xref: /spdk/module/bdev/raid/concat.c (revision 488570ebd418ba07c9e69e65106dcc964f3bb41b)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (c) Peng Yu yupeng0921@gmail.com.
3  *   All rights reserved.
4  */
5 
6 #include "bdev_raid.h"
7 
8 #include "spdk/env.h"
9 #include "spdk/thread.h"
10 #include "spdk/string.h"
11 #include "spdk/util.h"
12 
13 #include "spdk/log.h"
14 
15 struct concat_block_range {
16 	uint64_t start;
17 	uint64_t length;
18 };
19 
20 /*
21  * brief:
22  * concat_bdev_io_completion function is called by lower layers to notify raid
23  * module that particular bdev_io is completed.
24  * params:
25  * bdev_io - pointer to bdev io submitted to lower layers, like child io
26  * success - bdev_io status
27  * cb_arg - function callback context (parent raid_bdev_io)
28  * returns:
29  * none
30  */
31 static void
32 concat_bdev_io_completion(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
33 {
34 	struct raid_bdev_io *raid_io = cb_arg;
35 
36 	spdk_bdev_free_io(bdev_io);
37 
38 	if (success) {
39 		raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_SUCCESS);
40 	} else {
41 		raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_FAILED);
42 	}
43 }
44 
45 static void
46 concat_submit_rw_request(struct raid_bdev_io *raid_io);
47 
48 static void
49 _concat_submit_rw_request(void *_raid_io)
50 {
51 	struct raid_bdev_io *raid_io = _raid_io;
52 
53 	concat_submit_rw_request(raid_io);
54 }
55 
56 /*
57  * brief:
58  * concat_submit_rw_request function is used to submit I/O to the correct
59  * member disk for concat bdevs.
60  * params:
61  * raid_io
62  * returns:
63  * none
64  */
65 static void
66 concat_submit_rw_request(struct raid_bdev_io *raid_io)
67 {
68 	struct spdk_bdev_io		*bdev_io = spdk_bdev_io_from_ctx(raid_io);
69 	struct raid_bdev_io_channel	*raid_ch = raid_io->raid_ch;
70 	struct raid_bdev		*raid_bdev = raid_io->raid_bdev;
71 	struct concat_block_range	*block_range = raid_bdev->module_private;
72 	uint64_t			pd_lba;
73 	uint64_t			pd_blocks;
74 	int				pd_idx;
75 	int				ret = 0;
76 	struct raid_base_bdev_info	*base_info;
77 	struct spdk_io_channel		*base_ch;
78 	int i;
79 
80 	pd_idx = -1;
81 	for (i = 0; i < raid_bdev->num_base_bdevs; i++) {
82 		if (block_range[i].start > bdev_io->u.bdev.offset_blocks) {
83 			break;
84 		}
85 		pd_idx = i;
86 	}
87 	assert(pd_idx >= 0);
88 	assert(bdev_io->u.bdev.offset_blocks >= block_range[pd_idx].start);
89 	pd_lba = bdev_io->u.bdev.offset_blocks - block_range[pd_idx].start;
90 	pd_blocks = bdev_io->u.bdev.num_blocks;
91 	base_info = &raid_bdev->base_bdev_info[pd_idx];
92 	if (base_info->desc == NULL) {
93 		SPDK_ERRLOG("base bdev desc null for pd_idx %u\n", pd_idx);
94 		assert(0);
95 	}
96 
97 	/*
98 	 * Submit child io to bdev layer with using base bdev descriptors, base
99 	 * bdev lba, base bdev child io length in blocks, buffer, completion
100 	 * function and function callback context
101 	 */
102 	assert(raid_ch != NULL);
103 	assert(raid_ch->base_channel);
104 	base_ch = raid_ch->base_channel[pd_idx];
105 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
106 		ret = spdk_bdev_readv_blocks(base_info->desc, base_ch,
107 					     bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
108 					     pd_lba, pd_blocks, concat_bdev_io_completion,
109 					     raid_io);
110 	} else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
111 		ret = spdk_bdev_writev_blocks(base_info->desc, base_ch,
112 					      bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
113 					      pd_lba, pd_blocks, concat_bdev_io_completion,
114 					      raid_io);
115 	} else {
116 		SPDK_ERRLOG("Recvd not supported io type %u\n", bdev_io->type);
117 		assert(0);
118 	}
119 
120 	if (ret == -ENOMEM) {
121 		raid_bdev_queue_io_wait(raid_io, base_info->bdev, base_ch,
122 					_concat_submit_rw_request);
123 	} else if (ret != 0) {
124 		SPDK_ERRLOG("bdev io submit error not due to ENOMEM, it should not happen\n");
125 		assert(false);
126 		raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_FAILED);
127 	}
128 }
129 
130 static void
131 concat_submit_null_payload_request(struct raid_bdev_io *raid_io);
132 
133 static void
134 _concat_submit_null_payload_request(void *_raid_io)
135 {
136 	struct raid_bdev_io *raid_io = _raid_io;
137 
138 	concat_submit_null_payload_request(raid_io);
139 }
140 
141 static void
142 concat_base_io_complete(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
143 {
144 	struct raid_bdev_io *raid_io = cb_arg;
145 
146 	raid_bdev_io_complete_part(raid_io, 1, success ?
147 				   SPDK_BDEV_IO_STATUS_SUCCESS :
148 				   SPDK_BDEV_IO_STATUS_FAILED);
149 
150 	spdk_bdev_free_io(bdev_io);
151 }
152 
153 /*
154  * brief:
155  * concat_submit_null_payload_request function submits the next batch of
156  * io requests with range but without payload, like FLUSH and UNMAP, to member disks;
157  * it will submit as many as possible unless one base io request fails with -ENOMEM,
158  * in which case it will queue itself for later submission.
159  * params:
160  * bdev_io - pointer to parent bdev_io on raid bdev device
161  * returns:
162  * none
163  */
164 static void
165 concat_submit_null_payload_request(struct raid_bdev_io *raid_io)
166 {
167 	struct spdk_bdev_io		*bdev_io;
168 	struct raid_bdev		*raid_bdev;
169 	int				ret;
170 	struct raid_base_bdev_info	*base_info;
171 	struct spdk_io_channel		*base_ch;
172 	uint64_t			pd_lba;
173 	uint64_t			pd_blocks;
174 	uint64_t			offset_blocks;
175 	uint64_t			num_blocks;
176 	struct concat_block_range	*block_range;
177 	int				i, start_idx, stop_idx;
178 
179 	bdev_io = spdk_bdev_io_from_ctx(raid_io);
180 	raid_bdev = raid_io->raid_bdev;
181 	block_range = raid_bdev->module_private;
182 
183 	offset_blocks = bdev_io->u.bdev.offset_blocks;
184 	num_blocks = bdev_io->u.bdev.num_blocks;
185 	start_idx = -1;
186 	stop_idx = -1;
187 	/*
188 	 * Go through all base bdevs, find the first bdev and the last bdev
189 	 */
190 	for (i = 0; i < raid_bdev->num_base_bdevs; i++) {
191 		/* skip the bdevs before the offset_blocks */
192 		if (offset_blocks >= block_range[i].start + block_range[i].length) {
193 			continue;
194 		}
195 		if (start_idx == -1) {
196 			start_idx = i;
197 		} else {
198 			/*
199 			 * The offset_blocks might be at the middle of the first bdev.
200 			 * Besides the first bdev, the offset_blocks should be always
201 			 * at the start of the bdev.
202 			 */
203 			assert(offset_blocks == block_range[i].start);
204 		}
205 		pd_lba = offset_blocks - block_range[i].start;
206 		pd_blocks = spdk_min(num_blocks, block_range[i].length - pd_lba);
207 		offset_blocks += pd_blocks;
208 		num_blocks -= pd_blocks;
209 		if (num_blocks == 0) {
210 			stop_idx = i;
211 			break;
212 		}
213 	}
214 	assert(start_idx >= 0);
215 	assert(stop_idx >= 0);
216 
217 	if (raid_io->base_bdev_io_remaining == 0) {
218 		raid_io->base_bdev_io_remaining = stop_idx - start_idx + 1;
219 	}
220 	offset_blocks = bdev_io->u.bdev.offset_blocks;
221 	num_blocks = bdev_io->u.bdev.num_blocks;
222 	for (i = start_idx; i <= stop_idx; i++) {
223 		assert(offset_blocks >= block_range[i].start);
224 		assert(offset_blocks < block_range[i].start + block_range[i].length);
225 		pd_lba = offset_blocks -  block_range[i].start;
226 		pd_blocks = spdk_min(num_blocks, block_range[i].length - pd_lba);
227 		offset_blocks += pd_blocks;
228 		num_blocks -= pd_blocks;
229 		/*
230 		 * Skip the IOs we have submitted
231 		 */
232 		if (i < start_idx + raid_io->base_bdev_io_submitted) {
233 			continue;
234 		}
235 		base_info = &raid_bdev->base_bdev_info[i];
236 		base_ch = raid_io->raid_ch->base_channel[i];
237 		switch (bdev_io->type) {
238 		case SPDK_BDEV_IO_TYPE_UNMAP:
239 			ret = spdk_bdev_unmap_blocks(base_info->desc, base_ch,
240 						     pd_lba, pd_blocks,
241 						     concat_base_io_complete, raid_io);
242 			break;
243 		case SPDK_BDEV_IO_TYPE_FLUSH:
244 			ret = spdk_bdev_flush_blocks(base_info->desc, base_ch,
245 						     pd_lba, pd_blocks,
246 						     concat_base_io_complete, raid_io);
247 			break;
248 		default:
249 			SPDK_ERRLOG("submit request, invalid io type with null payload %u\n", bdev_io->type);
250 			assert(false);
251 			ret = -EIO;
252 		}
253 		if (ret == 0) {
254 			raid_io->base_bdev_io_submitted++;
255 		} else if (ret == -ENOMEM) {
256 			raid_bdev_queue_io_wait(raid_io, base_info->bdev, base_ch,
257 						_concat_submit_null_payload_request);
258 			return;
259 		} else {
260 			SPDK_ERRLOG("bdev io submit error not due to ENOMEM, it should not happen\n");
261 			assert(false);
262 			raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_FAILED);
263 			return;
264 		}
265 	}
266 }
267 
268 static int concat_start(struct raid_bdev *raid_bdev)
269 {
270 	uint64_t total_blockcnt = 0;
271 	struct raid_base_bdev_info *base_info;
272 	struct concat_block_range *block_range;
273 
274 	block_range = calloc(raid_bdev->num_base_bdevs, sizeof(struct concat_block_range));
275 	if (!block_range) {
276 		SPDK_ERRLOG("Can not allocate block_range, num_base_bdevs: %u",
277 			    raid_bdev->num_base_bdevs);
278 		return -ENOMEM;
279 	}
280 
281 	int idx = 0;
282 	RAID_FOR_EACH_BASE_BDEV(raid_bdev, base_info) {
283 		uint64_t strip_cnt = base_info->bdev->blockcnt >> raid_bdev->strip_size_shift;
284 		uint64_t pd_block_cnt = strip_cnt << raid_bdev->strip_size_shift;
285 
286 		block_range[idx].start = total_blockcnt;
287 		block_range[idx].length = pd_block_cnt;
288 		total_blockcnt += pd_block_cnt;
289 		idx++;
290 	}
291 
292 	raid_bdev->module_private = block_range;
293 
294 	SPDK_DEBUGLOG(bdev_concat, "total blockcount %" PRIu64 ",  numbasedev %u, strip size shift %u\n",
295 		      total_blockcnt, raid_bdev->num_base_bdevs, raid_bdev->strip_size_shift);
296 	raid_bdev->bdev.blockcnt = total_blockcnt;
297 
298 	raid_bdev->bdev.optimal_io_boundary = raid_bdev->strip_size;
299 	raid_bdev->bdev.split_on_optimal_io_boundary = true;
300 
301 	return 0;
302 }
303 
304 static void
305 concat_stop(struct raid_bdev *raid_bdev)
306 {
307 	struct concat_block_range *block_range = raid_bdev->module_private;
308 
309 	free(block_range);
310 }
311 
312 static struct raid_bdev_module g_concat_module = {
313 	.level = CONCAT,
314 	.base_bdevs_min = 1,
315 	.start = concat_start,
316 	.stop = concat_stop,
317 	.submit_rw_request = concat_submit_rw_request,
318 	.submit_null_payload_request = concat_submit_null_payload_request,
319 };
320 RAID_MODULE_REGISTER(&g_concat_module)
321 
322 SPDK_LOG_REGISTER_COMPONENT(bdev_concat)
323