1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright 2012 Garrett D'Amore <garrett@damore.org>. All rights reserved. 24 * Copyright 2012 Alexey Zaytsev <alexey.zaytsev@gmail.com> All rights reserved. 25 * Copyright 2017 The MathWorks, Inc. All rights reserved. 26 * Copyright 2020 Joyent, Inc. 27 * Copyright 2022 OmniOS Community Edition (OmniOSce) Association. 28 * Copyright 2022 Tintri by DDN, Inc. All rights reserved. 29 * Copyright 2023 Oxide Computer Company 30 */ 31 32 #include <sys/types.h> 33 #include <sys/ksynch.h> 34 #include <sys/kmem.h> 35 #include <sys/file.h> 36 #include <sys/errno.h> 37 #include <sys/open.h> 38 #include <sys/buf.h> 39 #include <sys/uio.h> 40 #include <sys/aio_req.h> 41 #include <sys/cred.h> 42 #include <sys/modctl.h> 43 #include <sys/cmlb.h> 44 #include <sys/conf.h> 45 #include <sys/devops.h> 46 #include <sys/list.h> 47 #include <sys/sysmacros.h> 48 #include <sys/dkio.h> 49 #include <sys/dkioc_free_util.h> 50 #include <sys/vtoc.h> 51 #include <sys/scsi/scsi.h> /* for DTYPE_DIRECT */ 52 #include <sys/kstat.h> 53 #include <sys/fs/dv_node.h> 54 #include <sys/ddi.h> 55 #include <sys/sunddi.h> 56 #include <sys/note.h> 57 #include <sys/blkdev.h> 58 #include <sys/scsi/impl/inquiry.h> 59 #include <sys/taskq.h> 60 #include <sys/taskq_impl.h> 61 #include <sys/disp.h> 62 #include <sys/sysevent/eventdefs.h> 63 #include <sys/sysevent/dev.h> 64 65 /* 66 * blkdev is a driver which provides a lot of the common functionality 67 * a block device driver may need and helps by removing code which 68 * is frequently duplicated in block device drivers. 69 * 70 * Within this driver all the struct cb_ops functions required for a 71 * block device driver are written with appropriate call back functions 72 * to be provided by the parent driver. 73 * 74 * To use blkdev, a driver needs to: 75 * 1. Create a bd_ops_t structure which has the call back operations 76 * blkdev will use. 77 * 2. Create a handle by calling bd_alloc_handle(). One of the 78 * arguments to this function is the bd_ops_t. 79 * 3. Call bd_attach_handle(). This will instantiate a blkdev device 80 * as a child device node of the calling driver. 81 * 82 * A parent driver is not restricted to just allocating and attaching a 83 * single instance, it may attach as many as it wishes. For each handle 84 * attached, appropriate entries in /dev/[r]dsk are created. 85 * 86 * The bd_ops_t routines that a parent of blkdev need to provide are: 87 * 88 * o_drive_info: Provide information to blkdev such as how many I/O queues 89 * to create and the size of those queues. Also some device 90 * specifics such as EUI, vendor, product, model, serial 91 * number .... 92 * 93 * o_media_info: Provide information about the media. Eg size and block size. 94 * 95 * o_devid_init: Creates and initializes the device id. Typically calls 96 * ddi_devid_init(). 97 * 98 * o_sync_cache: Issues a device appropriate command to flush any write 99 * caches. 100 * 101 * o_read: Read data as described by bd_xfer_t argument. 102 * 103 * o_write: Write data as described by bd_xfer_t argument. 104 * 105 * o_free_space: Free the space described by bd_xfer_t argument (optional). 106 * 107 * Queues 108 * ------ 109 * Part of the drive_info data is a queue count. blkdev will create 110 * "queue count" number of waitq/runq pairs. Each waitq/runq pair 111 * operates independently. As an I/O is scheduled up to the parent 112 * driver via o_read or o_write its queue number is given. If the 113 * parent driver supports multiple hardware queues it can then select 114 * where to submit the I/O request. 115 * 116 * Currently blkdev uses a simplistic round-robin queue selection method. 117 * It has the advantage that it is lockless. In the future it will be 118 * worthwhile reviewing this strategy for something which prioritizes queues 119 * depending on how busy they are. 120 * 121 * Each waitq/runq pair is protected by its mutex (q_iomutex). Incoming 122 * I/O requests are initially added to the waitq. They are taken off the 123 * waitq, added to the runq and submitted, providing the runq is less 124 * than the qsize as specified in the drive_info. As an I/O request 125 * completes, the parent driver is required to call bd_xfer_done(), which 126 * will remove the I/O request from the runq and pass I/O completion 127 * status up the stack. 128 * 129 * Locks 130 * ----- 131 * There are 5 instance global locks d_ocmutex, d_ksmutex, d_errmutex, 132 * d_statemutex and d_dle_mutex. As well a q_iomutex per waitq/runq pair. 133 * 134 * Lock Hierarchy 135 * -------------- 136 * The only two locks which may be held simultaneously are q_iomutex and 137 * d_ksmutex. In all cases q_iomutex must be acquired before d_ksmutex. 138 */ 139 140 #define BD_MAXPART 64 141 #define BDINST(dev) (getminor(dev) / BD_MAXPART) 142 #define BDPART(dev) (getminor(dev) % BD_MAXPART) 143 144 typedef struct bd bd_t; 145 typedef struct bd_xfer_impl bd_xfer_impl_t; 146 typedef struct bd_queue bd_queue_t; 147 148 typedef enum { 149 BD_DLE_PENDING = 1 << 0, 150 BD_DLE_RUNNING = 1 << 1 151 } bd_dle_state_t; 152 153 struct bd { 154 void *d_private; 155 dev_info_t *d_dip; 156 kmutex_t d_ocmutex; /* open/close */ 157 kmutex_t d_ksmutex; /* kstat */ 158 kmutex_t d_errmutex; 159 kmutex_t d_statemutex; 160 kcondvar_t d_statecv; 161 enum dkio_state d_state; 162 cmlb_handle_t d_cmlbh; 163 unsigned d_open_lyr[BD_MAXPART]; /* open count */ 164 uint64_t d_open_excl; /* bit mask indexed by partition */ 165 uint64_t d_open_reg[OTYPCNT]; /* bit mask */ 166 uint64_t d_io_counter; 167 168 uint32_t d_qcount; 169 uint32_t d_qactive; 170 uint32_t d_maxxfer; 171 uint32_t d_blkshift; 172 uint32_t d_pblkshift; 173 uint64_t d_numblks; 174 ddi_devid_t d_devid; 175 176 uint64_t d_max_free_seg; 177 uint64_t d_max_free_blks; 178 uint64_t d_max_free_seg_blks; 179 uint64_t d_free_align; 180 181 kmem_cache_t *d_cache; 182 bd_queue_t *d_queues; 183 kstat_t *d_ksp; 184 kstat_io_t *d_kiop; 185 kstat_t *d_errstats; 186 struct bd_errstats *d_kerr; 187 188 boolean_t d_rdonly; 189 boolean_t d_ssd; 190 boolean_t d_removable; 191 boolean_t d_hotpluggable; 192 boolean_t d_use_dma; 193 194 ddi_dma_attr_t d_dma; 195 bd_ops_t d_ops; 196 bd_handle_t d_handle; 197 198 kmutex_t d_dle_mutex; 199 taskq_ent_t d_dle_ent; 200 bd_dle_state_t d_dle_state; 201 }; 202 203 struct bd_handle { 204 bd_ops_t h_ops; 205 ddi_dma_attr_t *h_dma; 206 dev_info_t *h_parent; 207 dev_info_t *h_child; 208 void *h_private; 209 bd_t *h_bd; 210 char *h_name; 211 char h_addr[50]; /* enough for w%0.32x,%X */ 212 }; 213 214 struct bd_xfer_impl { 215 bd_xfer_t i_public; 216 list_node_t i_linkage; 217 bd_t *i_bd; 218 buf_t *i_bp; 219 bd_queue_t *i_bq; 220 uint_t i_num_win; 221 uint_t i_cur_win; 222 off_t i_offset; 223 int (*i_func)(void *, bd_xfer_t *); 224 uint32_t i_blkshift; 225 size_t i_len; 226 size_t i_resid; 227 }; 228 229 struct bd_queue { 230 kmutex_t q_iomutex; 231 uint32_t q_qsize; 232 uint32_t q_qactive; 233 list_t q_runq; 234 list_t q_waitq; 235 }; 236 237 #define i_dmah i_public.x_dmah 238 #define i_dmac i_public.x_dmac 239 #define i_ndmac i_public.x_ndmac 240 #define i_kaddr i_public.x_kaddr 241 #define i_nblks i_public.x_nblks 242 #define i_blkno i_public.x_blkno 243 #define i_flags i_public.x_flags 244 #define i_qnum i_public.x_qnum 245 #define i_dfl i_public.x_dfl 246 247 #define CAN_FREESPACE(bd) \ 248 (((bd)->d_ops.o_free_space == NULL) ? B_FALSE : B_TRUE) 249 250 /* 251 * Private prototypes. 252 */ 253 254 static void bd_prop_update_inqstring(dev_info_t *, char *, char *, size_t); 255 static void bd_create_inquiry_props(dev_info_t *, bd_drive_t *); 256 static void bd_create_errstats(bd_t *, int, bd_drive_t *); 257 static void bd_destroy_errstats(bd_t *); 258 static void bd_errstats_setstr(kstat_named_t *, char *, size_t, char *); 259 static void bd_init_errstats(bd_t *, bd_drive_t *); 260 static void bd_fini_errstats(bd_t *); 261 262 static int bd_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **); 263 static int bd_attach(dev_info_t *, ddi_attach_cmd_t); 264 static int bd_detach(dev_info_t *, ddi_detach_cmd_t); 265 266 static int bd_open(dev_t *, int, int, cred_t *); 267 static int bd_close(dev_t, int, int, cred_t *); 268 static int bd_strategy(struct buf *); 269 static int bd_ioctl(dev_t, int, intptr_t, int, cred_t *, int *); 270 static int bd_dump(dev_t, caddr_t, daddr_t, int); 271 static int bd_read(dev_t, struct uio *, cred_t *); 272 static int bd_write(dev_t, struct uio *, cred_t *); 273 static int bd_aread(dev_t, struct aio_req *, cred_t *); 274 static int bd_awrite(dev_t, struct aio_req *, cred_t *); 275 static int bd_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int, char *, 276 caddr_t, int *); 277 278 static int bd_tg_rdwr(dev_info_t *, uchar_t, void *, diskaddr_t, size_t, 279 void *); 280 static int bd_tg_getinfo(dev_info_t *, int, void *, void *); 281 static int bd_xfer_ctor(void *, void *, int); 282 static void bd_xfer_dtor(void *, void *); 283 static void bd_sched(bd_t *, bd_queue_t *); 284 static void bd_submit(bd_t *, bd_xfer_impl_t *); 285 static void bd_runq_exit(bd_xfer_impl_t *, int); 286 static void bd_update_state(bd_t *); 287 static int bd_check_state(bd_t *, enum dkio_state *); 288 static int bd_flush_write_cache(bd_t *, struct dk_callback *); 289 static int bd_check_uio(dev_t, struct uio *); 290 static int bd_free_space(dev_t, bd_t *, dkioc_free_list_t *); 291 292 struct cmlb_tg_ops bd_tg_ops = { 293 TG_DK_OPS_VERSION_1, 294 bd_tg_rdwr, 295 bd_tg_getinfo, 296 }; 297 298 static struct cb_ops bd_cb_ops = { 299 bd_open, /* open */ 300 bd_close, /* close */ 301 bd_strategy, /* strategy */ 302 nodev, /* print */ 303 bd_dump, /* dump */ 304 bd_read, /* read */ 305 bd_write, /* write */ 306 bd_ioctl, /* ioctl */ 307 nodev, /* devmap */ 308 nodev, /* mmap */ 309 nodev, /* segmap */ 310 nochpoll, /* poll */ 311 bd_prop_op, /* cb_prop_op */ 312 0, /* streamtab */ 313 D_64BIT | D_MP, /* Driver comaptibility flag */ 314 CB_REV, /* cb_rev */ 315 bd_aread, /* async read */ 316 bd_awrite /* async write */ 317 }; 318 319 struct dev_ops bd_dev_ops = { 320 DEVO_REV, /* devo_rev, */ 321 0, /* refcnt */ 322 bd_getinfo, /* getinfo */ 323 nulldev, /* identify */ 324 nulldev, /* probe */ 325 bd_attach, /* attach */ 326 bd_detach, /* detach */ 327 nodev, /* reset */ 328 &bd_cb_ops, /* driver operations */ 329 NULL, /* bus operations */ 330 NULL, /* power */ 331 ddi_quiesce_not_needed, /* quiesce */ 332 }; 333 334 static struct modldrv modldrv = { 335 &mod_driverops, 336 "Generic Block Device", 337 &bd_dev_ops, 338 }; 339 340 static struct modlinkage modlinkage = { 341 MODREV_1, { &modldrv, NULL } 342 }; 343 344 static void *bd_state; 345 static krwlock_t bd_lock; 346 static taskq_t *bd_taskq; 347 348 int 349 _init(void) 350 { 351 char taskq_name[TASKQ_NAMELEN]; 352 const char *name; 353 int rv; 354 355 rv = ddi_soft_state_init(&bd_state, sizeof (struct bd), 2); 356 if (rv != DDI_SUCCESS) 357 return (rv); 358 359 name = mod_modname(&modlinkage); 360 (void) snprintf(taskq_name, sizeof (taskq_name), "%s_taskq", name); 361 bd_taskq = taskq_create(taskq_name, 1, minclsyspri, 0, 0, 0); 362 if (bd_taskq == NULL) { 363 cmn_err(CE_WARN, "%s: unable to create %s", name, taskq_name); 364 ddi_soft_state_fini(&bd_state); 365 return (DDI_FAILURE); 366 } 367 368 rw_init(&bd_lock, NULL, RW_DRIVER, NULL); 369 370 rv = mod_install(&modlinkage); 371 if (rv != DDI_SUCCESS) { 372 rw_destroy(&bd_lock); 373 taskq_destroy(bd_taskq); 374 ddi_soft_state_fini(&bd_state); 375 } 376 return (rv); 377 } 378 379 int 380 _fini(void) 381 { 382 int rv; 383 384 rv = mod_remove(&modlinkage); 385 if (rv == DDI_SUCCESS) { 386 rw_destroy(&bd_lock); 387 taskq_destroy(bd_taskq); 388 ddi_soft_state_fini(&bd_state); 389 } 390 return (rv); 391 } 392 393 int 394 _info(struct modinfo *modinfop) 395 { 396 return (mod_info(&modlinkage, modinfop)); 397 } 398 399 static int 400 bd_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **resultp) 401 { 402 bd_t *bd; 403 minor_t inst; 404 405 _NOTE(ARGUNUSED(dip)); 406 407 inst = BDINST((dev_t)arg); 408 409 switch (cmd) { 410 case DDI_INFO_DEVT2DEVINFO: 411 bd = ddi_get_soft_state(bd_state, inst); 412 if (bd == NULL) { 413 return (DDI_FAILURE); 414 } 415 *resultp = (void *)bd->d_dip; 416 break; 417 418 case DDI_INFO_DEVT2INSTANCE: 419 *resultp = (void *)(intptr_t)inst; 420 break; 421 422 default: 423 return (DDI_FAILURE); 424 } 425 return (DDI_SUCCESS); 426 } 427 428 static void 429 bd_prop_update_inqstring(dev_info_t *dip, char *name, char *data, size_t len) 430 { 431 int ilen; 432 char *data_string; 433 434 ilen = scsi_ascii_inquiry_len(data, len); 435 ASSERT3U(ilen, <=, len); 436 if (ilen <= 0) 437 return; 438 /* ensure null termination */ 439 data_string = kmem_zalloc(ilen + 1, KM_SLEEP); 440 bcopy(data, data_string, ilen); 441 (void) ndi_prop_update_string(DDI_DEV_T_NONE, dip, name, data_string); 442 kmem_free(data_string, ilen + 1); 443 } 444 445 static void 446 bd_create_inquiry_props(dev_info_t *dip, bd_drive_t *drive) 447 { 448 if (drive->d_vendor_len > 0) 449 bd_prop_update_inqstring(dip, INQUIRY_VENDOR_ID, 450 drive->d_vendor, drive->d_vendor_len); 451 452 if (drive->d_product_len > 0) 453 bd_prop_update_inqstring(dip, INQUIRY_PRODUCT_ID, 454 drive->d_product, drive->d_product_len); 455 456 if (drive->d_serial_len > 0) 457 bd_prop_update_inqstring(dip, INQUIRY_SERIAL_NO, 458 drive->d_serial, drive->d_serial_len); 459 460 if (drive->d_revision_len > 0) 461 bd_prop_update_inqstring(dip, INQUIRY_REVISION_ID, 462 drive->d_revision, drive->d_revision_len); 463 } 464 465 static void 466 bd_create_errstats(bd_t *bd, int inst, bd_drive_t *drive) 467 { 468 char ks_module[KSTAT_STRLEN]; 469 char ks_name[KSTAT_STRLEN]; 470 int ndata = sizeof (struct bd_errstats) / sizeof (kstat_named_t); 471 472 if (bd->d_errstats != NULL) 473 return; 474 475 (void) snprintf(ks_module, sizeof (ks_module), "%serr", 476 ddi_driver_name(bd->d_dip)); 477 (void) snprintf(ks_name, sizeof (ks_name), "%s%d,err", 478 ddi_driver_name(bd->d_dip), inst); 479 480 bd->d_errstats = kstat_create(ks_module, inst, ks_name, "device_error", 481 KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT); 482 483 mutex_init(&bd->d_errmutex, NULL, MUTEX_DRIVER, NULL); 484 if (bd->d_errstats == NULL) { 485 /* 486 * Even if we cannot create the kstat, we create a 487 * scratch kstat. The reason for this is to ensure 488 * that we can update the kstat all of the time, 489 * without adding an extra branch instruction. 490 */ 491 bd->d_kerr = kmem_zalloc(sizeof (struct bd_errstats), 492 KM_SLEEP); 493 } else { 494 bd->d_errstats->ks_lock = &bd->d_errmutex; 495 bd->d_kerr = (struct bd_errstats *)bd->d_errstats->ks_data; 496 } 497 498 kstat_named_init(&bd->d_kerr->bd_softerrs, "Soft Errors", 499 KSTAT_DATA_UINT32); 500 kstat_named_init(&bd->d_kerr->bd_harderrs, "Hard Errors", 501 KSTAT_DATA_UINT32); 502 kstat_named_init(&bd->d_kerr->bd_transerrs, "Transport Errors", 503 KSTAT_DATA_UINT32); 504 505 if (drive->d_model_len > 0) { 506 kstat_named_init(&bd->d_kerr->bd_model, "Model", 507 KSTAT_DATA_STRING); 508 } else { 509 kstat_named_init(&bd->d_kerr->bd_vid, "Vendor", 510 KSTAT_DATA_STRING); 511 kstat_named_init(&bd->d_kerr->bd_pid, "Product", 512 KSTAT_DATA_STRING); 513 } 514 515 kstat_named_init(&bd->d_kerr->bd_revision, "Revision", 516 KSTAT_DATA_STRING); 517 kstat_named_init(&bd->d_kerr->bd_serial, "Serial No", 518 KSTAT_DATA_STRING); 519 kstat_named_init(&bd->d_kerr->bd_capacity, "Size", 520 KSTAT_DATA_ULONGLONG); 521 kstat_named_init(&bd->d_kerr->bd_rq_media_err, "Media Error", 522 KSTAT_DATA_UINT32); 523 kstat_named_init(&bd->d_kerr->bd_rq_ntrdy_err, "Device Not Ready", 524 KSTAT_DATA_UINT32); 525 kstat_named_init(&bd->d_kerr->bd_rq_nodev_err, "No Device", 526 KSTAT_DATA_UINT32); 527 kstat_named_init(&bd->d_kerr->bd_rq_recov_err, "Recoverable", 528 KSTAT_DATA_UINT32); 529 kstat_named_init(&bd->d_kerr->bd_rq_illrq_err, "Illegal Request", 530 KSTAT_DATA_UINT32); 531 kstat_named_init(&bd->d_kerr->bd_rq_pfa_err, 532 "Predictive Failure Analysis", KSTAT_DATA_UINT32); 533 534 bd->d_errstats->ks_private = bd; 535 536 kstat_install(bd->d_errstats); 537 bd_init_errstats(bd, drive); 538 } 539 540 static void 541 bd_destroy_errstats(bd_t *bd) 542 { 543 if (bd->d_errstats != NULL) { 544 bd_fini_errstats(bd); 545 kstat_delete(bd->d_errstats); 546 bd->d_errstats = NULL; 547 } else { 548 kmem_free(bd->d_kerr, sizeof (struct bd_errstats)); 549 bd->d_kerr = NULL; 550 mutex_destroy(&bd->d_errmutex); 551 } 552 } 553 554 static void 555 bd_errstats_setstr(kstat_named_t *k, char *str, size_t len, char *alt) 556 { 557 char *tmp; 558 size_t km_len; 559 560 if (KSTAT_NAMED_STR_PTR(k) == NULL) { 561 if (len > 0) 562 km_len = strnlen(str, len); 563 else if (alt != NULL) 564 km_len = strlen(alt); 565 else 566 return; 567 568 tmp = kmem_alloc(km_len + 1, KM_SLEEP); 569 bcopy(len > 0 ? str : alt, tmp, km_len); 570 tmp[km_len] = '\0'; 571 572 kstat_named_setstr(k, tmp); 573 } 574 } 575 576 static void 577 bd_errstats_clrstr(kstat_named_t *k) 578 { 579 if (KSTAT_NAMED_STR_PTR(k) == NULL) 580 return; 581 582 kmem_free(KSTAT_NAMED_STR_PTR(k), KSTAT_NAMED_STR_BUFLEN(k)); 583 kstat_named_setstr(k, NULL); 584 } 585 586 static void 587 bd_init_errstats(bd_t *bd, bd_drive_t *drive) 588 { 589 struct bd_errstats *est = bd->d_kerr; 590 591 mutex_enter(&bd->d_errmutex); 592 593 if (drive->d_model_len > 0 && 594 KSTAT_NAMED_STR_PTR(&est->bd_model) == NULL) { 595 bd_errstats_setstr(&est->bd_model, drive->d_model, 596 drive->d_model_len, NULL); 597 } else { 598 bd_errstats_setstr(&est->bd_vid, drive->d_vendor, 599 drive->d_vendor_len, "Unknown "); 600 bd_errstats_setstr(&est->bd_pid, drive->d_product, 601 drive->d_product_len, "Unknown "); 602 } 603 604 bd_errstats_setstr(&est->bd_revision, drive->d_revision, 605 drive->d_revision_len, "0001"); 606 bd_errstats_setstr(&est->bd_serial, drive->d_serial, 607 drive->d_serial_len, "0 "); 608 609 mutex_exit(&bd->d_errmutex); 610 } 611 612 static void 613 bd_fini_errstats(bd_t *bd) 614 { 615 struct bd_errstats *est = bd->d_kerr; 616 617 mutex_enter(&bd->d_errmutex); 618 619 bd_errstats_clrstr(&est->bd_model); 620 bd_errstats_clrstr(&est->bd_vid); 621 bd_errstats_clrstr(&est->bd_pid); 622 bd_errstats_clrstr(&est->bd_revision); 623 bd_errstats_clrstr(&est->bd_serial); 624 625 mutex_exit(&bd->d_errmutex); 626 } 627 628 static void 629 bd_queues_free(bd_t *bd) 630 { 631 uint32_t i; 632 633 for (i = 0; i < bd->d_qcount; i++) { 634 bd_queue_t *bq = &bd->d_queues[i]; 635 636 mutex_destroy(&bq->q_iomutex); 637 list_destroy(&bq->q_waitq); 638 list_destroy(&bq->q_runq); 639 } 640 641 kmem_free(bd->d_queues, sizeof (*bd->d_queues) * bd->d_qcount); 642 } 643 644 static int 645 bd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd) 646 { 647 int inst; 648 bd_handle_t hdl; 649 bd_t *bd; 650 bd_drive_t drive; 651 uint32_t i; 652 int rv; 653 char name[16]; 654 char kcache[32]; 655 char *node_type; 656 657 switch (cmd) { 658 case DDI_ATTACH: 659 break; 660 case DDI_RESUME: 661 /* We don't do anything native for suspend/resume */ 662 return (DDI_SUCCESS); 663 default: 664 return (DDI_FAILURE); 665 } 666 667 inst = ddi_get_instance(dip); 668 hdl = ddi_get_parent_data(dip); 669 670 (void) snprintf(name, sizeof (name), "%s%d", 671 ddi_driver_name(dip), ddi_get_instance(dip)); 672 (void) snprintf(kcache, sizeof (kcache), "%s_xfer", name); 673 674 if (hdl == NULL) { 675 cmn_err(CE_WARN, "%s: missing parent data!", name); 676 return (DDI_FAILURE); 677 } 678 679 if (ddi_soft_state_zalloc(bd_state, inst) != DDI_SUCCESS) { 680 cmn_err(CE_WARN, "%s: unable to zalloc soft state!", name); 681 return (DDI_FAILURE); 682 } 683 bd = ddi_get_soft_state(bd_state, inst); 684 685 if (hdl->h_dma) { 686 bd->d_dma = *(hdl->h_dma); 687 bd->d_dma.dma_attr_granular = 688 max(DEV_BSIZE, bd->d_dma.dma_attr_granular); 689 bd->d_use_dma = B_TRUE; 690 691 if (bd->d_maxxfer && 692 (bd->d_maxxfer != bd->d_dma.dma_attr_maxxfer)) { 693 cmn_err(CE_WARN, 694 "%s: inconsistent maximum transfer size!", 695 name); 696 /* We force it */ 697 bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer; 698 } else { 699 bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer; 700 } 701 } else { 702 bd->d_use_dma = B_FALSE; 703 if (bd->d_maxxfer == 0) { 704 bd->d_maxxfer = 1024 * 1024; 705 } 706 } 707 bd->d_ops = hdl->h_ops; 708 bd->d_private = hdl->h_private; 709 bd->d_blkshift = DEV_BSHIFT; /* 512 bytes, to start */ 710 711 if (bd->d_maxxfer % DEV_BSIZE) { 712 cmn_err(CE_WARN, "%s: maximum transfer misaligned!", name); 713 bd->d_maxxfer &= ~(DEV_BSIZE - 1); 714 } 715 if (bd->d_maxxfer < DEV_BSIZE) { 716 cmn_err(CE_WARN, "%s: maximum transfer size too small!", name); 717 ddi_soft_state_free(bd_state, inst); 718 return (DDI_FAILURE); 719 } 720 721 bd->d_dip = dip; 722 bd->d_handle = hdl; 723 ddi_set_driver_private(dip, bd); 724 725 mutex_init(&bd->d_ksmutex, NULL, MUTEX_DRIVER, NULL); 726 mutex_init(&bd->d_ocmutex, NULL, MUTEX_DRIVER, NULL); 727 mutex_init(&bd->d_statemutex, NULL, MUTEX_DRIVER, NULL); 728 cv_init(&bd->d_statecv, NULL, CV_DRIVER, NULL); 729 mutex_init(&bd->d_dle_mutex, NULL, MUTEX_DRIVER, NULL); 730 bd->d_dle_state = 0; 731 732 bd->d_cache = kmem_cache_create(kcache, sizeof (bd_xfer_impl_t), 8, 733 bd_xfer_ctor, bd_xfer_dtor, NULL, bd, NULL, 0); 734 735 bd->d_ksp = kstat_create(ddi_driver_name(dip), inst, NULL, "disk", 736 KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT); 737 if (bd->d_ksp != NULL) { 738 bd->d_ksp->ks_lock = &bd->d_ksmutex; 739 kstat_install(bd->d_ksp); 740 bd->d_kiop = bd->d_ksp->ks_data; 741 } else { 742 /* 743 * Even if we cannot create the kstat, we create a 744 * scratch kstat. The reason for this is to ensure 745 * that we can update the kstat all of the time, 746 * without adding an extra branch instruction. 747 */ 748 bd->d_kiop = kmem_zalloc(sizeof (kstat_io_t), KM_SLEEP); 749 } 750 751 cmlb_alloc_handle(&bd->d_cmlbh); 752 753 bd->d_state = DKIO_NONE; 754 755 bzero(&drive, sizeof (drive)); 756 /* 757 * Default to one queue, and no restrictions on free space requests 758 * (if driver provides method) parent driver can override. 759 */ 760 drive.d_qcount = 1; 761 drive.d_free_align = 1; 762 bd->d_ops.o_drive_info(bd->d_private, &drive); 763 764 /* 765 * Several checks to make sure o_drive_info() didn't return bad 766 * values: 767 * 768 * There must be at least one queue 769 */ 770 if (drive.d_qcount == 0) 771 goto fail_drive_info; 772 773 /* FREE/UNMAP/TRIM alignment needs to be at least 1 block */ 774 if (drive.d_free_align == 0) 775 goto fail_drive_info; 776 777 /* 778 * If d_max_free_blks is not unlimited (not 0), then we cannot allow 779 * an unlimited segment size. It is however permissible to not impose 780 * a limit on the total number of blocks freed while limiting the 781 * amount allowed in an individual segment. 782 */ 783 if ((drive.d_max_free_blks > 0 && drive.d_max_free_seg_blks == 0)) 784 goto fail_drive_info; 785 786 /* 787 * If a limit is set on d_max_free_blks (by the above check, we know 788 * if there's a limit on d_max_free_blks, d_max_free_seg_blks cannot 789 * be unlimited), it cannot be smaller than the limit on an individual 790 * segment. 791 */ 792 if ((drive.d_max_free_blks > 0 && 793 drive.d_max_free_seg_blks > drive.d_max_free_blks)) { 794 goto fail_drive_info; 795 } 796 797 bd->d_qcount = drive.d_qcount; 798 bd->d_removable = drive.d_removable; 799 bd->d_hotpluggable = drive.d_hotpluggable; 800 801 if (drive.d_maxxfer && drive.d_maxxfer < bd->d_maxxfer) 802 bd->d_maxxfer = drive.d_maxxfer; 803 804 bd->d_free_align = drive.d_free_align; 805 bd->d_max_free_seg = drive.d_max_free_seg; 806 bd->d_max_free_blks = drive.d_max_free_blks; 807 bd->d_max_free_seg_blks = drive.d_max_free_seg_blks; 808 809 bd_create_inquiry_props(dip, &drive); 810 bd_create_errstats(bd, inst, &drive); 811 bd_update_state(bd); 812 813 bd->d_queues = kmem_alloc(sizeof (*bd->d_queues) * bd->d_qcount, 814 KM_SLEEP); 815 for (i = 0; i < bd->d_qcount; i++) { 816 bd_queue_t *bq = &bd->d_queues[i]; 817 818 bq->q_qsize = drive.d_qsize; 819 bq->q_qactive = 0; 820 mutex_init(&bq->q_iomutex, NULL, MUTEX_DRIVER, NULL); 821 822 list_create(&bq->q_waitq, sizeof (bd_xfer_impl_t), 823 offsetof(struct bd_xfer_impl, i_linkage)); 824 list_create(&bq->q_runq, sizeof (bd_xfer_impl_t), 825 offsetof(struct bd_xfer_impl, i_linkage)); 826 } 827 828 if (*(uint64_t *)drive.d_eui64 != 0 || 829 *(uint64_t *)drive.d_guid != 0 || 830 *((uint64_t *)drive.d_guid + 1) != 0) 831 node_type = DDI_NT_BLOCK_BLKDEV; 832 else if (drive.d_lun >= 0) 833 node_type = DDI_NT_BLOCK_CHAN; 834 else 835 node_type = DDI_NT_BLOCK; 836 837 rv = cmlb_attach(dip, &bd_tg_ops, DTYPE_DIRECT, 838 bd->d_removable, bd->d_hotpluggable, node_type, 839 CMLB_FAKE_LABEL_ONE_PARTITION, bd->d_cmlbh, 0); 840 if (rv != 0) { 841 goto fail_cmlb_attach; 842 } 843 844 if (bd->d_ops.o_devid_init != NULL) { 845 rv = bd->d_ops.o_devid_init(bd->d_private, dip, &bd->d_devid); 846 if (rv == DDI_SUCCESS) { 847 if (ddi_devid_register(dip, bd->d_devid) != 848 DDI_SUCCESS) { 849 cmn_err(CE_WARN, 850 "%s: unable to register devid", name); 851 } 852 } 853 } 854 855 /* 856 * Add a zero-length attribute to tell the world we support 857 * kernel ioctls (for layered drivers). Also set up properties 858 * used by HAL to identify removable media. 859 */ 860 (void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP, 861 DDI_KERNEL_IOCTL, NULL, 0); 862 if (bd->d_removable) { 863 (void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP, 864 "removable-media", NULL, 0); 865 } 866 if (bd->d_hotpluggable) { 867 (void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP, 868 "hotpluggable", NULL, 0); 869 } 870 871 hdl->h_bd = bd; 872 ddi_report_dev(dip); 873 874 return (DDI_SUCCESS); 875 876 fail_cmlb_attach: 877 bd_queues_free(bd); 878 bd_destroy_errstats(bd); 879 880 fail_drive_info: 881 cmlb_free_handle(&bd->d_cmlbh); 882 883 if (bd->d_ksp != NULL) { 884 kstat_delete(bd->d_ksp); 885 bd->d_ksp = NULL; 886 } else { 887 kmem_free(bd->d_kiop, sizeof (kstat_io_t)); 888 } 889 890 kmem_cache_destroy(bd->d_cache); 891 cv_destroy(&bd->d_statecv); 892 mutex_destroy(&bd->d_statemutex); 893 mutex_destroy(&bd->d_ocmutex); 894 mutex_destroy(&bd->d_ksmutex); 895 mutex_destroy(&bd->d_dle_mutex); 896 ddi_soft_state_free(bd_state, inst); 897 return (DDI_FAILURE); 898 } 899 900 static int 901 bd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) 902 { 903 bd_handle_t hdl; 904 bd_t *bd; 905 906 bd = ddi_get_driver_private(dip); 907 hdl = ddi_get_parent_data(dip); 908 909 switch (cmd) { 910 case DDI_DETACH: 911 break; 912 case DDI_SUSPEND: 913 /* We don't suspend, but our parent does */ 914 return (DDI_SUCCESS); 915 default: 916 return (DDI_FAILURE); 917 } 918 919 hdl->h_bd = NULL; 920 921 if (bd->d_ksp != NULL) { 922 kstat_delete(bd->d_ksp); 923 bd->d_ksp = NULL; 924 } else { 925 kmem_free(bd->d_kiop, sizeof (kstat_io_t)); 926 } 927 928 bd_destroy_errstats(bd); 929 cmlb_detach(bd->d_cmlbh, 0); 930 cmlb_free_handle(&bd->d_cmlbh); 931 if (bd->d_devid) 932 ddi_devid_free(bd->d_devid); 933 kmem_cache_destroy(bd->d_cache); 934 mutex_destroy(&bd->d_ksmutex); 935 mutex_destroy(&bd->d_ocmutex); 936 mutex_destroy(&bd->d_statemutex); 937 cv_destroy(&bd->d_statecv); 938 mutex_destroy(&bd->d_dle_mutex); 939 bd_queues_free(bd); 940 ddi_soft_state_free(bd_state, ddi_get_instance(dip)); 941 return (DDI_SUCCESS); 942 } 943 944 static int 945 bd_xfer_ctor(void *buf, void *arg, int kmflag) 946 { 947 bd_xfer_impl_t *xi; 948 bd_t *bd = arg; 949 int (*dcb)(caddr_t); 950 951 if (kmflag == KM_PUSHPAGE || kmflag == KM_SLEEP) { 952 dcb = DDI_DMA_SLEEP; 953 } else { 954 dcb = DDI_DMA_DONTWAIT; 955 } 956 957 xi = buf; 958 bzero(xi, sizeof (*xi)); 959 xi->i_bd = bd; 960 961 if (bd->d_use_dma) { 962 if (ddi_dma_alloc_handle(bd->d_dip, &bd->d_dma, dcb, NULL, 963 &xi->i_dmah) != DDI_SUCCESS) { 964 return (-1); 965 } 966 } 967 968 return (0); 969 } 970 971 static void 972 bd_xfer_dtor(void *buf, void *arg) 973 { 974 bd_xfer_impl_t *xi = buf; 975 976 _NOTE(ARGUNUSED(arg)); 977 978 if (xi->i_dmah) 979 ddi_dma_free_handle(&xi->i_dmah); 980 xi->i_dmah = NULL; 981 } 982 983 static bd_xfer_impl_t * 984 bd_xfer_alloc(bd_t *bd, struct buf *bp, int (*func)(void *, bd_xfer_t *), 985 int kmflag) 986 { 987 bd_xfer_impl_t *xi; 988 int rv = 0; 989 int status; 990 unsigned dir; 991 int (*cb)(caddr_t); 992 size_t len; 993 uint32_t shift; 994 995 if (kmflag == KM_SLEEP) { 996 cb = DDI_DMA_SLEEP; 997 } else { 998 cb = DDI_DMA_DONTWAIT; 999 } 1000 1001 xi = kmem_cache_alloc(bd->d_cache, kmflag); 1002 if (xi == NULL) { 1003 bioerror(bp, ENOMEM); 1004 return (NULL); 1005 } 1006 1007 ASSERT(bp); 1008 1009 xi->i_bp = bp; 1010 xi->i_func = func; 1011 xi->i_blkno = bp->b_lblkno >> (bd->d_blkshift - DEV_BSHIFT); 1012 1013 if (bp->b_bcount == 0) { 1014 xi->i_len = 0; 1015 xi->i_nblks = 0; 1016 xi->i_kaddr = NULL; 1017 xi->i_resid = 0; 1018 xi->i_num_win = 0; 1019 goto done; 1020 } 1021 1022 if (bp->b_flags & B_READ) { 1023 dir = DDI_DMA_READ; 1024 xi->i_func = bd->d_ops.o_read; 1025 } else { 1026 dir = DDI_DMA_WRITE; 1027 xi->i_func = bd->d_ops.o_write; 1028 } 1029 1030 shift = bd->d_blkshift; 1031 xi->i_blkshift = shift; 1032 1033 if (!bd->d_use_dma) { 1034 bp_mapin(bp); 1035 rv = 0; 1036 xi->i_offset = 0; 1037 xi->i_num_win = 1038 (bp->b_bcount + (bd->d_maxxfer - 1)) / bd->d_maxxfer; 1039 xi->i_cur_win = 0; 1040 xi->i_len = min(bp->b_bcount, bd->d_maxxfer); 1041 xi->i_nblks = xi->i_len >> shift; 1042 xi->i_kaddr = bp->b_un.b_addr; 1043 xi->i_resid = bp->b_bcount; 1044 } else { 1045 1046 /* 1047 * We have to use consistent DMA if the address is misaligned. 1048 */ 1049 if (((bp->b_flags & (B_PAGEIO | B_REMAPPED)) != B_PAGEIO) && 1050 ((uintptr_t)bp->b_un.b_addr & 0x7)) { 1051 dir |= DDI_DMA_CONSISTENT | DDI_DMA_PARTIAL; 1052 } else { 1053 dir |= DDI_DMA_STREAMING | DDI_DMA_PARTIAL; 1054 } 1055 1056 status = ddi_dma_buf_bind_handle(xi->i_dmah, bp, dir, cb, 1057 NULL, &xi->i_dmac, &xi->i_ndmac); 1058 switch (status) { 1059 case DDI_DMA_MAPPED: 1060 xi->i_num_win = 1; 1061 xi->i_cur_win = 0; 1062 xi->i_offset = 0; 1063 xi->i_len = bp->b_bcount; 1064 xi->i_nblks = xi->i_len >> shift; 1065 xi->i_resid = bp->b_bcount; 1066 rv = 0; 1067 break; 1068 case DDI_DMA_PARTIAL_MAP: 1069 xi->i_cur_win = 0; 1070 1071 if ((ddi_dma_numwin(xi->i_dmah, &xi->i_num_win) != 1072 DDI_SUCCESS) || 1073 (ddi_dma_getwin(xi->i_dmah, 0, &xi->i_offset, 1074 &len, &xi->i_dmac, &xi->i_ndmac) != 1075 DDI_SUCCESS) || 1076 (P2PHASE(len, (1U << shift)) != 0)) { 1077 (void) ddi_dma_unbind_handle(xi->i_dmah); 1078 rv = EFAULT; 1079 goto done; 1080 } 1081 xi->i_len = len; 1082 xi->i_nblks = xi->i_len >> shift; 1083 xi->i_resid = bp->b_bcount; 1084 rv = 0; 1085 break; 1086 case DDI_DMA_NORESOURCES: 1087 rv = EAGAIN; 1088 goto done; 1089 case DDI_DMA_TOOBIG: 1090 rv = EINVAL; 1091 goto done; 1092 case DDI_DMA_NOMAPPING: 1093 case DDI_DMA_INUSE: 1094 default: 1095 rv = EFAULT; 1096 goto done; 1097 } 1098 } 1099 1100 done: 1101 if (rv != 0) { 1102 kmem_cache_free(bd->d_cache, xi); 1103 bioerror(bp, rv); 1104 return (NULL); 1105 } 1106 1107 return (xi); 1108 } 1109 1110 static void 1111 bd_xfer_free(bd_xfer_impl_t *xi) 1112 { 1113 if (xi->i_dmah) { 1114 (void) ddi_dma_unbind_handle(xi->i_dmah); 1115 } 1116 if (xi->i_dfl != NULL) { 1117 dfl_free((dkioc_free_list_t *)xi->i_dfl); 1118 xi->i_dfl = NULL; 1119 } 1120 kmem_cache_free(xi->i_bd->d_cache, xi); 1121 } 1122 1123 static int 1124 bd_open(dev_t *devp, int flag, int otyp, cred_t *credp) 1125 { 1126 dev_t dev = *devp; 1127 bd_t *bd; 1128 minor_t part; 1129 minor_t inst; 1130 uint64_t mask; 1131 boolean_t ndelay; 1132 int rv; 1133 diskaddr_t nblks; 1134 diskaddr_t lba; 1135 1136 _NOTE(ARGUNUSED(credp)); 1137 1138 part = BDPART(dev); 1139 inst = BDINST(dev); 1140 1141 if (otyp >= OTYPCNT) 1142 return (EINVAL); 1143 1144 ndelay = (flag & (FNDELAY | FNONBLOCK)) ? B_TRUE : B_FALSE; 1145 1146 /* 1147 * Block any DR events from changing the set of registered 1148 * devices while we function. 1149 */ 1150 rw_enter(&bd_lock, RW_READER); 1151 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) { 1152 rw_exit(&bd_lock); 1153 return (ENXIO); 1154 } 1155 1156 mutex_enter(&bd->d_ocmutex); 1157 1158 ASSERT(part < 64); 1159 mask = (1U << part); 1160 1161 bd_update_state(bd); 1162 1163 if (cmlb_validate(bd->d_cmlbh, 0, 0) != 0) { 1164 1165 /* non-blocking opens are allowed to succeed */ 1166 if (!ndelay) { 1167 rv = ENXIO; 1168 goto done; 1169 } 1170 } else if (cmlb_partinfo(bd->d_cmlbh, part, &nblks, &lba, 1171 NULL, NULL, 0) == 0) { 1172 1173 /* 1174 * We read the partinfo, verify valid ranges. If the 1175 * partition is invalid, and we aren't blocking or 1176 * doing a raw access, then fail. (Non-blocking and 1177 * raw accesses can still succeed to allow a disk with 1178 * bad partition data to opened by format and fdisk.) 1179 */ 1180 if ((!nblks) && ((!ndelay) || (otyp != OTYP_CHR))) { 1181 rv = ENXIO; 1182 goto done; 1183 } 1184 } else if (!ndelay) { 1185 /* 1186 * cmlb_partinfo failed -- invalid partition or no 1187 * disk label. 1188 */ 1189 rv = ENXIO; 1190 goto done; 1191 } 1192 1193 if ((flag & FWRITE) && bd->d_rdonly) { 1194 rv = EROFS; 1195 goto done; 1196 } 1197 1198 if ((bd->d_open_excl) & (mask)) { 1199 rv = EBUSY; 1200 goto done; 1201 } 1202 if (flag & FEXCL) { 1203 if (bd->d_open_lyr[part]) { 1204 rv = EBUSY; 1205 goto done; 1206 } 1207 for (int i = 0; i < OTYP_LYR; i++) { 1208 if (bd->d_open_reg[i] & mask) { 1209 rv = EBUSY; 1210 goto done; 1211 } 1212 } 1213 } 1214 1215 if (otyp == OTYP_LYR) { 1216 bd->d_open_lyr[part]++; 1217 } else { 1218 bd->d_open_reg[otyp] |= mask; 1219 } 1220 if (flag & FEXCL) { 1221 bd->d_open_excl |= mask; 1222 } 1223 1224 rv = 0; 1225 done: 1226 mutex_exit(&bd->d_ocmutex); 1227 rw_exit(&bd_lock); 1228 1229 return (rv); 1230 } 1231 1232 static int 1233 bd_close(dev_t dev, int flag, int otyp, cred_t *credp) 1234 { 1235 bd_t *bd; 1236 minor_t inst; 1237 minor_t part; 1238 uint64_t mask; 1239 boolean_t last = B_TRUE; 1240 1241 _NOTE(ARGUNUSED(flag)); 1242 _NOTE(ARGUNUSED(credp)); 1243 1244 part = BDPART(dev); 1245 inst = BDINST(dev); 1246 1247 ASSERT(part < 64); 1248 mask = (1U << part); 1249 1250 rw_enter(&bd_lock, RW_READER); 1251 1252 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) { 1253 rw_exit(&bd_lock); 1254 return (ENXIO); 1255 } 1256 1257 mutex_enter(&bd->d_ocmutex); 1258 if (bd->d_open_excl & mask) { 1259 bd->d_open_excl &= ~mask; 1260 } 1261 if (otyp == OTYP_LYR) { 1262 bd->d_open_lyr[part]--; 1263 } else { 1264 bd->d_open_reg[otyp] &= ~mask; 1265 } 1266 for (int i = 0; i < 64; i++) { 1267 if (bd->d_open_lyr[part]) { 1268 last = B_FALSE; 1269 } 1270 } 1271 for (int i = 0; last && (i < OTYP_LYR); i++) { 1272 if (bd->d_open_reg[i]) { 1273 last = B_FALSE; 1274 } 1275 } 1276 mutex_exit(&bd->d_ocmutex); 1277 1278 if (last) { 1279 cmlb_invalidate(bd->d_cmlbh, 0); 1280 } 1281 rw_exit(&bd_lock); 1282 1283 return (0); 1284 } 1285 1286 static int 1287 bd_dump(dev_t dev, caddr_t caddr, daddr_t blkno, int nblk) 1288 { 1289 minor_t inst; 1290 minor_t part; 1291 diskaddr_t pstart; 1292 diskaddr_t psize; 1293 bd_t *bd; 1294 bd_xfer_impl_t *xi; 1295 buf_t *bp; 1296 int rv; 1297 uint32_t shift; 1298 daddr_t d_blkno; 1299 int d_nblk; 1300 1301 rw_enter(&bd_lock, RW_READER); 1302 1303 part = BDPART(dev); 1304 inst = BDINST(dev); 1305 1306 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) { 1307 rw_exit(&bd_lock); 1308 return (ENXIO); 1309 } 1310 shift = bd->d_blkshift; 1311 d_blkno = blkno >> (shift - DEV_BSHIFT); 1312 d_nblk = nblk >> (shift - DEV_BSHIFT); 1313 /* 1314 * do cmlb, but do it synchronously unless we already have the 1315 * partition (which we probably should.) 1316 */ 1317 if (cmlb_partinfo(bd->d_cmlbh, part, &psize, &pstart, NULL, NULL, 1318 (void *)1)) { 1319 rw_exit(&bd_lock); 1320 return (ENXIO); 1321 } 1322 1323 if ((d_blkno + d_nblk) > psize) { 1324 rw_exit(&bd_lock); 1325 return (EINVAL); 1326 } 1327 bp = getrbuf(KM_NOSLEEP); 1328 if (bp == NULL) { 1329 rw_exit(&bd_lock); 1330 return (ENOMEM); 1331 } 1332 1333 bp->b_bcount = nblk << DEV_BSHIFT; 1334 bp->b_resid = bp->b_bcount; 1335 bp->b_lblkno = blkno; 1336 bp->b_un.b_addr = caddr; 1337 1338 xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_write, KM_NOSLEEP); 1339 if (xi == NULL) { 1340 rw_exit(&bd_lock); 1341 freerbuf(bp); 1342 return (ENOMEM); 1343 } 1344 xi->i_blkno = d_blkno + pstart; 1345 xi->i_flags = BD_XFER_POLL; 1346 bd_submit(bd, xi); 1347 rw_exit(&bd_lock); 1348 1349 /* 1350 * Generally, we should have run this entirely synchronously 1351 * at this point and the biowait call should be a no-op. If 1352 * it didn't happen this way, it's a bug in the underlying 1353 * driver not honoring BD_XFER_POLL. 1354 */ 1355 (void) biowait(bp); 1356 rv = geterror(bp); 1357 freerbuf(bp); 1358 return (rv); 1359 } 1360 1361 void 1362 bd_minphys(struct buf *bp) 1363 { 1364 minor_t inst; 1365 bd_t *bd; 1366 inst = BDINST(bp->b_edev); 1367 1368 bd = ddi_get_soft_state(bd_state, inst); 1369 1370 /* 1371 * In a non-debug kernel, bd_strategy will catch !bd as 1372 * well, and will fail nicely. 1373 */ 1374 ASSERT(bd); 1375 1376 if (bp->b_bcount > bd->d_maxxfer) 1377 bp->b_bcount = bd->d_maxxfer; 1378 } 1379 1380 static int 1381 bd_check_uio(dev_t dev, struct uio *uio) 1382 { 1383 bd_t *bd; 1384 uint32_t shift; 1385 1386 if ((bd = ddi_get_soft_state(bd_state, BDINST(dev))) == NULL) { 1387 return (ENXIO); 1388 } 1389 1390 shift = bd->d_blkshift; 1391 if ((P2PHASE(uio->uio_loffset, (1U << shift)) != 0) || 1392 (P2PHASE(uio->uio_iov->iov_len, (1U << shift)) != 0)) { 1393 return (EINVAL); 1394 } 1395 1396 return (0); 1397 } 1398 1399 static int 1400 bd_read(dev_t dev, struct uio *uio, cred_t *credp) 1401 { 1402 _NOTE(ARGUNUSED(credp)); 1403 int ret = bd_check_uio(dev, uio); 1404 if (ret != 0) { 1405 return (ret); 1406 } 1407 return (physio(bd_strategy, NULL, dev, B_READ, bd_minphys, uio)); 1408 } 1409 1410 static int 1411 bd_write(dev_t dev, struct uio *uio, cred_t *credp) 1412 { 1413 _NOTE(ARGUNUSED(credp)); 1414 int ret = bd_check_uio(dev, uio); 1415 if (ret != 0) { 1416 return (ret); 1417 } 1418 return (physio(bd_strategy, NULL, dev, B_WRITE, bd_minphys, uio)); 1419 } 1420 1421 static int 1422 bd_aread(dev_t dev, struct aio_req *aio, cred_t *credp) 1423 { 1424 _NOTE(ARGUNUSED(credp)); 1425 int ret = bd_check_uio(dev, aio->aio_uio); 1426 if (ret != 0) { 1427 return (ret); 1428 } 1429 return (aphysio(bd_strategy, anocancel, dev, B_READ, bd_minphys, aio)); 1430 } 1431 1432 static int 1433 bd_awrite(dev_t dev, struct aio_req *aio, cred_t *credp) 1434 { 1435 _NOTE(ARGUNUSED(credp)); 1436 int ret = bd_check_uio(dev, aio->aio_uio); 1437 if (ret != 0) { 1438 return (ret); 1439 } 1440 return (aphysio(bd_strategy, anocancel, dev, B_WRITE, bd_minphys, aio)); 1441 } 1442 1443 static int 1444 bd_strategy(struct buf *bp) 1445 { 1446 minor_t inst; 1447 minor_t part; 1448 bd_t *bd; 1449 diskaddr_t p_lba; 1450 diskaddr_t p_nblks; 1451 diskaddr_t b_nblks; 1452 bd_xfer_impl_t *xi; 1453 uint32_t shift; 1454 int (*func)(void *, bd_xfer_t *); 1455 diskaddr_t lblkno; 1456 1457 part = BDPART(bp->b_edev); 1458 inst = BDINST(bp->b_edev); 1459 1460 ASSERT(bp); 1461 1462 bp->b_resid = bp->b_bcount; 1463 1464 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) { 1465 bioerror(bp, ENXIO); 1466 biodone(bp); 1467 return (0); 1468 } 1469 1470 if (cmlb_partinfo(bd->d_cmlbh, part, &p_nblks, &p_lba, 1471 NULL, NULL, 0)) { 1472 bioerror(bp, ENXIO); 1473 biodone(bp); 1474 return (0); 1475 } 1476 1477 shift = bd->d_blkshift; 1478 lblkno = bp->b_lblkno >> (shift - DEV_BSHIFT); 1479 if ((P2PHASE(bp->b_lblkno, (1U << (shift - DEV_BSHIFT))) != 0) || 1480 (P2PHASE(bp->b_bcount, (1U << shift)) != 0) || 1481 (lblkno > p_nblks)) { 1482 bioerror(bp, EINVAL); 1483 biodone(bp); 1484 return (0); 1485 } 1486 b_nblks = bp->b_bcount >> shift; 1487 if ((lblkno == p_nblks) || (bp->b_bcount == 0)) { 1488 biodone(bp); 1489 return (0); 1490 } 1491 1492 if ((b_nblks + lblkno) > p_nblks) { 1493 bp->b_resid = ((lblkno + b_nblks - p_nblks) << shift); 1494 bp->b_bcount -= bp->b_resid; 1495 } else { 1496 bp->b_resid = 0; 1497 } 1498 func = (bp->b_flags & B_READ) ? bd->d_ops.o_read : bd->d_ops.o_write; 1499 1500 xi = bd_xfer_alloc(bd, bp, func, KM_NOSLEEP); 1501 if (xi == NULL) { 1502 xi = bd_xfer_alloc(bd, bp, func, KM_PUSHPAGE); 1503 } 1504 if (xi == NULL) { 1505 /* bd_request_alloc will have done bioerror */ 1506 biodone(bp); 1507 return (0); 1508 } 1509 xi->i_blkno = lblkno + p_lba; 1510 1511 bd_submit(bd, xi); 1512 1513 return (0); 1514 } 1515 1516 static int 1517 bd_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *credp, int *rvalp) 1518 { 1519 minor_t inst; 1520 uint16_t part; 1521 bd_t *bd; 1522 void *ptr = (void *)arg; 1523 int rv; 1524 1525 part = BDPART(dev); 1526 inst = BDINST(dev); 1527 1528 if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) { 1529 return (ENXIO); 1530 } 1531 1532 rv = cmlb_ioctl(bd->d_cmlbh, dev, cmd, arg, flag, credp, rvalp, 0); 1533 if (rv != ENOTTY) 1534 return (rv); 1535 1536 if (rvalp != NULL) { 1537 /* the return value of the ioctl is 0 by default */ 1538 *rvalp = 0; 1539 } 1540 1541 switch (cmd) { 1542 case DKIOCGMEDIAINFO: { 1543 struct dk_minfo minfo; 1544 1545 /* make sure our state information is current */ 1546 bd_update_state(bd); 1547 bzero(&minfo, sizeof (minfo)); 1548 minfo.dki_media_type = DK_FIXED_DISK; 1549 minfo.dki_lbsize = (1U << bd->d_blkshift); 1550 minfo.dki_capacity = bd->d_numblks; 1551 if (ddi_copyout(&minfo, ptr, sizeof (minfo), flag)) { 1552 return (EFAULT); 1553 } 1554 return (0); 1555 } 1556 case DKIOCGMEDIAINFOEXT: { 1557 struct dk_minfo_ext miext; 1558 size_t len; 1559 1560 /* make sure our state information is current */ 1561 bd_update_state(bd); 1562 bzero(&miext, sizeof (miext)); 1563 miext.dki_media_type = DK_FIXED_DISK; 1564 miext.dki_lbsize = (1U << bd->d_blkshift); 1565 miext.dki_pbsize = (1U << bd->d_pblkshift); 1566 miext.dki_capacity = bd->d_numblks; 1567 1568 switch (ddi_model_convert_from(flag & FMODELS)) { 1569 case DDI_MODEL_ILP32: 1570 len = sizeof (struct dk_minfo_ext32); 1571 break; 1572 default: 1573 len = sizeof (struct dk_minfo_ext); 1574 break; 1575 } 1576 1577 if (ddi_copyout(&miext, ptr, len, flag)) { 1578 return (EFAULT); 1579 } 1580 return (0); 1581 } 1582 case DKIOCINFO: { 1583 struct dk_cinfo cinfo; 1584 bzero(&cinfo, sizeof (cinfo)); 1585 cinfo.dki_ctype = DKC_BLKDEV; 1586 cinfo.dki_cnum = ddi_get_instance(ddi_get_parent(bd->d_dip)); 1587 (void) snprintf(cinfo.dki_cname, sizeof (cinfo.dki_cname), 1588 "%s", ddi_driver_name(ddi_get_parent(bd->d_dip))); 1589 (void) snprintf(cinfo.dki_dname, sizeof (cinfo.dki_dname), 1590 "%s", ddi_driver_name(bd->d_dip)); 1591 cinfo.dki_unit = inst; 1592 cinfo.dki_flags = DKI_FMTVOL; 1593 cinfo.dki_partition = part; 1594 cinfo.dki_maxtransfer = bd->d_maxxfer / DEV_BSIZE; 1595 cinfo.dki_addr = 0; 1596 cinfo.dki_slave = 0; 1597 cinfo.dki_space = 0; 1598 cinfo.dki_prio = 0; 1599 cinfo.dki_vec = 0; 1600 if (ddi_copyout(&cinfo, ptr, sizeof (cinfo), flag)) { 1601 return (EFAULT); 1602 } 1603 return (0); 1604 } 1605 case DKIOCREMOVABLE: { 1606 int i; 1607 i = bd->d_removable ? 1 : 0; 1608 if (ddi_copyout(&i, ptr, sizeof (i), flag)) { 1609 return (EFAULT); 1610 } 1611 return (0); 1612 } 1613 case DKIOCHOTPLUGGABLE: { 1614 int i; 1615 i = bd->d_hotpluggable ? 1 : 0; 1616 if (ddi_copyout(&i, ptr, sizeof (i), flag)) { 1617 return (EFAULT); 1618 } 1619 return (0); 1620 } 1621 case DKIOCREADONLY: { 1622 int i; 1623 i = bd->d_rdonly ? 1 : 0; 1624 if (ddi_copyout(&i, ptr, sizeof (i), flag)) { 1625 return (EFAULT); 1626 } 1627 return (0); 1628 } 1629 case DKIOCSOLIDSTATE: { 1630 int i; 1631 i = bd->d_ssd ? 1 : 0; 1632 if (ddi_copyout(&i, ptr, sizeof (i), flag)) { 1633 return (EFAULT); 1634 } 1635 return (0); 1636 } 1637 case DKIOCSTATE: { 1638 enum dkio_state state; 1639 if (ddi_copyin(ptr, &state, sizeof (state), flag)) { 1640 return (EFAULT); 1641 } 1642 if ((rv = bd_check_state(bd, &state)) != 0) { 1643 return (rv); 1644 } 1645 if (ddi_copyout(&state, ptr, sizeof (state), flag)) { 1646 return (EFAULT); 1647 } 1648 return (0); 1649 } 1650 case DKIOCFLUSHWRITECACHE: { 1651 struct dk_callback *dkc = NULL; 1652 1653 if (flag & FKIOCTL) 1654 dkc = (void *)arg; 1655 1656 rv = bd_flush_write_cache(bd, dkc); 1657 return (rv); 1658 } 1659 case DKIOCFREE: { 1660 dkioc_free_list_t *dfl = NULL; 1661 1662 /* 1663 * Check free space support early to avoid copyin/allocation 1664 * when unnecessary. 1665 */ 1666 if (!CAN_FREESPACE(bd)) 1667 return (ENOTSUP); 1668 1669 rv = dfl_copyin(ptr, &dfl, flag, KM_SLEEP); 1670 if (rv != 0) 1671 return (rv); 1672 1673 /* 1674 * bd_free_space() consumes 'dfl'. bd_free_space() will 1675 * call dfl_iter() which will normally try to pass dfl through 1676 * to bd_free_space_cb() which attaches dfl to the bd_xfer_t 1677 * that is then queued for the underlying driver. Once the 1678 * driver processes the request, the bd_xfer_t instance is 1679 * disposed of, including any attached dkioc_free_list_t. 1680 * 1681 * If dfl cannot be processed by the underlying driver due to 1682 * size or alignment requirements of the driver, dfl_iter() 1683 * will replace dfl with one or more new dkioc_free_list_t 1684 * instances with the correct alignment and sizes for the driver 1685 * (and free the original dkioc_free_list_t). 1686 */ 1687 rv = bd_free_space(dev, bd, dfl); 1688 return (rv); 1689 } 1690 1691 case DKIOC_CANFREE: { 1692 boolean_t supported = CAN_FREESPACE(bd); 1693 1694 if (ddi_copyout(&supported, (void *)arg, sizeof (supported), 1695 flag) != 0) { 1696 return (EFAULT); 1697 } 1698 1699 return (0); 1700 } 1701 1702 default: 1703 break; 1704 1705 } 1706 return (ENOTTY); 1707 } 1708 1709 static int 1710 bd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags, 1711 char *name, caddr_t valuep, int *lengthp) 1712 { 1713 bd_t *bd; 1714 1715 bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip)); 1716 if (bd == NULL) 1717 return (ddi_prop_op(dev, dip, prop_op, mod_flags, 1718 name, valuep, lengthp)); 1719 1720 return (cmlb_prop_op(bd->d_cmlbh, dev, dip, prop_op, mod_flags, name, 1721 valuep, lengthp, BDPART(dev), 0)); 1722 } 1723 1724 1725 static int 1726 bd_tg_rdwr(dev_info_t *dip, uchar_t cmd, void *bufaddr, diskaddr_t start, 1727 size_t length, void *tg_cookie) 1728 { 1729 bd_t *bd; 1730 buf_t *bp; 1731 bd_xfer_impl_t *xi; 1732 int rv; 1733 int (*func)(void *, bd_xfer_t *); 1734 int kmflag; 1735 1736 /* 1737 * If we are running in polled mode (such as during dump(9e) 1738 * execution), then we cannot sleep for kernel allocations. 1739 */ 1740 kmflag = tg_cookie ? KM_NOSLEEP : KM_SLEEP; 1741 1742 bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip)); 1743 1744 if (P2PHASE(length, (1U << bd->d_blkshift)) != 0) { 1745 /* We can only transfer whole blocks at a time! */ 1746 return (EINVAL); 1747 } 1748 1749 if ((bp = getrbuf(kmflag)) == NULL) { 1750 return (ENOMEM); 1751 } 1752 1753 switch (cmd) { 1754 case TG_READ: 1755 bp->b_flags = B_READ; 1756 func = bd->d_ops.o_read; 1757 break; 1758 case TG_WRITE: 1759 bp->b_flags = B_WRITE; 1760 func = bd->d_ops.o_write; 1761 break; 1762 default: 1763 freerbuf(bp); 1764 return (EINVAL); 1765 } 1766 1767 bp->b_un.b_addr = bufaddr; 1768 bp->b_bcount = length; 1769 xi = bd_xfer_alloc(bd, bp, func, kmflag); 1770 if (xi == NULL) { 1771 rv = geterror(bp); 1772 freerbuf(bp); 1773 return (rv); 1774 } 1775 xi->i_flags = tg_cookie ? BD_XFER_POLL : 0; 1776 xi->i_blkno = start; 1777 bd_submit(bd, xi); 1778 (void) biowait(bp); 1779 rv = geterror(bp); 1780 freerbuf(bp); 1781 1782 return (rv); 1783 } 1784 1785 static int 1786 bd_tg_getinfo(dev_info_t *dip, int cmd, void *arg, void *tg_cookie) 1787 { 1788 bd_t *bd; 1789 1790 _NOTE(ARGUNUSED(tg_cookie)); 1791 bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip)); 1792 1793 switch (cmd) { 1794 case TG_GETPHYGEOM: 1795 case TG_GETVIRTGEOM: 1796 /* 1797 * We don't have any "geometry" as such, let cmlb 1798 * fabricate something. 1799 */ 1800 return (ENOTTY); 1801 1802 case TG_GETCAPACITY: 1803 bd_update_state(bd); 1804 *(diskaddr_t *)arg = bd->d_numblks; 1805 return (0); 1806 1807 case TG_GETBLOCKSIZE: 1808 *(uint32_t *)arg = (1U << bd->d_blkshift); 1809 return (0); 1810 1811 case TG_GETATTR: 1812 /* 1813 * It turns out that cmlb really doesn't do much for 1814 * non-writable media, but lets make the information 1815 * available for it in case it does more in the 1816 * future. (The value is currently used for 1817 * triggering special behavior for CD-ROMs.) 1818 */ 1819 bd_update_state(bd); 1820 ((tg_attribute_t *)arg)->media_is_writable = 1821 bd->d_rdonly ? B_FALSE : B_TRUE; 1822 ((tg_attribute_t *)arg)->media_is_solid_state = bd->d_ssd; 1823 ((tg_attribute_t *)arg)->media_is_rotational = B_FALSE; 1824 return (0); 1825 1826 default: 1827 return (EINVAL); 1828 } 1829 } 1830 1831 1832 static void 1833 bd_sched(bd_t *bd, bd_queue_t *bq) 1834 { 1835 bd_xfer_impl_t *xi; 1836 struct buf *bp; 1837 int rv; 1838 1839 mutex_enter(&bq->q_iomutex); 1840 1841 while ((bq->q_qactive < bq->q_qsize) && 1842 ((xi = list_remove_head(&bq->q_waitq)) != NULL)) { 1843 mutex_enter(&bd->d_ksmutex); 1844 kstat_waitq_to_runq(bd->d_kiop); 1845 mutex_exit(&bd->d_ksmutex); 1846 1847 bq->q_qactive++; 1848 list_insert_tail(&bq->q_runq, xi); 1849 1850 /* 1851 * Submit the job to the driver. We drop the I/O mutex 1852 * so that we can deal with the case where the driver 1853 * completion routine calls back into us synchronously. 1854 */ 1855 1856 mutex_exit(&bq->q_iomutex); 1857 1858 rv = xi->i_func(bd->d_private, &xi->i_public); 1859 if (rv != 0) { 1860 bp = xi->i_bp; 1861 bioerror(bp, rv); 1862 biodone(bp); 1863 1864 atomic_inc_32(&bd->d_kerr->bd_transerrs.value.ui32); 1865 1866 mutex_enter(&bq->q_iomutex); 1867 1868 mutex_enter(&bd->d_ksmutex); 1869 kstat_runq_exit(bd->d_kiop); 1870 mutex_exit(&bd->d_ksmutex); 1871 1872 bq->q_qactive--; 1873 list_remove(&bq->q_runq, xi); 1874 bd_xfer_free(xi); 1875 } else { 1876 mutex_enter(&bq->q_iomutex); 1877 } 1878 } 1879 1880 mutex_exit(&bq->q_iomutex); 1881 } 1882 1883 static void 1884 bd_submit(bd_t *bd, bd_xfer_impl_t *xi) 1885 { 1886 uint64_t nv = atomic_inc_64_nv(&bd->d_io_counter); 1887 unsigned q = nv % bd->d_qcount; 1888 bd_queue_t *bq = &bd->d_queues[q]; 1889 1890 xi->i_bq = bq; 1891 xi->i_qnum = q; 1892 1893 mutex_enter(&bq->q_iomutex); 1894 1895 list_insert_tail(&bq->q_waitq, xi); 1896 1897 mutex_enter(&bd->d_ksmutex); 1898 kstat_waitq_enter(bd->d_kiop); 1899 mutex_exit(&bd->d_ksmutex); 1900 1901 mutex_exit(&bq->q_iomutex); 1902 1903 bd_sched(bd, bq); 1904 } 1905 1906 static void 1907 bd_runq_exit(bd_xfer_impl_t *xi, int err) 1908 { 1909 bd_t *bd = xi->i_bd; 1910 buf_t *bp = xi->i_bp; 1911 bd_queue_t *bq = xi->i_bq; 1912 1913 mutex_enter(&bq->q_iomutex); 1914 bq->q_qactive--; 1915 1916 mutex_enter(&bd->d_ksmutex); 1917 kstat_runq_exit(bd->d_kiop); 1918 mutex_exit(&bd->d_ksmutex); 1919 1920 list_remove(&bq->q_runq, xi); 1921 mutex_exit(&bq->q_iomutex); 1922 1923 if (err == 0) { 1924 if (bp->b_flags & B_READ) { 1925 atomic_inc_uint(&bd->d_kiop->reads); 1926 atomic_add_64((uint64_t *)&bd->d_kiop->nread, 1927 bp->b_bcount - xi->i_resid); 1928 } else { 1929 atomic_inc_uint(&bd->d_kiop->writes); 1930 atomic_add_64((uint64_t *)&bd->d_kiop->nwritten, 1931 bp->b_bcount - xi->i_resid); 1932 } 1933 } 1934 bd_sched(bd, bq); 1935 } 1936 1937 static void 1938 bd_dle_sysevent_task(void *arg) 1939 { 1940 nvlist_t *attr = NULL; 1941 char *path = NULL; 1942 bd_t *bd = arg; 1943 dev_info_t *dip = bd->d_dip; 1944 size_t n; 1945 1946 mutex_enter(&bd->d_dle_mutex); 1947 bd->d_dle_state &= ~BD_DLE_PENDING; 1948 bd->d_dle_state |= BD_DLE_RUNNING; 1949 mutex_exit(&bd->d_dle_mutex); 1950 1951 dev_err(dip, CE_NOTE, "!dynamic LUN expansion"); 1952 1953 if (nvlist_alloc(&attr, NV_UNIQUE_NAME_TYPE, KM_SLEEP) != 0) { 1954 mutex_enter(&bd->d_dle_mutex); 1955 bd->d_dle_state &= ~(BD_DLE_RUNNING|BD_DLE_PENDING); 1956 mutex_exit(&bd->d_dle_mutex); 1957 return; 1958 } 1959 1960 path = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 1961 1962 n = snprintf(path, MAXPATHLEN, "/devices"); 1963 (void) ddi_pathname(dip, path + n); 1964 n = strlen(path); 1965 n += snprintf(path + n, MAXPATHLEN - n, ":x"); 1966 1967 for (;;) { 1968 /* 1969 * On receipt of this event, the ZFS sysevent module will scan 1970 * active zpools for child vdevs matching this physical path. 1971 * In order to catch both whole disk pools and those with an 1972 * EFI boot partition, generate separate sysevents for minor 1973 * node 'a' and 'b'. 1974 */ 1975 for (char c = 'a'; c < 'c'; c++) { 1976 path[n - 1] = c; 1977 1978 if (nvlist_add_string(attr, DEV_PHYS_PATH, path) != 0) 1979 break; 1980 1981 (void) ddi_log_sysevent(dip, DDI_VENDOR_SUNW, 1982 EC_DEV_STATUS, ESC_DEV_DLE, attr, NULL, DDI_SLEEP); 1983 } 1984 1985 mutex_enter(&bd->d_dle_mutex); 1986 if ((bd->d_dle_state & BD_DLE_PENDING) == 0) { 1987 bd->d_dle_state &= ~BD_DLE_RUNNING; 1988 mutex_exit(&bd->d_dle_mutex); 1989 break; 1990 } 1991 bd->d_dle_state &= ~BD_DLE_PENDING; 1992 mutex_exit(&bd->d_dle_mutex); 1993 } 1994 1995 nvlist_free(attr); 1996 kmem_free(path, MAXPATHLEN); 1997 } 1998 1999 static void 2000 bd_update_state(bd_t *bd) 2001 { 2002 enum dkio_state state = DKIO_INSERTED; 2003 boolean_t docmlb = B_FALSE; 2004 bd_media_t media; 2005 2006 bzero(&media, sizeof (media)); 2007 2008 mutex_enter(&bd->d_statemutex); 2009 if (bd->d_ops.o_media_info(bd->d_private, &media) != 0) { 2010 bd->d_numblks = 0; 2011 state = DKIO_EJECTED; 2012 goto done; 2013 } 2014 2015 if ((media.m_blksize < 512) || 2016 (!ISP2(media.m_blksize)) || 2017 (P2PHASE(bd->d_maxxfer, media.m_blksize))) { 2018 dev_err(bd->d_dip, CE_WARN, "Invalid media block size (%d)", 2019 media.m_blksize); 2020 /* 2021 * We can't use the media, treat it as not present. 2022 */ 2023 state = DKIO_EJECTED; 2024 bd->d_numblks = 0; 2025 goto done; 2026 } 2027 2028 if (((1U << bd->d_blkshift) != media.m_blksize) || 2029 (bd->d_numblks != media.m_nblks)) { 2030 /* Device size changed */ 2031 docmlb = B_TRUE; 2032 } 2033 2034 bd->d_blkshift = ddi_ffs(media.m_blksize) - 1; 2035 bd->d_pblkshift = bd->d_blkshift; 2036 bd->d_numblks = media.m_nblks; 2037 bd->d_rdonly = media.m_readonly; 2038 bd->d_ssd = media.m_solidstate; 2039 2040 /* 2041 * Only use the supplied physical block size if it is non-zero, 2042 * greater or equal to the block size, and a power of 2. Ignore it 2043 * if not, it's just informational and we can still use the media. 2044 */ 2045 if ((media.m_pblksize != 0) && 2046 (media.m_pblksize >= media.m_blksize) && 2047 (ISP2(media.m_pblksize))) 2048 bd->d_pblkshift = ddi_ffs(media.m_pblksize) - 1; 2049 2050 done: 2051 if (state != bd->d_state) { 2052 bd->d_state = state; 2053 cv_broadcast(&bd->d_statecv); 2054 docmlb = B_TRUE; 2055 } 2056 mutex_exit(&bd->d_statemutex); 2057 2058 bd->d_kerr->bd_capacity.value.ui64 = bd->d_numblks << bd->d_blkshift; 2059 2060 if (docmlb) { 2061 if (state == DKIO_INSERTED) { 2062 (void) cmlb_validate(bd->d_cmlbh, 0, 0); 2063 2064 mutex_enter(&bd->d_dle_mutex); 2065 /* 2066 * If there is already an event pending, there's 2067 * nothing to do; we coalesce multiple events. 2068 */ 2069 if ((bd->d_dle_state & BD_DLE_PENDING) == 0) { 2070 if ((bd->d_dle_state & BD_DLE_RUNNING) == 0) { 2071 taskq_dispatch_ent(bd_taskq, 2072 bd_dle_sysevent_task, bd, 0, 2073 &bd->d_dle_ent); 2074 } 2075 bd->d_dle_state |= BD_DLE_PENDING; 2076 } 2077 mutex_exit(&bd->d_dle_mutex); 2078 } else { 2079 cmlb_invalidate(bd->d_cmlbh, 0); 2080 } 2081 } 2082 } 2083 2084 static int 2085 bd_check_state(bd_t *bd, enum dkio_state *state) 2086 { 2087 clock_t when; 2088 2089 for (;;) { 2090 2091 bd_update_state(bd); 2092 2093 mutex_enter(&bd->d_statemutex); 2094 2095 if (bd->d_state != *state) { 2096 *state = bd->d_state; 2097 mutex_exit(&bd->d_statemutex); 2098 break; 2099 } 2100 2101 when = drv_usectohz(1000000); 2102 if (cv_reltimedwait_sig(&bd->d_statecv, &bd->d_statemutex, 2103 when, TR_CLOCK_TICK) == 0) { 2104 mutex_exit(&bd->d_statemutex); 2105 return (EINTR); 2106 } 2107 2108 mutex_exit(&bd->d_statemutex); 2109 } 2110 2111 return (0); 2112 } 2113 2114 static int 2115 bd_flush_write_cache_done(struct buf *bp) 2116 { 2117 struct dk_callback *dc = (void *)bp->b_private; 2118 2119 (*dc->dkc_callback)(dc->dkc_cookie, geterror(bp)); 2120 kmem_free(dc, sizeof (*dc)); 2121 freerbuf(bp); 2122 return (0); 2123 } 2124 2125 static int 2126 bd_flush_write_cache(bd_t *bd, struct dk_callback *dkc) 2127 { 2128 buf_t *bp; 2129 struct dk_callback *dc; 2130 bd_xfer_impl_t *xi; 2131 int rv; 2132 2133 if (bd->d_ops.o_sync_cache == NULL) { 2134 return (ENOTSUP); 2135 } 2136 if ((bp = getrbuf(KM_SLEEP)) == NULL) { 2137 return (ENOMEM); 2138 } 2139 bp->b_resid = 0; 2140 bp->b_bcount = 0; 2141 2142 xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_sync_cache, KM_SLEEP); 2143 if (xi == NULL) { 2144 rv = geterror(bp); 2145 freerbuf(bp); 2146 return (rv); 2147 } 2148 2149 /* Make an asynchronous flush, but only if there is a callback */ 2150 if (dkc != NULL && dkc->dkc_callback != NULL) { 2151 /* Make a private copy of the callback structure */ 2152 dc = kmem_alloc(sizeof (*dc), KM_SLEEP); 2153 *dc = *dkc; 2154 bp->b_private = dc; 2155 bp->b_iodone = bd_flush_write_cache_done; 2156 2157 bd_submit(bd, xi); 2158 return (0); 2159 } 2160 2161 /* In case there is no callback, perform a synchronous flush */ 2162 bd_submit(bd, xi); 2163 (void) biowait(bp); 2164 rv = geterror(bp); 2165 freerbuf(bp); 2166 2167 return (rv); 2168 } 2169 2170 static int 2171 bd_free_space_done(struct buf *bp) 2172 { 2173 freerbuf(bp); 2174 return (0); 2175 } 2176 2177 static int 2178 bd_free_space_cb(dkioc_free_list_t *dfl, void *arg, int kmflag) 2179 { 2180 bd_t *bd = arg; 2181 buf_t *bp = NULL; 2182 bd_xfer_impl_t *xi = NULL; 2183 boolean_t sync = DFL_ISSYNC(dfl) ? B_TRUE : B_FALSE; 2184 int rv = 0; 2185 2186 bp = getrbuf(KM_SLEEP); 2187 bp->b_resid = 0; 2188 bp->b_bcount = 0; 2189 bp->b_lblkno = 0; 2190 2191 xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_free_space, kmflag); 2192 xi->i_dfl = dfl; 2193 2194 if (!sync) { 2195 bp->b_iodone = bd_free_space_done; 2196 bd_submit(bd, xi); 2197 return (0); 2198 } 2199 2200 xi->i_flags |= BD_XFER_POLL; 2201 bd_submit(bd, xi); 2202 2203 (void) biowait(bp); 2204 rv = geterror(bp); 2205 freerbuf(bp); 2206 2207 return (rv); 2208 } 2209 2210 static int 2211 bd_free_space(dev_t dev, bd_t *bd, dkioc_free_list_t *dfl) 2212 { 2213 diskaddr_t p_len, p_offset; 2214 uint64_t offset_bytes, len_bytes; 2215 minor_t part = BDPART(dev); 2216 const uint_t bshift = bd->d_blkshift; 2217 dkioc_free_info_t dfi = { 2218 .dfi_bshift = bshift, 2219 .dfi_align = bd->d_free_align << bshift, 2220 .dfi_max_bytes = bd->d_max_free_blks << bshift, 2221 .dfi_max_ext = bd->d_max_free_seg, 2222 .dfi_max_ext_bytes = bd->d_max_free_seg_blks << bshift, 2223 }; 2224 2225 if (cmlb_partinfo(bd->d_cmlbh, part, &p_len, &p_offset, NULL, 2226 NULL, 0) != 0) { 2227 dfl_free(dfl); 2228 return (ENXIO); 2229 } 2230 2231 /* 2232 * bd_ioctl created our own copy of dfl, so we can modify as 2233 * necessary 2234 */ 2235 offset_bytes = (uint64_t)p_offset << bshift; 2236 len_bytes = (uint64_t)p_len << bshift; 2237 2238 dfl->dfl_offset += offset_bytes; 2239 if (dfl->dfl_offset < offset_bytes) { 2240 dfl_free(dfl); 2241 return (EOVERFLOW); 2242 } 2243 2244 return (dfl_iter(dfl, &dfi, offset_bytes + len_bytes, bd_free_space_cb, 2245 bd, KM_SLEEP)); 2246 } 2247 2248 /* 2249 * Nexus support. 2250 */ 2251 int 2252 bd_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t ctlop, 2253 void *arg, void *result) 2254 { 2255 bd_handle_t hdl; 2256 2257 switch (ctlop) { 2258 case DDI_CTLOPS_REPORTDEV: 2259 cmn_err(CE_CONT, "?Block device: %s@%s, %s%d\n", 2260 ddi_node_name(rdip), ddi_get_name_addr(rdip), 2261 ddi_driver_name(rdip), ddi_get_instance(rdip)); 2262 return (DDI_SUCCESS); 2263 2264 case DDI_CTLOPS_INITCHILD: 2265 hdl = ddi_get_parent_data((dev_info_t *)arg); 2266 if (hdl == NULL) { 2267 return (DDI_NOT_WELL_FORMED); 2268 } 2269 ddi_set_name_addr((dev_info_t *)arg, hdl->h_addr); 2270 return (DDI_SUCCESS); 2271 2272 case DDI_CTLOPS_UNINITCHILD: 2273 ddi_set_name_addr((dev_info_t *)arg, NULL); 2274 ndi_prop_remove_all((dev_info_t *)arg); 2275 return (DDI_SUCCESS); 2276 2277 default: 2278 return (ddi_ctlops(dip, rdip, ctlop, arg, result)); 2279 } 2280 } 2281 2282 /* 2283 * Functions for device drivers. 2284 */ 2285 bd_handle_t 2286 bd_alloc_handle(void *private, bd_ops_t *ops, ddi_dma_attr_t *dma, int kmflag) 2287 { 2288 bd_handle_t hdl; 2289 2290 switch (ops->o_version) { 2291 case BD_OPS_VERSION_0: 2292 case BD_OPS_VERSION_1: 2293 case BD_OPS_VERSION_2: 2294 break; 2295 2296 default: 2297 /* Unsupported version */ 2298 return (NULL); 2299 } 2300 2301 hdl = kmem_zalloc(sizeof (*hdl), kmflag); 2302 if (hdl == NULL) { 2303 return (NULL); 2304 } 2305 2306 switch (ops->o_version) { 2307 case BD_OPS_VERSION_2: 2308 hdl->h_ops.o_free_space = ops->o_free_space; 2309 /*FALLTHRU*/ 2310 case BD_OPS_VERSION_1: 2311 case BD_OPS_VERSION_0: 2312 hdl->h_ops.o_drive_info = ops->o_drive_info; 2313 hdl->h_ops.o_media_info = ops->o_media_info; 2314 hdl->h_ops.o_devid_init = ops->o_devid_init; 2315 hdl->h_ops.o_sync_cache = ops->o_sync_cache; 2316 hdl->h_ops.o_read = ops->o_read; 2317 hdl->h_ops.o_write = ops->o_write; 2318 break; 2319 } 2320 2321 hdl->h_dma = dma; 2322 hdl->h_private = private; 2323 2324 return (hdl); 2325 } 2326 2327 void 2328 bd_free_handle(bd_handle_t hdl) 2329 { 2330 kmem_free(hdl, sizeof (*hdl)); 2331 } 2332 2333 int 2334 bd_attach_handle(dev_info_t *dip, bd_handle_t hdl) 2335 { 2336 bd_drive_t drive = { 0 }; 2337 dev_info_t *child; 2338 size_t len; 2339 2340 /* 2341 * It's not an error if bd_attach_handle() is called on a handle that 2342 * already is attached. We just ignore the request to attach and return. 2343 * This way drivers using blkdev don't have to keep track about blkdev 2344 * state, they can just call this function to make sure it attached. 2345 */ 2346 if (hdl->h_child != NULL) { 2347 return (DDI_SUCCESS); 2348 } 2349 2350 /* if drivers don't override this, make it assume none */ 2351 drive.d_lun = -1; 2352 hdl->h_ops.o_drive_info(hdl->h_private, &drive); 2353 2354 hdl->h_parent = dip; 2355 hdl->h_name = "blkdev"; 2356 2357 /* 2358 * Prefer the GUID over the EUI64. 2359 */ 2360 if (*(uint64_t *)drive.d_guid != 0 || 2361 *((uint64_t *)drive.d_guid + 1) != 0) { 2362 len = snprintf(hdl->h_addr, sizeof (hdl->h_addr), 2363 "w%02X%02X%02X%02X%02X%02X%02X%02X" 2364 "%02X%02X%02X%02X%02X%02X%02X%02X", 2365 drive.d_guid[0], drive.d_guid[1], drive.d_guid[2], 2366 drive.d_guid[3], drive.d_guid[4], drive.d_guid[5], 2367 drive.d_guid[6], drive.d_guid[7], drive.d_guid[8], 2368 drive.d_guid[9], drive.d_guid[10], drive.d_guid[11], 2369 drive.d_guid[12], drive.d_guid[13], drive.d_guid[14], 2370 drive.d_guid[15]); 2371 } else if (*(uint64_t *)drive.d_eui64 != 0) { 2372 len = snprintf(hdl->h_addr, sizeof (hdl->h_addr), 2373 "w%02X%02X%02X%02X%02X%02X%02X%02X", 2374 drive.d_eui64[0], drive.d_eui64[1], 2375 drive.d_eui64[2], drive.d_eui64[3], 2376 drive.d_eui64[4], drive.d_eui64[5], 2377 drive.d_eui64[6], drive.d_eui64[7]); 2378 } else { 2379 len = snprintf(hdl->h_addr, sizeof (hdl->h_addr), 2380 "%X", drive.d_target); 2381 } 2382 2383 VERIFY(len <= sizeof (hdl->h_addr)); 2384 2385 if (drive.d_lun >= 0) { 2386 (void) snprintf(hdl->h_addr + len, sizeof (hdl->h_addr) - len, 2387 ",%X", drive.d_lun); 2388 } 2389 2390 if (ndi_devi_alloc(dip, hdl->h_name, (pnode_t)DEVI_SID_NODEID, 2391 &child) != NDI_SUCCESS) { 2392 cmn_err(CE_WARN, "%s%d: unable to allocate node %s@%s", 2393 ddi_driver_name(dip), ddi_get_instance(dip), 2394 "blkdev", hdl->h_addr); 2395 return (DDI_FAILURE); 2396 } 2397 2398 ddi_set_parent_data(child, hdl); 2399 hdl->h_child = child; 2400 2401 if (ndi_devi_online(child, 0) != NDI_SUCCESS) { 2402 cmn_err(CE_WARN, "%s%d: failed bringing node %s@%s online", 2403 ddi_driver_name(dip), ddi_get_instance(dip), 2404 hdl->h_name, hdl->h_addr); 2405 (void) ndi_devi_free(child); 2406 hdl->h_child = NULL; 2407 return (DDI_FAILURE); 2408 } 2409 2410 return (DDI_SUCCESS); 2411 } 2412 2413 int 2414 bd_detach_handle(bd_handle_t hdl) 2415 { 2416 int rv; 2417 char *devnm; 2418 2419 /* 2420 * It's not an error if bd_detach_handle() is called on a handle that 2421 * already is detached. We just ignore the request to detach and return. 2422 * This way drivers using blkdev don't have to keep track about blkdev 2423 * state, they can just call this function to make sure it detached. 2424 */ 2425 if (hdl->h_child == NULL) { 2426 return (DDI_SUCCESS); 2427 } 2428 ndi_devi_enter(hdl->h_parent); 2429 if (i_ddi_node_state(hdl->h_child) < DS_INITIALIZED) { 2430 rv = ddi_remove_child(hdl->h_child, 0); 2431 } else { 2432 devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP); 2433 (void) ddi_deviname(hdl->h_child, devnm); 2434 (void) devfs_clean(hdl->h_parent, devnm + 1, DV_CLEAN_FORCE); 2435 rv = ndi_devi_unconfig_one(hdl->h_parent, devnm + 1, NULL, 2436 NDI_DEVI_REMOVE | NDI_UNCONFIG); 2437 kmem_free(devnm, MAXNAMELEN + 1); 2438 } 2439 if (rv == 0) { 2440 hdl->h_child = NULL; 2441 } 2442 2443 ndi_devi_exit(hdl->h_parent); 2444 return (rv == NDI_SUCCESS ? DDI_SUCCESS : DDI_FAILURE); 2445 } 2446 2447 void 2448 bd_xfer_done(bd_xfer_t *xfer, int err) 2449 { 2450 bd_xfer_impl_t *xi = (void *)xfer; 2451 buf_t *bp = xi->i_bp; 2452 int rv = DDI_SUCCESS; 2453 bd_t *bd = xi->i_bd; 2454 size_t len; 2455 2456 if (err != 0) { 2457 bd_runq_exit(xi, err); 2458 atomic_inc_32(&bd->d_kerr->bd_harderrs.value.ui32); 2459 2460 bp->b_resid += xi->i_resid; 2461 bd_xfer_free(xi); 2462 bioerror(bp, err); 2463 biodone(bp); 2464 return; 2465 } 2466 2467 xi->i_cur_win++; 2468 xi->i_resid -= xi->i_len; 2469 2470 if (xi->i_resid == 0) { 2471 /* Job completed succcessfully! */ 2472 bd_runq_exit(xi, 0); 2473 2474 bd_xfer_free(xi); 2475 biodone(bp); 2476 return; 2477 } 2478 2479 xi->i_blkno += xi->i_nblks; 2480 2481 if (bd->d_use_dma) { 2482 /* More transfer still pending... advance to next DMA window. */ 2483 rv = ddi_dma_getwin(xi->i_dmah, xi->i_cur_win, 2484 &xi->i_offset, &len, &xi->i_dmac, &xi->i_ndmac); 2485 } else { 2486 /* Advance memory window. */ 2487 xi->i_kaddr += xi->i_len; 2488 xi->i_offset += xi->i_len; 2489 len = min(bp->b_bcount - xi->i_offset, bd->d_maxxfer); 2490 } 2491 2492 2493 if ((rv != DDI_SUCCESS) || 2494 (P2PHASE(len, (1U << xi->i_blkshift)) != 0)) { 2495 bd_runq_exit(xi, EFAULT); 2496 2497 bp->b_resid += xi->i_resid; 2498 bd_xfer_free(xi); 2499 bioerror(bp, EFAULT); 2500 biodone(bp); 2501 return; 2502 } 2503 xi->i_len = len; 2504 xi->i_nblks = len >> xi->i_blkshift; 2505 2506 /* Submit next window to hardware. */ 2507 rv = xi->i_func(bd->d_private, &xi->i_public); 2508 if (rv != 0) { 2509 bd_runq_exit(xi, rv); 2510 2511 atomic_inc_32(&bd->d_kerr->bd_transerrs.value.ui32); 2512 2513 bp->b_resid += xi->i_resid; 2514 bd_xfer_free(xi); 2515 bioerror(bp, rv); 2516 biodone(bp); 2517 } 2518 } 2519 2520 void 2521 bd_error(bd_xfer_t *xfer, int error) 2522 { 2523 bd_xfer_impl_t *xi = (void *)xfer; 2524 bd_t *bd = xi->i_bd; 2525 2526 switch (error) { 2527 case BD_ERR_MEDIA: 2528 atomic_inc_32(&bd->d_kerr->bd_rq_media_err.value.ui32); 2529 break; 2530 case BD_ERR_NTRDY: 2531 atomic_inc_32(&bd->d_kerr->bd_rq_ntrdy_err.value.ui32); 2532 break; 2533 case BD_ERR_NODEV: 2534 atomic_inc_32(&bd->d_kerr->bd_rq_nodev_err.value.ui32); 2535 break; 2536 case BD_ERR_RECOV: 2537 atomic_inc_32(&bd->d_kerr->bd_rq_recov_err.value.ui32); 2538 break; 2539 case BD_ERR_ILLRQ: 2540 atomic_inc_32(&bd->d_kerr->bd_rq_illrq_err.value.ui32); 2541 break; 2542 case BD_ERR_PFA: 2543 atomic_inc_32(&bd->d_kerr->bd_rq_pfa_err.value.ui32); 2544 break; 2545 default: 2546 cmn_err(CE_PANIC, "bd_error: unknown error type %d", error); 2547 break; 2548 } 2549 } 2550 2551 void 2552 bd_state_change(bd_handle_t hdl) 2553 { 2554 bd_t *bd; 2555 2556 if ((bd = hdl->h_bd) != NULL) { 2557 bd_update_state(bd); 2558 } 2559 } 2560 2561 void 2562 bd_mod_init(struct dev_ops *devops) 2563 { 2564 static struct bus_ops bd_bus_ops = { 2565 BUSO_REV, /* busops_rev */ 2566 nullbusmap, /* bus_map */ 2567 NULL, /* bus_get_intrspec (OBSOLETE) */ 2568 NULL, /* bus_add_intrspec (OBSOLETE) */ 2569 NULL, /* bus_remove_intrspec (OBSOLETE) */ 2570 i_ddi_map_fault, /* bus_map_fault */ 2571 NULL, /* bus_dma_map (OBSOLETE) */ 2572 ddi_dma_allochdl, /* bus_dma_allochdl */ 2573 ddi_dma_freehdl, /* bus_dma_freehdl */ 2574 ddi_dma_bindhdl, /* bus_dma_bindhdl */ 2575 ddi_dma_unbindhdl, /* bus_dma_unbindhdl */ 2576 ddi_dma_flush, /* bus_dma_flush */ 2577 ddi_dma_win, /* bus_dma_win */ 2578 ddi_dma_mctl, /* bus_dma_ctl */ 2579 bd_bus_ctl, /* bus_ctl */ 2580 ddi_bus_prop_op, /* bus_prop_op */ 2581 NULL, /* bus_get_eventcookie */ 2582 NULL, /* bus_add_eventcall */ 2583 NULL, /* bus_remove_eventcall */ 2584 NULL, /* bus_post_event */ 2585 NULL, /* bus_intr_ctl (OBSOLETE) */ 2586 NULL, /* bus_config */ 2587 NULL, /* bus_unconfig */ 2588 NULL, /* bus_fm_init */ 2589 NULL, /* bus_fm_fini */ 2590 NULL, /* bus_fm_access_enter */ 2591 NULL, /* bus_fm_access_exit */ 2592 NULL, /* bus_power */ 2593 NULL, /* bus_intr_op */ 2594 }; 2595 2596 devops->devo_bus_ops = &bd_bus_ops; 2597 2598 /* 2599 * NB: The device driver is free to supply its own 2600 * character entry device support. 2601 */ 2602 } 2603 2604 void 2605 bd_mod_fini(struct dev_ops *devops) 2606 { 2607 devops->devo_bus_ops = NULL; 2608 } 2609