1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #ifndef SPDK_BDEV_RAID_INTERNAL_H 35 #define SPDK_BDEV_RAID_INTERNAL_H 36 37 #include "spdk/bdev_module.h" 38 39 enum raid_level { 40 INVALID_RAID_LEVEL = -1, 41 RAID0 = 0, 42 RAID5 = 5, 43 }; 44 45 /* 46 * Raid state describes the state of the raid. This raid bdev can be either in 47 * configured list or configuring list 48 */ 49 enum raid_bdev_state { 50 /* raid bdev is ready and is seen by upper layers */ 51 RAID_BDEV_STATE_ONLINE, 52 53 /* 54 * raid bdev is configuring, not all underlying bdevs are present. 55 * And can't be seen by upper layers. 56 */ 57 RAID_BDEV_STATE_CONFIGURING, 58 59 /* 60 * In offline state, raid bdev layer will complete all incoming commands without 61 * submitting to underlying base nvme bdevs 62 */ 63 RAID_BDEV_STATE_OFFLINE, 64 65 /* raid bdev max, new states should be added before this */ 66 RAID_BDEV_MAX 67 }; 68 69 /* 70 * raid_base_bdev_info contains information for the base bdevs which are part of some 71 * raid. This structure contains the per base bdev information. Whatever is 72 * required per base device for raid bdev will be kept here 73 */ 74 struct raid_base_bdev_info { 75 /* pointer to base spdk bdev */ 76 struct spdk_bdev *bdev; 77 78 /* pointer to base bdev descriptor opened by raid bdev */ 79 struct spdk_bdev_desc *desc; 80 81 /* 82 * When underlying base device calls the hot plug function on drive removal, 83 * this flag will be set and later after doing some processing, base device 84 * descriptor will be closed 85 */ 86 bool remove_scheduled; 87 88 /* thread where base device is opened */ 89 struct spdk_thread *thread; 90 }; 91 92 /* 93 * raid_bdev_io is the context part of bdev_io. It contains the information 94 * related to bdev_io for a raid bdev 95 */ 96 struct raid_bdev_io { 97 /* The raid bdev associated with this IO */ 98 struct raid_bdev *raid_bdev; 99 100 /* WaitQ entry, used only in waitq logic */ 101 struct spdk_bdev_io_wait_entry waitq_entry; 102 103 /* Context of the original channel for this IO */ 104 struct raid_bdev_io_channel *raid_ch; 105 106 /* Used for tracking progress on io requests sent to member disks. */ 107 uint64_t base_bdev_io_remaining; 108 uint8_t base_bdev_io_submitted; 109 uint8_t base_bdev_io_status; 110 }; 111 112 /* 113 * raid_bdev is the single entity structure which contains SPDK block device 114 * and the information related to any raid bdev either configured or 115 * in configuring list. io device is created on this. 116 */ 117 struct raid_bdev { 118 /* raid bdev device, this will get registered in bdev layer */ 119 struct spdk_bdev bdev; 120 121 /* link of raid bdev to link it to configured, configuring or offline list */ 122 TAILQ_ENTRY(raid_bdev) state_link; 123 124 /* link of raid bdev to link it to global raid bdev list */ 125 TAILQ_ENTRY(raid_bdev) global_link; 126 127 /* pointer to config file entry */ 128 struct raid_bdev_config *config; 129 130 /* array of base bdev info */ 131 struct raid_base_bdev_info *base_bdev_info; 132 133 /* strip size of raid bdev in blocks */ 134 uint32_t strip_size; 135 136 /* strip size of raid bdev in KB */ 137 uint32_t strip_size_kb; 138 139 /* strip size bit shift for optimized calculation */ 140 uint32_t strip_size_shift; 141 142 /* block length bit shift for optimized calculation */ 143 uint32_t blocklen_shift; 144 145 /* state of raid bdev */ 146 enum raid_bdev_state state; 147 148 /* number of base bdevs comprising raid bdev */ 149 uint8_t num_base_bdevs; 150 151 /* number of base bdevs discovered */ 152 uint8_t num_base_bdevs_discovered; 153 154 /* Raid Level of this raid bdev */ 155 enum raid_level level; 156 157 /* Set to true if destruct is called for this raid bdev */ 158 bool destruct_called; 159 160 /* Set to true if destroy of this raid bdev is started. */ 161 bool destroy_started; 162 163 /* Module for RAID-level specific operations */ 164 struct raid_bdev_module *module; 165 166 /* Private data for the raid module */ 167 void *module_private; 168 }; 169 170 #define RAID_FOR_EACH_BASE_BDEV(r, i) \ 171 for (i = r->base_bdev_info; i < r->base_bdev_info + r->num_base_bdevs; i++) 172 173 /* 174 * raid_base_bdev_config is the per base bdev data structure which contains 175 * information w.r.t to per base bdev during parsing config 176 */ 177 struct raid_base_bdev_config { 178 /* base bdev name from config file */ 179 char *name; 180 }; 181 182 /* 183 * raid_bdev_config contains the raid bdev config related information after 184 * parsing the config file 185 */ 186 struct raid_bdev_config { 187 /* base bdev config per underlying bdev */ 188 struct raid_base_bdev_config *base_bdev; 189 190 /* Points to already created raid bdev */ 191 struct raid_bdev *raid_bdev; 192 193 char *name; 194 195 /* strip size of this raid bdev in KB */ 196 uint32_t strip_size; 197 198 /* number of base bdevs */ 199 uint8_t num_base_bdevs; 200 201 /* raid level */ 202 enum raid_level level; 203 204 TAILQ_ENTRY(raid_bdev_config) link; 205 }; 206 207 /* 208 * raid_config is the top level structure representing the raid bdev config as read 209 * from config file for all raids 210 */ 211 struct raid_config { 212 /* raid bdev context from config file */ 213 TAILQ_HEAD(, raid_bdev_config) raid_bdev_config_head; 214 215 /* total raid bdev from config file */ 216 uint8_t total_raid_bdev; 217 }; 218 219 /* 220 * raid_bdev_io_channel is the context of spdk_io_channel for raid bdev device. It 221 * contains the relationship of raid bdev io channel with base bdev io channels. 222 */ 223 struct raid_bdev_io_channel { 224 /* Array of IO channels of base bdevs */ 225 struct spdk_io_channel **base_channel; 226 227 /* Number of IO channels */ 228 uint8_t num_channels; 229 }; 230 231 /* TAIL heads for various raid bdev lists */ 232 TAILQ_HEAD(raid_configured_tailq, raid_bdev); 233 TAILQ_HEAD(raid_configuring_tailq, raid_bdev); 234 TAILQ_HEAD(raid_all_tailq, raid_bdev); 235 TAILQ_HEAD(raid_offline_tailq, raid_bdev); 236 237 extern struct raid_configured_tailq g_raid_bdev_configured_list; 238 extern struct raid_configuring_tailq g_raid_bdev_configuring_list; 239 extern struct raid_all_tailq g_raid_bdev_list; 240 extern struct raid_offline_tailq g_raid_bdev_offline_list; 241 extern struct raid_config g_raid_config; 242 243 typedef void (*raid_bdev_destruct_cb)(void *cb_ctx, int rc); 244 245 int raid_bdev_create(struct raid_bdev_config *raid_cfg); 246 int raid_bdev_add_base_devices(struct raid_bdev_config *raid_cfg); 247 void raid_bdev_remove_base_devices(struct raid_bdev_config *raid_cfg, 248 raid_bdev_destruct_cb cb_fn, void *cb_ctx); 249 int raid_bdev_config_add(const char *raid_name, uint32_t strip_size, uint8_t num_base_bdevs, 250 enum raid_level level, struct raid_bdev_config **_raid_cfg); 251 int raid_bdev_config_add_base_bdev(struct raid_bdev_config *raid_cfg, 252 const char *base_bdev_name, uint8_t slot); 253 void raid_bdev_config_cleanup(struct raid_bdev_config *raid_cfg); 254 struct raid_bdev_config *raid_bdev_config_find_by_name(const char *raid_name); 255 enum raid_level raid_bdev_parse_raid_level(const char *str); 256 const char *raid_bdev_level_to_str(enum raid_level level); 257 258 /* 259 * RAID module descriptor 260 */ 261 struct raid_bdev_module { 262 /* RAID level implemented by this module */ 263 enum raid_level level; 264 265 /* Minimum required number of base bdevs. Must be > 0. */ 266 uint8_t base_bdevs_min; 267 268 /* 269 * Maximum number of base bdevs that can be removed without failing 270 * the array. 271 */ 272 uint8_t base_bdevs_max_degraded; 273 274 /* 275 * Called when the raid is starting, right before changing the state to 276 * online and registering the bdev. Parameters of the bdev like blockcnt 277 * should be set here. 278 * 279 * Non-zero return value will abort the startup process. 280 */ 281 int (*start)(struct raid_bdev *raid_bdev); 282 283 /* 284 * Called when the raid is stopping, right before changing the state to 285 * offline and unregistering the bdev. Optional. 286 */ 287 void (*stop)(struct raid_bdev *raid_bdev); 288 289 /* Handler for R/W requests */ 290 void (*submit_rw_request)(struct raid_bdev_io *raid_io); 291 292 /* Handler for requests without payload (flush, unmap). Optional. */ 293 void (*submit_null_payload_request)(struct raid_bdev_io *raid_io); 294 295 TAILQ_ENTRY(raid_bdev_module) link; 296 }; 297 298 void raid_bdev_module_list_add(struct raid_bdev_module *raid_module); 299 300 #define __RAID_MODULE_REGISTER(line) __RAID_MODULE_REGISTER_(line) 301 #define __RAID_MODULE_REGISTER_(line) raid_module_register_##line 302 303 #define RAID_MODULE_REGISTER(_module) \ 304 __attribute__((constructor)) static void \ 305 __RAID_MODULE_REGISTER(__LINE__)(void) \ 306 { \ 307 raid_bdev_module_list_add(_module); \ 308 } 309 310 bool 311 raid_bdev_io_complete_part(struct raid_bdev_io *raid_io, uint64_t completed, 312 enum spdk_bdev_io_status status); 313 void 314 raid_bdev_queue_io_wait(struct raid_bdev_io *raid_io, struct spdk_bdev *bdev, 315 struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn); 316 void 317 raid_bdev_io_complete(struct raid_bdev_io *raid_io, enum spdk_bdev_io_status status); 318 319 #endif /* SPDK_BDEV_RAID_INTERNAL_H */ 320