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 CONCAT = 99, 44 }; 45 46 /* 47 * Raid state describes the state of the raid. This raid bdev can be either in 48 * configured list or configuring list 49 */ 50 enum raid_bdev_state { 51 /* raid bdev is ready and is seen by upper layers */ 52 RAID_BDEV_STATE_ONLINE, 53 54 /* 55 * raid bdev is configuring, not all underlying bdevs are present. 56 * And can't be seen by upper layers. 57 */ 58 RAID_BDEV_STATE_CONFIGURING, 59 60 /* 61 * In offline state, raid bdev layer will complete all incoming commands without 62 * submitting to underlying base nvme bdevs 63 */ 64 RAID_BDEV_STATE_OFFLINE, 65 66 /* raid bdev max, new states should be added before this */ 67 RAID_BDEV_MAX 68 }; 69 70 /* 71 * raid_base_bdev_info contains information for the base bdevs which are part of some 72 * raid. This structure contains the per base bdev information. Whatever is 73 * required per base device for raid bdev will be kept here 74 */ 75 struct raid_base_bdev_info { 76 /* pointer to base spdk bdev */ 77 struct spdk_bdev *bdev; 78 79 /* pointer to base bdev descriptor opened by raid bdev */ 80 struct spdk_bdev_desc *desc; 81 82 /* 83 * When underlying base device calls the hot plug function on drive removal, 84 * this flag will be set and later after doing some processing, base device 85 * descriptor will be closed 86 */ 87 bool remove_scheduled; 88 89 /* thread where base device is opened */ 90 struct spdk_thread *thread; 91 }; 92 93 /* 94 * raid_bdev_io is the context part of bdev_io. It contains the information 95 * related to bdev_io for a raid bdev 96 */ 97 struct raid_bdev_io { 98 /* The raid bdev associated with this IO */ 99 struct raid_bdev *raid_bdev; 100 101 /* WaitQ entry, used only in waitq logic */ 102 struct spdk_bdev_io_wait_entry waitq_entry; 103 104 /* Context of the original channel for this IO */ 105 struct raid_bdev_io_channel *raid_ch; 106 107 /* Used for tracking progress on io requests sent to member disks. */ 108 uint64_t base_bdev_io_remaining; 109 uint8_t base_bdev_io_submitted; 110 uint8_t base_bdev_io_status; 111 }; 112 113 /* 114 * raid_bdev is the single entity structure which contains SPDK block device 115 * and the information related to any raid bdev either configured or 116 * in configuring list. io device is created on this. 117 */ 118 struct raid_bdev { 119 /* raid bdev device, this will get registered in bdev layer */ 120 struct spdk_bdev bdev; 121 122 /* link of raid bdev to link it to configured, configuring or offline list */ 123 TAILQ_ENTRY(raid_bdev) state_link; 124 125 /* link of raid bdev to link it to global raid bdev list */ 126 TAILQ_ENTRY(raid_bdev) global_link; 127 128 /* pointer to config file entry */ 129 struct raid_bdev_config *config; 130 131 /* array of base bdev info */ 132 struct raid_base_bdev_info *base_bdev_info; 133 134 /* strip size of raid bdev in blocks */ 135 uint32_t strip_size; 136 137 /* strip size of raid bdev in KB */ 138 uint32_t strip_size_kb; 139 140 /* strip size bit shift for optimized calculation */ 141 uint32_t strip_size_shift; 142 143 /* block length bit shift for optimized calculation */ 144 uint32_t blocklen_shift; 145 146 /* state of raid bdev */ 147 enum raid_bdev_state state; 148 149 /* number of base bdevs comprising raid bdev */ 150 uint8_t num_base_bdevs; 151 152 /* number of base bdevs discovered */ 153 uint8_t num_base_bdevs_discovered; 154 155 /* Raid Level of this raid bdev */ 156 enum raid_level level; 157 158 /* Set to true if destruct is called for this raid bdev */ 159 bool destruct_called; 160 161 /* Set to true if destroy of this raid bdev is started. */ 162 bool destroy_started; 163 164 /* Module for RAID-level specific operations */ 165 struct raid_bdev_module *module; 166 167 /* Private data for the raid module */ 168 void *module_private; 169 }; 170 171 #define RAID_FOR_EACH_BASE_BDEV(r, i) \ 172 for (i = r->base_bdev_info; i < r->base_bdev_info + r->num_base_bdevs; i++) 173 174 /* 175 * raid_base_bdev_config is the per base bdev data structure which contains 176 * information w.r.t to per base bdev during parsing config 177 */ 178 struct raid_base_bdev_config { 179 /* base bdev name from config file */ 180 char *name; 181 }; 182 183 /* 184 * raid_bdev_config contains the raid bdev config related information after 185 * parsing the config file 186 */ 187 struct raid_bdev_config { 188 /* base bdev config per underlying bdev */ 189 struct raid_base_bdev_config *base_bdev; 190 191 /* Points to already created raid bdev */ 192 struct raid_bdev *raid_bdev; 193 194 char *name; 195 196 /* strip size of this raid bdev in KB */ 197 uint32_t strip_size; 198 199 /* number of base bdevs */ 200 uint8_t num_base_bdevs; 201 202 /* raid level */ 203 enum raid_level level; 204 205 TAILQ_ENTRY(raid_bdev_config) link; 206 }; 207 208 /* 209 * raid_config is the top level structure representing the raid bdev config as read 210 * from config file for all raids 211 */ 212 struct raid_config { 213 /* raid bdev context from config file */ 214 TAILQ_HEAD(, raid_bdev_config) raid_bdev_config_head; 215 216 /* total raid bdev from config file */ 217 uint8_t total_raid_bdev; 218 }; 219 220 /* 221 * raid_bdev_io_channel is the context of spdk_io_channel for raid bdev device. It 222 * contains the relationship of raid bdev io channel with base bdev io channels. 223 */ 224 struct raid_bdev_io_channel { 225 /* Array of IO channels of base bdevs */ 226 struct spdk_io_channel **base_channel; 227 228 /* Number of IO channels */ 229 uint8_t num_channels; 230 }; 231 232 /* TAIL heads for various raid bdev lists */ 233 TAILQ_HEAD(raid_configured_tailq, raid_bdev); 234 TAILQ_HEAD(raid_configuring_tailq, raid_bdev); 235 TAILQ_HEAD(raid_all_tailq, raid_bdev); 236 TAILQ_HEAD(raid_offline_tailq, raid_bdev); 237 238 extern struct raid_configured_tailq g_raid_bdev_configured_list; 239 extern struct raid_configuring_tailq g_raid_bdev_configuring_list; 240 extern struct raid_all_tailq g_raid_bdev_list; 241 extern struct raid_offline_tailq g_raid_bdev_offline_list; 242 extern struct raid_config g_raid_config; 243 244 typedef void (*raid_bdev_destruct_cb)(void *cb_ctx, int rc); 245 246 int raid_bdev_create(struct raid_bdev_config *raid_cfg); 247 int raid_bdev_add_base_devices(struct raid_bdev_config *raid_cfg); 248 void raid_bdev_remove_base_devices(struct raid_bdev_config *raid_cfg, 249 raid_bdev_destruct_cb cb_fn, void *cb_ctx); 250 int raid_bdev_config_add(const char *raid_name, uint32_t strip_size, uint8_t num_base_bdevs, 251 enum raid_level level, struct raid_bdev_config **_raid_cfg); 252 int raid_bdev_config_add_base_bdev(struct raid_bdev_config *raid_cfg, 253 const char *base_bdev_name, uint8_t slot); 254 void raid_bdev_config_cleanup(struct raid_bdev_config *raid_cfg); 255 struct raid_bdev_config *raid_bdev_config_find_by_name(const char *raid_name); 256 enum raid_level raid_bdev_parse_raid_level(const char *str); 257 const char *raid_bdev_level_to_str(enum raid_level level); 258 259 /* 260 * RAID module descriptor 261 */ 262 struct raid_bdev_module { 263 /* RAID level implemented by this module */ 264 enum raid_level level; 265 266 /* Minimum required number of base bdevs. Must be > 0. */ 267 uint8_t base_bdevs_min; 268 269 /* 270 * Maximum number of base bdevs that can be removed without failing 271 * the array. 272 */ 273 uint8_t base_bdevs_max_degraded; 274 275 /* 276 * Called when the raid is starting, right before changing the state to 277 * online and registering the bdev. Parameters of the bdev like blockcnt 278 * should be set here. 279 * 280 * Non-zero return value will abort the startup process. 281 */ 282 int (*start)(struct raid_bdev *raid_bdev); 283 284 /* 285 * Called when the raid is stopping, right before changing the state to 286 * offline and unregistering the bdev. Optional. 287 */ 288 void (*stop)(struct raid_bdev *raid_bdev); 289 290 /* Handler for R/W requests */ 291 void (*submit_rw_request)(struct raid_bdev_io *raid_io); 292 293 /* Handler for requests without payload (flush, unmap). Optional. */ 294 void (*submit_null_payload_request)(struct raid_bdev_io *raid_io); 295 296 TAILQ_ENTRY(raid_bdev_module) link; 297 }; 298 299 void raid_bdev_module_list_add(struct raid_bdev_module *raid_module); 300 301 #define __RAID_MODULE_REGISTER(line) __RAID_MODULE_REGISTER_(line) 302 #define __RAID_MODULE_REGISTER_(line) raid_module_register_##line 303 304 #define RAID_MODULE_REGISTER(_module) \ 305 __attribute__((constructor)) static void \ 306 __RAID_MODULE_REGISTER(__LINE__)(void) \ 307 { \ 308 raid_bdev_module_list_add(_module); \ 309 } 310 311 bool 312 raid_bdev_io_complete_part(struct raid_bdev_io *raid_io, uint64_t completed, 313 enum spdk_bdev_io_status status); 314 void 315 raid_bdev_queue_io_wait(struct raid_bdev_io *raid_io, struct spdk_bdev *bdev, 316 struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn); 317 void 318 raid_bdev_io_complete(struct raid_bdev_io *raid_io, enum spdk_bdev_io_status status); 319 320 #endif /* SPDK_BDEV_RAID_INTERNAL_H */ 321