1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (C) 2012-2013 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 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 #include <sys/param.h> 31 #include <sys/bio.h> 32 #include <sys/bus.h> 33 #include <sys/conf.h> 34 #include <sys/disk.h> 35 #include <sys/fcntl.h> 36 #include <sys/ioccom.h> 37 #include <sys/malloc.h> 38 #include <sys/module.h> 39 #include <sys/proc.h> 40 #include <sys/systm.h> 41 42 #include <dev/pci/pcivar.h> 43 44 #include <geom/geom.h> 45 46 #include "nvme_private.h" 47 48 static void nvme_bio_child_inbed(struct bio *parent, int bio_error); 49 static void nvme_bio_child_done(void *arg, 50 const struct nvme_completion *cpl); 51 static uint32_t nvme_get_num_segments(uint64_t addr, uint64_t size, 52 uint32_t alignment); 53 static void nvme_free_child_bios(int num_bios, 54 struct bio **child_bios); 55 static struct bio ** nvme_allocate_child_bios(int num_bios); 56 static struct bio ** nvme_construct_child_bios(struct bio *bp, 57 uint32_t alignment, 58 int *num_bios); 59 static int nvme_ns_split_bio(struct nvme_namespace *ns, 60 struct bio *bp, 61 uint32_t alignment); 62 63 static int 64 nvme_ns_ioctl(struct cdev *cdev, u_long cmd, caddr_t arg, int flag, 65 struct thread *td) 66 { 67 struct nvme_namespace *ns; 68 struct nvme_controller *ctrlr; 69 struct nvme_pt_command *pt; 70 71 ns = cdev->si_drv1; 72 ctrlr = ns->ctrlr; 73 74 switch (cmd) { 75 case NVME_IO_TEST: 76 case NVME_BIO_TEST: 77 nvme_ns_test(ns, cmd, arg); 78 break; 79 case NVME_PASSTHROUGH_CMD: 80 pt = (struct nvme_pt_command *)arg; 81 return (nvme_ctrlr_passthrough_cmd(ctrlr, pt, ns->id, 82 1 /* is_user_buffer */, 0 /* is_admin_cmd */)); 83 case NVME_GET_NSID: 84 { 85 struct nvme_get_nsid *gnsid = (struct nvme_get_nsid *)arg; 86 strncpy(gnsid->cdev, device_get_nameunit(ctrlr->dev), 87 sizeof(gnsid->cdev)); 88 gnsid->cdev[sizeof(gnsid->cdev) - 1] = '\0'; 89 gnsid->nsid = ns->id; 90 break; 91 } 92 case DIOCGMEDIASIZE: 93 *(off_t *)arg = (off_t)nvme_ns_get_size(ns); 94 break; 95 case DIOCGSECTORSIZE: 96 *(u_int *)arg = nvme_ns_get_sector_size(ns); 97 break; 98 default: 99 return (ENOTTY); 100 } 101 102 return (0); 103 } 104 105 static int 106 nvme_ns_open(struct cdev *dev __unused, int flags, int fmt __unused, 107 struct thread *td) 108 { 109 int error = 0; 110 111 if (flags & FWRITE) 112 error = securelevel_gt(td->td_ucred, 0); 113 114 return (error); 115 } 116 117 static int 118 nvme_ns_close(struct cdev *dev __unused, int flags, int fmt __unused, 119 struct thread *td) 120 { 121 122 return (0); 123 } 124 125 static void 126 nvme_ns_strategy_done(void *arg, const struct nvme_completion *cpl) 127 { 128 struct bio *bp = arg; 129 130 /* 131 * TODO: add more extensive translation of NVMe status codes 132 * to different bio error codes (i.e. EIO, EINVAL, etc.) 133 */ 134 if (nvme_completion_is_error(cpl)) { 135 bp->bio_error = EIO; 136 bp->bio_flags |= BIO_ERROR; 137 bp->bio_resid = bp->bio_bcount; 138 } else 139 bp->bio_resid = 0; 140 141 biodone(bp); 142 } 143 144 static void 145 nvme_ns_strategy(struct bio *bp) 146 { 147 struct nvme_namespace *ns; 148 int err; 149 150 ns = bp->bio_dev->si_drv1; 151 err = nvme_ns_bio_process(ns, bp, nvme_ns_strategy_done); 152 153 if (err) { 154 bp->bio_error = err; 155 bp->bio_flags |= BIO_ERROR; 156 bp->bio_resid = bp->bio_bcount; 157 biodone(bp); 158 } 159 160 } 161 162 static struct cdevsw nvme_ns_cdevsw = { 163 .d_version = D_VERSION, 164 .d_flags = D_DISK, 165 .d_read = physread, 166 .d_write = physwrite, 167 .d_open = nvme_ns_open, 168 .d_close = nvme_ns_close, 169 .d_strategy = nvme_ns_strategy, 170 .d_ioctl = nvme_ns_ioctl 171 }; 172 173 uint32_t 174 nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns) 175 { 176 return ns->ctrlr->max_xfer_size; 177 } 178 179 uint32_t 180 nvme_ns_get_sector_size(struct nvme_namespace *ns) 181 { 182 uint8_t flbas_fmt, lbads; 183 184 flbas_fmt = NVMEV(NVME_NS_DATA_FLBAS_FORMAT, ns->data.flbas); 185 lbads = NVMEV(NVME_NS_DATA_LBAF_LBADS, ns->data.lbaf[flbas_fmt]); 186 187 return (1 << lbads); 188 } 189 190 uint64_t 191 nvme_ns_get_num_sectors(struct nvme_namespace *ns) 192 { 193 return (ns->data.nsze); 194 } 195 196 uint64_t 197 nvme_ns_get_size(struct nvme_namespace *ns) 198 { 199 return (nvme_ns_get_num_sectors(ns) * nvme_ns_get_sector_size(ns)); 200 } 201 202 uint32_t 203 nvme_ns_get_flags(struct nvme_namespace *ns) 204 { 205 return (ns->flags); 206 } 207 208 const char * 209 nvme_ns_get_serial_number(struct nvme_namespace *ns) 210 { 211 return ((const char *)ns->ctrlr->cdata.sn); 212 } 213 214 const char * 215 nvme_ns_get_model_number(struct nvme_namespace *ns) 216 { 217 return ((const char *)ns->ctrlr->cdata.mn); 218 } 219 220 const struct nvme_namespace_data * 221 nvme_ns_get_data(struct nvme_namespace *ns) 222 { 223 224 return (&ns->data); 225 } 226 227 uint32_t 228 nvme_ns_get_stripesize(struct nvme_namespace *ns) 229 { 230 uint32_t ss; 231 232 if (NVMEV(NVME_NS_DATA_NSFEAT_NPVALID, ns->data.nsfeat) != 0) { 233 ss = nvme_ns_get_sector_size(ns); 234 if (ns->data.npwa != 0) 235 return ((ns->data.npwa + 1) * ss); 236 else if (ns->data.npwg != 0) 237 return ((ns->data.npwg + 1) * ss); 238 } 239 return (ns->boundary); 240 } 241 242 static void 243 nvme_ns_bio_done(void *arg, const struct nvme_completion *status) 244 { 245 struct bio *bp = arg; 246 nvme_cb_fn_t bp_cb_fn; 247 248 bp_cb_fn = bp->bio_driver1; 249 250 if (bp->bio_driver2) 251 free(bp->bio_driver2, M_NVME); 252 253 if (nvme_completion_is_error(status)) { 254 bp->bio_flags |= BIO_ERROR; 255 if (bp->bio_error == 0) 256 bp->bio_error = EIO; 257 } 258 259 if ((bp->bio_flags & BIO_ERROR) == 0) 260 bp->bio_resid = 0; 261 else 262 bp->bio_resid = bp->bio_bcount; 263 264 bp_cb_fn(bp, status); 265 } 266 267 static void 268 nvme_bio_child_inbed(struct bio *parent, int bio_error) 269 { 270 struct nvme_completion parent_cpl; 271 int children, inbed; 272 273 if (bio_error != 0) { 274 parent->bio_flags |= BIO_ERROR; 275 parent->bio_error = bio_error; 276 } 277 278 /* 279 * atomic_fetchadd will return value before adding 1, so we still 280 * must add 1 to get the updated inbed number. Save bio_children 281 * before incrementing to guard against race conditions when 282 * two children bios complete on different queues. 283 */ 284 children = atomic_load_acq_int(&parent->bio_children); 285 inbed = atomic_fetchadd_int(&parent->bio_inbed, 1) + 1; 286 if (inbed == children) { 287 bzero(&parent_cpl, sizeof(parent_cpl)); 288 if (parent->bio_flags & BIO_ERROR) { 289 parent_cpl.status &= ~NVMEM(NVME_STATUS_SC); 290 parent_cpl.status |= NVMEF(NVME_STATUS_SC, 291 NVME_SC_DATA_TRANSFER_ERROR); 292 } 293 nvme_ns_bio_done(parent, &parent_cpl); 294 } 295 } 296 297 static void 298 nvme_bio_child_done(void *arg, const struct nvme_completion *cpl) 299 { 300 struct bio *child = arg; 301 struct bio *parent; 302 int bio_error; 303 304 parent = child->bio_parent; 305 g_destroy_bio(child); 306 bio_error = nvme_completion_is_error(cpl) ? EIO : 0; 307 nvme_bio_child_inbed(parent, bio_error); 308 } 309 310 static uint32_t 311 nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t align) 312 { 313 uint32_t num_segs, offset, remainder; 314 315 if (align == 0) 316 return (1); 317 318 KASSERT((align & (align - 1)) == 0, ("alignment not power of 2\n")); 319 320 num_segs = size / align; 321 remainder = size & (align - 1); 322 offset = addr & (align - 1); 323 if (remainder > 0 || offset > 0) 324 num_segs += 1 + (remainder + offset - 1) / align; 325 return (num_segs); 326 } 327 328 static void 329 nvme_free_child_bios(int num_bios, struct bio **child_bios) 330 { 331 int i; 332 333 for (i = 0; i < num_bios; i++) { 334 if (child_bios[i] != NULL) 335 g_destroy_bio(child_bios[i]); 336 } 337 338 free(child_bios, M_NVME); 339 } 340 341 static struct bio ** 342 nvme_allocate_child_bios(int num_bios) 343 { 344 struct bio **child_bios; 345 int err = 0, i; 346 347 child_bios = malloc(num_bios * sizeof(struct bio *), M_NVME, M_NOWAIT); 348 if (child_bios == NULL) 349 return (NULL); 350 351 for (i = 0; i < num_bios; i++) { 352 child_bios[i] = g_new_bio(); 353 if (child_bios[i] == NULL) 354 err = ENOMEM; 355 } 356 357 if (err == ENOMEM) { 358 nvme_free_child_bios(num_bios, child_bios); 359 return (NULL); 360 } 361 362 return (child_bios); 363 } 364 365 static struct bio ** 366 nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios) 367 { 368 struct bio **child_bios; 369 struct bio *child; 370 uint64_t cur_offset; 371 caddr_t data; 372 uint32_t rem_bcount; 373 int i; 374 struct vm_page **ma; 375 uint32_t ma_offset; 376 377 *num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount, 378 alignment); 379 child_bios = nvme_allocate_child_bios(*num_bios); 380 if (child_bios == NULL) 381 return (NULL); 382 383 bp->bio_children = *num_bios; 384 bp->bio_inbed = 0; 385 cur_offset = bp->bio_offset; 386 rem_bcount = bp->bio_bcount; 387 data = bp->bio_data; 388 ma_offset = bp->bio_ma_offset; 389 ma = bp->bio_ma; 390 391 for (i = 0; i < *num_bios; i++) { 392 child = child_bios[i]; 393 child->bio_parent = bp; 394 child->bio_cmd = bp->bio_cmd; 395 child->bio_offset = cur_offset; 396 child->bio_bcount = min(rem_bcount, 397 alignment - (cur_offset & (alignment - 1))); 398 child->bio_flags = bp->bio_flags; 399 if (bp->bio_flags & BIO_UNMAPPED) { 400 child->bio_ma_offset = ma_offset; 401 child->bio_ma = ma; 402 child->bio_ma_n = 403 nvme_get_num_segments(child->bio_ma_offset, 404 child->bio_bcount, PAGE_SIZE); 405 ma_offset = (ma_offset + child->bio_bcount) & 406 PAGE_MASK; 407 ma += child->bio_ma_n; 408 if (ma_offset != 0) 409 ma -= 1; 410 } else { 411 child->bio_data = data; 412 data += child->bio_bcount; 413 } 414 cur_offset += child->bio_bcount; 415 rem_bcount -= child->bio_bcount; 416 } 417 418 return (child_bios); 419 } 420 421 static int 422 nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp, 423 uint32_t alignment) 424 { 425 struct bio *child; 426 struct bio **child_bios; 427 int err, i, num_bios; 428 429 child_bios = nvme_construct_child_bios(bp, alignment, &num_bios); 430 if (child_bios == NULL) 431 return (ENOMEM); 432 433 for (i = 0; i < num_bios; i++) { 434 child = child_bios[i]; 435 err = nvme_ns_bio_process(ns, child, nvme_bio_child_done); 436 if (err != 0) { 437 nvme_bio_child_inbed(bp, err); 438 g_destroy_bio(child); 439 } 440 } 441 442 free(child_bios, M_NVME); 443 return (0); 444 } 445 446 int 447 nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp, 448 nvme_cb_fn_t cb_fn) 449 { 450 struct nvme_dsm_range *dsm_range; 451 uint32_t num_bios; 452 int err; 453 454 bp->bio_driver1 = cb_fn; 455 456 if (ns->boundary > 0 && 457 (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) { 458 num_bios = nvme_get_num_segments(bp->bio_offset, 459 bp->bio_bcount, ns->boundary); 460 if (num_bios > 1) 461 return (nvme_ns_split_bio(ns, bp, ns->boundary)); 462 } 463 464 switch (bp->bio_cmd) { 465 case BIO_READ: 466 err = nvme_ns_cmd_read_bio(ns, bp, nvme_ns_bio_done, bp); 467 break; 468 case BIO_WRITE: 469 err = nvme_ns_cmd_write_bio(ns, bp, nvme_ns_bio_done, bp); 470 break; 471 case BIO_FLUSH: 472 err = nvme_ns_cmd_flush(ns, nvme_ns_bio_done, bp); 473 break; 474 case BIO_DELETE: 475 dsm_range = 476 malloc(sizeof(struct nvme_dsm_range), M_NVME, 477 M_ZERO | M_NOWAIT); 478 if (!dsm_range) { 479 err = ENOMEM; 480 break; 481 } 482 dsm_range->length = 483 htole32(bp->bio_bcount/nvme_ns_get_sector_size(ns)); 484 dsm_range->starting_lba = 485 htole64(bp->bio_offset/nvme_ns_get_sector_size(ns)); 486 bp->bio_driver2 = dsm_range; 487 err = nvme_ns_cmd_deallocate(ns, dsm_range, 1, 488 nvme_ns_bio_done, bp); 489 if (err != 0) 490 free(dsm_range, M_NVME); 491 break; 492 default: 493 err = EOPNOTSUPP; 494 break; 495 } 496 497 return (err); 498 } 499 500 int 501 nvme_ns_ioctl_process(struct nvme_namespace *ns, u_long cmd, caddr_t arg, 502 int flag, struct thread *td) 503 { 504 return (nvme_ns_ioctl(ns->cdev, cmd, arg, flag, td)); 505 } 506 507 int 508 nvme_ns_construct(struct nvme_namespace *ns, uint32_t id, 509 struct nvme_controller *ctrlr) 510 { 511 struct make_dev_args md_args; 512 struct nvme_completion_poll_status status; 513 int res; 514 int unit; 515 uint8_t flbas_fmt; 516 uint8_t vwc_present; 517 518 ns->ctrlr = ctrlr; 519 ns->id = id; 520 521 /* 522 * Namespaces are reconstructed after a controller reset, so check 523 * to make sure we only call mtx_init once on each mtx. 524 * 525 * TODO: Move this somewhere where it gets called at controller 526 * construction time, which is not invoked as part of each 527 * controller reset. 528 */ 529 if (!mtx_initialized(&ns->lock)) 530 mtx_init(&ns->lock, "nvme ns lock", NULL, MTX_DEF); 531 532 status.done = 0; 533 nvme_ctrlr_cmd_identify_namespace(ctrlr, id, &ns->data, 534 nvme_completion_poll_cb, &status); 535 nvme_completion_poll(&status); 536 if (nvme_completion_is_error(&status.cpl)) { 537 nvme_printf(ctrlr, "nvme_identify_namespace failed\n"); 538 return (ENXIO); 539 } 540 541 /* Convert data to host endian */ 542 nvme_namespace_data_swapbytes(&ns->data); 543 544 /* 545 * If the size of is zero, chances are this isn't a valid 546 * namespace (eg one that's not been configured yet). The 547 * standard says the entire id will be zeros, so this is a 548 * cheap way to test for that. 549 */ 550 if (ns->data.nsze == 0) 551 return (ENXIO); 552 553 flbas_fmt = NVMEV(NVME_NS_DATA_FLBAS_FORMAT, ns->data.flbas); 554 555 /* 556 * Note: format is a 0-based value, so > is appropriate here, 557 * not >=. 558 */ 559 if (flbas_fmt > ns->data.nlbaf) { 560 nvme_printf(ctrlr, 561 "lba format %d exceeds number supported (%d)\n", 562 flbas_fmt, ns->data.nlbaf + 1); 563 return (ENXIO); 564 } 565 566 /* 567 * Older Intel devices (like the PC35xxx and P45xx series) advertise in 568 * vendor specific space an alignment that improves performance. If 569 * present use for the stripe size. NVMe 1.3 standardized this as 570 * NOIOB, and newer Intel drives use that. 571 */ 572 if ((ctrlr->quirks & QUIRK_INTEL_ALIGNMENT) != 0) { 573 if (ctrlr->cdata.vs[3] != 0) 574 ns->boundary = 575 1 << (ctrlr->cdata.vs[3] + NVME_MPS_SHIFT + 576 NVME_CAP_HI_MPSMIN(ctrlr->cap_hi)); 577 else 578 ns->boundary = 0; 579 } else { 580 ns->boundary = ns->data.noiob * nvme_ns_get_sector_size(ns); 581 } 582 583 if (nvme_ctrlr_has_dataset_mgmt(&ctrlr->cdata)) 584 ns->flags |= NVME_NS_DEALLOCATE_SUPPORTED; 585 586 vwc_present = NVMEV(NVME_CTRLR_DATA_VWC_PRESENT, ctrlr->cdata.vwc); 587 if (vwc_present) 588 ns->flags |= NVME_NS_FLUSH_SUPPORTED; 589 590 /* 591 * cdev may have already been created, if we are reconstructing the 592 * namespace after a controller-level reset. 593 */ 594 if (ns->cdev != NULL) 595 return (0); 596 597 /* 598 * Namespace IDs start at 1, so we need to subtract 1 to create a 599 * correct unit number. 600 */ 601 unit = device_get_unit(ctrlr->dev) * NVME_MAX_NAMESPACES + ns->id - 1; 602 603 make_dev_args_init(&md_args); 604 md_args.mda_devsw = &nvme_ns_cdevsw; 605 md_args.mda_unit = unit; 606 md_args.mda_mode = 0600; 607 md_args.mda_si_drv1 = ns; 608 res = make_dev_s(&md_args, &ns->cdev, "nvme%dns%d", 609 device_get_unit(ctrlr->dev), ns->id); 610 if (res != 0) 611 return (ENXIO); 612 613 ns->cdev->si_flags |= SI_UNMAPPED; 614 615 return (0); 616 } 617 618 void 619 nvme_ns_destruct(struct nvme_namespace *ns) 620 { 621 622 if (ns->cdev != NULL) 623 destroy_dev(ns->cdev); 624 } 625