1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (C) 2012-2016 Intel Corporation 5 * All rights reserved. 6 * Copyright (C) 2018 Alexander Motin <mav@FreeBSD.org> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 #include <sys/param.h> 34 #include <sys/bio.h> 35 #include <sys/kernel.h> 36 #include <sys/malloc.h> 37 #include <sys/module.h> 38 #include <sys/queue.h> 39 #include <sys/sysctl.h> 40 #include <sys/systm.h> 41 #include <sys/taskqueue.h> 42 #include <machine/atomic.h> 43 44 #include <geom/geom.h> 45 #include <geom/geom_disk.h> 46 47 #include <dev/nvme/nvme.h> 48 49 #define NVD_STR "nvd" 50 51 struct nvd_disk; 52 struct nvd_controller; 53 54 static disk_ioctl_t nvd_ioctl; 55 static disk_strategy_t nvd_strategy; 56 static dumper_t nvd_dump; 57 58 static void nvd_done(void *arg, const struct nvme_completion *cpl); 59 static void nvd_gone(struct nvd_disk *ndisk); 60 61 static void *nvd_new_disk(struct nvme_namespace *ns, void *ctrlr); 62 63 static void *nvd_new_controller(struct nvme_controller *ctrlr); 64 static void nvd_controller_fail(void *ctrlr); 65 66 static int nvd_load(void); 67 static void nvd_unload(void); 68 69 MALLOC_DEFINE(M_NVD, "nvd", "nvd(4) allocations"); 70 71 struct nvme_consumer *consumer_handle; 72 73 struct nvd_disk { 74 struct nvd_controller *ctrlr; 75 76 struct bio_queue_head bioq; 77 struct task bioqtask; 78 struct mtx bioqlock; 79 80 struct disk *disk; 81 struct taskqueue *tq; 82 struct nvme_namespace *ns; 83 84 uint32_t cur_depth; 85 uint32_t ordered_in_flight; 86 u_int unit; 87 88 TAILQ_ENTRY(nvd_disk) global_tailq; 89 TAILQ_ENTRY(nvd_disk) ctrlr_tailq; 90 }; 91 92 struct nvd_controller { 93 94 TAILQ_ENTRY(nvd_controller) tailq; 95 TAILQ_HEAD(, nvd_disk) disk_head; 96 }; 97 98 static struct mtx nvd_lock; 99 static TAILQ_HEAD(, nvd_controller) ctrlr_head; 100 static TAILQ_HEAD(disk_list, nvd_disk) disk_head; 101 102 static SYSCTL_NODE(_hw, OID_AUTO, nvd, CTLFLAG_RD, 0, "nvd driver parameters"); 103 /* 104 * The NVMe specification does not define a maximum or optimal delete size, so 105 * technically max delete size is min(full size of the namespace, 2^32 - 1 106 * LBAs). A single delete for a multi-TB NVMe namespace though may take much 107 * longer to complete than the nvme(4) I/O timeout period. So choose a sensible 108 * default here that is still suitably large to minimize the number of overall 109 * delete operations. 110 */ 111 static uint64_t nvd_delete_max = (1024 * 1024 * 1024); /* 1GB */ 112 SYSCTL_UQUAD(_hw_nvd, OID_AUTO, delete_max, CTLFLAG_RDTUN, &nvd_delete_max, 0, 113 "nvd maximum BIO_DELETE size in bytes"); 114 115 static int nvd_modevent(module_t mod, int type, void *arg) 116 { 117 int error = 0; 118 119 switch (type) { 120 case MOD_LOAD: 121 error = nvd_load(); 122 break; 123 case MOD_UNLOAD: 124 nvd_unload(); 125 break; 126 default: 127 break; 128 } 129 130 return (error); 131 } 132 133 moduledata_t nvd_mod = { 134 NVD_STR, 135 (modeventhand_t)nvd_modevent, 136 0 137 }; 138 139 DECLARE_MODULE(nvd, nvd_mod, SI_SUB_DRIVERS, SI_ORDER_ANY); 140 MODULE_VERSION(nvd, 1); 141 MODULE_DEPEND(nvd, nvme, 1, 1, 1); 142 143 static int 144 nvd_load() 145 { 146 if (!nvme_use_nvd) 147 return 0; 148 149 mtx_init(&nvd_lock, "nvd_lock", NULL, MTX_DEF); 150 TAILQ_INIT(&ctrlr_head); 151 TAILQ_INIT(&disk_head); 152 153 consumer_handle = nvme_register_consumer(nvd_new_disk, 154 nvd_new_controller, NULL, nvd_controller_fail); 155 156 return (consumer_handle != NULL ? 0 : -1); 157 } 158 159 static void 160 nvd_unload() 161 { 162 struct nvd_controller *ctrlr; 163 struct nvd_disk *ndisk; 164 165 if (!nvme_use_nvd) 166 return; 167 168 mtx_lock(&nvd_lock); 169 while ((ctrlr = TAILQ_FIRST(&ctrlr_head)) != NULL) { 170 TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq); 171 TAILQ_FOREACH(ndisk, &ctrlr->disk_head, ctrlr_tailq) 172 nvd_gone(ndisk); 173 while (!TAILQ_EMPTY(&ctrlr->disk_head)) 174 msleep(&ctrlr->disk_head, &nvd_lock, 0, "nvd_unload",0); 175 free(ctrlr, M_NVD); 176 } 177 mtx_unlock(&nvd_lock); 178 179 nvme_unregister_consumer(consumer_handle); 180 181 mtx_destroy(&nvd_lock); 182 } 183 184 static int 185 nvd_bio_submit(struct nvd_disk *ndisk, struct bio *bp) 186 { 187 int err; 188 189 bp->bio_driver1 = NULL; 190 atomic_add_int(&ndisk->cur_depth, 1); 191 err = nvme_ns_bio_process(ndisk->ns, bp, nvd_done); 192 if (err) { 193 atomic_add_int(&ndisk->cur_depth, -1); 194 if (__predict_false(bp->bio_flags & BIO_ORDERED)) 195 atomic_add_int(&ndisk->ordered_in_flight, -1); 196 bp->bio_error = err; 197 bp->bio_flags |= BIO_ERROR; 198 bp->bio_resid = bp->bio_bcount; 199 biodone(bp); 200 return (-1); 201 } 202 203 return (0); 204 } 205 206 static void 207 nvd_strategy(struct bio *bp) 208 { 209 struct nvd_disk *ndisk; 210 211 ndisk = (struct nvd_disk *)bp->bio_disk->d_drv1; 212 213 if (__predict_false(bp->bio_flags & BIO_ORDERED)) 214 atomic_add_int(&ndisk->ordered_in_flight, 1); 215 216 if (__predict_true(ndisk->ordered_in_flight == 0)) { 217 nvd_bio_submit(ndisk, bp); 218 return; 219 } 220 221 /* 222 * There are ordered bios in flight, so we need to submit 223 * bios through the task queue to enforce ordering. 224 */ 225 mtx_lock(&ndisk->bioqlock); 226 bioq_insert_tail(&ndisk->bioq, bp); 227 mtx_unlock(&ndisk->bioqlock); 228 taskqueue_enqueue(ndisk->tq, &ndisk->bioqtask); 229 } 230 231 static void 232 nvd_gone(struct nvd_disk *ndisk) 233 { 234 struct bio *bp; 235 236 printf(NVD_STR"%u: detached\n", ndisk->unit); 237 mtx_lock(&ndisk->bioqlock); 238 disk_gone(ndisk->disk); 239 while ((bp = bioq_takefirst(&ndisk->bioq)) != NULL) { 240 if (__predict_false(bp->bio_flags & BIO_ORDERED)) 241 atomic_add_int(&ndisk->ordered_in_flight, -1); 242 bp->bio_error = ENXIO; 243 bp->bio_flags |= BIO_ERROR; 244 bp->bio_resid = bp->bio_bcount; 245 biodone(bp); 246 } 247 mtx_unlock(&ndisk->bioqlock); 248 } 249 250 static void 251 nvd_gonecb(struct disk *dp) 252 { 253 struct nvd_disk *ndisk = (struct nvd_disk *)dp->d_drv1; 254 255 disk_destroy(ndisk->disk); 256 mtx_lock(&nvd_lock); 257 TAILQ_REMOVE(&disk_head, ndisk, global_tailq); 258 TAILQ_REMOVE(&ndisk->ctrlr->disk_head, ndisk, ctrlr_tailq); 259 if (TAILQ_EMPTY(&ndisk->ctrlr->disk_head)) 260 wakeup(&ndisk->ctrlr->disk_head); 261 mtx_unlock(&nvd_lock); 262 taskqueue_free(ndisk->tq); 263 mtx_destroy(&ndisk->bioqlock); 264 free(ndisk, M_NVD); 265 } 266 267 static int 268 nvd_ioctl(struct disk *ndisk, u_long cmd, void *data, int fflag, 269 struct thread *td) 270 { 271 int ret = 0; 272 273 switch (cmd) { 274 default: 275 ret = EIO; 276 } 277 278 return (ret); 279 } 280 281 static int 282 nvd_dump(void *arg, void *virt, vm_offset_t phys, off_t offset, size_t len) 283 { 284 struct nvd_disk *ndisk; 285 struct disk *dp; 286 287 dp = arg; 288 ndisk = dp->d_drv1; 289 290 return (nvme_ns_dump(ndisk->ns, virt, offset, len)); 291 } 292 293 static void 294 nvd_done(void *arg, const struct nvme_completion *cpl) 295 { 296 struct bio *bp; 297 struct nvd_disk *ndisk; 298 299 bp = (struct bio *)arg; 300 301 ndisk = bp->bio_disk->d_drv1; 302 303 atomic_add_int(&ndisk->cur_depth, -1); 304 if (__predict_false(bp->bio_flags & BIO_ORDERED)) 305 atomic_add_int(&ndisk->ordered_in_flight, -1); 306 307 biodone(bp); 308 } 309 310 static void 311 nvd_bioq_process(void *arg, int pending) 312 { 313 struct nvd_disk *ndisk = arg; 314 struct bio *bp; 315 316 for (;;) { 317 mtx_lock(&ndisk->bioqlock); 318 bp = bioq_takefirst(&ndisk->bioq); 319 mtx_unlock(&ndisk->bioqlock); 320 if (bp == NULL) 321 break; 322 323 if (nvd_bio_submit(ndisk, bp) != 0) { 324 continue; 325 } 326 327 #ifdef BIO_ORDERED 328 /* 329 * BIO_ORDERED flag dictates that the bio with BIO_ORDERED 330 * flag set must be completed before proceeding with 331 * additional bios. 332 */ 333 if (bp->bio_flags & BIO_ORDERED) { 334 while (ndisk->cur_depth > 0) { 335 pause("nvd flush", 1); 336 } 337 } 338 #endif 339 } 340 } 341 342 static void * 343 nvd_new_controller(struct nvme_controller *ctrlr) 344 { 345 struct nvd_controller *nvd_ctrlr; 346 347 nvd_ctrlr = malloc(sizeof(struct nvd_controller), M_NVD, 348 M_ZERO | M_WAITOK); 349 350 TAILQ_INIT(&nvd_ctrlr->disk_head); 351 mtx_lock(&nvd_lock); 352 TAILQ_INSERT_TAIL(&ctrlr_head, nvd_ctrlr, tailq); 353 mtx_unlock(&nvd_lock); 354 355 return (nvd_ctrlr); 356 } 357 358 static void * 359 nvd_new_disk(struct nvme_namespace *ns, void *ctrlr_arg) 360 { 361 uint8_t descr[NVME_MODEL_NUMBER_LENGTH+1]; 362 struct nvd_disk *ndisk, *tnd; 363 struct disk *disk; 364 struct nvd_controller *ctrlr = ctrlr_arg; 365 int unit; 366 367 ndisk = malloc(sizeof(struct nvd_disk), M_NVD, M_ZERO | M_WAITOK); 368 ndisk->ctrlr = ctrlr; 369 ndisk->ns = ns; 370 ndisk->cur_depth = 0; 371 ndisk->ordered_in_flight = 0; 372 mtx_init(&ndisk->bioqlock, "nvd bioq lock", NULL, MTX_DEF); 373 bioq_init(&ndisk->bioq); 374 TASK_INIT(&ndisk->bioqtask, 0, nvd_bioq_process, ndisk); 375 376 mtx_lock(&nvd_lock); 377 unit = 0; 378 TAILQ_FOREACH(tnd, &disk_head, global_tailq) { 379 if (tnd->unit > unit) 380 break; 381 unit = tnd->unit + 1; 382 } 383 ndisk->unit = unit; 384 if (tnd != NULL) 385 TAILQ_INSERT_BEFORE(tnd, ndisk, global_tailq); 386 else 387 TAILQ_INSERT_TAIL(&disk_head, ndisk, global_tailq); 388 TAILQ_INSERT_TAIL(&ctrlr->disk_head, ndisk, ctrlr_tailq); 389 mtx_unlock(&nvd_lock); 390 391 ndisk->tq = taskqueue_create("nvd_taskq", M_WAITOK, 392 taskqueue_thread_enqueue, &ndisk->tq); 393 taskqueue_start_threads(&ndisk->tq, 1, PI_DISK, "nvd taskq"); 394 395 disk = ndisk->disk = disk_alloc(); 396 disk->d_strategy = nvd_strategy; 397 disk->d_ioctl = nvd_ioctl; 398 disk->d_dump = nvd_dump; 399 disk->d_gone = nvd_gonecb; 400 disk->d_name = NVD_STR; 401 disk->d_unit = ndisk->unit; 402 disk->d_drv1 = ndisk; 403 404 disk->d_sectorsize = nvme_ns_get_sector_size(ns); 405 disk->d_mediasize = (off_t)nvme_ns_get_size(ns); 406 disk->d_maxsize = nvme_ns_get_max_io_xfer_size(ns); 407 disk->d_delmaxsize = (off_t)nvme_ns_get_size(ns); 408 if (disk->d_delmaxsize > nvd_delete_max) 409 disk->d_delmaxsize = nvd_delete_max; 410 disk->d_stripesize = nvme_ns_get_stripesize(ns); 411 disk->d_flags = DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION; 412 if (nvme_ns_get_flags(ns) & NVME_NS_DEALLOCATE_SUPPORTED) 413 disk->d_flags |= DISKFLAG_CANDELETE; 414 if (nvme_ns_get_flags(ns) & NVME_NS_FLUSH_SUPPORTED) 415 disk->d_flags |= DISKFLAG_CANFLUSHCACHE; 416 417 /* 418 * d_ident and d_descr are both far bigger than the length of either 419 * the serial or model number strings. 420 */ 421 nvme_strvis(disk->d_ident, nvme_ns_get_serial_number(ns), 422 sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH); 423 nvme_strvis(descr, nvme_ns_get_model_number(ns), sizeof(descr), 424 NVME_MODEL_NUMBER_LENGTH); 425 strlcpy(disk->d_descr, descr, sizeof(descr)); 426 427 disk->d_rotation_rate = DISK_RR_NON_ROTATING; 428 429 disk_create(disk, DISK_VERSION); 430 431 printf(NVD_STR"%u: <%s> NVMe namespace\n", disk->d_unit, descr); 432 printf(NVD_STR"%u: %juMB (%ju %u byte sectors)\n", disk->d_unit, 433 (uintmax_t)disk->d_mediasize / (1024*1024), 434 (uintmax_t)disk->d_mediasize / disk->d_sectorsize, 435 disk->d_sectorsize); 436 437 return (ndisk); 438 } 439 440 static void 441 nvd_controller_fail(void *ctrlr_arg) 442 { 443 struct nvd_controller *ctrlr = ctrlr_arg; 444 struct nvd_disk *ndisk; 445 446 mtx_lock(&nvd_lock); 447 TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq); 448 TAILQ_FOREACH(ndisk, &ctrlr->disk_head, ctrlr_tailq) 449 nvd_gone(ndisk); 450 while (!TAILQ_EMPTY(&ctrlr->disk_head)) 451 msleep(&ctrlr->disk_head, &nvd_lock, 0, "nvd_fail", 0); 452 mtx_unlock(&nvd_lock); 453 free(ctrlr, M_NVD); 454 } 455 456