]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/lib/bdev/part.c
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / spdk / lib / bdev / part.c
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 /*
35 * Common code for partition-like virtual bdevs.
36 */
37
38 #include "spdk/bdev.h"
39 #include "spdk/log.h"
40 #include "spdk/string.h"
41
42 #include "spdk/bdev_module.h"
43
44 struct spdk_bdev_part_base {
45 struct spdk_bdev *bdev;
46 struct spdk_bdev_desc *desc;
47 uint32_t ref;
48 uint32_t channel_size;
49 spdk_bdev_part_base_free_fn base_free_fn;
50 void *ctx;
51 bool claimed;
52 struct spdk_bdev_module *module;
53 struct spdk_bdev_fn_table *fn_table;
54 struct bdev_part_tailq *tailq;
55 spdk_io_channel_create_cb ch_create_cb;
56 spdk_io_channel_destroy_cb ch_destroy_cb;
57 };
58
59 struct spdk_bdev *
60 spdk_bdev_part_base_get_bdev(struct spdk_bdev_part_base *part_base)
61 {
62 return part_base->bdev;
63 }
64
65 struct spdk_bdev_desc *
66 spdk_bdev_part_base_get_desc(struct spdk_bdev_part_base *part_base)
67 {
68 return part_base->desc;
69 }
70
71 struct bdev_part_tailq *
72 spdk_bdev_part_base_get_tailq(struct spdk_bdev_part_base *part_base)
73 {
74 return part_base->tailq;
75 }
76
77 void *
78 spdk_bdev_part_base_get_ctx(struct spdk_bdev_part_base *part_base)
79 {
80 return part_base->ctx;
81 }
82
83 void
84 spdk_bdev_part_base_free(struct spdk_bdev_part_base *base)
85 {
86 if (base->desc) {
87 spdk_bdev_close(base->desc);
88 base->desc = NULL;
89 }
90
91 if (base->base_free_fn != NULL) {
92 base->base_free_fn(base->ctx);
93 }
94
95 free(base);
96 }
97
98 static void
99 spdk_bdev_part_free_cb(void *io_device)
100 {
101 struct spdk_bdev_part *part = io_device;
102 struct spdk_bdev_part_base *base;
103
104 assert(part);
105 assert(part->internal.base);
106
107 base = part->internal.base;
108
109 TAILQ_REMOVE(base->tailq, part, tailq);
110
111 if (__sync_sub_and_fetch(&base->ref, 1) == 0) {
112 spdk_bdev_module_release_bdev(base->bdev);
113 spdk_bdev_part_base_free(base);
114 }
115
116 spdk_bdev_destruct_done(&part->internal.bdev, 0);
117 free(part->internal.bdev.name);
118 free(part->internal.bdev.product_name);
119 free(part);
120 }
121
122 int
123 spdk_bdev_part_free(struct spdk_bdev_part *part)
124 {
125 spdk_io_device_unregister(part, spdk_bdev_part_free_cb);
126
127 /* Return 1 to indicate that this is an asynchronous operation that isn't complete
128 * until spdk_bdev_destruct_done is called */
129 return 1;
130 }
131
132 void
133 spdk_bdev_part_base_hotremove(struct spdk_bdev *base_bdev, struct bdev_part_tailq *tailq)
134 {
135 struct spdk_bdev_part *part, *tmp;
136
137 TAILQ_FOREACH_SAFE(part, tailq, tailq, tmp) {
138 if (part->internal.base->bdev == base_bdev) {
139 spdk_bdev_unregister(&part->internal.bdev, NULL, NULL);
140 }
141 }
142 }
143
144 static bool
145 spdk_bdev_part_io_type_supported(void *_part, enum spdk_bdev_io_type io_type)
146 {
147 struct spdk_bdev_part *part = _part;
148
149 return part->internal.base->bdev->fn_table->io_type_supported(part->internal.base->bdev->ctxt,
150 io_type);
151 }
152
153 static struct spdk_io_channel *
154 spdk_bdev_part_get_io_channel(void *_part)
155 {
156 struct spdk_bdev_part *part = _part;
157
158 return spdk_get_io_channel(part);
159 }
160
161 struct spdk_bdev *
162 spdk_bdev_part_get_bdev(struct spdk_bdev_part *part)
163 {
164 return &part->internal.bdev;
165 }
166
167 struct spdk_bdev_part_base *
168 spdk_bdev_part_get_base(struct spdk_bdev_part *part)
169 {
170 return part->internal.base;
171 }
172
173 struct spdk_bdev *
174 spdk_bdev_part_get_base_bdev(struct spdk_bdev_part *part)
175 {
176 return part->internal.base->bdev;
177 }
178
179 uint64_t
180 spdk_bdev_part_get_offset_blocks(struct spdk_bdev_part *part)
181 {
182 return part->internal.offset_blocks;
183 }
184
185 static void
186 spdk_bdev_part_complete_io(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
187 {
188 struct spdk_bdev_io *part_io = cb_arg;
189 int status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
190
191 spdk_bdev_io_complete(part_io, status);
192 spdk_bdev_free_io(bdev_io);
193 }
194
195 int
196 spdk_bdev_part_submit_request(struct spdk_bdev_part_channel *ch, struct spdk_bdev_io *bdev_io)
197 {
198 struct spdk_bdev_part *part = ch->part;
199 struct spdk_io_channel *base_ch = ch->base_ch;
200 struct spdk_bdev_desc *base_desc = part->internal.base->desc;
201 uint64_t offset;
202 int rc = 0;
203
204 /* Modify the I/O to adjust for the offset within the base bdev. */
205 switch (bdev_io->type) {
206 case SPDK_BDEV_IO_TYPE_READ:
207 offset = bdev_io->u.bdev.offset_blocks + part->internal.offset_blocks;
208 rc = spdk_bdev_readv_blocks(base_desc, base_ch, bdev_io->u.bdev.iovs,
209 bdev_io->u.bdev.iovcnt, offset,
210 bdev_io->u.bdev.num_blocks, spdk_bdev_part_complete_io,
211 bdev_io);
212 break;
213 case SPDK_BDEV_IO_TYPE_WRITE:
214 offset = bdev_io->u.bdev.offset_blocks + part->internal.offset_blocks;
215 rc = spdk_bdev_writev_blocks(base_desc, base_ch, bdev_io->u.bdev.iovs,
216 bdev_io->u.bdev.iovcnt, offset,
217 bdev_io->u.bdev.num_blocks, spdk_bdev_part_complete_io,
218 bdev_io);
219 break;
220 case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
221 offset = bdev_io->u.bdev.offset_blocks + part->internal.offset_blocks;
222 rc = spdk_bdev_write_zeroes_blocks(base_desc, base_ch, offset, bdev_io->u.bdev.num_blocks,
223 spdk_bdev_part_complete_io, bdev_io);
224 break;
225 case SPDK_BDEV_IO_TYPE_UNMAP:
226 offset = bdev_io->u.bdev.offset_blocks + part->internal.offset_blocks;
227 rc = spdk_bdev_unmap_blocks(base_desc, base_ch, offset, bdev_io->u.bdev.num_blocks,
228 spdk_bdev_part_complete_io, bdev_io);
229 break;
230 case SPDK_BDEV_IO_TYPE_FLUSH:
231 offset = bdev_io->u.bdev.offset_blocks + part->internal.offset_blocks;
232 rc = spdk_bdev_flush_blocks(base_desc, base_ch, offset, bdev_io->u.bdev.num_blocks,
233 spdk_bdev_part_complete_io, bdev_io);
234 break;
235 case SPDK_BDEV_IO_TYPE_RESET:
236 rc = spdk_bdev_reset(base_desc, base_ch,
237 spdk_bdev_part_complete_io, bdev_io);
238 break;
239 default:
240 SPDK_ERRLOG("split: unknown I/O type %d\n", bdev_io->type);
241 return SPDK_BDEV_IO_STATUS_FAILED;
242 }
243
244 return rc;
245 }
246
247 static int
248 spdk_bdev_part_channel_create_cb(void *io_device, void *ctx_buf)
249 {
250 struct spdk_bdev_part *part = (struct spdk_bdev_part *)io_device;
251 struct spdk_bdev_part_channel *ch = ctx_buf;
252
253 ch->part = part;
254 ch->base_ch = spdk_bdev_get_io_channel(part->internal.base->desc);
255 if (ch->base_ch == NULL) {
256 return -1;
257 }
258
259 if (part->internal.base->ch_create_cb) {
260 return part->internal.base->ch_create_cb(io_device, ctx_buf);
261 } else {
262 return 0;
263 }
264 }
265
266 static void
267 spdk_bdev_part_channel_destroy_cb(void *io_device, void *ctx_buf)
268 {
269 struct spdk_bdev_part *part = (struct spdk_bdev_part *)io_device;
270 struct spdk_bdev_part_channel *ch = ctx_buf;
271
272 if (part->internal.base->ch_destroy_cb) {
273 part->internal.base->ch_destroy_cb(io_device, ctx_buf);
274 }
275 spdk_put_io_channel(ch->base_ch);
276 }
277
278 struct spdk_bdev_part_base *
279 spdk_bdev_part_base_construct(struct spdk_bdev *bdev,
280 spdk_bdev_remove_cb_t remove_cb, struct spdk_bdev_module *module,
281 struct spdk_bdev_fn_table *fn_table, struct bdev_part_tailq *tailq,
282 spdk_bdev_part_base_free_fn free_fn, void *ctx,
283 uint32_t channel_size, spdk_io_channel_create_cb ch_create_cb,
284 spdk_io_channel_destroy_cb ch_destroy_cb)
285 {
286 int rc;
287 struct spdk_bdev_part_base *base;
288
289 base = calloc(1, sizeof(*base));
290 if (!base) {
291 SPDK_ERRLOG("Memory allocation failure\n");
292 return NULL;
293 }
294 fn_table->get_io_channel = spdk_bdev_part_get_io_channel;
295 fn_table->io_type_supported = spdk_bdev_part_io_type_supported;
296
297 base->bdev = bdev;
298 base->desc = NULL;
299 base->ref = 0;
300 base->module = module;
301 base->fn_table = fn_table;
302 base->tailq = tailq;
303 base->base_free_fn = free_fn;
304 base->ctx = ctx;
305 base->claimed = false;
306 base->channel_size = channel_size;
307 base->ch_create_cb = ch_create_cb;
308 base->ch_destroy_cb = ch_destroy_cb;
309
310 rc = spdk_bdev_open(bdev, false, remove_cb, bdev, &base->desc);
311 if (rc) {
312 spdk_bdev_part_base_free(base);
313 SPDK_ERRLOG("could not open bdev %s\n", spdk_bdev_get_name(bdev));
314 return NULL;
315 }
316
317 return base;
318 }
319
320 int
321 spdk_bdev_part_construct(struct spdk_bdev_part *part, struct spdk_bdev_part_base *base,
322 char *name, uint64_t offset_blocks, uint64_t num_blocks,
323 char *product_name)
324 {
325 part->internal.bdev.blocklen = base->bdev->blocklen;
326 part->internal.bdev.blockcnt = num_blocks;
327 part->internal.offset_blocks = offset_blocks;
328
329 part->internal.bdev.write_cache = base->bdev->write_cache;
330 part->internal.bdev.need_aligned_buffer = base->bdev->need_aligned_buffer;
331 part->internal.bdev.ctxt = part;
332 part->internal.bdev.module = base->module;
333 part->internal.bdev.fn_table = base->fn_table;
334
335 part->internal.bdev.name = strdup(name);
336 part->internal.bdev.product_name = strdup(product_name);
337
338 if (part->internal.bdev.name == NULL) {
339 SPDK_ERRLOG("Failed to allocate name for new part of bdev %s\n", spdk_bdev_get_name(base->bdev));
340 return -1;
341 } else if (part->internal.bdev.product_name == NULL) {
342 free(part->internal.bdev.name);
343 SPDK_ERRLOG("Failed to allocate product name for new part of bdev %s\n",
344 spdk_bdev_get_name(base->bdev));
345 return -1;
346 }
347
348 __sync_fetch_and_add(&base->ref, 1);
349 part->internal.base = base;
350
351 if (!base->claimed) {
352 int rc;
353
354 rc = spdk_bdev_module_claim_bdev(base->bdev, base->desc, base->module);
355 if (rc) {
356 SPDK_ERRLOG("could not claim bdev %s\n", spdk_bdev_get_name(base->bdev));
357 free(part->internal.bdev.name);
358 free(part->internal.bdev.product_name);
359 return -1;
360 }
361 base->claimed = true;
362 }
363
364 spdk_io_device_register(part, spdk_bdev_part_channel_create_cb,
365 spdk_bdev_part_channel_destroy_cb,
366 base->channel_size,
367 name);
368
369 spdk_vbdev_register(&part->internal.bdev, &base->bdev, 1);
370 TAILQ_INSERT_TAIL(base->tailq, part, tailq);
371
372 return 0;
373 }