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/param.h> 30 #include <sys/bio.h> 31 #include <sys/bus.h> 32 #include <sys/conf.h> 33 #include <sys/disk.h> 34 #include <sys/fcntl.h> 35 #include <sys/ioccom.h> 36 #include <sys/malloc.h> 37 #include <sys/module.h> 38 #include <sys/proc.h> 39 #include <sys/systm.h> 40 41 #include <dev/pci/pcivar.h> 42 43 #include <geom/geom.h> 44 45 #include "nvme_private.h" 46 #include "nvme_linux.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 strlcpy(gnsid->cdev, device_get_nameunit(ctrlr->dev), 87 sizeof(gnsid->cdev)); 88 gnsid->nsid = ns->id; 89 break; 90 } 91 case DIOCGMEDIASIZE: 92 *(off_t *)arg = (off_t)nvme_ns_get_size(ns); 93 break; 94 case DIOCGSECTORSIZE: 95 *(u_int *)arg = nvme_ns_get_sector_size(ns); 96 break; 97 /* Linux Compatible (see nvme_linux.h) */ 98 case NVME_IOCTL_ID: 99 td->td_retval[0] = ns->id; 100 return (0); 101 102 case NVME_IOCTL_ADMIN_CMD: 103 case NVME_IOCTL_IO_CMD: { 104 struct nvme_passthru_cmd *npc = (struct nvme_passthru_cmd *)arg; 105 106 return (nvme_ctrlr_linux_passthru_cmd(ctrlr, npc, ns->id, true, 107 cmd == NVME_IOCTL_ADMIN_CMD)); 108 } 109 default: 110 return (ENOTTY); 111 } 112 113 return (0); 114 } 115 116 static int 117 nvme_ns_open(struct cdev *dev __unused, int flags, int fmt __unused, 118 struct thread *td) 119 { 120 int error = 0; 121 122 if (flags & FWRITE) 123 error = securelevel_gt(td->td_ucred, 0); 124 125 return (error); 126 } 127 128 static int 129 nvme_ns_close(struct cdev *dev __unused, int flags, int fmt __unused, 130 struct thread *td) 131 { 132 return (0); 133 } 134 135 static void 136 nvme_ns_strategy_done(void *arg, const struct nvme_completion *cpl) 137 { 138 struct bio *bp = arg; 139 140 /* 141 * TODO: add more extensive translation of NVMe status codes 142 * to different bio error codes (i.e. EIO, EINVAL, etc.) 143 */ 144 if (nvme_completion_is_error(cpl)) { 145 bp->bio_error = EIO; 146 bp->bio_flags |= BIO_ERROR; 147 bp->bio_resid = bp->bio_bcount; 148 } else 149 bp->bio_resid = 0; 150 151 biodone(bp); 152 } 153 154 static void 155 nvme_ns_strategy(struct bio *bp) 156 { 157 struct nvme_namespace *ns; 158 int err; 159 160 ns = bp->bio_dev->si_drv1; 161 err = nvme_ns_bio_process(ns, bp, nvme_ns_strategy_done); 162 163 if (err) { 164 bp->bio_error = err; 165 bp->bio_flags |= BIO_ERROR; 166 bp->bio_resid = bp->bio_bcount; 167 biodone(bp); 168 } 169 170 } 171 172 static struct cdevsw nvme_ns_cdevsw = { 173 .d_version = D_VERSION, 174 .d_flags = D_DISK, 175 .d_read = physread, 176 .d_write = physwrite, 177 .d_open = nvme_ns_open, 178 .d_close = nvme_ns_close, 179 .d_strategy = nvme_ns_strategy, 180 .d_ioctl = nvme_ns_ioctl 181 }; 182 183 uint32_t 184 nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns) 185 { 186 return ns->ctrlr->max_xfer_size; 187 } 188 189 uint32_t 190 nvme_ns_get_sector_size(struct nvme_namespace *ns) 191 { 192 uint8_t flbas_fmt, lbads; 193 194 flbas_fmt = NVMEV(NVME_NS_DATA_FLBAS_FORMAT, ns->data.flbas); 195 lbads = NVMEV(NVME_NS_DATA_LBAF_LBADS, ns->data.lbaf[flbas_fmt]); 196 197 return (1 << lbads); 198 } 199 200 uint64_t 201 nvme_ns_get_num_sectors(struct nvme_namespace *ns) 202 { 203 return (ns->data.nsze); 204 } 205 206 uint64_t 207 nvme_ns_get_size(struct nvme_namespace *ns) 208 { 209 return (nvme_ns_get_num_sectors(ns) * nvme_ns_get_sector_size(ns)); 210 } 211 212 uint32_t 213 nvme_ns_get_flags(struct nvme_namespace *ns) 214 { 215 return (ns->flags); 216 } 217 218 const char * 219 nvme_ns_get_serial_number(struct nvme_namespace *ns) 220 { 221 return ((const char *)ns->ctrlr->cdata.sn); 222 } 223 224 const char * 225 nvme_ns_get_model_number(struct nvme_namespace *ns) 226 { 227 return ((const char *)ns->ctrlr->cdata.mn); 228 } 229 230 const struct nvme_namespace_data * 231 nvme_ns_get_data(struct nvme_namespace *ns) 232 { 233 return (&ns->data); 234 } 235 236 uint32_t 237 nvme_ns_get_stripesize(struct nvme_namespace *ns) 238 { 239 uint32_t ss; 240 241 if (NVMEV(NVME_NS_DATA_NSFEAT_NPVALID, ns->data.nsfeat) != 0) { 242 ss = nvme_ns_get_sector_size(ns); 243 if (ns->data.npwa != 0) 244 return ((ns->data.npwa + 1) * ss); 245 else if (ns->data.npwg != 0) 246 return ((ns->data.npwg + 1) * ss); 247 } 248 return (ns->boundary); 249 } 250 251 static void 252 nvme_ns_bio_done(void *arg, const struct nvme_completion *status) 253 { 254 struct bio *bp = arg; 255 nvme_cb_fn_t bp_cb_fn; 256 257 bp_cb_fn = bp->bio_driver1; 258 259 if (bp->bio_driver2) 260 free(bp->bio_driver2, M_NVME); 261 262 if (nvme_completion_is_error(status)) { 263 bp->bio_flags |= BIO_ERROR; 264 if (bp->bio_error == 0) 265 bp->bio_error = EIO; 266 } 267 268 if ((bp->bio_flags & BIO_ERROR) == 0) 269 bp->bio_resid = 0; 270 else 271 bp->bio_resid = bp->bio_bcount; 272 273 bp_cb_fn(bp, status); 274 } 275 276 static void 277 nvme_bio_child_inbed(struct bio *parent, int bio_error) 278 { 279 struct nvme_completion parent_cpl; 280 int children, inbed; 281 282 if (bio_error != 0) { 283 parent->bio_flags |= BIO_ERROR; 284 parent->bio_error = bio_error; 285 } 286 287 /* 288 * atomic_fetchadd will return value before adding 1, so we still 289 * must add 1 to get the updated inbed number. Save bio_children 290 * before incrementing to guard against race conditions when 291 * two children bios complete on different queues. 292 */ 293 children = atomic_load_acq_int(&parent->bio_children); 294 inbed = atomic_fetchadd_int(&parent->bio_inbed, 1) + 1; 295 if (inbed == children) { 296 bzero(&parent_cpl, sizeof(parent_cpl)); 297 if (parent->bio_flags & BIO_ERROR) { 298 parent_cpl.status &= ~NVMEM(NVME_STATUS_SC); 299 parent_cpl.status |= NVMEF(NVME_STATUS_SC, 300 NVME_SC_DATA_TRANSFER_ERROR); 301 } 302 nvme_ns_bio_done(parent, &parent_cpl); 303 } 304 } 305 306 static void 307 nvme_bio_child_done(void *arg, const struct nvme_completion *cpl) 308 { 309 struct bio *child = arg; 310 struct bio *parent; 311 int bio_error; 312 313 parent = child->bio_parent; 314 g_destroy_bio(child); 315 bio_error = nvme_completion_is_error(cpl) ? EIO : 0; 316 nvme_bio_child_inbed(parent, bio_error); 317 } 318 319 static uint32_t 320 nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t align) 321 { 322 uint32_t num_segs, offset, remainder; 323 324 if (align == 0) 325 return (1); 326 327 KASSERT((align & (align - 1)) == 0, ("alignment not power of 2\n")); 328 329 num_segs = size / align; 330 remainder = size & (align - 1); 331 offset = addr & (align - 1); 332 if (remainder > 0 || offset > 0) 333 num_segs += 1 + (remainder + offset - 1) / align; 334 return (num_segs); 335 } 336 337 static void 338 nvme_free_child_bios(int num_bios, struct bio **child_bios) 339 { 340 int i; 341 342 for (i = 0; i < num_bios; i++) { 343 if (child_bios[i] != NULL) 344 g_destroy_bio(child_bios[i]); 345 } 346 347 free(child_bios, M_NVME); 348 } 349 350 static struct bio ** 351 nvme_allocate_child_bios(int num_bios) 352 { 353 struct bio **child_bios; 354 int err = 0, i; 355 356 child_bios = malloc(num_bios * sizeof(struct bio *), M_NVME, M_NOWAIT); 357 if (child_bios == NULL) 358 return (NULL); 359 360 for (i = 0; i < num_bios; i++) { 361 child_bios[i] = g_new_bio(); 362 if (child_bios[i] == NULL) 363 err = ENOMEM; 364 } 365 366 if (err == ENOMEM) { 367 nvme_free_child_bios(num_bios, child_bios); 368 return (NULL); 369 } 370 371 return (child_bios); 372 } 373 374 static struct bio ** 375 nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios) 376 { 377 struct bio **child_bios; 378 struct bio *child; 379 uint64_t cur_offset; 380 caddr_t data; 381 uint32_t rem_bcount; 382 int i; 383 struct vm_page **ma; 384 uint32_t ma_offset; 385 386 *num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount, 387 alignment); 388 child_bios = nvme_allocate_child_bios(*num_bios); 389 if (child_bios == NULL) 390 return (NULL); 391 392 bp->bio_children = *num_bios; 393 bp->bio_inbed = 0; 394 cur_offset = bp->bio_offset; 395 rem_bcount = bp->bio_bcount; 396 data = bp->bio_data; 397 ma_offset = bp->bio_ma_offset; 398 ma = bp->bio_ma; 399 400 for (i = 0; i < *num_bios; i++) { 401 child = child_bios[i]; 402 child->bio_parent = bp; 403 child->bio_cmd = bp->bio_cmd; 404 child->bio_offset = cur_offset; 405 child->bio_bcount = min(rem_bcount, 406 alignment - (cur_offset & (alignment - 1))); 407 child->bio_flags = bp->bio_flags; 408 if (bp->bio_flags & BIO_UNMAPPED) { 409 child->bio_ma_offset = ma_offset; 410 child->bio_ma = ma; 411 child->bio_ma_n = 412 nvme_get_num_segments(child->bio_ma_offset, 413 child->bio_bcount, PAGE_SIZE); 414 ma_offset = (ma_offset + child->bio_bcount) & 415 PAGE_MASK; 416 ma += child->bio_ma_n; 417 if (ma_offset != 0) 418 ma -= 1; 419 } else { 420 child->bio_data = data; 421 data += child->bio_bcount; 422 } 423 cur_offset += child->bio_bcount; 424 rem_bcount -= child->bio_bcount; 425 } 426 427 return (child_bios); 428 } 429 430 static int 431 nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp, 432 uint32_t alignment) 433 { 434 struct bio *child; 435 struct bio **child_bios; 436 int err, i, num_bios; 437 438 child_bios = nvme_construct_child_bios(bp, alignment, &num_bios); 439 if (child_bios == NULL) 440 return (ENOMEM); 441 442 counter_u64_add(ns->ctrlr->alignment_splits, 1); 443 for (i = 0; i < num_bios; i++) { 444 child = child_bios[i]; 445 err = nvme_ns_bio_process(ns, child, nvme_bio_child_done); 446 if (err != 0) { 447 nvme_bio_child_inbed(bp, err); 448 g_destroy_bio(child); 449 } 450 } 451 452 free(child_bios, M_NVME); 453 return (0); 454 } 455 456 int 457 nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp, 458 nvme_cb_fn_t cb_fn) 459 { 460 struct nvme_dsm_range *dsm_range; 461 uint32_t num_bios; 462 int err; 463 464 bp->bio_driver1 = cb_fn; 465 466 if (ns->boundary > 0 && 467 (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) { 468 num_bios = nvme_get_num_segments(bp->bio_offset, 469 bp->bio_bcount, ns->boundary); 470 if (num_bios > 1) 471 return (nvme_ns_split_bio(ns, bp, ns->boundary)); 472 } 473 474 switch (bp->bio_cmd) { 475 case BIO_READ: 476 err = nvme_ns_cmd_read_bio(ns, bp, nvme_ns_bio_done, bp); 477 break; 478 case BIO_WRITE: 479 err = nvme_ns_cmd_write_bio(ns, bp, nvme_ns_bio_done, bp); 480 break; 481 case BIO_FLUSH: 482 err = nvme_ns_cmd_flush(ns, nvme_ns_bio_done, bp); 483 break; 484 case BIO_DELETE: 485 dsm_range = 486 malloc(sizeof(struct nvme_dsm_range), M_NVME, 487 M_ZERO | M_NOWAIT); 488 if (!dsm_range) { 489 err = ENOMEM; 490 break; 491 } 492 dsm_range->length = 493 htole32(bp->bio_bcount/nvme_ns_get_sector_size(ns)); 494 dsm_range->starting_lba = 495 htole64(bp->bio_offset/nvme_ns_get_sector_size(ns)); 496 bp->bio_driver2 = dsm_range; 497 err = nvme_ns_cmd_deallocate(ns, dsm_range, 1, 498 nvme_ns_bio_done, bp); 499 if (err != 0) 500 free(dsm_range, M_NVME); 501 break; 502 default: 503 err = EOPNOTSUPP; 504 break; 505 } 506 507 return (err); 508 } 509 510 int 511 nvme_ns_ioctl_process(struct nvme_namespace *ns, u_long cmd, caddr_t arg, 512 int flag, struct thread *td) 513 { 514 return (nvme_ns_ioctl(ns->cdev, cmd, arg, flag, td)); 515 } 516 517 int 518 nvme_ns_construct(struct nvme_namespace *ns, uint32_t id, 519 struct nvme_controller *ctrlr) 520 { 521 struct make_dev_args md_args; 522 struct nvme_completion_poll_status status; 523 int res; 524 int unit; 525 uint8_t flbas_fmt; 526 uint8_t vwc_present; 527 528 ns->ctrlr = ctrlr; 529 ns->id = id; 530 531 /* 532 * Namespaces are reconstructed after a controller reset, so check 533 * to make sure we only call mtx_init once on each mtx. 534 * 535 * TODO: Move this somewhere where it gets called at controller 536 * construction time, which is not invoked as part of each 537 * controller reset. 538 */ 539 if (!mtx_initialized(&ns->lock)) 540 mtx_init(&ns->lock, "nvme ns lock", NULL, MTX_DEF); 541 542 status.done = 0; 543 nvme_ctrlr_cmd_identify_namespace(ctrlr, id, &ns->data, 544 nvme_completion_poll_cb, &status); 545 nvme_completion_poll(&status); 546 if (nvme_completion_is_error(&status.cpl)) { 547 nvme_printf(ctrlr, "nvme_identify_namespace failed\n"); 548 return (ENXIO); 549 } 550 551 /* Convert data to host endian */ 552 nvme_namespace_data_swapbytes(&ns->data); 553 554 /* 555 * If the size of is zero, chances are this isn't a valid 556 * namespace (eg one that's not been configured yet). The 557 * standard says the entire id will be zeros, so this is a 558 * cheap way to test for that. 559 */ 560 if (ns->data.nsze == 0) 561 return (ENXIO); 562 563 flbas_fmt = NVMEV(NVME_NS_DATA_FLBAS_FORMAT, ns->data.flbas); 564 565 /* 566 * Note: format is a 0-based value, so > is appropriate here, 567 * not >=. 568 */ 569 if (flbas_fmt > ns->data.nlbaf) { 570 nvme_printf(ctrlr, 571 "lba format %d exceeds number supported (%d)\n", 572 flbas_fmt, ns->data.nlbaf + 1); 573 return (ENXIO); 574 } 575 576 /* 577 * Older Intel devices (like the PC35xxx and P45xx series) advertise in 578 * vendor specific space an alignment that improves performance. If 579 * present use for the stripe size. NVMe 1.3 standardized this as 580 * NOIOB, and newer Intel drives use that. 581 */ 582 if ((ctrlr->quirks & QUIRK_INTEL_ALIGNMENT) != 0) { 583 if (ctrlr->cdata.vs[3] != 0) 584 ns->boundary = 585 1 << (ctrlr->cdata.vs[3] + NVME_MPS_SHIFT + 586 NVME_CAP_HI_MPSMIN(ctrlr->cap_hi)); 587 else 588 ns->boundary = 0; 589 } else { 590 ns->boundary = ns->data.noiob * nvme_ns_get_sector_size(ns); 591 } 592 593 if (nvme_ctrlr_has_dataset_mgmt(&ctrlr->cdata)) 594 ns->flags |= NVME_NS_DEALLOCATE_SUPPORTED; 595 596 vwc_present = NVMEV(NVME_CTRLR_DATA_VWC_PRESENT, ctrlr->cdata.vwc); 597 if (vwc_present) 598 ns->flags |= NVME_NS_FLUSH_SUPPORTED; 599 600 /* 601 * cdev may have already been created, if we are reconstructing the 602 * namespace after a controller-level reset. 603 */ 604 if (ns->cdev != NULL) 605 return (0); 606 607 /* 608 * Namespace IDs start at 1, so we need to subtract 1 to create a 609 * correct unit number. 610 */ 611 unit = device_get_unit(ctrlr->dev) * NVME_MAX_NAMESPACES + ns->id - 1; 612 613 make_dev_args_init(&md_args); 614 md_args.mda_devsw = &nvme_ns_cdevsw; 615 md_args.mda_unit = unit; 616 md_args.mda_mode = 0600; 617 md_args.mda_si_drv1 = ns; 618 res = make_dev_s(&md_args, &ns->cdev, "%sn%d", 619 device_get_nameunit(ctrlr->dev), ns->id); 620 if (res != 0) 621 return (ENXIO); 622 ns->cdev->si_drv2 = make_dev_alias(ns->cdev, "%sns%d", 623 device_get_nameunit(ctrlr->dev), ns->id); 624 ns->cdev->si_flags |= SI_UNMAPPED; 625 626 return (0); 627 } 628 629 void 630 nvme_ns_destruct(struct nvme_namespace *ns) 631 { 632 if (ns->cdev != NULL) { 633 if (ns->cdev->si_drv2 != NULL) 634 destroy_dev(ns->cdev->si_drv2); 635 destroy_dev(ns->cdev); 636 } 637 } 638