page: The read/write units of blobs and metadata. This is
347  *         an offset into a blob in units of 4KiB.
348  * - Cluster Index: The disk is broken into a sequential list of
349  *		    clusters. This is the offset from the beginning.
350  *
351  * NOTE: These conversions all act on simple magnitudes, not with any sort
352  *        of knowledge about the blobs themselves. For instance, converting
353  *        a page to an lba with the conversion function below simply converts
354  *        a number of pages to an equivalent number of lbas, but that
355  *        lba certainly isn't the right lba that corresponds to a page offset
356  *        for a particular blob.
357  */
358 static inline uint64_t
359 _spdk_bs_byte_to_lba(struct spdk_blob_store *bs, uint64_t length)
360 {
361 	assert(length % bs->dev->blocklen == 0);
362 
363 	return length / bs->dev->blocklen;
364 }
365 
366 static inline uint64_t
367 _spdk_bs_dev_byte_to_lba(struct spdk_bs_dev *bs_dev, uint64_t length)
368 {
369 	assert(length % bs_dev->blocklen == 0);
370 
371 	return length / bs_dev->blocklen;
372 }
373 
374 static inline uint64_t
375 _spdk_bs_lba_to_byte(struct spdk_blob_store *bs, uint64_t lba)
376 {
377 	return lba * bs->dev->blocklen;
378 }
379 
380 static inline uint64_t
381 _spdk_bs_page_to_lba(struct spdk_blob_store *bs, uint64_t page)
382 {
383 	return page * SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
384 }
385 
386 static inline uint64_t
387 _spdk_bs_dev_page_to_lba(struct spdk_bs_dev *bs_dev, uint64_t page)
388 {
389 	return page * SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
390 }
391 
392 static inline uint32_t
393 _spdk_bs_lba_to_page(struct spdk_blob_store *bs, uint64_t lba)
394 {
395 	uint64_t	lbas_per_page;
396 
397 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
398 
399 	assert(lba % lbas_per_page == 0);
400 
401 	return lba / lbas_per_page;
402 }
403 
404 static inline uint64_t
405 _spdk_bs_dev_lba_to_page(struct spdk_bs_dev *bs_dev, 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_cluster_to_page(struct spdk_blob_store *bs, uint32_t cluster)
418 {
419 	return cluster * bs->pages_per_cluster;
420 }
421 
422 static inline uint32_t
423 _spdk_bs_page_to_cluster(struct spdk_blob_store *bs, uint64_t page)
424 {
425 	assert(page % bs->pages_per_cluster == 0);
426 
427 	return page / bs->pages_per_cluster;
428 }
429 
430 static inline uint64_t
431 _spdk_bs_cluster_to_lba(struct spdk_blob_store *bs, uint32_t cluster)
432 {
433 	return cluster * (bs->cluster_sz / bs->dev->blocklen);
434 }
435 
436 static inline uint32_t
437 _spdk_bs_lba_to_cluster(struct spdk_blob_store *bs, uint64_t lba)
438 {
439 	assert(lba % (bs->cluster_sz / bs->dev->blocklen) == 0);
440 
441 	return lba / (bs->cluster_sz / bs->dev->blocklen);
442 }
443 
444 static inline uint64_t
445 _spdk_bs_blob_lba_to_back_dev_lba(struct spdk_blob_data *blob, uint64_t lba)
446 {
447 	return lba * blob->bs->dev->blocklen / blob->back_bs_dev->blocklen;
448 }
449 
450 static inline uint64_t
451 _spdk_bs_blob_lba_from_back_dev_lba(struct spdk_blob_data *blob, uint64_t lba)
452 {
453 	return lba * blob->back_bs_dev->blocklen / blob->bs->dev->blocklen;
454 }
455 
456 /* End basic conversions */
457 
458 static inline uint32_t
459 _spdk_bs_blobid_to_page(spdk_blob_id id)
460 {
461 	return id & 0xFFFFFFFF;
462 }
463 
464 /* The blob id is a 64 bit number. The lower 32 bits are the page_idx. The upper
465  * 32 bits are not currently used. Stick a 1 there just to catch bugs where the
466  * code assumes blob id == page_idx.
467  */
468 static inline spdk_blob_id
469 _spdk_bs_page_to_blobid(uint32_t page_idx)
470 {
471 	return SPDK_BLOB_BLOBID_HIGH_BIT | page_idx;
472 }
473 
474 /* Given a page offset into a blob, look up the LBA for the
475  * start of that page.
476  */
477 static inline uint64_t
478 _spdk_bs_blob_page_to_lba(struct spdk_blob_data *blob, uint32_t page)
479 {
480 	uint64_t	lba;
481 	uint32_t	pages_per_cluster;
482 
483 	pages_per_cluster = blob->bs->pages_per_cluster;
484 
485 	assert(page < blob->active.num_clusters * pages_per_cluster);
486 
487 	lba = blob->active.clusters[page / pages_per_cluster];
488 	lba += _spdk_bs_page_to_lba(blob->bs, page % pages_per_cluster);
489 
490 	return lba;
491 }
492 
493 /* Given a page offset into a blob, look up the number of pages until the
494  * next cluster boundary.
495  */
496 static inline uint32_t
497 _spdk_bs_num_pages_to_cluster_boundary(struct spdk_blob_data *blob, uint32_t page)
498 {
499 	uint32_t	pages_per_cluster;
500 
501 	pages_per_cluster = blob->bs->pages_per_cluster;
502 
503 	return pages_per_cluster - (page % pages_per_cluster);
504 }
505 
506 /* Given a page offset into a blob, look up the number of pages into blob to beginning of current cluster */
507 static inline uint32_t
508 _spdk_bs_page_to_cluster_start(struct spdk_blob_data *blob, uint32_t page)
509 {
510 	uint32_t	pages_per_cluster;
511 
512 	pages_per_cluster = blob->bs->pages_per_cluster;
513 
514 	return page - (page % pages_per_cluster);
515 }
516 
517 /* Given a page offset into a blob, look up if it is from allocated cluster. */
518 static inline bool
519 _spdk_bs_page_is_allocated(struct spdk_blob_data *blob, uint32_t page)
520 {
521 	uint64_t	lba;
522 	uint32_t	pages_per_cluster;
523 
524 	pages_per_cluster = blob->bs->pages_per_cluster;
525 
526 	assert(page < blob->active.num_clusters * pages_per_cluster);
527 
528 	lba = blob->active.clusters[page / pages_per_cluster];
529 
530 	if (lba == 0) {
531 		assert(spdk_blob_is_thin_provisioned(blob));
532 		return false;
533 	} else {
534 		return true;
535 	}
536 }
537 
538 #endif
539