Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2018 Intel Corporation.
3 : * All rights reserved.
4 : * Copyright (c) 2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5 : */
6 :
7 : #include "spdk/stdinc.h"
8 :
9 : #include "queue_internal.h"
10 :
11 : #include "spdk/reduce.h"
12 : #include "spdk/env.h"
13 : #include "spdk/string.h"
14 : #include "spdk/bit_array.h"
15 : #include "spdk/util.h"
16 : #include "spdk/log.h"
17 : #include "spdk/memory.h"
18 : #include "spdk/tree.h"
19 :
20 : #include "libpmem.h"
21 :
22 : /* Always round up the size of the PM region to the nearest cacheline. */
23 : #define REDUCE_PM_SIZE_ALIGNMENT 64
24 :
25 : /* Offset into the backing device where the persistent memory file's path is stored. */
26 : #define REDUCE_BACKING_DEV_PATH_OFFSET 4096
27 :
28 : #define REDUCE_EMPTY_MAP_ENTRY -1ULL
29 :
30 : #define REDUCE_NUM_VOL_REQUESTS 256
31 :
32 : /* Structure written to offset 0 of both the pm file and the backing device. */
33 : struct spdk_reduce_vol_superblock {
34 : uint8_t signature[8];
35 : struct spdk_reduce_vol_params params;
36 : uint8_t reserved[4040];
37 : };
38 : SPDK_STATIC_ASSERT(sizeof(struct spdk_reduce_vol_superblock) == 4096, "size incorrect");
39 :
40 : #define SPDK_REDUCE_SIGNATURE "SPDKREDU"
41 : /* null terminator counts one */
42 : SPDK_STATIC_ASSERT(sizeof(SPDK_REDUCE_SIGNATURE) - 1 ==
43 : SPDK_SIZEOF_MEMBER(struct spdk_reduce_vol_superblock, signature), "size incorrect");
44 :
45 : #define REDUCE_PATH_MAX 4096
46 :
47 : #define REDUCE_ZERO_BUF_SIZE 0x100000
48 :
49 : /**
50 : * Describes a persistent memory file used to hold metadata associated with a
51 : * compressed volume.
52 : */
53 : struct spdk_reduce_pm_file {
54 : char path[REDUCE_PATH_MAX];
55 : void *pm_buf;
56 : int pm_is_pmem;
57 : uint64_t size;
58 : };
59 :
60 : #define REDUCE_IO_READV 1
61 : #define REDUCE_IO_WRITEV 2
62 : #define REDUCE_IO_UNMAP 3
63 :
64 : struct spdk_reduce_chunk_map {
65 : uint32_t compressed_size;
66 : uint32_t reserved;
67 : uint64_t io_unit_index[0];
68 : };
69 :
70 : struct spdk_reduce_vol_request {
71 : /**
72 : * Scratch buffer used for uncompressed chunk. This is used for:
73 : * 1) source buffer for compression operations
74 : * 2) destination buffer for decompression operations
75 : * 3) data buffer when writing uncompressed chunk to disk
76 : * 4) data buffer when reading uncompressed chunk from disk
77 : */
78 : uint8_t *decomp_buf;
79 : struct iovec *decomp_buf_iov;
80 :
81 : /**
82 : * These are used to construct the iovecs that are sent to
83 : * the decomp engine, they point to a mix of the scratch buffer
84 : * and user buffer
85 : */
86 : struct iovec decomp_iov[REDUCE_MAX_IOVECS + 2];
87 : int decomp_iovcnt;
88 :
89 : /**
90 : * Scratch buffer used for compressed chunk. This is used for:
91 : * 1) destination buffer for compression operations
92 : * 2) source buffer for decompression operations
93 : * 3) data buffer when writing compressed chunk to disk
94 : * 4) data buffer when reading compressed chunk from disk
95 : */
96 : uint8_t *comp_buf;
97 : struct iovec *comp_buf_iov;
98 : struct iovec *iov;
99 : bool rmw;
100 : struct spdk_reduce_vol *vol;
101 : int type;
102 : int reduce_errno;
103 : int iovcnt;
104 : int num_backing_ops;
105 : uint32_t num_io_units;
106 : struct spdk_reduce_backing_io *backing_io;
107 : bool chunk_is_compressed;
108 : bool copy_after_decompress;
109 : uint64_t offset;
110 : uint64_t logical_map_index;
111 : uint64_t length;
112 : uint64_t chunk_map_index;
113 : struct spdk_reduce_chunk_map *chunk;
114 : spdk_reduce_vol_op_complete cb_fn;
115 : void *cb_arg;
116 : TAILQ_ENTRY(spdk_reduce_vol_request) tailq;
117 : RB_ENTRY(spdk_reduce_vol_request) rbnode;
118 : struct spdk_reduce_vol_cb_args backing_cb_args;
119 : };
120 :
121 : struct spdk_reduce_vol {
122 : struct spdk_reduce_vol_params params;
123 : struct spdk_reduce_vol_info info;
124 : uint32_t backing_io_units_per_chunk;
125 : uint32_t backing_lba_per_io_unit;
126 : uint32_t logical_blocks_per_chunk;
127 : struct spdk_reduce_pm_file pm_file;
128 : struct spdk_reduce_backing_dev *backing_dev;
129 : struct spdk_reduce_vol_superblock *backing_super;
130 : struct spdk_reduce_vol_superblock *pm_super;
131 : uint64_t *pm_logical_map;
132 : uint64_t *pm_chunk_maps;
133 :
134 : struct spdk_bit_array *allocated_chunk_maps;
135 : /* The starting position when looking for a block from allocated_chunk_maps */
136 : uint64_t find_chunk_offset;
137 : /* Cache free chunks to speed up lookup of free chunk. */
138 : struct reduce_queue free_chunks_queue;
139 : struct spdk_bit_array *allocated_backing_io_units;
140 : /* The starting position when looking for a block from allocated_backing_io_units */
141 : uint64_t find_block_offset;
142 : /* Cache free blocks for backing bdev to speed up lookup of free backing blocks. */
143 : struct reduce_queue free_backing_blocks_queue;
144 :
145 : struct spdk_reduce_vol_request *request_mem;
146 : TAILQ_HEAD(, spdk_reduce_vol_request) free_requests;
147 : RB_HEAD(executing_req_tree, spdk_reduce_vol_request) executing_requests;
148 : TAILQ_HEAD(, spdk_reduce_vol_request) queued_requests;
149 :
150 : /* Single contiguous buffer used for all request buffers for this volume. */
151 : uint8_t *buf_mem;
152 : struct iovec *buf_iov_mem;
153 : /* Single contiguous buffer used for backing io buffers for this volume. */
154 : uint8_t *buf_backing_io_mem;
155 : };
156 :
157 : static void _start_readv_request(struct spdk_reduce_vol_request *req);
158 : static void _start_writev_request(struct spdk_reduce_vol_request *req);
159 : static uint8_t *g_zero_buf;
160 : static int g_vol_count = 0;
161 :
162 : /*
163 : * Allocate extra metadata chunks and corresponding backing io units to account for
164 : * outstanding IO in worst case scenario where logical map is completely allocated
165 : * and no data can be compressed. We need extra chunks in this case to handle
166 : * in-flight writes since reduce never writes data in place.
167 : */
168 : #define REDUCE_NUM_EXTRA_CHUNKS 128
169 :
170 : static void
171 48 : _reduce_persist(struct spdk_reduce_vol *vol, const void *addr, size_t len)
172 : {
173 48 : if (vol->pm_file.pm_is_pmem) {
174 48 : pmem_persist(addr, len);
175 48 : } else {
176 0 : pmem_msync(addr, len);
177 : }
178 48 : }
179 :
180 : static uint64_t
181 49 : _get_pm_logical_map_size(uint64_t vol_size, uint64_t chunk_size)
182 : {
183 49 : uint64_t chunks_in_logical_map, logical_map_size;
184 :
185 49 : chunks_in_logical_map = vol_size / chunk_size;
186 49 : logical_map_size = chunks_in_logical_map * sizeof(uint64_t);
187 :
188 : /* Round up to next cacheline. */
189 98 : return spdk_divide_round_up(logical_map_size, REDUCE_PM_SIZE_ALIGNMENT) *
190 : REDUCE_PM_SIZE_ALIGNMENT;
191 49 : }
192 :
193 : static uint64_t
194 349 : _get_total_chunks(uint64_t vol_size, uint64_t chunk_size)
195 : {
196 349 : uint64_t num_chunks;
197 :
198 349 : num_chunks = vol_size / chunk_size;
199 349 : num_chunks += REDUCE_NUM_EXTRA_CHUNKS;
200 :
201 698 : return num_chunks;
202 349 : }
203 :
204 : static inline uint32_t
205 341 : _reduce_vol_get_chunk_struct_size(uint64_t backing_io_units_per_chunk)
206 : {
207 341 : return sizeof(struct spdk_reduce_chunk_map) + sizeof(uint64_t) * backing_io_units_per_chunk;
208 : }
209 :
210 : static uint64_t
211 25 : _get_pm_total_chunks_size(uint64_t vol_size, uint64_t chunk_size, uint64_t backing_io_unit_size)
212 : {
213 25 : uint64_t io_units_per_chunk, num_chunks, total_chunks_size;
214 :
215 25 : num_chunks = _get_total_chunks(vol_size, chunk_size);
216 25 : io_units_per_chunk = chunk_size / backing_io_unit_size;
217 :
218 25 : total_chunks_size = num_chunks * _reduce_vol_get_chunk_struct_size(io_units_per_chunk);
219 :
220 50 : return spdk_divide_round_up(total_chunks_size, REDUCE_PM_SIZE_ALIGNMENT) *
221 : REDUCE_PM_SIZE_ALIGNMENT;
222 25 : }
223 :
224 : static struct spdk_reduce_chunk_map *
225 300 : _reduce_vol_get_chunk_map(struct spdk_reduce_vol *vol, uint64_t chunk_map_index)
226 : {
227 300 : uintptr_t chunk_map_addr;
228 :
229 300 : assert(chunk_map_index < _get_total_chunks(vol->params.vol_size, vol->params.chunk_size));
230 :
231 300 : chunk_map_addr = (uintptr_t)vol->pm_chunk_maps;
232 600 : chunk_map_addr += chunk_map_index *
233 300 : _reduce_vol_get_chunk_struct_size(vol->backing_io_units_per_chunk);
234 :
235 600 : return (struct spdk_reduce_chunk_map *)chunk_map_addr;
236 300 : }
237 :
238 : static int
239 21 : _validate_vol_params(struct spdk_reduce_vol_params *params)
240 : {
241 21 : if (params->vol_size > 0) {
242 : /**
243 : * User does not pass in the vol size - it gets calculated by libreduce from
244 : * values in this structure plus the size of the backing device.
245 : */
246 0 : return -EINVAL;
247 : }
248 :
249 42 : if (params->chunk_size == 0 || params->backing_io_unit_size == 0 ||
250 21 : params->logical_block_size == 0) {
251 0 : return -EINVAL;
252 : }
253 :
254 : /* Chunk size must be an even multiple of the backing io unit size. */
255 21 : if ((params->chunk_size % params->backing_io_unit_size) != 0) {
256 0 : return -EINVAL;
257 : }
258 :
259 : /* Chunk size must be an even multiple of the logical block size. */
260 21 : if ((params->chunk_size % params->logical_block_size) != 0) {
261 0 : return -1;
262 : }
263 :
264 21 : return 0;
265 21 : }
266 :
267 : static uint64_t
268 27 : _get_vol_size(uint64_t chunk_size, uint64_t backing_dev_size)
269 : {
270 27 : uint64_t num_chunks;
271 :
272 27 : num_chunks = backing_dev_size / chunk_size;
273 27 : if (num_chunks <= REDUCE_NUM_EXTRA_CHUNKS) {
274 1 : return 0;
275 : }
276 :
277 26 : num_chunks -= REDUCE_NUM_EXTRA_CHUNKS;
278 26 : return num_chunks * chunk_size;
279 27 : }
280 :
281 : static uint64_t
282 25 : _get_pm_file_size(struct spdk_reduce_vol_params *params)
283 : {
284 25 : uint64_t total_pm_size;
285 :
286 25 : total_pm_size = sizeof(struct spdk_reduce_vol_superblock);
287 25 : total_pm_size += _get_pm_logical_map_size(params->vol_size, params->chunk_size);
288 50 : total_pm_size += _get_pm_total_chunks_size(params->vol_size, params->chunk_size,
289 25 : params->backing_io_unit_size);
290 50 : return total_pm_size;
291 25 : }
292 :
293 : const struct spdk_uuid *
294 1 : spdk_reduce_vol_get_uuid(struct spdk_reduce_vol *vol)
295 : {
296 1 : return &vol->params.uuid;
297 : }
298 :
299 : static void
300 24 : _initialize_vol_pm_pointers(struct spdk_reduce_vol *vol)
301 : {
302 24 : uint64_t logical_map_size;
303 :
304 : /* Superblock is at the beginning of the pm file. */
305 24 : vol->pm_super = (struct spdk_reduce_vol_superblock *)vol->pm_file.pm_buf;
306 :
307 : /* Logical map immediately follows the super block. */
308 24 : vol->pm_logical_map = (uint64_t *)(vol->pm_super + 1);
309 :
310 : /* Chunks maps follow the logical map. */
311 24 : logical_map_size = _get_pm_logical_map_size(vol->params.vol_size, vol->params.chunk_size);
312 24 : vol->pm_chunk_maps = (uint64_t *)((uint8_t *)vol->pm_logical_map + logical_map_size);
313 24 : }
314 :
315 : /* We need 2 iovs during load - one for the superblock, another for the path */
316 : #define LOAD_IOV_COUNT 2
317 :
318 : struct reduce_init_load_ctx {
319 : struct spdk_reduce_vol *vol;
320 : struct spdk_reduce_vol_cb_args backing_cb_args;
321 : spdk_reduce_vol_op_with_handle_complete cb_fn;
322 : void *cb_arg;
323 : struct iovec iov[LOAD_IOV_COUNT];
324 : void *path;
325 : struct spdk_reduce_backing_io *backing_io;
326 : };
327 :
328 : static inline bool
329 14342 : _addr_crosses_huge_page(const void *addr, size_t *size)
330 : {
331 14342 : size_t _size;
332 14342 : uint64_t rc;
333 :
334 14342 : assert(size);
335 :
336 14342 : _size = *size;
337 14342 : rc = spdk_vtophys(addr, size);
338 :
339 14342 : return rc == SPDK_VTOPHYS_ERROR || _size != *size;
340 14342 : }
341 :
342 : static inline int
343 14336 : _set_buffer(uint8_t **vol_buffer, uint8_t **_addr, uint8_t *addr_range, size_t buffer_size)
344 : {
345 14336 : uint8_t *addr;
346 14336 : size_t size_tmp = buffer_size;
347 :
348 14336 : addr = *_addr;
349 :
350 : /* Verify that addr + buffer_size doesn't cross huge page boundary */
351 14336 : if (_addr_crosses_huge_page(addr, &size_tmp)) {
352 : /* Memory start is aligned on 2MiB, so buffer should be located at the end of the page.
353 : * Skip remaining bytes and continue from the beginning of the next page */
354 6 : addr += size_tmp;
355 6 : }
356 :
357 14336 : if (addr + buffer_size > addr_range) {
358 0 : SPDK_ERRLOG("Vol buffer %p out of range %p\n", addr, addr_range);
359 0 : return -ERANGE;
360 : }
361 :
362 14336 : *vol_buffer = addr;
363 14336 : *_addr = addr + buffer_size;
364 :
365 14336 : return 0;
366 14336 : }
367 :
368 : static int
369 28 : _allocate_vol_requests(struct spdk_reduce_vol *vol)
370 : {
371 28 : struct spdk_reduce_vol_request *req;
372 28 : struct spdk_reduce_backing_dev *backing_dev = vol->backing_dev;
373 28 : uint32_t reqs_in_2mb_page, huge_pages_needed;
374 28 : uint8_t *buffer, *buffer_end;
375 28 : int i = 0;
376 28 : int rc = 0;
377 :
378 : /* It is needed to allocate comp and decomp buffers so that they do not cross physical
379 : * page boundaries. Assume that the system uses default 2MiB pages and chunk_size is not
380 : * necessarily power of 2
381 : * Allocate 2x since we need buffers for both read/write and compress/decompress
382 : * intermediate buffers. */
383 28 : reqs_in_2mb_page = VALUE_2MB / (vol->params.chunk_size * 2);
384 28 : if (!reqs_in_2mb_page) {
385 0 : return -EINVAL;
386 : }
387 28 : huge_pages_needed = SPDK_CEIL_DIV(REDUCE_NUM_VOL_REQUESTS, reqs_in_2mb_page);
388 :
389 28 : vol->buf_mem = spdk_dma_malloc(VALUE_2MB * huge_pages_needed, VALUE_2MB, NULL);
390 28 : if (vol->buf_mem == NULL) {
391 0 : return -ENOMEM;
392 : }
393 :
394 28 : vol->request_mem = calloc(REDUCE_NUM_VOL_REQUESTS, sizeof(*req));
395 28 : if (vol->request_mem == NULL) {
396 0 : spdk_free(vol->buf_mem);
397 0 : vol->buf_mem = NULL;
398 0 : return -ENOMEM;
399 : }
400 :
401 : /* Allocate 2x since we need iovs for both read/write and compress/decompress intermediate
402 : * buffers.
403 : */
404 28 : vol->buf_iov_mem = calloc(REDUCE_NUM_VOL_REQUESTS,
405 28 : 2 * sizeof(struct iovec) * vol->backing_io_units_per_chunk);
406 28 : if (vol->buf_iov_mem == NULL) {
407 0 : free(vol->request_mem);
408 0 : spdk_free(vol->buf_mem);
409 0 : vol->request_mem = NULL;
410 0 : vol->buf_mem = NULL;
411 0 : return -ENOMEM;
412 : }
413 :
414 56 : vol->buf_backing_io_mem = calloc(REDUCE_NUM_VOL_REQUESTS, (sizeof(struct spdk_reduce_backing_io) +
415 28 : backing_dev->user_ctx_size) * vol->backing_io_units_per_chunk);
416 28 : if (vol->buf_backing_io_mem == NULL) {
417 0 : free(vol->request_mem);
418 0 : free(vol->buf_iov_mem);
419 0 : spdk_free(vol->buf_mem);
420 0 : vol->request_mem = NULL;
421 0 : vol->buf_iov_mem = NULL;
422 0 : vol->buf_mem = NULL;
423 0 : return -ENOMEM;
424 : }
425 :
426 28 : buffer = vol->buf_mem;
427 28 : buffer_end = buffer + VALUE_2MB * huge_pages_needed;
428 :
429 7196 : for (i = 0; i < REDUCE_NUM_VOL_REQUESTS; i++) {
430 7168 : req = &vol->request_mem[i];
431 7168 : TAILQ_INSERT_HEAD(&vol->free_requests, req, tailq);
432 21504 : req->backing_io = (struct spdk_reduce_backing_io *)(vol->buf_backing_io_mem + i *
433 14336 : (sizeof(struct spdk_reduce_backing_io) + backing_dev->user_ctx_size) *
434 7168 : vol->backing_io_units_per_chunk);
435 :
436 7168 : req->decomp_buf_iov = &vol->buf_iov_mem[(2 * i) * vol->backing_io_units_per_chunk];
437 7168 : req->comp_buf_iov = &vol->buf_iov_mem[(2 * i + 1) * vol->backing_io_units_per_chunk];
438 :
439 7168 : rc = _set_buffer(&req->comp_buf, &buffer, buffer_end, vol->params.chunk_size);
440 7168 : if (rc) {
441 0 : SPDK_ERRLOG("Failed to set comp buffer for req idx %u, addr %p, start %p, end %p\n", i, buffer,
442 : vol->buf_mem, buffer_end);
443 0 : break;
444 : }
445 7168 : rc = _set_buffer(&req->decomp_buf, &buffer, buffer_end, vol->params.chunk_size);
446 7168 : if (rc) {
447 0 : SPDK_ERRLOG("Failed to set decomp buffer for req idx %u, addr %p, start %p, end %p\n", i, buffer,
448 : vol->buf_mem, buffer_end);
449 0 : break;
450 : }
451 7168 : }
452 :
453 28 : if (rc) {
454 0 : free(vol->buf_backing_io_mem);
455 0 : free(vol->buf_iov_mem);
456 0 : free(vol->request_mem);
457 0 : spdk_free(vol->buf_mem);
458 0 : vol->buf_mem = NULL;
459 0 : vol->buf_backing_io_mem = NULL;
460 0 : vol->buf_iov_mem = NULL;
461 0 : vol->request_mem = NULL;
462 0 : }
463 :
464 28 : return rc;
465 28 : }
466 :
467 : const struct spdk_reduce_vol_info *
468 0 : spdk_reduce_vol_get_info(const struct spdk_reduce_vol *vol)
469 : {
470 0 : return &vol->info;
471 : }
472 :
473 : static void
474 55 : _init_load_cleanup(struct spdk_reduce_vol *vol, struct reduce_init_load_ctx *ctx)
475 : {
476 55 : if (ctx != NULL) {
477 26 : spdk_free(ctx->path);
478 26 : free(ctx->backing_io);
479 26 : free(ctx);
480 26 : }
481 :
482 55 : if (vol != NULL) {
483 30 : if (vol->pm_file.pm_buf != NULL) {
484 24 : pmem_unmap(vol->pm_file.pm_buf, vol->pm_file.size);
485 24 : }
486 :
487 30 : spdk_free(vol->backing_super);
488 30 : spdk_bit_array_free(&vol->allocated_chunk_maps);
489 30 : spdk_bit_array_free(&vol->allocated_backing_io_units);
490 30 : free(vol->request_mem);
491 30 : free(vol->buf_backing_io_mem);
492 30 : free(vol->buf_iov_mem);
493 30 : spdk_free(vol->buf_mem);
494 30 : free(vol);
495 30 : }
496 55 : }
497 :
498 : static int
499 26 : _alloc_zero_buff(void)
500 : {
501 26 : int rc = 0;
502 :
503 : /* The zero buffer is shared between all volumes and just used
504 : * for reads so allocate one global instance here if not already
505 : * allocated when another vol init'd or loaded.
506 : */
507 26 : if (g_vol_count++ == 0) {
508 24 : g_zero_buf = spdk_zmalloc(REDUCE_ZERO_BUF_SIZE,
509 : 64, NULL, SPDK_ENV_LCORE_ID_ANY,
510 : SPDK_MALLOC_DMA);
511 24 : if (g_zero_buf == NULL) {
512 0 : g_vol_count--;
513 0 : rc = -ENOMEM;
514 0 : }
515 24 : }
516 52 : return rc;
517 26 : }
518 :
519 : static void
520 15 : _init_write_super_cpl(void *cb_arg, int reduce_errno)
521 : {
522 15 : struct reduce_init_load_ctx *init_ctx = cb_arg;
523 15 : int rc = 0;
524 :
525 15 : if (reduce_errno != 0) {
526 0 : rc = reduce_errno;
527 0 : goto err;
528 : }
529 :
530 15 : rc = _allocate_vol_requests(init_ctx->vol);
531 15 : if (rc != 0) {
532 0 : goto err;
533 : }
534 :
535 15 : rc = _alloc_zero_buff();
536 15 : if (rc != 0) {
537 0 : goto err;
538 : }
539 :
540 15 : init_ctx->cb_fn(init_ctx->cb_arg, init_ctx->vol, rc);
541 : /* Only clean up the ctx - the vol has been passed to the application
542 : * for use now that initialization was successful.
543 : */
544 15 : _init_load_cleanup(NULL, init_ctx);
545 :
546 15 : return;
547 : err:
548 0 : if (unlink(init_ctx->path)) {
549 0 : SPDK_ERRLOG("%s could not be unlinked: %s\n",
550 : (char *)init_ctx->path, spdk_strerror(errno));
551 0 : }
552 :
553 0 : init_ctx->cb_fn(init_ctx->cb_arg, NULL, rc);
554 0 : _init_load_cleanup(init_ctx->vol, init_ctx);
555 15 : }
556 :
557 : static void
558 15 : _init_write_path_cpl(void *cb_arg, int reduce_errno)
559 : {
560 15 : struct reduce_init_load_ctx *init_ctx = cb_arg;
561 15 : struct spdk_reduce_vol *vol = init_ctx->vol;
562 15 : struct spdk_reduce_backing_io *backing_io = init_ctx->backing_io;
563 :
564 15 : if (reduce_errno != 0) {
565 0 : _init_write_super_cpl(cb_arg, reduce_errno);
566 0 : return;
567 : }
568 :
569 15 : init_ctx->iov[0].iov_base = vol->backing_super;
570 15 : init_ctx->iov[0].iov_len = sizeof(*vol->backing_super);
571 15 : init_ctx->backing_cb_args.cb_fn = _init_write_super_cpl;
572 15 : init_ctx->backing_cb_args.cb_arg = init_ctx;
573 :
574 15 : backing_io->dev = vol->backing_dev;
575 15 : backing_io->iov = init_ctx->iov;
576 15 : backing_io->iovcnt = 1;
577 15 : backing_io->lba = 0;
578 15 : backing_io->lba_count = sizeof(*vol->backing_super) / vol->backing_dev->blocklen;
579 15 : backing_io->backing_cb_args = &init_ctx->backing_cb_args;
580 15 : backing_io->backing_io_type = SPDK_REDUCE_BACKING_IO_WRITE;
581 :
582 15 : vol->backing_dev->submit_backing_io(backing_io);
583 15 : }
584 :
585 : static int
586 24 : _allocate_bit_arrays(struct spdk_reduce_vol *vol)
587 : {
588 24 : uint64_t total_chunks, total_backing_io_units;
589 24 : uint32_t i, num_metadata_io_units;
590 :
591 24 : total_chunks = _get_total_chunks(vol->params.vol_size, vol->params.chunk_size);
592 24 : vol->allocated_chunk_maps = spdk_bit_array_create(total_chunks);
593 24 : vol->find_chunk_offset = 0;
594 24 : total_backing_io_units = total_chunks * (vol->params.chunk_size / vol->params.backing_io_unit_size);
595 24 : vol->allocated_backing_io_units = spdk_bit_array_create(total_backing_io_units);
596 24 : vol->find_block_offset = 0;
597 :
598 24 : if (vol->allocated_chunk_maps == NULL || vol->allocated_backing_io_units == NULL) {
599 0 : return -ENOMEM;
600 : }
601 :
602 : /* Set backing io unit bits associated with metadata. */
603 24 : num_metadata_io_units = (sizeof(*vol->backing_super) + REDUCE_PATH_MAX) /
604 24 : vol->params.backing_io_unit_size;
605 198 : for (i = 0; i < num_metadata_io_units; i++) {
606 174 : spdk_bit_array_set(vol->allocated_backing_io_units, i);
607 174 : vol->info.allocated_io_units++;
608 174 : }
609 :
610 24 : return 0;
611 24 : }
612 :
613 : static int
614 1 : overlap_cmp(struct spdk_reduce_vol_request *req1, struct spdk_reduce_vol_request *req2)
615 : {
616 2 : return (req1->logical_map_index < req2->logical_map_index ? -1 : req1->logical_map_index >
617 1 : req2->logical_map_index);
618 : }
619 1145 : RB_GENERATE_STATIC(executing_req_tree, spdk_reduce_vol_request, rbnode, overlap_cmp);
620 :
621 :
622 : void
623 17 : spdk_reduce_vol_init(struct spdk_reduce_vol_params *params,
624 : struct spdk_reduce_backing_dev *backing_dev,
625 : const char *pm_file_dir,
626 : spdk_reduce_vol_op_with_handle_complete cb_fn, void *cb_arg)
627 : {
628 17 : struct spdk_reduce_vol *vol;
629 17 : struct reduce_init_load_ctx *init_ctx;
630 17 : struct spdk_reduce_backing_io *backing_io;
631 17 : uint64_t backing_dev_size;
632 17 : size_t mapped_len;
633 17 : int dir_len, max_dir_len, rc;
634 :
635 : /* We need to append a path separator and the UUID to the supplied
636 : * path.
637 : */
638 17 : max_dir_len = REDUCE_PATH_MAX - SPDK_UUID_STRING_LEN - 1;
639 17 : dir_len = strnlen(pm_file_dir, max_dir_len);
640 : /* Strip trailing slash if the user provided one - we will add it back
641 : * later when appending the filename.
642 : */
643 17 : if (pm_file_dir[dir_len - 1] == '/') {
644 0 : dir_len--;
645 0 : }
646 17 : if (dir_len == max_dir_len) {
647 0 : SPDK_ERRLOG("pm_file_dir (%s) too long\n", pm_file_dir);
648 0 : cb_fn(cb_arg, NULL, -EINVAL);
649 0 : return;
650 : }
651 :
652 17 : rc = _validate_vol_params(params);
653 17 : if (rc != 0) {
654 0 : SPDK_ERRLOG("invalid vol params\n");
655 0 : cb_fn(cb_arg, NULL, rc);
656 0 : return;
657 : }
658 :
659 17 : backing_dev_size = backing_dev->blockcnt * backing_dev->blocklen;
660 17 : params->vol_size = _get_vol_size(params->chunk_size, backing_dev_size);
661 17 : if (params->vol_size == 0) {
662 1 : SPDK_ERRLOG("backing device is too small\n");
663 1 : cb_fn(cb_arg, NULL, -EINVAL);
664 1 : return;
665 : }
666 :
667 16 : if (backing_dev->submit_backing_io == NULL) {
668 1 : SPDK_ERRLOG("backing_dev function pointer not specified\n");
669 1 : cb_fn(cb_arg, NULL, -EINVAL);
670 1 : return;
671 : }
672 :
673 15 : vol = calloc(1, sizeof(*vol));
674 15 : if (vol == NULL) {
675 0 : cb_fn(cb_arg, NULL, -ENOMEM);
676 0 : return;
677 : }
678 :
679 15 : TAILQ_INIT(&vol->free_requests);
680 15 : RB_INIT(&vol->executing_requests);
681 15 : TAILQ_INIT(&vol->queued_requests);
682 15 : queue_init(&vol->free_chunks_queue);
683 15 : queue_init(&vol->free_backing_blocks_queue);
684 :
685 15 : vol->backing_super = spdk_zmalloc(sizeof(*vol->backing_super), 0, NULL,
686 : SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
687 15 : if (vol->backing_super == NULL) {
688 0 : cb_fn(cb_arg, NULL, -ENOMEM);
689 0 : _init_load_cleanup(vol, NULL);
690 0 : return;
691 : }
692 :
693 15 : init_ctx = calloc(1, sizeof(*init_ctx));
694 15 : if (init_ctx == NULL) {
695 0 : cb_fn(cb_arg, NULL, -ENOMEM);
696 0 : _init_load_cleanup(vol, NULL);
697 0 : return;
698 : }
699 :
700 15 : backing_io = calloc(1, sizeof(*backing_io) + backing_dev->user_ctx_size);
701 15 : if (backing_io == NULL) {
702 0 : cb_fn(cb_arg, NULL, -ENOMEM);
703 0 : _init_load_cleanup(vol, init_ctx);
704 0 : return;
705 : }
706 15 : init_ctx->backing_io = backing_io;
707 :
708 15 : init_ctx->path = spdk_zmalloc(REDUCE_PATH_MAX, 0, NULL,
709 : SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
710 15 : if (init_ctx->path == NULL) {
711 0 : cb_fn(cb_arg, NULL, -ENOMEM);
712 0 : _init_load_cleanup(vol, init_ctx);
713 0 : return;
714 : }
715 :
716 15 : if (spdk_uuid_is_null(¶ms->uuid)) {
717 1 : spdk_uuid_generate(¶ms->uuid);
718 1 : }
719 :
720 15 : memcpy(vol->pm_file.path, pm_file_dir, dir_len);
721 15 : vol->pm_file.path[dir_len] = '/';
722 30 : spdk_uuid_fmt_lower(&vol->pm_file.path[dir_len + 1], SPDK_UUID_STRING_LEN,
723 15 : ¶ms->uuid);
724 15 : vol->pm_file.size = _get_pm_file_size(params);
725 30 : vol->pm_file.pm_buf = pmem_map_file(vol->pm_file.path, vol->pm_file.size,
726 15 : PMEM_FILE_CREATE | PMEM_FILE_EXCL, 0600,
727 15 : &mapped_len, &vol->pm_file.pm_is_pmem);
728 15 : if (vol->pm_file.pm_buf == NULL) {
729 0 : SPDK_ERRLOG("could not pmem_map_file(%s): %s\n",
730 : vol->pm_file.path, strerror(errno));
731 0 : cb_fn(cb_arg, NULL, -errno);
732 0 : _init_load_cleanup(vol, init_ctx);
733 0 : return;
734 : }
735 :
736 15 : if (vol->pm_file.size != mapped_len) {
737 0 : SPDK_ERRLOG("could not map entire pmem file (size=%" PRIu64 " mapped=%" PRIu64 ")\n",
738 : vol->pm_file.size, mapped_len);
739 0 : cb_fn(cb_arg, NULL, -ENOMEM);
740 0 : _init_load_cleanup(vol, init_ctx);
741 0 : return;
742 : }
743 :
744 15 : vol->backing_io_units_per_chunk = params->chunk_size / params->backing_io_unit_size;
745 15 : vol->logical_blocks_per_chunk = params->chunk_size / params->logical_block_size;
746 15 : vol->backing_lba_per_io_unit = params->backing_io_unit_size / backing_dev->blocklen;
747 15 : memcpy(&vol->params, params, sizeof(*params));
748 :
749 15 : vol->backing_dev = backing_dev;
750 :
751 15 : rc = _allocate_bit_arrays(vol);
752 15 : if (rc != 0) {
753 0 : cb_fn(cb_arg, NULL, rc);
754 0 : _init_load_cleanup(vol, init_ctx);
755 0 : return;
756 : }
757 :
758 15 : memcpy(vol->backing_super->signature, SPDK_REDUCE_SIGNATURE,
759 : sizeof(vol->backing_super->signature));
760 15 : memcpy(&vol->backing_super->params, params, sizeof(*params));
761 :
762 15 : _initialize_vol_pm_pointers(vol);
763 :
764 15 : memcpy(vol->pm_super, vol->backing_super, sizeof(*vol->backing_super));
765 : /* Writing 0xFF's is equivalent of filling it all with SPDK_EMPTY_MAP_ENTRY.
766 : * Note that this writes 0xFF to not just the logical map but the chunk maps as well.
767 : */
768 15 : memset(vol->pm_logical_map, 0xFF, vol->pm_file.size - sizeof(*vol->backing_super));
769 15 : _reduce_persist(vol, vol->pm_file.pm_buf, vol->pm_file.size);
770 :
771 15 : init_ctx->vol = vol;
772 15 : init_ctx->cb_fn = cb_fn;
773 15 : init_ctx->cb_arg = cb_arg;
774 :
775 15 : memcpy(init_ctx->path, vol->pm_file.path, REDUCE_PATH_MAX);
776 15 : init_ctx->iov[0].iov_base = init_ctx->path;
777 15 : init_ctx->iov[0].iov_len = REDUCE_PATH_MAX;
778 15 : init_ctx->backing_cb_args.cb_fn = _init_write_path_cpl;
779 15 : init_ctx->backing_cb_args.cb_arg = init_ctx;
780 : /* Write path to offset 4K on backing device - just after where the super
781 : * block will be written. We wait until this is committed before writing the
782 : * super block to guarantee we don't get the super block written without the
783 : * the path if the system crashed in the middle of a write operation.
784 : */
785 15 : backing_io->dev = vol->backing_dev;
786 15 : backing_io->iov = init_ctx->iov;
787 15 : backing_io->iovcnt = 1;
788 15 : backing_io->lba = REDUCE_BACKING_DEV_PATH_OFFSET / vol->backing_dev->blocklen;
789 15 : backing_io->lba_count = REDUCE_PATH_MAX / vol->backing_dev->blocklen;
790 15 : backing_io->backing_cb_args = &init_ctx->backing_cb_args;
791 15 : backing_io->backing_io_type = SPDK_REDUCE_BACKING_IO_WRITE;
792 :
793 15 : vol->backing_dev->submit_backing_io(backing_io);
794 17 : }
795 :
796 : static void destroy_load_cb(void *cb_arg, struct spdk_reduce_vol *vol, int reduce_errno);
797 :
798 : static void
799 11 : _load_read_super_and_path_cpl(void *cb_arg, int reduce_errno)
800 : {
801 11 : struct reduce_init_load_ctx *load_ctx = cb_arg;
802 11 : struct spdk_reduce_vol *vol = load_ctx->vol;
803 11 : uint64_t backing_dev_size;
804 11 : uint64_t i, num_chunks, logical_map_index;
805 11 : struct spdk_reduce_chunk_map *chunk;
806 11 : size_t mapped_len;
807 11 : uint32_t j;
808 11 : int rc;
809 :
810 11 : if (reduce_errno != 0) {
811 0 : rc = reduce_errno;
812 0 : goto error;
813 : }
814 :
815 11 : rc = _alloc_zero_buff();
816 11 : if (rc) {
817 0 : goto error;
818 : }
819 :
820 11 : if (memcmp(vol->backing_super->signature,
821 : SPDK_REDUCE_SIGNATURE,
822 11 : sizeof(vol->backing_super->signature)) != 0) {
823 : /* This backing device isn't a libreduce backing device. */
824 1 : rc = -EILSEQ;
825 1 : goto error;
826 : }
827 :
828 : /* If the cb_fn is destroy_load_cb, it means we are wanting to destroy this compress bdev.
829 : * So don't bother getting the volume ready to use - invoke the callback immediately
830 : * so destroy_load_cb can delete the metadata off of the block device and delete the
831 : * persistent memory file if it exists.
832 : */
833 10 : memcpy(vol->pm_file.path, load_ctx->path, sizeof(vol->pm_file.path));
834 10 : if (load_ctx->cb_fn == (*destroy_load_cb)) {
835 1 : load_ctx->cb_fn(load_ctx->cb_arg, vol, 0);
836 1 : _init_load_cleanup(NULL, load_ctx);
837 1 : return;
838 : }
839 :
840 9 : memcpy(&vol->params, &vol->backing_super->params, sizeof(vol->params));
841 9 : vol->backing_io_units_per_chunk = vol->params.chunk_size / vol->params.backing_io_unit_size;
842 9 : vol->logical_blocks_per_chunk = vol->params.chunk_size / vol->params.logical_block_size;
843 9 : vol->backing_lba_per_io_unit = vol->params.backing_io_unit_size / vol->backing_dev->blocklen;
844 :
845 9 : rc = _allocate_bit_arrays(vol);
846 9 : if (rc != 0) {
847 0 : goto error;
848 : }
849 :
850 9 : backing_dev_size = vol->backing_dev->blockcnt * vol->backing_dev->blocklen;
851 9 : if (_get_vol_size(vol->params.chunk_size, backing_dev_size) < vol->params.vol_size) {
852 0 : SPDK_ERRLOG("backing device size %" PRIi64 " smaller than expected\n",
853 : backing_dev_size);
854 0 : rc = -EILSEQ;
855 0 : goto error;
856 : }
857 :
858 9 : vol->pm_file.size = _get_pm_file_size(&vol->params);
859 18 : vol->pm_file.pm_buf = pmem_map_file(vol->pm_file.path, 0, 0, 0, &mapped_len,
860 9 : &vol->pm_file.pm_is_pmem);
861 9 : if (vol->pm_file.pm_buf == NULL) {
862 0 : SPDK_ERRLOG("could not pmem_map_file(%s): %s\n", vol->pm_file.path, strerror(errno));
863 0 : rc = -errno;
864 0 : goto error;
865 : }
866 :
867 9 : if (vol->pm_file.size != mapped_len) {
868 0 : SPDK_ERRLOG("could not map entire pmem file (size=%" PRIu64 " mapped=%" PRIu64 ")\n",
869 : vol->pm_file.size, mapped_len);
870 0 : rc = -ENOMEM;
871 0 : goto error;
872 : }
873 :
874 9 : rc = _allocate_vol_requests(vol);
875 9 : if (rc != 0) {
876 0 : goto error;
877 : }
878 :
879 9 : _initialize_vol_pm_pointers(vol);
880 :
881 9 : num_chunks = vol->params.vol_size / vol->params.chunk_size;
882 1161 : for (i = 0; i < num_chunks; i++) {
883 1152 : logical_map_index = vol->pm_logical_map[i];
884 1152 : if (logical_map_index == REDUCE_EMPTY_MAP_ENTRY) {
885 1146 : continue;
886 : }
887 6 : spdk_bit_array_set(vol->allocated_chunk_maps, logical_map_index);
888 6 : chunk = _reduce_vol_get_chunk_map(vol, logical_map_index);
889 30 : for (j = 0; j < vol->backing_io_units_per_chunk; j++) {
890 24 : if (chunk->io_unit_index[j] != REDUCE_EMPTY_MAP_ENTRY) {
891 12 : spdk_bit_array_set(vol->allocated_backing_io_units, chunk->io_unit_index[j]);
892 12 : vol->info.allocated_io_units++;
893 12 : }
894 24 : }
895 6 : }
896 :
897 9 : load_ctx->cb_fn(load_ctx->cb_arg, vol, 0);
898 : /* Only clean up the ctx - the vol has been passed to the application
899 : * for use now that volume load was successful.
900 : */
901 9 : _init_load_cleanup(NULL, load_ctx);
902 9 : return;
903 :
904 : error:
905 1 : load_ctx->cb_fn(load_ctx->cb_arg, NULL, rc);
906 1 : _init_load_cleanup(vol, load_ctx);
907 11 : }
908 :
909 : void
910 11 : spdk_reduce_vol_load(struct spdk_reduce_backing_dev *backing_dev,
911 : spdk_reduce_vol_op_with_handle_complete cb_fn, void *cb_arg)
912 : {
913 11 : struct spdk_reduce_vol *vol;
914 11 : struct reduce_init_load_ctx *load_ctx;
915 11 : struct spdk_reduce_backing_io *backing_io;
916 :
917 11 : if (backing_dev->submit_backing_io == NULL) {
918 0 : SPDK_ERRLOG("backing_dev function pointer not specified\n");
919 0 : cb_fn(cb_arg, NULL, -EINVAL);
920 0 : return;
921 : }
922 :
923 11 : vol = calloc(1, sizeof(*vol));
924 11 : if (vol == NULL) {
925 0 : cb_fn(cb_arg, NULL, -ENOMEM);
926 0 : return;
927 : }
928 :
929 11 : TAILQ_INIT(&vol->free_requests);
930 11 : RB_INIT(&vol->executing_requests);
931 11 : TAILQ_INIT(&vol->queued_requests);
932 11 : queue_init(&vol->free_chunks_queue);
933 11 : queue_init(&vol->free_backing_blocks_queue);
934 :
935 11 : vol->backing_super = spdk_zmalloc(sizeof(*vol->backing_super), 64, NULL,
936 : SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
937 11 : if (vol->backing_super == NULL) {
938 0 : _init_load_cleanup(vol, NULL);
939 0 : cb_fn(cb_arg, NULL, -ENOMEM);
940 0 : return;
941 : }
942 :
943 11 : vol->backing_dev = backing_dev;
944 :
945 11 : load_ctx = calloc(1, sizeof(*load_ctx));
946 11 : if (load_ctx == NULL) {
947 0 : _init_load_cleanup(vol, NULL);
948 0 : cb_fn(cb_arg, NULL, -ENOMEM);
949 0 : return;
950 : }
951 :
952 11 : backing_io = calloc(1, sizeof(*backing_io) + backing_dev->user_ctx_size);
953 11 : if (backing_io == NULL) {
954 0 : _init_load_cleanup(vol, load_ctx);
955 0 : cb_fn(cb_arg, NULL, -ENOMEM);
956 0 : return;
957 : }
958 :
959 11 : load_ctx->backing_io = backing_io;
960 :
961 11 : load_ctx->path = spdk_zmalloc(REDUCE_PATH_MAX, 64, NULL,
962 : SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
963 11 : if (load_ctx->path == NULL) {
964 0 : _init_load_cleanup(vol, load_ctx);
965 0 : cb_fn(cb_arg, NULL, -ENOMEM);
966 0 : return;
967 : }
968 :
969 11 : load_ctx->vol = vol;
970 11 : load_ctx->cb_fn = cb_fn;
971 11 : load_ctx->cb_arg = cb_arg;
972 :
973 11 : load_ctx->iov[0].iov_base = vol->backing_super;
974 11 : load_ctx->iov[0].iov_len = sizeof(*vol->backing_super);
975 11 : load_ctx->iov[1].iov_base = load_ctx->path;
976 11 : load_ctx->iov[1].iov_len = REDUCE_PATH_MAX;
977 11 : backing_io->dev = vol->backing_dev;
978 11 : backing_io->iov = load_ctx->iov;
979 11 : backing_io->iovcnt = LOAD_IOV_COUNT;
980 11 : backing_io->lba = 0;
981 11 : backing_io->lba_count = (sizeof(*vol->backing_super) + REDUCE_PATH_MAX) /
982 11 : vol->backing_dev->blocklen;
983 11 : backing_io->backing_cb_args = &load_ctx->backing_cb_args;
984 11 : backing_io->backing_io_type = SPDK_REDUCE_BACKING_IO_READ;
985 :
986 11 : load_ctx->backing_cb_args.cb_fn = _load_read_super_and_path_cpl;
987 11 : load_ctx->backing_cb_args.cb_arg = load_ctx;
988 11 : vol->backing_dev->submit_backing_io(backing_io);
989 11 : }
990 :
991 : void
992 25 : spdk_reduce_vol_unload(struct spdk_reduce_vol *vol,
993 : spdk_reduce_vol_op_complete cb_fn, void *cb_arg)
994 : {
995 25 : if (vol == NULL) {
996 : /* This indicates a programming error. */
997 0 : assert(false);
998 : cb_fn(cb_arg, -EINVAL);
999 : return;
1000 : }
1001 :
1002 25 : if (--g_vol_count == 0) {
1003 23 : spdk_free(g_zero_buf);
1004 23 : }
1005 25 : assert(g_vol_count >= 0);
1006 25 : _init_load_cleanup(vol, NULL);
1007 25 : cb_fn(cb_arg, 0);
1008 25 : }
1009 :
1010 : struct reduce_destroy_ctx {
1011 : spdk_reduce_vol_op_complete cb_fn;
1012 : void *cb_arg;
1013 : struct spdk_reduce_vol *vol;
1014 : struct spdk_reduce_vol_superblock *super;
1015 : struct iovec iov;
1016 : struct spdk_reduce_vol_cb_args backing_cb_args;
1017 : int reduce_errno;
1018 : char pm_path[REDUCE_PATH_MAX];
1019 : struct spdk_reduce_backing_io *backing_io;
1020 : };
1021 :
1022 : static void
1023 1 : destroy_unload_cpl(void *cb_arg, int reduce_errno)
1024 : {
1025 1 : struct reduce_destroy_ctx *destroy_ctx = cb_arg;
1026 :
1027 1 : if (destroy_ctx->reduce_errno == 0) {
1028 1 : if (unlink(destroy_ctx->pm_path)) {
1029 0 : SPDK_ERRLOG("%s could not be unlinked: %s\n",
1030 : destroy_ctx->pm_path, strerror(errno));
1031 0 : }
1032 1 : }
1033 :
1034 : /* Even if the unload somehow failed, we still pass the destroy_ctx
1035 : * reduce_errno since that indicates whether or not the volume was
1036 : * actually destroyed.
1037 : */
1038 1 : destroy_ctx->cb_fn(destroy_ctx->cb_arg, destroy_ctx->reduce_errno);
1039 1 : spdk_free(destroy_ctx->super);
1040 1 : free(destroy_ctx->backing_io);
1041 1 : free(destroy_ctx);
1042 1 : }
1043 :
1044 : static void
1045 1 : _destroy_zero_super_cpl(void *cb_arg, int reduce_errno)
1046 : {
1047 1 : struct reduce_destroy_ctx *destroy_ctx = cb_arg;
1048 1 : struct spdk_reduce_vol *vol = destroy_ctx->vol;
1049 :
1050 1 : destroy_ctx->reduce_errno = reduce_errno;
1051 1 : spdk_reduce_vol_unload(vol, destroy_unload_cpl, destroy_ctx);
1052 1 : }
1053 :
1054 : static void
1055 1 : destroy_load_cb(void *cb_arg, struct spdk_reduce_vol *vol, int reduce_errno)
1056 : {
1057 1 : struct reduce_destroy_ctx *destroy_ctx = cb_arg;
1058 1 : struct spdk_reduce_backing_io *backing_io = destroy_ctx->backing_io;
1059 :
1060 1 : if (reduce_errno != 0) {
1061 0 : destroy_ctx->cb_fn(destroy_ctx->cb_arg, reduce_errno);
1062 0 : spdk_free(destroy_ctx->super);
1063 0 : free(destroy_ctx);
1064 0 : return;
1065 : }
1066 :
1067 1 : destroy_ctx->vol = vol;
1068 1 : memcpy(destroy_ctx->pm_path, vol->pm_file.path, sizeof(destroy_ctx->pm_path));
1069 1 : destroy_ctx->iov.iov_base = destroy_ctx->super;
1070 1 : destroy_ctx->iov.iov_len = sizeof(*destroy_ctx->super);
1071 1 : destroy_ctx->backing_cb_args.cb_fn = _destroy_zero_super_cpl;
1072 1 : destroy_ctx->backing_cb_args.cb_arg = destroy_ctx;
1073 :
1074 1 : backing_io->dev = vol->backing_dev;
1075 1 : backing_io->iov = &destroy_ctx->iov;
1076 1 : backing_io->iovcnt = 1;
1077 1 : backing_io->lba = 0;
1078 1 : backing_io->lba_count = sizeof(*destroy_ctx->super) / vol->backing_dev->blocklen;
1079 1 : backing_io->backing_cb_args = &destroy_ctx->backing_cb_args;
1080 1 : backing_io->backing_io_type = SPDK_REDUCE_BACKING_IO_WRITE;
1081 :
1082 1 : vol->backing_dev->submit_backing_io(backing_io);
1083 1 : }
1084 :
1085 : void
1086 1 : spdk_reduce_vol_destroy(struct spdk_reduce_backing_dev *backing_dev,
1087 : spdk_reduce_vol_op_complete cb_fn, void *cb_arg)
1088 : {
1089 1 : struct reduce_destroy_ctx *destroy_ctx;
1090 1 : struct spdk_reduce_backing_io *backing_io;
1091 :
1092 1 : destroy_ctx = calloc(1, sizeof(*destroy_ctx));
1093 1 : if (destroy_ctx == NULL) {
1094 0 : cb_fn(cb_arg, -ENOMEM);
1095 0 : return;
1096 : }
1097 :
1098 1 : backing_io = calloc(1, sizeof(*backing_io) + backing_dev->user_ctx_size);
1099 1 : if (backing_io == NULL) {
1100 0 : free(destroy_ctx);
1101 0 : cb_fn(cb_arg, -ENOMEM);
1102 0 : return;
1103 : }
1104 :
1105 1 : destroy_ctx->backing_io = backing_io;
1106 :
1107 1 : destroy_ctx->super = spdk_zmalloc(sizeof(*destroy_ctx->super), 64, NULL,
1108 : SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
1109 1 : if (destroy_ctx->super == NULL) {
1110 0 : free(destroy_ctx);
1111 0 : free(backing_io);
1112 0 : cb_fn(cb_arg, -ENOMEM);
1113 0 : return;
1114 : }
1115 1 : destroy_ctx->cb_fn = cb_fn;
1116 1 : destroy_ctx->cb_arg = cb_arg;
1117 1 : spdk_reduce_vol_load(backing_dev, destroy_load_cb, destroy_ctx);
1118 1 : }
1119 :
1120 : static bool
1121 283 : _request_spans_chunk_boundary(struct spdk_reduce_vol *vol, uint64_t offset, uint64_t length)
1122 : {
1123 283 : uint64_t start_chunk, end_chunk;
1124 :
1125 283 : start_chunk = offset / vol->logical_blocks_per_chunk;
1126 283 : end_chunk = (offset + length - 1) / vol->logical_blocks_per_chunk;
1127 :
1128 566 : return (start_chunk != end_chunk);
1129 283 : }
1130 :
1131 : typedef void (*reduce_request_fn)(void *_req, int reduce_errno);
1132 : static void _start_unmap_request_full_chunk(void *ctx);
1133 :
1134 : static void
1135 288 : _reduce_vol_complete_req(struct spdk_reduce_vol_request *req, int reduce_errno)
1136 : {
1137 288 : struct spdk_reduce_vol_request *next_req;
1138 288 : struct spdk_reduce_vol *vol = req->vol;
1139 :
1140 288 : req->cb_fn(req->cb_arg, reduce_errno);
1141 288 : RB_REMOVE(executing_req_tree, &vol->executing_requests, req);
1142 :
1143 288 : TAILQ_FOREACH(next_req, &vol->queued_requests, tailq) {
1144 1 : if (next_req->logical_map_index == req->logical_map_index) {
1145 1 : TAILQ_REMOVE(&vol->queued_requests, next_req, tailq);
1146 1 : if (next_req->type == REDUCE_IO_READV) {
1147 0 : _start_readv_request(next_req);
1148 1 : } else if (next_req->type == REDUCE_IO_WRITEV) {
1149 1 : _start_writev_request(next_req);
1150 1 : } else {
1151 0 : assert(next_req->type == REDUCE_IO_UNMAP);
1152 0 : _start_unmap_request_full_chunk(next_req);
1153 : }
1154 1 : break;
1155 : }
1156 0 : }
1157 :
1158 288 : TAILQ_INSERT_HEAD(&vol->free_requests, req, tailq);
1159 288 : }
1160 :
1161 : static void
1162 7 : _reduce_vol_reset_chunk(struct spdk_reduce_vol *vol, uint64_t chunk_map_index)
1163 : {
1164 7 : struct spdk_reduce_chunk_map *chunk;
1165 7 : uint64_t index;
1166 7 : bool success;
1167 7 : uint32_t i;
1168 :
1169 7 : chunk = _reduce_vol_get_chunk_map(vol, chunk_map_index);
1170 20 : for (i = 0; i < vol->backing_io_units_per_chunk; i++) {
1171 18 : index = chunk->io_unit_index[i];
1172 18 : if (index == REDUCE_EMPTY_MAP_ENTRY) {
1173 5 : break;
1174 : }
1175 13 : assert(spdk_bit_array_get(vol->allocated_backing_io_units,
1176 : index) == true);
1177 13 : spdk_bit_array_clear(vol->allocated_backing_io_units, index);
1178 13 : vol->info.allocated_io_units--;
1179 13 : success = queue_enqueue(&vol->free_backing_blocks_queue, index);
1180 13 : if (!success && index < vol->find_block_offset) {
1181 0 : vol->find_block_offset = index;
1182 0 : }
1183 13 : chunk->io_unit_index[i] = REDUCE_EMPTY_MAP_ENTRY;
1184 13 : }
1185 7 : success = queue_enqueue(&vol->free_chunks_queue, chunk_map_index);
1186 7 : if (!success && chunk_map_index < vol->find_chunk_offset) {
1187 0 : vol->find_chunk_offset = chunk_map_index;
1188 0 : }
1189 7 : spdk_bit_array_clear(vol->allocated_chunk_maps, chunk_map_index);
1190 7 : }
1191 :
1192 : static void
1193 16 : _write_write_done(void *_req, int reduce_errno)
1194 : {
1195 16 : struct spdk_reduce_vol_request *req = _req;
1196 16 : struct spdk_reduce_vol *vol = req->vol;
1197 16 : uint64_t old_chunk_map_index;
1198 :
1199 16 : if (reduce_errno != 0) {
1200 0 : req->reduce_errno = reduce_errno;
1201 0 : }
1202 :
1203 16 : assert(req->num_backing_ops > 0);
1204 16 : if (--req->num_backing_ops > 0) {
1205 0 : return;
1206 : }
1207 :
1208 16 : if (req->reduce_errno != 0) {
1209 0 : _reduce_vol_reset_chunk(vol, req->chunk_map_index);
1210 0 : _reduce_vol_complete_req(req, req->reduce_errno);
1211 0 : return;
1212 : }
1213 :
1214 16 : old_chunk_map_index = vol->pm_logical_map[req->logical_map_index];
1215 16 : if (old_chunk_map_index != REDUCE_EMPTY_MAP_ENTRY) {
1216 6 : _reduce_vol_reset_chunk(vol, old_chunk_map_index);
1217 6 : }
1218 :
1219 : /*
1220 : * We don't need to persist the clearing of the old chunk map here. The old chunk map
1221 : * becomes invalid after we update the logical map, since the old chunk map will no
1222 : * longer have a reference to it in the logical map.
1223 : */
1224 :
1225 : /* Persist the new chunk map. This must be persisted before we update the logical map. */
1226 32 : _reduce_persist(vol, req->chunk,
1227 16 : _reduce_vol_get_chunk_struct_size(vol->backing_io_units_per_chunk));
1228 :
1229 16 : vol->pm_logical_map[req->logical_map_index] = req->chunk_map_index;
1230 :
1231 16 : _reduce_persist(vol, &vol->pm_logical_map[req->logical_map_index], sizeof(uint64_t));
1232 :
1233 16 : _reduce_vol_complete_req(req, 0);
1234 16 : }
1235 :
1236 : static struct spdk_reduce_backing_io *
1237 283 : _reduce_vol_req_get_backing_io(struct spdk_reduce_vol_request *req, uint32_t index)
1238 : {
1239 283 : struct spdk_reduce_backing_dev *backing_dev = req->vol->backing_dev;
1240 283 : struct spdk_reduce_backing_io *backing_io;
1241 :
1242 566 : backing_io = (struct spdk_reduce_backing_io *)((uint8_t *)req->backing_io +
1243 283 : (sizeof(*backing_io) + backing_dev->user_ctx_size) * index);
1244 :
1245 566 : return backing_io;
1246 :
1247 283 : }
1248 :
1249 : struct reduce_merged_io_desc {
1250 : uint64_t io_unit_index;
1251 : uint32_t num_io_units;
1252 : };
1253 :
1254 : static void
1255 0 : _issue_backing_ops_without_merge(struct spdk_reduce_vol_request *req, struct spdk_reduce_vol *vol,
1256 : reduce_request_fn next_fn, bool is_write)
1257 : {
1258 0 : struct iovec *iov;
1259 0 : struct spdk_reduce_backing_io *backing_io;
1260 0 : uint8_t *buf;
1261 0 : uint32_t i;
1262 :
1263 0 : if (req->chunk_is_compressed) {
1264 0 : iov = req->comp_buf_iov;
1265 0 : buf = req->comp_buf;
1266 0 : } else {
1267 0 : iov = req->decomp_buf_iov;
1268 0 : buf = req->decomp_buf;
1269 : }
1270 :
1271 0 : req->num_backing_ops = req->num_io_units;
1272 0 : req->backing_cb_args.cb_fn = next_fn;
1273 0 : req->backing_cb_args.cb_arg = req;
1274 0 : for (i = 0; i < req->num_io_units; i++) {
1275 0 : backing_io = _reduce_vol_req_get_backing_io(req, i);
1276 0 : iov[i].iov_base = buf + i * vol->params.backing_io_unit_size;
1277 0 : iov[i].iov_len = vol->params.backing_io_unit_size;
1278 0 : backing_io->dev = vol->backing_dev;
1279 0 : backing_io->iov = &iov[i];
1280 0 : backing_io->iovcnt = 1;
1281 0 : backing_io->lba = req->chunk->io_unit_index[i] * vol->backing_lba_per_io_unit;
1282 0 : backing_io->lba_count = vol->backing_lba_per_io_unit;
1283 0 : backing_io->backing_cb_args = &req->backing_cb_args;
1284 0 : if (is_write) {
1285 0 : backing_io->backing_io_type = SPDK_REDUCE_BACKING_IO_WRITE;
1286 0 : } else {
1287 0 : backing_io->backing_io_type = SPDK_REDUCE_BACKING_IO_READ;
1288 : }
1289 0 : vol->backing_dev->submit_backing_io(backing_io);
1290 0 : }
1291 0 : }
1292 :
1293 : static void
1294 283 : _issue_backing_ops(struct spdk_reduce_vol_request *req, struct spdk_reduce_vol *vol,
1295 : reduce_request_fn next_fn, bool is_write)
1296 : {
1297 283 : struct iovec *iov;
1298 283 : struct spdk_reduce_backing_io *backing_io;
1299 283 : struct reduce_merged_io_desc merged_io_desc[4];
1300 283 : uint8_t *buf;
1301 283 : bool merge = false;
1302 283 : uint32_t num_io = 0;
1303 283 : uint32_t io_unit_counts = 0;
1304 283 : uint32_t merged_io_idx = 0;
1305 283 : uint32_t i;
1306 :
1307 : /* The merged_io_desc value is defined here to contain four elements,
1308 : * and the chunk size must be four times the maximum of the io unit.
1309 : * if chunk size is too big, don't merge IO.
1310 : */
1311 283 : if (vol->backing_io_units_per_chunk > 4) {
1312 0 : _issue_backing_ops_without_merge(req, vol, next_fn, is_write);
1313 0 : return;
1314 : }
1315 :
1316 283 : if (req->chunk_is_compressed) {
1317 279 : iov = req->comp_buf_iov;
1318 279 : buf = req->comp_buf;
1319 279 : } else {
1320 4 : iov = req->decomp_buf_iov;
1321 4 : buf = req->decomp_buf;
1322 : }
1323 :
1324 295 : for (i = 0; i < req->num_io_units; i++) {
1325 295 : if (!merge) {
1326 283 : merged_io_desc[merged_io_idx].io_unit_index = req->chunk->io_unit_index[i];
1327 283 : merged_io_desc[merged_io_idx].num_io_units = 1;
1328 283 : num_io++;
1329 283 : }
1330 :
1331 295 : if (i + 1 == req->num_io_units) {
1332 283 : break;
1333 : }
1334 :
1335 12 : if (req->chunk->io_unit_index[i] + 1 == req->chunk->io_unit_index[i + 1]) {
1336 12 : merged_io_desc[merged_io_idx].num_io_units += 1;
1337 12 : merge = true;
1338 12 : continue;
1339 : }
1340 0 : merge = false;
1341 0 : merged_io_idx++;
1342 0 : }
1343 :
1344 283 : req->num_backing_ops = num_io;
1345 283 : req->backing_cb_args.cb_fn = next_fn;
1346 283 : req->backing_cb_args.cb_arg = req;
1347 566 : for (i = 0; i < num_io; i++) {
1348 283 : backing_io = _reduce_vol_req_get_backing_io(req, i);
1349 283 : iov[i].iov_base = buf + io_unit_counts * vol->params.backing_io_unit_size;
1350 283 : iov[i].iov_len = vol->params.backing_io_unit_size * merged_io_desc[i].num_io_units;
1351 283 : backing_io->dev = vol->backing_dev;
1352 283 : backing_io->iov = &iov[i];
1353 283 : backing_io->iovcnt = 1;
1354 283 : backing_io->lba = merged_io_desc[i].io_unit_index * vol->backing_lba_per_io_unit;
1355 283 : backing_io->lba_count = vol->backing_lba_per_io_unit * merged_io_desc[i].num_io_units;
1356 283 : backing_io->backing_cb_args = &req->backing_cb_args;
1357 283 : if (is_write) {
1358 16 : backing_io->backing_io_type = SPDK_REDUCE_BACKING_IO_WRITE;
1359 16 : } else {
1360 267 : backing_io->backing_io_type = SPDK_REDUCE_BACKING_IO_READ;
1361 : }
1362 283 : vol->backing_dev->submit_backing_io(backing_io);
1363 :
1364 : /* Collects the number of processed I/O. */
1365 283 : io_unit_counts += merged_io_desc[i].num_io_units;
1366 283 : }
1367 283 : }
1368 :
1369 : static void
1370 16 : _reduce_vol_write_chunk(struct spdk_reduce_vol_request *req, reduce_request_fn next_fn,
1371 : uint32_t compressed_size)
1372 : {
1373 16 : struct spdk_reduce_vol *vol = req->vol;
1374 16 : uint32_t i;
1375 16 : uint64_t chunk_offset, remainder, free_index, total_len = 0;
1376 16 : uint8_t *buf;
1377 16 : bool success;
1378 16 : int j;
1379 :
1380 16 : success = queue_dequeue(&vol->free_chunks_queue, &free_index);
1381 16 : if (success) {
1382 0 : req->chunk_map_index = free_index;
1383 0 : } else {
1384 32 : req->chunk_map_index = spdk_bit_array_find_first_clear(vol->allocated_chunk_maps,
1385 16 : vol->find_chunk_offset);
1386 16 : vol->find_chunk_offset = req->chunk_map_index + 1;
1387 : }
1388 :
1389 : /* TODO: fail if no chunk map found - but really this should not happen if we
1390 : * size the number of requests similarly to number of extra chunk maps
1391 : */
1392 16 : assert(req->chunk_map_index != REDUCE_EMPTY_MAP_ENTRY);
1393 16 : spdk_bit_array_set(vol->allocated_chunk_maps, req->chunk_map_index);
1394 :
1395 16 : req->chunk = _reduce_vol_get_chunk_map(vol, req->chunk_map_index);
1396 32 : req->num_io_units = spdk_divide_round_up(compressed_size,
1397 16 : vol->params.backing_io_unit_size);
1398 16 : req->chunk_is_compressed = (req->num_io_units != vol->backing_io_units_per_chunk);
1399 16 : req->chunk->compressed_size =
1400 16 : req->chunk_is_compressed ? compressed_size : vol->params.chunk_size;
1401 :
1402 : /* if the chunk is uncompressed we need to copy the data from the host buffers. */
1403 16 : if (req->chunk_is_compressed == false) {
1404 4 : chunk_offset = req->offset % vol->logical_blocks_per_chunk;
1405 4 : buf = req->decomp_buf;
1406 4 : total_len = chunk_offset * vol->params.logical_block_size;
1407 :
1408 : /* zero any offset into chunk */
1409 4 : if (req->rmw == false && chunk_offset) {
1410 0 : memset(buf, 0, total_len);
1411 0 : }
1412 4 : buf += total_len;
1413 :
1414 : /* copy the data */
1415 8 : for (j = 0; j < req->iovcnt; j++) {
1416 4 : memcpy(buf, req->iov[j].iov_base, req->iov[j].iov_len);
1417 4 : buf += req->iov[j].iov_len;
1418 4 : total_len += req->iov[j].iov_len;
1419 4 : }
1420 :
1421 : /* zero any remainder */
1422 4 : remainder = vol->params.chunk_size - total_len;
1423 4 : total_len += remainder;
1424 4 : if (req->rmw == false && remainder) {
1425 0 : memset(buf, 0, remainder);
1426 0 : }
1427 4 : assert(total_len == vol->params.chunk_size);
1428 4 : }
1429 :
1430 44 : for (i = 0; i < req->num_io_units; i++) {
1431 28 : success = queue_dequeue(&vol->free_backing_blocks_queue, &free_index);
1432 28 : if (success) {
1433 0 : req->chunk->io_unit_index[i] = free_index;
1434 0 : } else {
1435 56 : req->chunk->io_unit_index[i] = spdk_bit_array_find_first_clear(vol->allocated_backing_io_units,
1436 28 : vol->find_block_offset);
1437 28 : vol->find_block_offset = req->chunk->io_unit_index[i] + 1;
1438 : }
1439 : /* TODO: fail if no backing block found - but really this should also not
1440 : * happen (see comment above).
1441 : */
1442 28 : assert(req->chunk->io_unit_index[i] != REDUCE_EMPTY_MAP_ENTRY);
1443 28 : spdk_bit_array_set(vol->allocated_backing_io_units, req->chunk->io_unit_index[i]);
1444 28 : vol->info.allocated_io_units++;
1445 28 : }
1446 :
1447 16 : _issue_backing_ops(req, vol, next_fn, true /* write */);
1448 16 : }
1449 :
1450 : static void
1451 16 : _write_compress_done(void *_req, int reduce_errno)
1452 : {
1453 16 : struct spdk_reduce_vol_request *req = _req;
1454 :
1455 : /* Negative reduce_errno indicates failure for compression operations.
1456 : * Just write the uncompressed data instead. Force this to happen
1457 : * by just passing the full chunk size to _reduce_vol_write_chunk.
1458 : * When it sees the data couldn't be compressed, it will just write
1459 : * the uncompressed buffer to disk.
1460 : */
1461 16 : if (reduce_errno < 0) {
1462 4 : req->backing_cb_args.output_size = req->vol->params.chunk_size;
1463 4 : }
1464 :
1465 16 : _reduce_vol_write_chunk(req, _write_write_done, req->backing_cb_args.output_size);
1466 16 : }
1467 :
1468 : static void
1469 16 : _reduce_vol_compress_chunk(struct spdk_reduce_vol_request *req, reduce_request_fn next_fn)
1470 : {
1471 16 : struct spdk_reduce_vol *vol = req->vol;
1472 :
1473 16 : req->backing_cb_args.cb_fn = next_fn;
1474 16 : req->backing_cb_args.cb_arg = req;
1475 16 : req->comp_buf_iov[0].iov_base = req->comp_buf;
1476 16 : req->comp_buf_iov[0].iov_len = vol->params.chunk_size;
1477 32 : vol->backing_dev->compress(vol->backing_dev,
1478 16 : req->decomp_iov, req->decomp_iovcnt, req->comp_buf_iov, 1,
1479 16 : &req->backing_cb_args);
1480 16 : }
1481 :
1482 : static void
1483 4 : _reduce_vol_decompress_chunk_scratch(struct spdk_reduce_vol_request *req, reduce_request_fn next_fn)
1484 : {
1485 4 : struct spdk_reduce_vol *vol = req->vol;
1486 :
1487 4 : req->backing_cb_args.cb_fn = next_fn;
1488 4 : req->backing_cb_args.cb_arg = req;
1489 4 : req->comp_buf_iov[0].iov_base = req->comp_buf;
1490 4 : req->comp_buf_iov[0].iov_len = req->chunk->compressed_size;
1491 4 : req->decomp_buf_iov[0].iov_base = req->decomp_buf;
1492 4 : req->decomp_buf_iov[0].iov_len = vol->params.chunk_size;
1493 8 : vol->backing_dev->decompress(vol->backing_dev,
1494 4 : req->comp_buf_iov, 1, req->decomp_buf_iov, 1,
1495 4 : &req->backing_cb_args);
1496 4 : }
1497 :
1498 : static void
1499 271 : _reduce_vol_decompress_chunk(struct spdk_reduce_vol_request *req, reduce_request_fn next_fn)
1500 : {
1501 271 : struct spdk_reduce_vol *vol = req->vol;
1502 271 : uint64_t chunk_offset, remainder = 0;
1503 271 : uint64_t ttl_len = 0;
1504 271 : size_t iov_len;
1505 271 : int i;
1506 :
1507 271 : req->decomp_iovcnt = 0;
1508 271 : chunk_offset = req->offset % vol->logical_blocks_per_chunk;
1509 :
1510 : /* If backing device doesn't support SGL output then we should copy the result of decompression to user's buffer
1511 : * if at least one of the conditions below is true:
1512 : * 1. User's buffer is fragmented
1513 : * 2. Length of the user's buffer is less than the chunk
1514 : * 3. User's buffer is contig, equals chunk_size but crosses huge page boundary */
1515 271 : iov_len = req->iov[0].iov_len;
1516 276 : req->copy_after_decompress = !vol->backing_dev->sgl_out && (req->iovcnt > 1 ||
1517 4 : req->iov[0].iov_len < vol->params.chunk_size ||
1518 2 : _addr_crosses_huge_page(req->iov[0].iov_base, &iov_len));
1519 271 : if (req->copy_after_decompress) {
1520 4 : req->decomp_iov[0].iov_base = req->decomp_buf;
1521 4 : req->decomp_iov[0].iov_len = vol->params.chunk_size;
1522 4 : req->decomp_iovcnt = 1;
1523 4 : goto decompress;
1524 : }
1525 :
1526 267 : if (chunk_offset) {
1527 : /* first iov point to our scratch buffer for any offset into the chunk */
1528 249 : req->decomp_iov[0].iov_base = req->decomp_buf;
1529 249 : req->decomp_iov[0].iov_len = chunk_offset * vol->params.logical_block_size;
1530 249 : ttl_len += req->decomp_iov[0].iov_len;
1531 249 : req->decomp_iovcnt = 1;
1532 249 : }
1533 :
1534 : /* now the user data iov, direct to the user buffer */
1535 537 : for (i = 0; i < req->iovcnt; i++) {
1536 270 : req->decomp_iov[i + req->decomp_iovcnt].iov_base = req->iov[i].iov_base;
1537 270 : req->decomp_iov[i + req->decomp_iovcnt].iov_len = req->iov[i].iov_len;
1538 270 : ttl_len += req->decomp_iov[i + req->decomp_iovcnt].iov_len;
1539 270 : }
1540 267 : req->decomp_iovcnt += req->iovcnt;
1541 :
1542 : /* send the rest of the chunk to our scratch buffer */
1543 267 : remainder = vol->params.chunk_size - ttl_len;
1544 267 : if (remainder) {
1545 252 : req->decomp_iov[req->decomp_iovcnt].iov_base = req->decomp_buf + ttl_len;
1546 252 : req->decomp_iov[req->decomp_iovcnt].iov_len = remainder;
1547 252 : ttl_len += req->decomp_iov[req->decomp_iovcnt].iov_len;
1548 252 : req->decomp_iovcnt++;
1549 252 : }
1550 267 : assert(ttl_len == vol->params.chunk_size);
1551 :
1552 : decompress:
1553 271 : assert(!req->copy_after_decompress || (req->copy_after_decompress && req->decomp_iovcnt == 1));
1554 271 : req->backing_cb_args.cb_fn = next_fn;
1555 271 : req->backing_cb_args.cb_arg = req;
1556 271 : req->comp_buf_iov[0].iov_base = req->comp_buf;
1557 271 : req->comp_buf_iov[0].iov_len = req->chunk->compressed_size;
1558 542 : vol->backing_dev->decompress(vol->backing_dev,
1559 271 : req->comp_buf_iov, 1, req->decomp_iov, req->decomp_iovcnt,
1560 271 : &req->backing_cb_args);
1561 271 : }
1562 :
1563 : static inline void
1564 8 : _prepare_compress_chunk_copy_user_buffers(struct spdk_reduce_vol_request *req, bool zero_paddings)
1565 : {
1566 8 : struct spdk_reduce_vol *vol = req->vol;
1567 8 : uint64_t chunk_offset, ttl_len = 0;
1568 8 : uint64_t remainder = 0;
1569 8 : char *copy_offset = NULL;
1570 8 : uint32_t lbsize = vol->params.logical_block_size;
1571 8 : int i;
1572 :
1573 8 : req->decomp_iov[0].iov_base = req->decomp_buf;
1574 8 : req->decomp_iov[0].iov_len = vol->params.chunk_size;
1575 8 : req->decomp_iovcnt = 1;
1576 8 : copy_offset = req->decomp_iov[0].iov_base;
1577 8 : chunk_offset = req->offset % vol->logical_blocks_per_chunk;
1578 :
1579 8 : if (chunk_offset) {
1580 2 : ttl_len += chunk_offset * lbsize;
1581 : /* copy_offset already points to the correct buffer if zero_paddings=false */
1582 2 : if (zero_paddings) {
1583 1 : memset(copy_offset, 0, ttl_len);
1584 1 : }
1585 2 : copy_offset += ttl_len;
1586 2 : }
1587 :
1588 : /* now the user data iov, direct from the user buffer */
1589 22 : for (i = 0; i < req->iovcnt; i++) {
1590 14 : memcpy(copy_offset, req->iov[i].iov_base, req->iov[i].iov_len);
1591 14 : copy_offset += req->iov[i].iov_len;
1592 14 : ttl_len += req->iov[i].iov_len;
1593 14 : }
1594 :
1595 8 : remainder = vol->params.chunk_size - ttl_len;
1596 8 : if (remainder) {
1597 : /* copy_offset already points to the correct buffer if zero_paddings=false */
1598 4 : if (zero_paddings) {
1599 2 : memset(copy_offset, 0, remainder);
1600 2 : }
1601 4 : ttl_len += remainder;
1602 4 : }
1603 :
1604 8 : assert(ttl_len == req->vol->params.chunk_size);
1605 8 : }
1606 :
1607 : /* This function can be called when we are compressing a new data or in case of read-modify-write
1608 : * In the first case possible paddings should be filled with zeroes, in the second case the paddings
1609 : * should point to already read and decompressed buffer */
1610 : static inline void
1611 33 : _prepare_compress_chunk(struct spdk_reduce_vol_request *req, bool zero_paddings)
1612 : {
1613 33 : struct spdk_reduce_vol *vol = req->vol;
1614 33 : char *padding_buffer = zero_paddings ? g_zero_buf : req->decomp_buf;
1615 33 : uint64_t chunk_offset, ttl_len = 0;
1616 33 : uint64_t remainder = 0;
1617 33 : uint32_t lbsize = vol->params.logical_block_size;
1618 33 : size_t iov_len;
1619 33 : int i;
1620 :
1621 : /* If backing device doesn't support SGL input then we should copy user's buffer into decomp_buf
1622 : * if at least one of the conditions below is true:
1623 : * 1. User's buffer is fragmented
1624 : * 2. Length of the user's buffer is less than the chunk
1625 : * 3. User's buffer is contig, equals chunk_size but crosses huge page boundary */
1626 33 : iov_len = req->iov[0].iov_len;
1627 37 : if (!vol->backing_dev->sgl_in && (req->iovcnt > 1 ||
1628 4 : req->iov[0].iov_len < vol->params.chunk_size ||
1629 4 : _addr_crosses_huge_page(req->iov[0].iov_base, &iov_len))) {
1630 8 : _prepare_compress_chunk_copy_user_buffers(req, zero_paddings);
1631 8 : return;
1632 : }
1633 :
1634 25 : req->decomp_iovcnt = 0;
1635 25 : chunk_offset = req->offset % vol->logical_blocks_per_chunk;
1636 :
1637 25 : if (chunk_offset != 0) {
1638 11 : ttl_len += chunk_offset * lbsize;
1639 11 : req->decomp_iov[0].iov_base = padding_buffer;
1640 11 : req->decomp_iov[0].iov_len = ttl_len;
1641 11 : req->decomp_iovcnt = 1;
1642 11 : }
1643 :
1644 : /* now the user data iov, direct from the user buffer */
1645 57 : for (i = 0; i < req->iovcnt; i++) {
1646 32 : req->decomp_iov[i + req->decomp_iovcnt].iov_base = req->iov[i].iov_base;
1647 32 : req->decomp_iov[i + req->decomp_iovcnt].iov_len = req->iov[i].iov_len;
1648 32 : ttl_len += req->iov[i].iov_len;
1649 32 : }
1650 25 : req->decomp_iovcnt += req->iovcnt;
1651 :
1652 25 : remainder = vol->params.chunk_size - ttl_len;
1653 25 : if (remainder) {
1654 14 : req->decomp_iov[req->decomp_iovcnt].iov_base = padding_buffer + ttl_len;
1655 14 : req->decomp_iov[req->decomp_iovcnt].iov_len = remainder;
1656 14 : req->decomp_iovcnt++;
1657 14 : ttl_len += remainder;
1658 14 : }
1659 25 : assert(ttl_len == req->vol->params.chunk_size);
1660 33 : }
1661 :
1662 : static void
1663 4 : _write_decompress_done(void *_req, int reduce_errno)
1664 : {
1665 4 : struct spdk_reduce_vol_request *req = _req;
1666 :
1667 : /* Negative reduce_errno indicates failure for compression operations. */
1668 4 : if (reduce_errno < 0) {
1669 0 : _reduce_vol_complete_req(req, reduce_errno);
1670 0 : return;
1671 : }
1672 :
1673 : /* Positive reduce_errno indicates that the output size field in the backing_cb_args
1674 : * represents the output_size.
1675 : */
1676 4 : if (req->backing_cb_args.output_size != req->vol->params.chunk_size) {
1677 0 : _reduce_vol_complete_req(req, -EIO);
1678 0 : return;
1679 : }
1680 :
1681 4 : _prepare_compress_chunk(req, false);
1682 4 : _reduce_vol_compress_chunk(req, _write_compress_done);
1683 4 : }
1684 :
1685 : static void
1686 4 : _write_read_done(void *_req, int reduce_errno)
1687 : {
1688 4 : struct spdk_reduce_vol_request *req = _req;
1689 :
1690 4 : if (reduce_errno != 0) {
1691 0 : req->reduce_errno = reduce_errno;
1692 0 : }
1693 :
1694 4 : assert(req->num_backing_ops > 0);
1695 4 : if (--req->num_backing_ops > 0) {
1696 0 : return;
1697 : }
1698 :
1699 4 : if (req->reduce_errno != 0) {
1700 0 : _reduce_vol_complete_req(req, req->reduce_errno);
1701 0 : return;
1702 : }
1703 :
1704 4 : if (req->chunk_is_compressed) {
1705 4 : _reduce_vol_decompress_chunk_scratch(req, _write_decompress_done);
1706 4 : } else {
1707 0 : req->backing_cb_args.output_size = req->chunk->compressed_size;
1708 :
1709 0 : _write_decompress_done(req, 0);
1710 : }
1711 4 : }
1712 :
1713 : static void
1714 271 : _read_decompress_done(void *_req, int reduce_errno)
1715 : {
1716 271 : struct spdk_reduce_vol_request *req = _req;
1717 271 : struct spdk_reduce_vol *vol = req->vol;
1718 :
1719 : /* Negative reduce_errno indicates failure for compression operations. */
1720 271 : if (reduce_errno < 0) {
1721 0 : _reduce_vol_complete_req(req, reduce_errno);
1722 0 : return;
1723 : }
1724 :
1725 : /* Positive reduce_errno indicates that the output size field in the backing_cb_args
1726 : * represents the output_size.
1727 : */
1728 271 : if (req->backing_cb_args.output_size != vol->params.chunk_size) {
1729 0 : _reduce_vol_complete_req(req, -EIO);
1730 0 : return;
1731 : }
1732 :
1733 271 : if (req->copy_after_decompress) {
1734 4 : uint64_t chunk_offset = req->offset % vol->logical_blocks_per_chunk;
1735 4 : char *decomp_buffer = (char *)req->decomp_buf + chunk_offset * vol->params.logical_block_size;
1736 4 : int i;
1737 :
1738 11 : for (i = 0; i < req->iovcnt; i++) {
1739 7 : memcpy(req->iov[i].iov_base, decomp_buffer, req->iov[i].iov_len);
1740 7 : decomp_buffer += req->iov[i].iov_len;
1741 7 : assert(decomp_buffer <= (char *)req->decomp_buf + vol->params.chunk_size);
1742 7 : }
1743 4 : }
1744 :
1745 271 : _reduce_vol_complete_req(req, 0);
1746 271 : }
1747 :
1748 : static void
1749 263 : _read_read_done(void *_req, int reduce_errno)
1750 : {
1751 263 : struct spdk_reduce_vol_request *req = _req;
1752 263 : uint64_t chunk_offset;
1753 263 : uint8_t *buf;
1754 263 : int i;
1755 :
1756 263 : if (reduce_errno != 0) {
1757 0 : req->reduce_errno = reduce_errno;
1758 0 : }
1759 :
1760 263 : assert(req->num_backing_ops > 0);
1761 263 : if (--req->num_backing_ops > 0) {
1762 0 : return;
1763 : }
1764 :
1765 263 : if (req->reduce_errno != 0) {
1766 0 : _reduce_vol_complete_req(req, req->reduce_errno);
1767 0 : return;
1768 : }
1769 :
1770 263 : if (req->chunk_is_compressed) {
1771 263 : _reduce_vol_decompress_chunk(req, _read_decompress_done);
1772 263 : } else {
1773 :
1774 : /* If the chunk was compressed, the data would have been sent to the
1775 : * host buffers by the decompression operation, if not we need to memcpy here.
1776 : */
1777 0 : chunk_offset = req->offset % req->vol->logical_blocks_per_chunk;
1778 0 : buf = req->decomp_buf + chunk_offset * req->vol->params.logical_block_size;
1779 0 : for (i = 0; i < req->iovcnt; i++) {
1780 0 : memcpy(req->iov[i].iov_base, buf, req->iov[i].iov_len);
1781 0 : buf += req->iov[i].iov_len;
1782 0 : }
1783 :
1784 0 : req->backing_cb_args.output_size = req->chunk->compressed_size;
1785 :
1786 0 : _read_decompress_done(req, 0);
1787 : }
1788 263 : }
1789 :
1790 : static void
1791 267 : _reduce_vol_read_chunk(struct spdk_reduce_vol_request *req, reduce_request_fn next_fn)
1792 : {
1793 267 : struct spdk_reduce_vol *vol = req->vol;
1794 :
1795 267 : req->chunk_map_index = vol->pm_logical_map[req->logical_map_index];
1796 267 : assert(req->chunk_map_index != REDUCE_EMPTY_MAP_ENTRY);
1797 :
1798 267 : req->chunk = _reduce_vol_get_chunk_map(vol, req->chunk_map_index);
1799 534 : req->num_io_units = spdk_divide_round_up(req->chunk->compressed_size,
1800 267 : vol->params.backing_io_unit_size);
1801 267 : req->chunk_is_compressed = (req->num_io_units != vol->backing_io_units_per_chunk);
1802 :
1803 267 : _issue_backing_ops(req, vol, next_fn, false /* read */);
1804 267 : }
1805 :
1806 : static bool
1807 280 : _iov_array_is_valid(struct spdk_reduce_vol *vol, struct iovec *iov, int iovcnt,
1808 : uint64_t length)
1809 : {
1810 280 : uint64_t size = 0;
1811 280 : int i;
1812 :
1813 280 : if (iovcnt > REDUCE_MAX_IOVECS) {
1814 0 : return false;
1815 : }
1816 :
1817 560 : for (i = 0; i < iovcnt; i++) {
1818 280 : size += iov[i].iov_len;
1819 280 : }
1820 :
1821 280 : return size == (length * vol->params.logical_block_size);
1822 280 : }
1823 :
1824 : static bool
1825 281 : _check_overlap(struct spdk_reduce_vol *vol, uint64_t logical_map_index)
1826 : {
1827 281 : struct spdk_reduce_vol_request req;
1828 :
1829 281 : req.logical_map_index = logical_map_index;
1830 :
1831 281 : return (NULL != RB_FIND(executing_req_tree, &vol->executing_requests, &req));
1832 281 : }
1833 :
1834 : static void
1835 263 : _start_readv_request(struct spdk_reduce_vol_request *req)
1836 : {
1837 263 : RB_INSERT(executing_req_tree, &req->vol->executing_requests, req);
1838 263 : _reduce_vol_read_chunk(req, _read_read_done);
1839 263 : }
1840 :
1841 : void
1842 264 : spdk_reduce_vol_readv(struct spdk_reduce_vol *vol,
1843 : struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
1844 : spdk_reduce_vol_op_complete cb_fn, void *cb_arg)
1845 : {
1846 264 : struct spdk_reduce_vol_request *req;
1847 264 : uint64_t logical_map_index;
1848 264 : bool overlapped;
1849 264 : int i;
1850 :
1851 264 : if (length == 0) {
1852 0 : cb_fn(cb_arg, 0);
1853 0 : return;
1854 : }
1855 :
1856 264 : if (_request_spans_chunk_boundary(vol, offset, length)) {
1857 0 : cb_fn(cb_arg, -EINVAL);
1858 0 : return;
1859 : }
1860 :
1861 264 : if (!_iov_array_is_valid(vol, iov, iovcnt, length)) {
1862 0 : cb_fn(cb_arg, -EINVAL);
1863 0 : return;
1864 : }
1865 :
1866 264 : logical_map_index = offset / vol->logical_blocks_per_chunk;
1867 264 : overlapped = _check_overlap(vol, logical_map_index);
1868 :
1869 264 : if (!overlapped && vol->pm_logical_map[logical_map_index] == REDUCE_EMPTY_MAP_ENTRY) {
1870 : /*
1871 : * This chunk hasn't been allocated. So treat the data as all
1872 : * zeroes for this chunk - do the memset and immediately complete
1873 : * the operation.
1874 : */
1875 2 : for (i = 0; i < iovcnt; i++) {
1876 1 : memset(iov[i].iov_base, 0, iov[i].iov_len);
1877 1 : }
1878 1 : cb_fn(cb_arg, 0);
1879 1 : return;
1880 : }
1881 :
1882 263 : req = TAILQ_FIRST(&vol->free_requests);
1883 263 : if (req == NULL) {
1884 0 : cb_fn(cb_arg, -ENOMEM);
1885 0 : return;
1886 : }
1887 :
1888 263 : TAILQ_REMOVE(&vol->free_requests, req, tailq);
1889 263 : req->type = REDUCE_IO_READV;
1890 263 : req->vol = vol;
1891 263 : req->iov = iov;
1892 263 : req->iovcnt = iovcnt;
1893 263 : req->offset = offset;
1894 263 : req->logical_map_index = logical_map_index;
1895 263 : req->length = length;
1896 263 : req->copy_after_decompress = false;
1897 263 : req->cb_fn = cb_fn;
1898 263 : req->cb_arg = cb_arg;
1899 263 : req->reduce_errno = 0;
1900 :
1901 263 : if (!overlapped) {
1902 263 : _start_readv_request(req);
1903 263 : } else {
1904 0 : TAILQ_INSERT_TAIL(&vol->queued_requests, req, tailq);
1905 : }
1906 264 : }
1907 :
1908 : static void
1909 16 : _start_writev_request(struct spdk_reduce_vol_request *req)
1910 : {
1911 16 : struct spdk_reduce_vol *vol = req->vol;
1912 :
1913 16 : RB_INSERT(executing_req_tree, &req->vol->executing_requests, req);
1914 16 : if (vol->pm_logical_map[req->logical_map_index] != REDUCE_EMPTY_MAP_ENTRY) {
1915 6 : if ((req->length * vol->params.logical_block_size) < vol->params.chunk_size) {
1916 : /* Read old chunk, then overwrite with data from this write
1917 : * operation.
1918 : */
1919 4 : req->rmw = true;
1920 4 : _reduce_vol_read_chunk(req, _write_read_done);
1921 4 : return;
1922 : }
1923 2 : }
1924 :
1925 12 : req->rmw = false;
1926 :
1927 12 : _prepare_compress_chunk(req, true);
1928 12 : _reduce_vol_compress_chunk(req, _write_compress_done);
1929 16 : }
1930 :
1931 : void
1932 18 : spdk_reduce_vol_writev(struct spdk_reduce_vol *vol,
1933 : struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
1934 : spdk_reduce_vol_op_complete cb_fn, void *cb_arg)
1935 : {
1936 18 : struct spdk_reduce_vol_request *req;
1937 18 : uint64_t logical_map_index;
1938 18 : bool overlapped;
1939 :
1940 18 : if (length == 0) {
1941 0 : cb_fn(cb_arg, 0);
1942 0 : return;
1943 : }
1944 :
1945 18 : if (_request_spans_chunk_boundary(vol, offset, length)) {
1946 2 : cb_fn(cb_arg, -EINVAL);
1947 2 : return;
1948 : }
1949 :
1950 16 : if (!_iov_array_is_valid(vol, iov, iovcnt, length)) {
1951 0 : cb_fn(cb_arg, -EINVAL);
1952 0 : return;
1953 : }
1954 :
1955 16 : logical_map_index = offset / vol->logical_blocks_per_chunk;
1956 16 : overlapped = _check_overlap(vol, logical_map_index);
1957 :
1958 16 : req = TAILQ_FIRST(&vol->free_requests);
1959 16 : if (req == NULL) {
1960 0 : cb_fn(cb_arg, -ENOMEM);
1961 0 : return;
1962 : }
1963 :
1964 16 : TAILQ_REMOVE(&vol->free_requests, req, tailq);
1965 16 : req->type = REDUCE_IO_WRITEV;
1966 16 : req->vol = vol;
1967 16 : req->iov = iov;
1968 16 : req->iovcnt = iovcnt;
1969 16 : req->offset = offset;
1970 16 : req->logical_map_index = logical_map_index;
1971 16 : req->length = length;
1972 16 : req->copy_after_decompress = false;
1973 16 : req->cb_fn = cb_fn;
1974 16 : req->cb_arg = cb_arg;
1975 16 : req->reduce_errno = 0;
1976 :
1977 16 : if (!overlapped) {
1978 15 : _start_writev_request(req);
1979 15 : } else {
1980 1 : TAILQ_INSERT_TAIL(&vol->queued_requests, req, tailq);
1981 : }
1982 18 : }
1983 :
1984 : static void
1985 1 : _start_unmap_request_full_chunk(void *ctx)
1986 : {
1987 1 : struct spdk_reduce_vol_request *req = ctx;
1988 1 : struct spdk_reduce_vol *vol = req->vol;
1989 1 : uint64_t chunk_map_index;
1990 :
1991 1 : RB_INSERT(executing_req_tree, &req->vol->executing_requests, req);
1992 :
1993 1 : chunk_map_index = vol->pm_logical_map[req->logical_map_index];
1994 1 : if (chunk_map_index != REDUCE_EMPTY_MAP_ENTRY) {
1995 1 : _reduce_vol_reset_chunk(vol, chunk_map_index);
1996 1 : vol->pm_logical_map[req->logical_map_index] = REDUCE_EMPTY_MAP_ENTRY;
1997 1 : _reduce_persist(vol, &vol->pm_logical_map[req->logical_map_index], sizeof(uint64_t));
1998 1 : }
1999 1 : _reduce_vol_complete_req(req, 0);
2000 1 : }
2001 :
2002 : static void
2003 1 : _reduce_vol_unmap_full_chunk(struct spdk_reduce_vol *vol,
2004 : uint64_t offset, uint64_t length,
2005 : spdk_reduce_vol_op_complete cb_fn, void *cb_arg)
2006 : {
2007 1 : struct spdk_reduce_vol_request *req;
2008 1 : uint64_t logical_map_index;
2009 1 : bool overlapped;
2010 :
2011 1 : if (_request_spans_chunk_boundary(vol, offset, length)) {
2012 0 : cb_fn(cb_arg, -EINVAL);
2013 0 : return;
2014 : }
2015 :
2016 1 : logical_map_index = offset / vol->logical_blocks_per_chunk;
2017 1 : overlapped = _check_overlap(vol, logical_map_index);
2018 :
2019 1 : req = TAILQ_FIRST(&vol->free_requests);
2020 1 : if (req == NULL) {
2021 0 : cb_fn(cb_arg, -ENOMEM);
2022 0 : return;
2023 : }
2024 :
2025 1 : TAILQ_REMOVE(&vol->free_requests, req, tailq);
2026 1 : req->type = REDUCE_IO_UNMAP;
2027 1 : req->vol = vol;
2028 1 : req->iov = NULL;
2029 1 : req->iovcnt = 0;
2030 1 : req->offset = offset;
2031 1 : req->logical_map_index = logical_map_index;
2032 1 : req->length = length;
2033 1 : req->copy_after_decompress = false;
2034 1 : req->cb_fn = cb_fn;
2035 1 : req->cb_arg = cb_arg;
2036 1 : req->reduce_errno = 0;
2037 :
2038 1 : if (!overlapped) {
2039 1 : _start_unmap_request_full_chunk(req);
2040 1 : } else {
2041 0 : TAILQ_INSERT_TAIL(&vol->queued_requests, req, tailq);
2042 : }
2043 1 : }
2044 :
2045 : struct unmap_partial_chunk_ctx {
2046 : struct spdk_reduce_vol *vol;
2047 : struct iovec iov;
2048 : spdk_reduce_vol_op_complete cb_fn;
2049 : void *cb_arg;
2050 : };
2051 :
2052 : static void
2053 1 : _reduce_unmap_partial_chunk_complete(void *_ctx, int reduce_errno)
2054 : {
2055 1 : struct unmap_partial_chunk_ctx *ctx = _ctx;
2056 :
2057 1 : ctx->cb_fn(ctx->cb_arg, reduce_errno);
2058 1 : free(ctx);
2059 1 : }
2060 :
2061 : static void
2062 1 : _reduce_vol_unmap_partial_chunk(struct spdk_reduce_vol *vol, uint64_t offset, uint64_t length,
2063 : spdk_reduce_vol_op_complete cb_fn, void *cb_arg)
2064 : {
2065 1 : struct unmap_partial_chunk_ctx *ctx;
2066 :
2067 1 : ctx = calloc(1, sizeof(struct unmap_partial_chunk_ctx));
2068 1 : if (ctx == NULL) {
2069 0 : cb_fn(cb_arg, -ENOMEM);
2070 0 : return;
2071 : }
2072 :
2073 1 : ctx->vol = vol;
2074 1 : ctx->iov.iov_base = g_zero_buf;
2075 1 : ctx->iov.iov_len = length * vol->params.logical_block_size;
2076 1 : ctx->cb_fn = cb_fn;
2077 1 : ctx->cb_arg = cb_arg;
2078 :
2079 1 : spdk_reduce_vol_writev(vol, &ctx->iov, 1, offset, length, _reduce_unmap_partial_chunk_complete,
2080 1 : ctx);
2081 1 : }
2082 :
2083 : void
2084 3 : spdk_reduce_vol_unmap(struct spdk_reduce_vol *vol,
2085 : uint64_t offset, uint64_t length,
2086 : spdk_reduce_vol_op_complete cb_fn, void *cb_arg)
2087 : {
2088 3 : if (length < vol->logical_blocks_per_chunk) {
2089 1 : _reduce_vol_unmap_partial_chunk(vol, offset, length, cb_fn, cb_arg);
2090 3 : } else if (length == vol->logical_blocks_per_chunk) {
2091 1 : _reduce_vol_unmap_full_chunk(vol, offset, length, cb_fn, cb_arg);
2092 1 : } else {
2093 1 : cb_fn(cb_arg, -EINVAL);
2094 : }
2095 3 : }
2096 :
2097 : const struct spdk_reduce_vol_params *
2098 0 : spdk_reduce_vol_get_params(struct spdk_reduce_vol *vol)
2099 : {
2100 0 : return &vol->params;
2101 : }
2102 :
2103 : const char *
2104 0 : spdk_reduce_vol_get_pm_path(const struct spdk_reduce_vol *vol)
2105 : {
2106 0 : return vol->pm_file.path;
2107 : }
2108 :
2109 : void
2110 0 : spdk_reduce_vol_print_info(struct spdk_reduce_vol *vol)
2111 : {
2112 0 : uint64_t logical_map_size, num_chunks, ttl_chunk_sz;
2113 0 : uint32_t struct_size;
2114 0 : uint64_t chunk_map_size;
2115 :
2116 0 : SPDK_NOTICELOG("vol info:\n");
2117 0 : SPDK_NOTICELOG("\tvol->params.backing_io_unit_size = 0x%x\n", vol->params.backing_io_unit_size);
2118 0 : SPDK_NOTICELOG("\tvol->params.logical_block_size = 0x%x\n", vol->params.logical_block_size);
2119 0 : SPDK_NOTICELOG("\tvol->params.chunk_size = 0x%x\n", vol->params.chunk_size);
2120 0 : SPDK_NOTICELOG("\tvol->params.vol_size = 0x%" PRIx64 "\n", vol->params.vol_size);
2121 0 : num_chunks = _get_total_chunks(vol->params.vol_size, vol->params.chunk_size);
2122 0 : SPDK_NOTICELOG("\ttotal chunks (including extra) = 0x%" PRIx64 "\n", num_chunks);
2123 0 : SPDK_NOTICELOG("\ttotal chunks (excluding extra) = 0x%" PRIx64 "\n",
2124 : vol->params.vol_size / vol->params.chunk_size);
2125 0 : ttl_chunk_sz = _get_pm_total_chunks_size(vol->params.vol_size, vol->params.chunk_size,
2126 0 : vol->params.backing_io_unit_size);
2127 0 : SPDK_NOTICELOG("\ttotal_chunks_size = 0x%" PRIx64 "\n", ttl_chunk_sz);
2128 0 : struct_size = _reduce_vol_get_chunk_struct_size(vol->backing_io_units_per_chunk);
2129 0 : SPDK_NOTICELOG("\tchunk_struct_size = 0x%x\n", struct_size);
2130 :
2131 0 : SPDK_NOTICELOG("pmem info:\n");
2132 0 : SPDK_NOTICELOG("\tvol->pm_file.size = 0x%" PRIx64 "\n", vol->pm_file.size);
2133 0 : SPDK_NOTICELOG("\tvol->pm_file.pm_buf = %p\n", (void *)vol->pm_file.pm_buf);
2134 0 : SPDK_NOTICELOG("\tvol->pm_super = %p\n", (void *)vol->pm_super);
2135 0 : SPDK_NOTICELOG("\tvol->pm_logical_map = %p\n", (void *)vol->pm_logical_map);
2136 0 : logical_map_size = _get_pm_logical_map_size(vol->params.vol_size,
2137 0 : vol->params.chunk_size);
2138 0 : SPDK_NOTICELOG("\tlogical_map_size = 0x%" PRIx64 "\n", logical_map_size);
2139 0 : SPDK_NOTICELOG("\tvol->pm_chunk_maps = %p\n", (void *)vol->pm_chunk_maps);
2140 0 : chunk_map_size = _get_pm_total_chunks_size(vol->params.vol_size, vol->params.chunk_size,
2141 0 : vol->params.backing_io_unit_size);
2142 0 : SPDK_NOTICELOG("\tchunk_map_size = 0x%" PRIx64 "\n", chunk_map_size);
2143 0 : }
2144 :
2145 1 : SPDK_LOG_REGISTER_COMPONENT(reduce)
|
