1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2003 Silicon Graphics International Corp. 5 * Copyright (c) 2009-2011 Spectra Logic Corporation 6 * Copyright (c) 2012,2021 The FreeBSD Foundation 7 * Copyright (c) 2014-2021 Alexander Motin <mav@FreeBSD.org> 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Edward Tomasz Napierala 11 * under sponsorship from the FreeBSD Foundation. 12 * 13 * Portions of this software were developed by Ka Ho Ng <khng@FreeBSD.org> 14 * under sponsorship from the FreeBSD Foundation. 15 * 16 * Redistribution and use in source and binary forms, with or without 17 * modification, are permitted provided that the following conditions 18 * are met: 19 * 1. Redistributions of source code must retain the above copyright 20 * notice, this list of conditions, and the following disclaimer, 21 * without modification. 22 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 23 * substantially similar to the "NO WARRANTY" disclaimer below 24 * ("Disclaimer") and any redistribution must be conditioned upon 25 * including a substantially similar Disclaimer requirement for further 26 * binary redistribution. 27 * 28 * NO WARRANTY 29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 33 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 37 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 38 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 39 * POSSIBILITY OF SUCH DAMAGES. 40 * 41 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 42 */ 43 /* 44 * CAM Target Layer driver backend for block devices. 45 * 46 * Author: Ken Merry <ken@FreeBSD.org> 47 */ 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/kernel.h> 51 #include <sys/types.h> 52 #include <sys/kthread.h> 53 #include <sys/bio.h> 54 #include <sys/fcntl.h> 55 #include <sys/limits.h> 56 #include <sys/lock.h> 57 #include <sys/mutex.h> 58 #include <sys/condvar.h> 59 #include <sys/malloc.h> 60 #include <sys/conf.h> 61 #include <sys/ioccom.h> 62 #include <sys/queue.h> 63 #include <sys/sbuf.h> 64 #include <sys/endian.h> 65 #include <sys/uio.h> 66 #include <sys/buf.h> 67 #include <sys/taskqueue.h> 68 #include <sys/vnode.h> 69 #include <sys/namei.h> 70 #include <sys/mount.h> 71 #include <sys/disk.h> 72 #include <sys/fcntl.h> 73 #include <sys/filedesc.h> 74 #include <sys/filio.h> 75 #include <sys/proc.h> 76 #include <sys/pcpu.h> 77 #include <sys/module.h> 78 #include <sys/sdt.h> 79 #include <sys/devicestat.h> 80 #include <sys/sysctl.h> 81 #include <sys/nv.h> 82 #include <sys/dnv.h> 83 #include <sys/sx.h> 84 #include <sys/unistd.h> 85 86 #include <geom/geom.h> 87 88 #include <cam/cam.h> 89 #include <cam/scsi/scsi_all.h> 90 #include <cam/scsi/scsi_da.h> 91 #include <cam/ctl/ctl_io.h> 92 #include <cam/ctl/ctl.h> 93 #include <cam/ctl/ctl_backend.h> 94 #include <cam/ctl/ctl_ioctl.h> 95 #include <cam/ctl/ctl_ha.h> 96 #include <cam/ctl/ctl_scsi_all.h> 97 #include <cam/ctl/ctl_private.h> 98 #include <cam/ctl/ctl_error.h> 99 100 /* 101 * The idea here is to allocate enough S/G space to handle at least 1MB I/Os. 102 * On systems with small maxphys it can be 8 128KB segments. On large systems 103 * it can be up to 8 1MB segments. I/Os larger than that we'll split. 104 */ 105 #define CTLBLK_MAX_SEGS 8 106 #define CTLBLK_HALF_SEGS (CTLBLK_MAX_SEGS / 2) 107 #define CTLBLK_MIN_SEG (128 * 1024) 108 #define CTLBLK_MAX_SEG MIN(1024 * 1024, MAX(CTLBLK_MIN_SEG, maxphys)) 109 #define CTLBLK_MAX_IO_SIZE (CTLBLK_MAX_SEG * CTLBLK_MAX_SEGS) 110 111 #ifdef CTLBLK_DEBUG 112 #define DPRINTF(fmt, args...) \ 113 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 114 #else 115 #define DPRINTF(fmt, args...) do {} while(0) 116 #endif 117 118 #define PRIV(io) \ 119 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 120 #define ARGS(io) \ 121 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 122 123 SDT_PROVIDER_DEFINE(cbb); 124 125 typedef enum { 126 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 127 CTL_BE_BLOCK_LUN_WAITING = 0x04, 128 } ctl_be_block_lun_flags; 129 130 typedef enum { 131 CTL_BE_BLOCK_NONE, 132 CTL_BE_BLOCK_DEV, 133 CTL_BE_BLOCK_FILE 134 } ctl_be_block_type; 135 136 struct ctl_be_block_filedata { 137 struct ucred *cred; 138 }; 139 140 union ctl_be_block_bedata { 141 struct ctl_be_block_filedata file; 142 }; 143 144 struct ctl_be_block_io; 145 struct ctl_be_block_lun; 146 147 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 148 struct ctl_be_block_io *beio); 149 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 150 const char *attrname); 151 152 /* 153 * Backend LUN structure. There is a 1:1 mapping between a block device 154 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 155 */ 156 struct ctl_be_block_lun { 157 struct ctl_be_lun cbe_lun; /* Must be first element. */ 158 struct ctl_lun_create_params params; 159 char *dev_path; 160 ctl_be_block_type dev_type; 161 struct vnode *vn; 162 union ctl_be_block_bedata backend; 163 cbb_dispatch_t dispatch; 164 cbb_dispatch_t lun_flush; 165 cbb_dispatch_t unmap; 166 cbb_dispatch_t get_lba_status; 167 cbb_getattr_t getattr; 168 uint64_t size_blocks; 169 uint64_t size_bytes; 170 struct ctl_be_block_softc *softc; 171 struct devstat *disk_stats; 172 ctl_be_block_lun_flags flags; 173 SLIST_ENTRY(ctl_be_block_lun) links; 174 struct taskqueue *io_taskqueue; 175 struct task io_task; 176 int num_threads; 177 STAILQ_HEAD(, ctl_io_hdr) input_queue; 178 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 179 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 180 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 181 struct mtx_padalign io_lock; 182 struct mtx_padalign queue_lock; 183 }; 184 185 /* 186 * Overall softc structure for the block backend module. 187 */ 188 struct ctl_be_block_softc { 189 struct sx modify_lock; 190 struct mtx lock; 191 int num_luns; 192 SLIST_HEAD(, ctl_be_block_lun) lun_list; 193 uma_zone_t beio_zone; 194 uma_zone_t bufmin_zone; 195 uma_zone_t bufmax_zone; 196 }; 197 198 static struct ctl_be_block_softc backend_block_softc; 199 200 /* 201 * Per-I/O information. 202 */ 203 struct ctl_be_block_io { 204 union ctl_io *io; 205 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 206 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 207 int refcnt; 208 int bio_cmd; 209 int two_sglists; 210 int num_segs; 211 int num_bios_sent; 212 int num_bios_done; 213 int send_complete; 214 int first_error; 215 uint64_t first_error_offset; 216 struct bintime ds_t0; 217 devstat_tag_type ds_tag_type; 218 devstat_trans_flags ds_trans_type; 219 uint64_t io_len; 220 uint64_t io_offset; 221 int io_arg; 222 struct ctl_be_block_softc *softc; 223 struct ctl_be_block_lun *lun; 224 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 225 }; 226 227 extern struct ctl_softc *control_softc; 228 229 static int cbb_num_threads = 32; 230 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 231 "CAM Target Layer Block Backend"); 232 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 233 &cbb_num_threads, 0, "Number of threads per backing file"); 234 235 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 236 static void ctl_free_beio(struct ctl_be_block_io *beio); 237 static void ctl_complete_beio(struct ctl_be_block_io *beio); 238 static int ctl_be_block_move_done(union ctl_io *io, bool samethr); 239 static void ctl_be_block_biodone(struct bio *bio); 240 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 241 struct ctl_be_block_io *beio); 242 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 243 struct ctl_be_block_io *beio); 244 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 245 struct ctl_be_block_io *beio); 246 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 247 const char *attrname); 248 static void ctl_be_block_unmap_file(struct ctl_be_block_lun *be_lun, 249 struct ctl_be_block_io *beio); 250 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 251 struct ctl_be_block_io *beio); 252 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 253 struct ctl_be_block_io *beio); 254 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 255 struct ctl_be_block_io *beio); 256 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 257 const char *attrname); 258 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 259 union ctl_io *io); 260 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 261 union ctl_io *io); 262 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 263 union ctl_io *io); 264 static void ctl_be_block_worker(void *context, int pending); 265 static int ctl_be_block_submit(union ctl_io *io); 266 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 267 int flag, struct thread *td); 268 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 269 struct ctl_lun_req *req); 270 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 271 struct ctl_lun_req *req); 272 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 273 static int ctl_be_block_open(struct ctl_be_block_lun *be_lun, 274 struct ctl_lun_req *req); 275 static int ctl_be_block_create(struct ctl_be_block_softc *softc, 276 struct ctl_lun_req *req); 277 static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 278 struct ctl_lun_req *req); 279 static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 280 struct ctl_lun_req *req); 281 static void ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun); 282 static int ctl_be_block_config_write(union ctl_io *io); 283 static int ctl_be_block_config_read(union ctl_io *io); 284 static int ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb); 285 static uint64_t ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname); 286 static int ctl_be_block_init(void); 287 static int ctl_be_block_shutdown(void); 288 289 static struct ctl_backend_driver ctl_be_block_driver = 290 { 291 .name = "block", 292 .flags = CTL_BE_FLAG_HAS_CONFIG, 293 .init = ctl_be_block_init, 294 .shutdown = ctl_be_block_shutdown, 295 .data_submit = ctl_be_block_submit, 296 .config_read = ctl_be_block_config_read, 297 .config_write = ctl_be_block_config_write, 298 .ioctl = ctl_be_block_ioctl, 299 .lun_info = ctl_be_block_lun_info, 300 .lun_attr = ctl_be_block_lun_attr 301 }; 302 303 MALLOC_DEFINE(M_CTLBLK, "ctlblock", "Memory used for CTL block backend"); 304 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 305 306 static void 307 ctl_alloc_seg(struct ctl_be_block_softc *softc, struct ctl_sg_entry *sg, 308 size_t len) 309 { 310 311 if (len <= CTLBLK_MIN_SEG) { 312 sg->addr = uma_zalloc(softc->bufmin_zone, M_WAITOK); 313 } else { 314 KASSERT(len <= CTLBLK_MAX_SEG, 315 ("Too large alloc %zu > %lu", len, CTLBLK_MAX_SEG)); 316 sg->addr = uma_zalloc(softc->bufmax_zone, M_WAITOK); 317 } 318 sg->len = len; 319 } 320 321 static void 322 ctl_free_seg(struct ctl_be_block_softc *softc, struct ctl_sg_entry *sg) 323 { 324 325 if (sg->len <= CTLBLK_MIN_SEG) { 326 uma_zfree(softc->bufmin_zone, sg->addr); 327 } else { 328 KASSERT(sg->len <= CTLBLK_MAX_SEG, 329 ("Too large free %zu > %lu", sg->len, CTLBLK_MAX_SEG)); 330 uma_zfree(softc->bufmax_zone, sg->addr); 331 } 332 } 333 334 static struct ctl_be_block_io * 335 ctl_alloc_beio(struct ctl_be_block_softc *softc) 336 { 337 struct ctl_be_block_io *beio; 338 339 beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO); 340 beio->softc = softc; 341 beio->refcnt = 1; 342 return (beio); 343 } 344 345 static void 346 ctl_real_free_beio(struct ctl_be_block_io *beio) 347 { 348 struct ctl_be_block_softc *softc = beio->softc; 349 int i; 350 351 for (i = 0; i < beio->num_segs; i++) { 352 ctl_free_seg(softc, &beio->sg_segs[i]); 353 354 /* For compare we had two equal S/G lists. */ 355 if (beio->two_sglists) { 356 ctl_free_seg(softc, 357 &beio->sg_segs[i + CTLBLK_HALF_SEGS]); 358 } 359 } 360 361 uma_zfree(softc->beio_zone, beio); 362 } 363 364 static void 365 ctl_refcnt_beio(void *arg, int diff) 366 { 367 struct ctl_be_block_io *beio = arg; 368 369 if (atomic_fetchadd_int(&beio->refcnt, diff) + diff == 0) 370 ctl_real_free_beio(beio); 371 } 372 373 static void 374 ctl_free_beio(struct ctl_be_block_io *beio) 375 { 376 377 ctl_refcnt_beio(beio, -1); 378 } 379 380 static void 381 ctl_complete_beio(struct ctl_be_block_io *beio) 382 { 383 union ctl_io *io = beio->io; 384 385 if (beio->beio_cont != NULL) { 386 beio->beio_cont(beio); 387 } else { 388 ctl_free_beio(beio); 389 ctl_data_submit_done(io); 390 } 391 } 392 393 static size_t 394 cmp(uint8_t *a, uint8_t *b, size_t size) 395 { 396 size_t i; 397 398 for (i = 0; i < size; i++) { 399 if (a[i] != b[i]) 400 break; 401 } 402 return (i); 403 } 404 405 static void 406 ctl_be_block_compare(union ctl_io *io) 407 { 408 struct ctl_be_block_io *beio; 409 uint64_t off, res; 410 int i; 411 uint8_t info[8]; 412 413 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 414 off = 0; 415 for (i = 0; i < beio->num_segs; i++) { 416 res = cmp(beio->sg_segs[i].addr, 417 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 418 beio->sg_segs[i].len); 419 off += res; 420 if (res < beio->sg_segs[i].len) 421 break; 422 } 423 if (i < beio->num_segs) { 424 scsi_u64to8b(off, info); 425 ctl_set_sense(&io->scsiio, /*current_error*/ 1, 426 /*sense_key*/ SSD_KEY_MISCOMPARE, 427 /*asc*/ 0x1D, /*ascq*/ 0x00, 428 /*type*/ SSD_ELEM_INFO, 429 /*size*/ sizeof(info), /*data*/ &info, 430 /*type*/ SSD_ELEM_NONE); 431 } else 432 ctl_set_success(&io->scsiio); 433 } 434 435 static int 436 ctl_be_block_move_done(union ctl_io *io, bool samethr) 437 { 438 struct ctl_be_block_io *beio; 439 struct ctl_be_block_lun *be_lun; 440 struct ctl_lba_len_flags *lbalen; 441 442 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 443 444 DPRINTF("entered\n"); 445 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 446 447 /* 448 * We set status at this point for read and compare commands. 449 */ 450 if ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0 && 451 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE) { 452 lbalen = ARGS(io); 453 if (lbalen->flags & CTL_LLF_READ) { 454 ctl_set_success(&io->scsiio); 455 } else if (lbalen->flags & CTL_LLF_COMPARE) { 456 /* We have two data blocks ready for comparison. */ 457 ctl_be_block_compare(io); 458 } 459 } 460 461 /* 462 * If this is a read, or a write with errors, it is done. 463 */ 464 if ((beio->bio_cmd == BIO_READ) 465 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 466 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 467 ctl_complete_beio(beio); 468 return (0); 469 } 470 471 /* 472 * At this point, we have a write and the DMA completed successfully. 473 * If we were called synchronously in the original thread then just 474 * dispatch, otherwise we now have to queue it to the task queue to 475 * execute the backend I/O. That is because we do blocking 476 * memory allocations, and in the file backing case, blocking I/O. 477 * This move done routine is generally called in the SIM's 478 * interrupt context, and therefore we cannot block. 479 */ 480 be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io); 481 if (samethr) { 482 be_lun->dispatch(be_lun, beio); 483 } else { 484 mtx_lock(&be_lun->queue_lock); 485 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 486 mtx_unlock(&be_lun->queue_lock); 487 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 488 } 489 return (0); 490 } 491 492 static void 493 ctl_be_block_biodone(struct bio *bio) 494 { 495 struct ctl_be_block_io *beio = bio->bio_caller1; 496 struct ctl_be_block_lun *be_lun = beio->lun; 497 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 498 union ctl_io *io; 499 int error; 500 501 io = beio->io; 502 503 DPRINTF("entered\n"); 504 505 error = bio->bio_error; 506 mtx_lock(&be_lun->io_lock); 507 if (error != 0 && 508 (beio->first_error == 0 || 509 bio->bio_offset < beio->first_error_offset)) { 510 beio->first_error = error; 511 beio->first_error_offset = bio->bio_offset; 512 } 513 514 beio->num_bios_done++; 515 516 /* 517 * XXX KDM will this cause WITNESS to complain? Holding a lock 518 * during the free might cause it to complain. 519 */ 520 g_destroy_bio(bio); 521 522 /* 523 * If the send complete bit isn't set, or we aren't the last I/O to 524 * complete, then we're done. 525 */ 526 if ((beio->send_complete == 0) 527 || (beio->num_bios_done < beio->num_bios_sent)) { 528 mtx_unlock(&be_lun->io_lock); 529 return; 530 } 531 532 /* 533 * At this point, we've verified that we are the last I/O to 534 * complete, so it's safe to drop the lock. 535 */ 536 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 537 beio->ds_tag_type, beio->ds_trans_type, 538 /*now*/ NULL, /*then*/&beio->ds_t0); 539 mtx_unlock(&be_lun->io_lock); 540 541 /* 542 * If there are any errors from the backing device, we fail the 543 * entire I/O with a medium error. 544 */ 545 error = beio->first_error; 546 if (error != 0) { 547 if (error == EOPNOTSUPP) { 548 ctl_set_invalid_opcode(&io->scsiio); 549 } else if (error == ENOSPC || error == EDQUOT) { 550 ctl_set_space_alloc_fail(&io->scsiio); 551 } else if (error == EROFS || error == EACCES) { 552 ctl_set_hw_write_protected(&io->scsiio); 553 } else if (beio->bio_cmd == BIO_FLUSH) { 554 /* XXX KDM is there is a better error here? */ 555 ctl_set_internal_failure(&io->scsiio, 556 /*sks_valid*/ 1, 557 /*retry_count*/ 0xbad2); 558 } else { 559 ctl_set_medium_error(&io->scsiio, 560 beio->bio_cmd == BIO_READ); 561 } 562 ctl_complete_beio(beio); 563 return; 564 } 565 566 /* 567 * If this is a write, a flush, a delete or verify, we're all done. 568 * If this is a read, we can now send the data to the user. 569 */ 570 if ((beio->bio_cmd == BIO_WRITE) 571 || (beio->bio_cmd == BIO_FLUSH) 572 || (beio->bio_cmd == BIO_DELETE) 573 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 574 ctl_set_success(&io->scsiio); 575 ctl_complete_beio(beio); 576 } else { 577 if ((ARGS(io)->flags & CTL_LLF_READ) && 578 beio->beio_cont == NULL) { 579 ctl_set_success(&io->scsiio); 580 if (cbe_lun->serseq >= CTL_LUN_SERSEQ_SOFT) 581 ctl_serseq_done(io); 582 } 583 ctl_datamove(io); 584 } 585 } 586 587 static void 588 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 589 struct ctl_be_block_io *beio) 590 { 591 union ctl_io *io = beio->io; 592 struct mount *mountpoint; 593 int error; 594 595 DPRINTF("entered\n"); 596 597 binuptime(&beio->ds_t0); 598 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 599 600 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 601 602 vn_lock(be_lun->vn, vn_lktype_write(mountpoint, be_lun->vn) | 603 LK_RETRY); 604 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 605 curthread); 606 VOP_UNLOCK(be_lun->vn); 607 608 vn_finished_write(mountpoint); 609 610 mtx_lock(&be_lun->io_lock); 611 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 612 beio->ds_tag_type, beio->ds_trans_type, 613 /*now*/ NULL, /*then*/&beio->ds_t0); 614 mtx_unlock(&be_lun->io_lock); 615 616 if (error == 0) 617 ctl_set_success(&io->scsiio); 618 else { 619 /* XXX KDM is there is a better error here? */ 620 ctl_set_internal_failure(&io->scsiio, 621 /*sks_valid*/ 1, 622 /*retry_count*/ 0xbad1); 623 } 624 625 ctl_complete_beio(beio); 626 } 627 628 SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t"); 629 SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t"); 630 SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t"); 631 SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t"); 632 633 static void 634 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 635 struct ctl_be_block_io *beio) 636 { 637 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 638 struct ctl_be_block_filedata *file_data; 639 union ctl_io *io; 640 struct uio xuio; 641 struct iovec *xiovec; 642 size_t s; 643 int error, flags, i; 644 645 DPRINTF("entered\n"); 646 647 file_data = &be_lun->backend.file; 648 io = beio->io; 649 flags = 0; 650 if (ARGS(io)->flags & CTL_LLF_DPO) 651 flags |= IO_DIRECT; 652 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 653 flags |= IO_SYNC; 654 655 bzero(&xuio, sizeof(xuio)); 656 if (beio->bio_cmd == BIO_READ) { 657 SDT_PROBE0(cbb, , read, file_start); 658 xuio.uio_rw = UIO_READ; 659 } else { 660 SDT_PROBE0(cbb, , write, file_start); 661 xuio.uio_rw = UIO_WRITE; 662 } 663 xuio.uio_offset = beio->io_offset; 664 xuio.uio_resid = beio->io_len; 665 xuio.uio_segflg = UIO_SYSSPACE; 666 xuio.uio_iov = beio->xiovecs; 667 xuio.uio_iovcnt = beio->num_segs; 668 xuio.uio_td = curthread; 669 670 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 671 xiovec->iov_base = beio->sg_segs[i].addr; 672 xiovec->iov_len = beio->sg_segs[i].len; 673 } 674 675 binuptime(&beio->ds_t0); 676 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 677 678 if (beio->bio_cmd == BIO_READ) { 679 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 680 681 if (beio->beio_cont == NULL && 682 cbe_lun->serseq == CTL_LUN_SERSEQ_SOFT) 683 ctl_serseq_done(io); 684 /* 685 * UFS pays attention to IO_DIRECT for reads. If the 686 * DIRECTIO option is configured into the kernel, it calls 687 * ffs_rawread(). But that only works for single-segment 688 * uios with user space addresses. In our case, with a 689 * kernel uio, it still reads into the buffer cache, but it 690 * will just try to release the buffer from the cache later 691 * on in ffs_read(). 692 * 693 * ZFS does not pay attention to IO_DIRECT for reads. 694 * 695 * UFS does not pay attention to IO_SYNC for reads. 696 * 697 * ZFS pays attention to IO_SYNC (which translates into the 698 * Solaris define FRSYNC for zfs_read()) for reads. It 699 * attempts to sync the file before reading. 700 */ 701 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 702 703 VOP_UNLOCK(be_lun->vn); 704 SDT_PROBE0(cbb, , read, file_done); 705 if (error == 0 && xuio.uio_resid > 0) { 706 /* 707 * If we red less then requested (EOF), then 708 * we should clean the rest of the buffer. 709 */ 710 s = beio->io_len - xuio.uio_resid; 711 for (i = 0; i < beio->num_segs; i++) { 712 if (s >= beio->sg_segs[i].len) { 713 s -= beio->sg_segs[i].len; 714 continue; 715 } 716 bzero((uint8_t *)beio->sg_segs[i].addr + s, 717 beio->sg_segs[i].len - s); 718 s = 0; 719 } 720 } 721 } else { 722 struct mount *mountpoint; 723 724 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 725 vn_lock(be_lun->vn, vn_lktype_write(mountpoint, 726 be_lun->vn) | LK_RETRY); 727 728 /* 729 * UFS pays attention to IO_DIRECT for writes. The write 730 * is done asynchronously. (Normally the write would just 731 * get put into cache. 732 * 733 * UFS pays attention to IO_SYNC for writes. It will 734 * attempt to write the buffer out synchronously if that 735 * flag is set. 736 * 737 * ZFS does not pay attention to IO_DIRECT for writes. 738 * 739 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 740 * for writes. It will flush the transaction from the 741 * cache before returning. 742 */ 743 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 744 VOP_UNLOCK(be_lun->vn); 745 746 vn_finished_write(mountpoint); 747 SDT_PROBE0(cbb, , write, file_done); 748 } 749 750 mtx_lock(&be_lun->io_lock); 751 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 752 beio->ds_tag_type, beio->ds_trans_type, 753 /*now*/ NULL, /*then*/&beio->ds_t0); 754 mtx_unlock(&be_lun->io_lock); 755 756 /* 757 * If we got an error, set the sense data to "MEDIUM ERROR" and 758 * return the I/O to the user. 759 */ 760 if (error != 0) { 761 if (error == ENOSPC || error == EDQUOT) { 762 ctl_set_space_alloc_fail(&io->scsiio); 763 } else if (error == EROFS || error == EACCES) { 764 ctl_set_hw_write_protected(&io->scsiio); 765 } else { 766 ctl_set_medium_error(&io->scsiio, 767 beio->bio_cmd == BIO_READ); 768 } 769 ctl_complete_beio(beio); 770 return; 771 } 772 773 /* 774 * If this is a write or a verify, we're all done. 775 * If this is a read, we can now send the data to the user. 776 */ 777 if ((beio->bio_cmd == BIO_WRITE) || 778 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 779 ctl_set_success(&io->scsiio); 780 ctl_complete_beio(beio); 781 } else { 782 if ((ARGS(io)->flags & CTL_LLF_READ) && 783 beio->beio_cont == NULL) { 784 ctl_set_success(&io->scsiio); 785 if (cbe_lun->serseq > CTL_LUN_SERSEQ_SOFT) 786 ctl_serseq_done(io); 787 } 788 ctl_datamove(io); 789 } 790 } 791 792 static void 793 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 794 struct ctl_be_block_io *beio) 795 { 796 union ctl_io *io = beio->io; 797 struct ctl_lba_len_flags *lbalen = ARGS(io); 798 struct scsi_get_lba_status_data *data; 799 off_t roff, off; 800 int error, status; 801 802 DPRINTF("entered\n"); 803 804 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 805 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 806 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 807 0, curthread->td_ucred, curthread); 808 if (error == 0 && off > roff) 809 status = 0; /* mapped up to off */ 810 else { 811 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 812 0, curthread->td_ucred, curthread); 813 if (error == 0 && off > roff) 814 status = 1; /* deallocated up to off */ 815 else { 816 status = 0; /* unknown up to the end */ 817 off = be_lun->size_bytes; 818 } 819 } 820 VOP_UNLOCK(be_lun->vn); 821 822 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 823 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 824 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 825 lbalen->lba), data->descr[0].length); 826 data->descr[0].status = status; 827 828 ctl_complete_beio(beio); 829 } 830 831 static uint64_t 832 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 833 { 834 struct vattr vattr; 835 struct statfs statfs; 836 uint64_t val; 837 int error; 838 839 val = UINT64_MAX; 840 if (be_lun->vn == NULL) 841 return (val); 842 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 843 if (strcmp(attrname, "blocksused") == 0) { 844 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 845 if (error == 0) 846 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 847 } 848 if (strcmp(attrname, "blocksavail") == 0 && 849 !VN_IS_DOOMED(be_lun->vn)) { 850 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 851 if (error == 0) 852 val = statfs.f_bavail * statfs.f_bsize / 853 be_lun->cbe_lun.blocksize; 854 } 855 VOP_UNLOCK(be_lun->vn); 856 return (val); 857 } 858 859 static void 860 ctl_be_block_unmap_file(struct ctl_be_block_lun *be_lun, 861 struct ctl_be_block_io *beio) 862 { 863 struct ctl_be_block_filedata *file_data; 864 union ctl_io *io; 865 struct ctl_ptr_len_flags *ptrlen; 866 struct scsi_unmap_desc *buf, *end; 867 struct mount *mp; 868 off_t off, len; 869 int error; 870 871 io = beio->io; 872 file_data = &be_lun->backend.file; 873 mp = NULL; 874 error = 0; 875 876 binuptime(&beio->ds_t0); 877 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 878 879 (void)vn_start_write(be_lun->vn, &mp, V_WAIT); 880 vn_lock(be_lun->vn, vn_lktype_write(mp, be_lun->vn) | LK_RETRY); 881 if (beio->io_offset == -1) { 882 beio->io_len = 0; 883 ptrlen = (struct ctl_ptr_len_flags *) 884 &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 885 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 886 end = buf + ptrlen->len / sizeof(*buf); 887 for (; buf < end; buf++) { 888 off = (off_t)scsi_8btou64(buf->lba) * 889 be_lun->cbe_lun.blocksize; 890 len = (off_t)scsi_4btoul(buf->length) * 891 be_lun->cbe_lun.blocksize; 892 beio->io_len += len; 893 error = vn_deallocate(be_lun->vn, &off, &len, 894 0, IO_NOMACCHECK | IO_NODELOCKED, file_data->cred, 895 NOCRED); 896 if (error != 0) 897 break; 898 } 899 } else { 900 /* WRITE_SAME */ 901 off = beio->io_offset; 902 len = beio->io_len; 903 error = vn_deallocate(be_lun->vn, &off, &len, 0, 904 IO_NOMACCHECK | IO_NODELOCKED, file_data->cred, NOCRED); 905 } 906 VOP_UNLOCK(be_lun->vn); 907 vn_finished_write(mp); 908 909 mtx_lock(&be_lun->io_lock); 910 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 911 beio->ds_tag_type, beio->ds_trans_type, 912 /*now*/ NULL, /*then*/&beio->ds_t0); 913 mtx_unlock(&be_lun->io_lock); 914 915 /* 916 * If we got an error, set the sense data to "MEDIUM ERROR" and 917 * return the I/O to the user. 918 */ 919 switch (error) { 920 case 0: 921 ctl_set_success(&io->scsiio); 922 break; 923 case ENOSPC: 924 case EDQUOT: 925 ctl_set_space_alloc_fail(&io->scsiio); 926 break; 927 case EROFS: 928 case EACCES: 929 ctl_set_hw_write_protected(&io->scsiio); 930 break; 931 default: 932 ctl_set_medium_error(&io->scsiio, false); 933 } 934 ctl_complete_beio(beio); 935 } 936 937 static void 938 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 939 struct ctl_be_block_io *beio) 940 { 941 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 942 union ctl_io *io; 943 struct cdevsw *csw; 944 struct cdev *dev; 945 struct uio xuio; 946 struct iovec *xiovec; 947 int error, flags, i, ref; 948 949 DPRINTF("entered\n"); 950 951 io = beio->io; 952 flags = 0; 953 if (ARGS(io)->flags & CTL_LLF_DPO) 954 flags |= IO_DIRECT; 955 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 956 flags |= IO_SYNC; 957 958 bzero(&xuio, sizeof(xuio)); 959 if (beio->bio_cmd == BIO_READ) { 960 SDT_PROBE0(cbb, , read, file_start); 961 xuio.uio_rw = UIO_READ; 962 } else { 963 SDT_PROBE0(cbb, , write, file_start); 964 xuio.uio_rw = UIO_WRITE; 965 } 966 xuio.uio_offset = beio->io_offset; 967 xuio.uio_resid = beio->io_len; 968 xuio.uio_segflg = UIO_SYSSPACE; 969 xuio.uio_iov = beio->xiovecs; 970 xuio.uio_iovcnt = beio->num_segs; 971 xuio.uio_td = curthread; 972 973 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 974 xiovec->iov_base = beio->sg_segs[i].addr; 975 xiovec->iov_len = beio->sg_segs[i].len; 976 } 977 978 binuptime(&beio->ds_t0); 979 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 980 981 csw = devvn_refthread(be_lun->vn, &dev, &ref); 982 if (csw) { 983 if (beio->bio_cmd == BIO_READ) { 984 if (beio->beio_cont == NULL && 985 cbe_lun->serseq == CTL_LUN_SERSEQ_SOFT) 986 ctl_serseq_done(io); 987 error = csw->d_read(dev, &xuio, flags); 988 } else 989 error = csw->d_write(dev, &xuio, flags); 990 dev_relthread(dev, ref); 991 } else 992 error = ENXIO; 993 994 if (beio->bio_cmd == BIO_READ) 995 SDT_PROBE0(cbb, , read, file_done); 996 else 997 SDT_PROBE0(cbb, , write, file_done); 998 999 mtx_lock(&be_lun->io_lock); 1000 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 1001 beio->ds_tag_type, beio->ds_trans_type, 1002 /*now*/ NULL, /*then*/&beio->ds_t0); 1003 mtx_unlock(&be_lun->io_lock); 1004 1005 /* 1006 * If we got an error, set the sense data to "MEDIUM ERROR" and 1007 * return the I/O to the user. 1008 */ 1009 if (error != 0) { 1010 if (error == ENOSPC || error == EDQUOT) { 1011 ctl_set_space_alloc_fail(&io->scsiio); 1012 } else if (error == EROFS || error == EACCES) { 1013 ctl_set_hw_write_protected(&io->scsiio); 1014 } else { 1015 ctl_set_medium_error(&io->scsiio, 1016 beio->bio_cmd == BIO_READ); 1017 } 1018 ctl_complete_beio(beio); 1019 return; 1020 } 1021 1022 /* 1023 * If this is a write or a verify, we're all done. 1024 * If this is a read, we can now send the data to the user. 1025 */ 1026 if ((beio->bio_cmd == BIO_WRITE) || 1027 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 1028 ctl_set_success(&io->scsiio); 1029 ctl_complete_beio(beio); 1030 } else { 1031 if ((ARGS(io)->flags & CTL_LLF_READ) && 1032 beio->beio_cont == NULL) { 1033 ctl_set_success(&io->scsiio); 1034 if (cbe_lun->serseq > CTL_LUN_SERSEQ_SOFT) 1035 ctl_serseq_done(io); 1036 } 1037 ctl_datamove(io); 1038 } 1039 } 1040 1041 static void 1042 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 1043 struct ctl_be_block_io *beio) 1044 { 1045 union ctl_io *io = beio->io; 1046 struct cdevsw *csw; 1047 struct cdev *dev; 1048 struct ctl_lba_len_flags *lbalen = ARGS(io); 1049 struct scsi_get_lba_status_data *data; 1050 off_t roff, off; 1051 int error, ref, status; 1052 1053 DPRINTF("entered\n"); 1054 1055 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1056 if (csw == NULL) { 1057 status = 0; /* unknown up to the end */ 1058 off = be_lun->size_bytes; 1059 goto done; 1060 } 1061 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 1062 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD, 1063 curthread); 1064 if (error == 0 && off > roff) 1065 status = 0; /* mapped up to off */ 1066 else { 1067 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD, 1068 curthread); 1069 if (error == 0 && off > roff) 1070 status = 1; /* deallocated up to off */ 1071 else { 1072 status = 0; /* unknown up to the end */ 1073 off = be_lun->size_bytes; 1074 } 1075 } 1076 dev_relthread(dev, ref); 1077 1078 done: 1079 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 1080 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 1081 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 1082 lbalen->lba), data->descr[0].length); 1083 data->descr[0].status = status; 1084 1085 ctl_complete_beio(beio); 1086 } 1087 1088 static void 1089 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 1090 struct ctl_be_block_io *beio) 1091 { 1092 struct bio *bio; 1093 struct cdevsw *csw; 1094 struct cdev *dev; 1095 int ref; 1096 1097 DPRINTF("entered\n"); 1098 1099 /* This can't fail, it's a blocking allocation. */ 1100 bio = g_alloc_bio(); 1101 1102 bio->bio_cmd = BIO_FLUSH; 1103 bio->bio_offset = 0; 1104 bio->bio_data = 0; 1105 bio->bio_done = ctl_be_block_biodone; 1106 bio->bio_caller1 = beio; 1107 bio->bio_pblkno = 0; 1108 1109 /* 1110 * We don't need to acquire the LUN lock here, because we are only 1111 * sending one bio, and so there is no other context to synchronize 1112 * with. 1113 */ 1114 beio->num_bios_sent = 1; 1115 beio->send_complete = 1; 1116 1117 binuptime(&beio->ds_t0); 1118 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1119 1120 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1121 if (csw) { 1122 bio->bio_dev = dev; 1123 csw->d_strategy(bio); 1124 dev_relthread(dev, ref); 1125 } else { 1126 bio->bio_error = ENXIO; 1127 ctl_be_block_biodone(bio); 1128 } 1129 } 1130 1131 static void 1132 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1133 struct ctl_be_block_io *beio, 1134 uint64_t off, uint64_t len, int last) 1135 { 1136 struct bio *bio; 1137 uint64_t maxlen; 1138 struct cdevsw *csw; 1139 struct cdev *dev; 1140 int ref; 1141 1142 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1143 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1144 while (len > 0) { 1145 bio = g_alloc_bio(); 1146 bio->bio_cmd = BIO_DELETE; 1147 bio->bio_dev = dev; 1148 bio->bio_offset = off; 1149 bio->bio_length = MIN(len, maxlen); 1150 bio->bio_data = 0; 1151 bio->bio_done = ctl_be_block_biodone; 1152 bio->bio_caller1 = beio; 1153 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1154 1155 off += bio->bio_length; 1156 len -= bio->bio_length; 1157 1158 mtx_lock(&be_lun->io_lock); 1159 beio->num_bios_sent++; 1160 if (last && len == 0) 1161 beio->send_complete = 1; 1162 mtx_unlock(&be_lun->io_lock); 1163 1164 if (csw) { 1165 csw->d_strategy(bio); 1166 } else { 1167 bio->bio_error = ENXIO; 1168 ctl_be_block_biodone(bio); 1169 } 1170 } 1171 if (csw) 1172 dev_relthread(dev, ref); 1173 } 1174 1175 static void 1176 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1177 struct ctl_be_block_io *beio) 1178 { 1179 union ctl_io *io; 1180 struct ctl_ptr_len_flags *ptrlen; 1181 struct scsi_unmap_desc *buf, *end; 1182 uint64_t len; 1183 1184 io = beio->io; 1185 1186 DPRINTF("entered\n"); 1187 1188 binuptime(&beio->ds_t0); 1189 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1190 1191 if (beio->io_offset == -1) { 1192 beio->io_len = 0; 1193 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1194 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1195 end = buf + ptrlen->len / sizeof(*buf); 1196 for (; buf < end; buf++) { 1197 len = (uint64_t)scsi_4btoul(buf->length) * 1198 be_lun->cbe_lun.blocksize; 1199 beio->io_len += len; 1200 ctl_be_block_unmap_dev_range(be_lun, beio, 1201 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1202 len, (end - buf < 2) ? TRUE : FALSE); 1203 } 1204 } else 1205 ctl_be_block_unmap_dev_range(be_lun, beio, 1206 beio->io_offset, beio->io_len, TRUE); 1207 } 1208 1209 static void 1210 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1211 struct ctl_be_block_io *beio) 1212 { 1213 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1214 struct bio *bio; 1215 struct cdevsw *csw; 1216 struct cdev *dev; 1217 off_t cur_offset; 1218 int i, max_iosize, ref; 1219 1220 DPRINTF("entered\n"); 1221 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1222 1223 /* 1224 * We have to limit our I/O size to the maximum supported by the 1225 * backend device. 1226 */ 1227 if (csw) { 1228 max_iosize = dev->si_iosize_max; 1229 if (max_iosize <= 0) 1230 max_iosize = DFLTPHYS; 1231 } else 1232 max_iosize = maxphys; 1233 1234 cur_offset = beio->io_offset; 1235 for (i = 0; i < beio->num_segs; i++) { 1236 size_t cur_size; 1237 uint8_t *cur_ptr; 1238 1239 cur_size = beio->sg_segs[i].len; 1240 cur_ptr = beio->sg_segs[i].addr; 1241 1242 while (cur_size > 0) { 1243 /* This can't fail, it's a blocking allocation. */ 1244 bio = g_alloc_bio(); 1245 1246 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1247 1248 bio->bio_cmd = beio->bio_cmd; 1249 bio->bio_dev = dev; 1250 bio->bio_caller1 = beio; 1251 bio->bio_length = min(cur_size, max_iosize); 1252 bio->bio_offset = cur_offset; 1253 bio->bio_data = cur_ptr; 1254 bio->bio_done = ctl_be_block_biodone; 1255 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1256 1257 cur_offset += bio->bio_length; 1258 cur_ptr += bio->bio_length; 1259 cur_size -= bio->bio_length; 1260 1261 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1262 beio->num_bios_sent++; 1263 } 1264 } 1265 beio->send_complete = 1; 1266 binuptime(&beio->ds_t0); 1267 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1268 1269 /* 1270 * Fire off all allocated requests! 1271 */ 1272 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1273 TAILQ_REMOVE(&queue, bio, bio_queue); 1274 if (csw) 1275 csw->d_strategy(bio); 1276 else { 1277 bio->bio_error = ENXIO; 1278 ctl_be_block_biodone(bio); 1279 } 1280 } 1281 if (csw) 1282 dev_relthread(dev, ref); 1283 } 1284 1285 static uint64_t 1286 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1287 { 1288 struct diocgattr_arg arg; 1289 struct cdevsw *csw; 1290 struct cdev *dev; 1291 int error, ref; 1292 1293 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1294 if (csw == NULL) 1295 return (UINT64_MAX); 1296 strlcpy(arg.name, attrname, sizeof(arg.name)); 1297 arg.len = sizeof(arg.value.off); 1298 if (csw->d_ioctl) { 1299 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 1300 curthread); 1301 } else 1302 error = ENODEV; 1303 dev_relthread(dev, ref); 1304 if (error != 0) 1305 return (UINT64_MAX); 1306 return (arg.value.off); 1307 } 1308 1309 static void 1310 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1311 union ctl_io *io) 1312 { 1313 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1314 struct ctl_be_block_io *beio; 1315 struct ctl_lba_len_flags *lbalen; 1316 1317 DPRINTF("entered\n"); 1318 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1319 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1320 1321 beio->io_len = lbalen->len * cbe_lun->blocksize; 1322 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1323 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1324 beio->bio_cmd = BIO_FLUSH; 1325 beio->ds_trans_type = DEVSTAT_NO_DATA; 1326 DPRINTF("SYNC\n"); 1327 be_lun->lun_flush(be_lun, beio); 1328 } 1329 1330 static void 1331 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1332 { 1333 union ctl_io *io; 1334 1335 io = beio->io; 1336 ctl_free_beio(beio); 1337 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1338 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1339 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1340 ctl_config_write_done(io); 1341 return; 1342 } 1343 1344 ctl_be_block_config_write(io); 1345 } 1346 1347 static void 1348 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1349 union ctl_io *io) 1350 { 1351 struct ctl_be_block_softc *softc = be_lun->softc; 1352 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1353 struct ctl_be_block_io *beio; 1354 struct ctl_lba_len_flags *lbalen; 1355 uint64_t len_left, lba; 1356 uint32_t pb, pbo, adj; 1357 int i, seglen; 1358 uint8_t *buf, *end; 1359 1360 DPRINTF("entered\n"); 1361 1362 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1363 lbalen = ARGS(io); 1364 1365 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1366 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1367 ctl_free_beio(beio); 1368 ctl_set_invalid_field(&io->scsiio, 1369 /*sks_valid*/ 1, 1370 /*command*/ 1, 1371 /*field*/ 1, 1372 /*bit_valid*/ 0, 1373 /*bit*/ 0); 1374 ctl_config_write_done(io); 1375 return; 1376 } 1377 1378 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1379 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1380 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1381 beio->bio_cmd = BIO_DELETE; 1382 beio->ds_trans_type = DEVSTAT_FREE; 1383 1384 be_lun->unmap(be_lun, beio); 1385 return; 1386 } 1387 1388 beio->bio_cmd = BIO_WRITE; 1389 beio->ds_trans_type = DEVSTAT_WRITE; 1390 1391 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1392 (uintmax_t)lbalen->lba, lbalen->len); 1393 1394 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1395 if (be_lun->cbe_lun.pblockoff > 0) 1396 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1397 else 1398 pbo = 0; 1399 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1400 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1401 /* 1402 * Setup the S/G entry for this chunk. 1403 */ 1404 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1405 if (pb > cbe_lun->blocksize) { 1406 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1407 seglen - pbo) % pb; 1408 if (seglen > adj) 1409 seglen -= adj; 1410 else 1411 seglen -= seglen % cbe_lun->blocksize; 1412 } else 1413 seglen -= seglen % cbe_lun->blocksize; 1414 ctl_alloc_seg(softc, &beio->sg_segs[i], seglen); 1415 1416 DPRINTF("segment %d addr %p len %zd\n", i, 1417 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1418 1419 beio->num_segs++; 1420 len_left -= seglen; 1421 1422 buf = beio->sg_segs[i].addr; 1423 end = buf + seglen; 1424 for (; buf < end; buf += cbe_lun->blocksize) { 1425 if (lbalen->flags & SWS_NDOB) { 1426 memset(buf, 0, cbe_lun->blocksize); 1427 } else { 1428 memcpy(buf, io->scsiio.kern_data_ptr, 1429 cbe_lun->blocksize); 1430 } 1431 if (lbalen->flags & SWS_LBDATA) 1432 scsi_ulto4b(lbalen->lba + lba, buf); 1433 lba++; 1434 } 1435 } 1436 1437 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1438 beio->io_len = lba * cbe_lun->blocksize; 1439 1440 /* We can not do all in one run. Correct and schedule rerun. */ 1441 if (len_left > 0) { 1442 lbalen->lba += lba; 1443 lbalen->len -= lba; 1444 beio->beio_cont = ctl_be_block_cw_done_ws; 1445 } 1446 1447 be_lun->dispatch(be_lun, beio); 1448 } 1449 1450 static void 1451 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1452 union ctl_io *io) 1453 { 1454 struct ctl_be_block_io *beio; 1455 struct ctl_ptr_len_flags *ptrlen; 1456 1457 DPRINTF("entered\n"); 1458 1459 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1460 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1461 1462 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1463 ctl_free_beio(beio); 1464 ctl_set_invalid_field(&io->scsiio, 1465 /*sks_valid*/ 0, 1466 /*command*/ 1, 1467 /*field*/ 0, 1468 /*bit_valid*/ 0, 1469 /*bit*/ 0); 1470 ctl_config_write_done(io); 1471 return; 1472 } 1473 1474 beio->io_len = 0; 1475 beio->io_offset = -1; 1476 beio->bio_cmd = BIO_DELETE; 1477 beio->ds_trans_type = DEVSTAT_FREE; 1478 DPRINTF("UNMAP\n"); 1479 be_lun->unmap(be_lun, beio); 1480 } 1481 1482 static void 1483 ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1484 { 1485 union ctl_io *io; 1486 1487 io = beio->io; 1488 ctl_free_beio(beio); 1489 ctl_config_read_done(io); 1490 } 1491 1492 static void 1493 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1494 union ctl_io *io) 1495 { 1496 struct ctl_be_block_io *beio; 1497 struct ctl_be_block_softc *softc; 1498 1499 DPRINTF("entered\n"); 1500 1501 softc = be_lun->softc; 1502 beio = ctl_alloc_beio(softc); 1503 beio->io = io; 1504 beio->lun = be_lun; 1505 beio->beio_cont = ctl_be_block_cr_done; 1506 PRIV(io)->ptr = (void *)beio; 1507 1508 switch (io->scsiio.cdb[0]) { 1509 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1510 beio->bio_cmd = -1; 1511 beio->ds_trans_type = DEVSTAT_NO_DATA; 1512 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1513 beio->io_len = 0; 1514 if (be_lun->get_lba_status) 1515 be_lun->get_lba_status(be_lun, beio); 1516 else 1517 ctl_be_block_cr_done(beio); 1518 break; 1519 default: 1520 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1521 break; 1522 } 1523 } 1524 1525 static void 1526 ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1527 { 1528 union ctl_io *io; 1529 1530 io = beio->io; 1531 ctl_free_beio(beio); 1532 ctl_config_write_done(io); 1533 } 1534 1535 static void 1536 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1537 union ctl_io *io) 1538 { 1539 struct ctl_be_block_io *beio; 1540 struct ctl_be_block_softc *softc; 1541 1542 DPRINTF("entered\n"); 1543 1544 softc = be_lun->softc; 1545 beio = ctl_alloc_beio(softc); 1546 beio->io = io; 1547 beio->lun = be_lun; 1548 beio->beio_cont = ctl_be_block_cw_done; 1549 switch (io->scsiio.tag_type) { 1550 case CTL_TAG_ORDERED: 1551 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1552 break; 1553 case CTL_TAG_HEAD_OF_QUEUE: 1554 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1555 break; 1556 case CTL_TAG_UNTAGGED: 1557 case CTL_TAG_SIMPLE: 1558 case CTL_TAG_ACA: 1559 default: 1560 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1561 break; 1562 } 1563 PRIV(io)->ptr = (void *)beio; 1564 1565 switch (io->scsiio.cdb[0]) { 1566 case SYNCHRONIZE_CACHE: 1567 case SYNCHRONIZE_CACHE_16: 1568 ctl_be_block_cw_dispatch_sync(be_lun, io); 1569 break; 1570 case WRITE_SAME_10: 1571 case WRITE_SAME_16: 1572 ctl_be_block_cw_dispatch_ws(be_lun, io); 1573 break; 1574 case UNMAP: 1575 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1576 break; 1577 default: 1578 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1579 break; 1580 } 1581 } 1582 1583 SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t"); 1584 SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t"); 1585 SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t"); 1586 SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t"); 1587 1588 static void 1589 ctl_be_block_next(struct ctl_be_block_io *beio) 1590 { 1591 struct ctl_be_block_lun *be_lun; 1592 union ctl_io *io; 1593 1594 io = beio->io; 1595 be_lun = beio->lun; 1596 ctl_free_beio(beio); 1597 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1598 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1599 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1600 ctl_data_submit_done(io); 1601 return; 1602 } 1603 1604 io->io_hdr.status &= ~CTL_STATUS_MASK; 1605 io->io_hdr.status |= CTL_STATUS_NONE; 1606 1607 mtx_lock(&be_lun->queue_lock); 1608 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1609 mtx_unlock(&be_lun->queue_lock); 1610 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1611 } 1612 1613 static void 1614 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1615 union ctl_io *io) 1616 { 1617 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1618 struct ctl_be_block_io *beio; 1619 struct ctl_be_block_softc *softc; 1620 struct ctl_lba_len_flags *lbalen; 1621 struct ctl_ptr_len_flags *bptrlen; 1622 uint64_t len_left, lbas; 1623 int i; 1624 1625 softc = be_lun->softc; 1626 1627 DPRINTF("entered\n"); 1628 1629 lbalen = ARGS(io); 1630 if (lbalen->flags & CTL_LLF_WRITE) { 1631 SDT_PROBE0(cbb, , write, start); 1632 } else { 1633 SDT_PROBE0(cbb, , read, start); 1634 } 1635 1636 beio = ctl_alloc_beio(softc); 1637 beio->io = io; 1638 beio->lun = be_lun; 1639 bptrlen = PRIV(io); 1640 bptrlen->ptr = (void *)beio; 1641 1642 switch (io->scsiio.tag_type) { 1643 case CTL_TAG_ORDERED: 1644 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1645 break; 1646 case CTL_TAG_HEAD_OF_QUEUE: 1647 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1648 break; 1649 case CTL_TAG_UNTAGGED: 1650 case CTL_TAG_SIMPLE: 1651 case CTL_TAG_ACA: 1652 default: 1653 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1654 break; 1655 } 1656 1657 if (lbalen->flags & CTL_LLF_WRITE) { 1658 beio->bio_cmd = BIO_WRITE; 1659 beio->ds_trans_type = DEVSTAT_WRITE; 1660 } else { 1661 beio->bio_cmd = BIO_READ; 1662 beio->ds_trans_type = DEVSTAT_READ; 1663 } 1664 1665 DPRINTF("%s at LBA %jx len %u @%ju\n", 1666 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1667 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1668 lbas = CTLBLK_MAX_IO_SIZE; 1669 if (lbalen->flags & CTL_LLF_COMPARE) { 1670 beio->two_sglists = 1; 1671 lbas /= 2; 1672 } 1673 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1674 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1675 beio->io_len = lbas * cbe_lun->blocksize; 1676 bptrlen->len += lbas; 1677 1678 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1679 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1680 i, CTLBLK_MAX_SEGS)); 1681 1682 /* 1683 * Setup the S/G entry for this chunk. 1684 */ 1685 ctl_alloc_seg(softc, &beio->sg_segs[i], 1686 MIN(CTLBLK_MAX_SEG, len_left)); 1687 1688 DPRINTF("segment %d addr %p len %zd\n", i, 1689 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1690 1691 /* Set up second segment for compare operation. */ 1692 if (beio->two_sglists) { 1693 ctl_alloc_seg(softc, 1694 &beio->sg_segs[i + CTLBLK_HALF_SEGS], 1695 beio->sg_segs[i].len); 1696 } 1697 1698 beio->num_segs++; 1699 len_left -= beio->sg_segs[i].len; 1700 } 1701 if (bptrlen->len < lbalen->len) 1702 beio->beio_cont = ctl_be_block_next; 1703 io->scsiio.be_move_done = ctl_be_block_move_done; 1704 /* For compare we have separate S/G lists for read and datamove. */ 1705 if (beio->two_sglists) 1706 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1707 else 1708 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1709 io->scsiio.kern_data_len = beio->io_len; 1710 io->scsiio.kern_sg_entries = beio->num_segs; 1711 io->scsiio.kern_data_ref = ctl_refcnt_beio; 1712 io->scsiio.kern_data_arg = beio; 1713 io->io_hdr.flags |= CTL_FLAG_ALLOCATED; 1714 1715 /* 1716 * For the read case, we need to read the data into our buffers and 1717 * then we can send it back to the user. For the write case, we 1718 * need to get the data from the user first. 1719 */ 1720 if (beio->bio_cmd == BIO_READ) { 1721 SDT_PROBE0(cbb, , read, alloc_done); 1722 be_lun->dispatch(be_lun, beio); 1723 } else { 1724 SDT_PROBE0(cbb, , write, alloc_done); 1725 ctl_datamove(io); 1726 } 1727 } 1728 1729 static void 1730 ctl_be_block_worker(void *context, int pending) 1731 { 1732 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; 1733 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1734 union ctl_io *io; 1735 struct ctl_be_block_io *beio; 1736 1737 DPRINTF("entered\n"); 1738 /* 1739 * Fetch and process I/Os from all queues. If we detect LUN 1740 * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race, 1741 * so make response maximally opaque to not confuse initiator. 1742 */ 1743 for (;;) { 1744 mtx_lock(&be_lun->queue_lock); 1745 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1746 if (io != NULL) { 1747 DPRINTF("datamove queue\n"); 1748 STAILQ_REMOVE_HEAD(&be_lun->datamove_queue, links); 1749 mtx_unlock(&be_lun->queue_lock); 1750 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1751 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1752 ctl_set_busy(&io->scsiio); 1753 ctl_complete_beio(beio); 1754 continue; 1755 } 1756 be_lun->dispatch(be_lun, beio); 1757 continue; 1758 } 1759 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1760 if (io != NULL) { 1761 DPRINTF("config write queue\n"); 1762 STAILQ_REMOVE_HEAD(&be_lun->config_write_queue, links); 1763 mtx_unlock(&be_lun->queue_lock); 1764 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1765 ctl_set_busy(&io->scsiio); 1766 ctl_config_write_done(io); 1767 continue; 1768 } 1769 ctl_be_block_cw_dispatch(be_lun, io); 1770 continue; 1771 } 1772 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1773 if (io != NULL) { 1774 DPRINTF("config read queue\n"); 1775 STAILQ_REMOVE_HEAD(&be_lun->config_read_queue, links); 1776 mtx_unlock(&be_lun->queue_lock); 1777 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1778 ctl_set_busy(&io->scsiio); 1779 ctl_config_read_done(io); 1780 continue; 1781 } 1782 ctl_be_block_cr_dispatch(be_lun, io); 1783 continue; 1784 } 1785 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1786 if (io != NULL) { 1787 DPRINTF("input queue\n"); 1788 STAILQ_REMOVE_HEAD(&be_lun->input_queue, links); 1789 mtx_unlock(&be_lun->queue_lock); 1790 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1791 ctl_set_busy(&io->scsiio); 1792 ctl_data_submit_done(io); 1793 continue; 1794 } 1795 ctl_be_block_dispatch(be_lun, io); 1796 continue; 1797 } 1798 1799 /* 1800 * If we get here, there is no work left in the queues, so 1801 * just break out and let the task queue go to sleep. 1802 */ 1803 mtx_unlock(&be_lun->queue_lock); 1804 break; 1805 } 1806 } 1807 1808 /* 1809 * Entry point from CTL to the backend for I/O. We queue everything to a 1810 * work thread, so this just puts the I/O on a queue and wakes up the 1811 * thread. 1812 */ 1813 static int 1814 ctl_be_block_submit(union ctl_io *io) 1815 { 1816 struct ctl_be_block_lun *be_lun; 1817 1818 DPRINTF("entered\n"); 1819 1820 be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io); 1821 1822 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, 1823 ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type)); 1824 1825 PRIV(io)->len = 0; 1826 1827 mtx_lock(&be_lun->queue_lock); 1828 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1829 mtx_unlock(&be_lun->queue_lock); 1830 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1831 1832 return (CTL_RETVAL_COMPLETE); 1833 } 1834 1835 static int 1836 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1837 int flag, struct thread *td) 1838 { 1839 struct ctl_be_block_softc *softc = &backend_block_softc; 1840 int error; 1841 1842 error = 0; 1843 switch (cmd) { 1844 case CTL_LUN_REQ: { 1845 struct ctl_lun_req *lun_req; 1846 1847 lun_req = (struct ctl_lun_req *)addr; 1848 1849 switch (lun_req->reqtype) { 1850 case CTL_LUNREQ_CREATE: 1851 error = ctl_be_block_create(softc, lun_req); 1852 break; 1853 case CTL_LUNREQ_RM: 1854 error = ctl_be_block_rm(softc, lun_req); 1855 break; 1856 case CTL_LUNREQ_MODIFY: 1857 error = ctl_be_block_modify(softc, lun_req); 1858 break; 1859 default: 1860 lun_req->status = CTL_LUN_ERROR; 1861 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1862 "invalid LUN request type %d", 1863 lun_req->reqtype); 1864 break; 1865 } 1866 break; 1867 } 1868 default: 1869 error = ENOTTY; 1870 break; 1871 } 1872 1873 return (error); 1874 } 1875 1876 static int 1877 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1878 { 1879 struct ctl_be_lun *cbe_lun; 1880 struct ctl_be_block_filedata *file_data; 1881 struct ctl_lun_create_params *params; 1882 const char *value; 1883 struct vattr vattr; 1884 off_t ps, pss, po, pos, us, uss, uo, uos; 1885 int error; 1886 long pconf; 1887 1888 cbe_lun = &be_lun->cbe_lun; 1889 file_data = &be_lun->backend.file; 1890 params = &be_lun->params; 1891 1892 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1893 be_lun->dispatch = ctl_be_block_dispatch_file; 1894 be_lun->lun_flush = ctl_be_block_flush_file; 1895 be_lun->get_lba_status = ctl_be_block_gls_file; 1896 be_lun->getattr = ctl_be_block_getattr_file; 1897 be_lun->unmap = ctl_be_block_unmap_file; 1898 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1899 1900 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1901 if (error != 0) { 1902 snprintf(req->error_str, sizeof(req->error_str), 1903 "error calling VOP_GETATTR() for file %s", 1904 be_lun->dev_path); 1905 return (error); 1906 } 1907 1908 error = VOP_PATHCONF(be_lun->vn, _PC_DEALLOC_PRESENT, &pconf); 1909 if (error != 0) { 1910 snprintf(req->error_str, sizeof(req->error_str), 1911 "error calling VOP_PATHCONF() for file %s", 1912 be_lun->dev_path); 1913 return (error); 1914 } 1915 if (pconf == 1) 1916 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 1917 1918 file_data->cred = crhold(curthread->td_ucred); 1919 if (params->lun_size_bytes != 0) 1920 be_lun->size_bytes = params->lun_size_bytes; 1921 else 1922 be_lun->size_bytes = vattr.va_size; 1923 1924 /* 1925 * For files we can use any logical block size. Prefer 512 bytes 1926 * for compatibility reasons. If file's vattr.va_blocksize 1927 * (preferred I/O block size) is bigger and multiple to chosen 1928 * logical block size -- report it as physical block size. 1929 */ 1930 if (params->blocksize_bytes != 0) 1931 cbe_lun->blocksize = params->blocksize_bytes; 1932 else if (cbe_lun->lun_type == T_CDROM) 1933 cbe_lun->blocksize = 2048; 1934 else 1935 cbe_lun->blocksize = 512; 1936 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1937 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1938 0 : (be_lun->size_blocks - 1); 1939 1940 us = ps = vattr.va_blocksize; 1941 uo = po = 0; 1942 1943 value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL); 1944 if (value != NULL) 1945 ctl_expand_number(value, &ps); 1946 value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL); 1947 if (value != NULL) 1948 ctl_expand_number(value, &po); 1949 pss = ps / cbe_lun->blocksize; 1950 pos = po / cbe_lun->blocksize; 1951 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1952 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1953 cbe_lun->pblockexp = fls(pss) - 1; 1954 cbe_lun->pblockoff = (pss - pos) % pss; 1955 } 1956 1957 value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL); 1958 if (value != NULL) 1959 ctl_expand_number(value, &us); 1960 value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL); 1961 if (value != NULL) 1962 ctl_expand_number(value, &uo); 1963 uss = us / cbe_lun->blocksize; 1964 uos = uo / cbe_lun->blocksize; 1965 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1966 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1967 cbe_lun->ublockexp = fls(uss) - 1; 1968 cbe_lun->ublockoff = (uss - uos) % uss; 1969 } 1970 1971 /* 1972 * Sanity check. The media size has to be at least one 1973 * sector long. 1974 */ 1975 if (be_lun->size_bytes < cbe_lun->blocksize) { 1976 error = EINVAL; 1977 snprintf(req->error_str, sizeof(req->error_str), 1978 "file %s size %ju < block size %u", be_lun->dev_path, 1979 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1980 } 1981 1982 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1983 return (error); 1984 } 1985 1986 static int 1987 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1988 { 1989 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1990 struct ctl_lun_create_params *params; 1991 struct cdevsw *csw; 1992 struct cdev *dev; 1993 const char *value; 1994 int error, atomic, maxio, ref, unmap, tmp; 1995 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1996 1997 params = &be_lun->params; 1998 1999 be_lun->dev_type = CTL_BE_BLOCK_DEV; 2000 csw = devvn_refthread(be_lun->vn, &dev, &ref); 2001 if (csw == NULL) 2002 return (ENXIO); 2003 if (strcmp(csw->d_name, "zvol") == 0) { 2004 be_lun->dispatch = ctl_be_block_dispatch_zvol; 2005 be_lun->get_lba_status = ctl_be_block_gls_zvol; 2006 atomic = maxio = CTLBLK_MAX_IO_SIZE; 2007 } else { 2008 be_lun->dispatch = ctl_be_block_dispatch_dev; 2009 be_lun->get_lba_status = NULL; 2010 atomic = 0; 2011 maxio = dev->si_iosize_max; 2012 if (maxio <= 0) 2013 maxio = DFLTPHYS; 2014 if (maxio > CTLBLK_MAX_SEG) 2015 maxio = CTLBLK_MAX_SEG; 2016 } 2017 be_lun->lun_flush = ctl_be_block_flush_dev; 2018 be_lun->getattr = ctl_be_block_getattr_dev; 2019 be_lun->unmap = ctl_be_block_unmap_dev; 2020 2021 if (!csw->d_ioctl) { 2022 dev_relthread(dev, ref); 2023 snprintf(req->error_str, sizeof(req->error_str), 2024 "no d_ioctl for device %s!", be_lun->dev_path); 2025 return (ENODEV); 2026 } 2027 2028 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 2029 curthread); 2030 if (error) { 2031 dev_relthread(dev, ref); 2032 snprintf(req->error_str, sizeof(req->error_str), 2033 "error %d returned for DIOCGSECTORSIZE ioctl " 2034 "on %s!", error, be_lun->dev_path); 2035 return (error); 2036 } 2037 2038 /* 2039 * If the user has asked for a blocksize that is greater than the 2040 * backing device's blocksize, we can do it only if the blocksize 2041 * the user is asking for is an even multiple of the underlying 2042 * device's blocksize. 2043 */ 2044 if ((params->blocksize_bytes != 0) && 2045 (params->blocksize_bytes >= tmp)) { 2046 if (params->blocksize_bytes % tmp == 0) { 2047 cbe_lun->blocksize = params->blocksize_bytes; 2048 } else { 2049 dev_relthread(dev, ref); 2050 snprintf(req->error_str, sizeof(req->error_str), 2051 "requested blocksize %u is not an even " 2052 "multiple of backing device blocksize %u", 2053 params->blocksize_bytes, tmp); 2054 return (EINVAL); 2055 } 2056 } else if (params->blocksize_bytes != 0) { 2057 dev_relthread(dev, ref); 2058 snprintf(req->error_str, sizeof(req->error_str), 2059 "requested blocksize %u < backing device " 2060 "blocksize %u", params->blocksize_bytes, tmp); 2061 return (EINVAL); 2062 } else if (cbe_lun->lun_type == T_CDROM) 2063 cbe_lun->blocksize = MAX(tmp, 2048); 2064 else 2065 cbe_lun->blocksize = tmp; 2066 2067 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 2068 curthread); 2069 if (error) { 2070 dev_relthread(dev, ref); 2071 snprintf(req->error_str, sizeof(req->error_str), 2072 "error %d returned for DIOCGMEDIASIZE " 2073 " ioctl on %s!", error, 2074 be_lun->dev_path); 2075 return (error); 2076 } 2077 2078 if (params->lun_size_bytes != 0) { 2079 if (params->lun_size_bytes > otmp) { 2080 dev_relthread(dev, ref); 2081 snprintf(req->error_str, sizeof(req->error_str), 2082 "requested LUN size %ju > backing device " 2083 "size %ju", 2084 (uintmax_t)params->lun_size_bytes, 2085 (uintmax_t)otmp); 2086 return (EINVAL); 2087 } 2088 2089 be_lun->size_bytes = params->lun_size_bytes; 2090 } else 2091 be_lun->size_bytes = otmp; 2092 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2093 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2094 0 : (be_lun->size_blocks - 1); 2095 2096 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, 2097 curthread); 2098 if (error) 2099 ps = po = 0; 2100 else { 2101 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, 2102 FREAD, curthread); 2103 if (error) 2104 po = 0; 2105 } 2106 us = ps; 2107 uo = po; 2108 2109 value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL); 2110 if (value != NULL) 2111 ctl_expand_number(value, &ps); 2112 value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL); 2113 if (value != NULL) 2114 ctl_expand_number(value, &po); 2115 pss = ps / cbe_lun->blocksize; 2116 pos = po / cbe_lun->blocksize; 2117 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 2118 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 2119 cbe_lun->pblockexp = fls(pss) - 1; 2120 cbe_lun->pblockoff = (pss - pos) % pss; 2121 } 2122 2123 value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL); 2124 if (value != NULL) 2125 ctl_expand_number(value, &us); 2126 value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL); 2127 if (value != NULL) 2128 ctl_expand_number(value, &uo); 2129 uss = us / cbe_lun->blocksize; 2130 uos = uo / cbe_lun->blocksize; 2131 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2132 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2133 cbe_lun->ublockexp = fls(uss) - 1; 2134 cbe_lun->ublockoff = (uss - uos) % uss; 2135 } 2136 2137 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2138 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2139 2140 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2141 unmap = 1; 2142 } else { 2143 struct diocgattr_arg arg; 2144 2145 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2146 arg.len = sizeof(arg.value.i); 2147 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 2148 curthread); 2149 unmap = (error == 0) ? arg.value.i : 0; 2150 } 2151 value = dnvlist_get_string(cbe_lun->options, "unmap", NULL); 2152 if (value != NULL) 2153 unmap = (strcmp(value, "on") == 0); 2154 if (unmap) 2155 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2156 else 2157 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2158 2159 dev_relthread(dev, ref); 2160 return (0); 2161 } 2162 2163 static int 2164 ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2165 { 2166 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2167 int flags; 2168 2169 if (be_lun->vn) { 2170 flags = FREAD; 2171 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2172 flags |= FWRITE; 2173 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2174 be_lun->vn = NULL; 2175 2176 switch (be_lun->dev_type) { 2177 case CTL_BE_BLOCK_DEV: 2178 break; 2179 case CTL_BE_BLOCK_FILE: 2180 if (be_lun->backend.file.cred != NULL) { 2181 crfree(be_lun->backend.file.cred); 2182 be_lun->backend.file.cred = NULL; 2183 } 2184 break; 2185 case CTL_BE_BLOCK_NONE: 2186 break; 2187 default: 2188 panic("Unexpected backend type %d", be_lun->dev_type); 2189 break; 2190 } 2191 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2192 } 2193 return (0); 2194 } 2195 2196 static int 2197 ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2198 { 2199 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2200 struct nameidata nd; 2201 const char *value; 2202 int error, flags; 2203 2204 error = 0; 2205 if (rootvnode == NULL) { 2206 snprintf(req->error_str, sizeof(req->error_str), 2207 "Root filesystem is not mounted"); 2208 return (1); 2209 } 2210 pwd_ensure_dirs(); 2211 2212 value = dnvlist_get_string(cbe_lun->options, "file", NULL); 2213 if (value == NULL) { 2214 snprintf(req->error_str, sizeof(req->error_str), 2215 "no file argument specified"); 2216 return (1); 2217 } 2218 free(be_lun->dev_path, M_CTLBLK); 2219 be_lun->dev_path = strdup(value, M_CTLBLK); 2220 2221 flags = FREAD; 2222 value = dnvlist_get_string(cbe_lun->options, "readonly", NULL); 2223 if (value != NULL) { 2224 if (strcmp(value, "on") != 0) 2225 flags |= FWRITE; 2226 } else if (cbe_lun->lun_type == T_DIRECT) 2227 flags |= FWRITE; 2228 2229 again: 2230 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path); 2231 error = vn_open(&nd, &flags, 0, NULL); 2232 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2233 flags &= ~FWRITE; 2234 goto again; 2235 } 2236 if (error) { 2237 /* 2238 * This is the only reasonable guess we can make as far as 2239 * path if the user doesn't give us a fully qualified path. 2240 * If they want to specify a file, they need to specify the 2241 * full path. 2242 */ 2243 if (be_lun->dev_path[0] != '/') { 2244 char *dev_name; 2245 2246 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2247 be_lun->dev_path); 2248 free(be_lun->dev_path, M_CTLBLK); 2249 be_lun->dev_path = dev_name; 2250 goto again; 2251 } 2252 snprintf(req->error_str, sizeof(req->error_str), 2253 "error opening %s: %d", be_lun->dev_path, error); 2254 return (error); 2255 } 2256 if (flags & FWRITE) 2257 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2258 else 2259 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2260 2261 NDFREE_PNBUF(&nd); 2262 be_lun->vn = nd.ni_vp; 2263 2264 /* We only support disks and files. */ 2265 if (vn_isdisk_error(be_lun->vn, &error)) { 2266 error = ctl_be_block_open_dev(be_lun, req); 2267 } else if (be_lun->vn->v_type == VREG) { 2268 error = ctl_be_block_open_file(be_lun, req); 2269 } else { 2270 error = EINVAL; 2271 snprintf(req->error_str, sizeof(req->error_str), 2272 "%s is not a disk or plain file", be_lun->dev_path); 2273 } 2274 VOP_UNLOCK(be_lun->vn); 2275 2276 if (error != 0) 2277 ctl_be_block_close(be_lun); 2278 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2279 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2280 cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT; 2281 value = dnvlist_get_string(cbe_lun->options, "serseq", NULL); 2282 if (value != NULL && strcmp(value, "on") == 0) 2283 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2284 else if (value != NULL && strcmp(value, "read") == 0) 2285 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2286 else if (value != NULL && strcmp(value, "soft") == 0) 2287 cbe_lun->serseq = CTL_LUN_SERSEQ_SOFT; 2288 else if (value != NULL && strcmp(value, "off") == 0) 2289 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2290 return (0); 2291 } 2292 2293 static int 2294 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2295 { 2296 struct ctl_be_lun *cbe_lun; 2297 struct ctl_be_block_lun *be_lun; 2298 struct ctl_lun_create_params *params; 2299 char num_thread_str[16]; 2300 char tmpstr[32]; 2301 const char *value; 2302 int retval, num_threads; 2303 int tmp_num_threads; 2304 2305 params = &req->reqdata.create; 2306 retval = 0; 2307 req->status = CTL_LUN_OK; 2308 2309 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2310 cbe_lun = &be_lun->cbe_lun; 2311 be_lun->params = req->reqdata.create; 2312 be_lun->softc = softc; 2313 STAILQ_INIT(&be_lun->input_queue); 2314 STAILQ_INIT(&be_lun->config_read_queue); 2315 STAILQ_INIT(&be_lun->config_write_queue); 2316 STAILQ_INIT(&be_lun->datamove_queue); 2317 mtx_init(&be_lun->io_lock, "ctlblock io", NULL, MTX_DEF); 2318 mtx_init(&be_lun->queue_lock, "ctlblock queue", NULL, MTX_DEF); 2319 cbe_lun->options = nvlist_clone(req->args_nvl); 2320 2321 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2322 cbe_lun->lun_type = params->device_type; 2323 else 2324 cbe_lun->lun_type = T_DIRECT; 2325 be_lun->flags = 0; 2326 cbe_lun->flags = 0; 2327 value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL); 2328 if (value != NULL) { 2329 if (strcmp(value, "primary") == 0) 2330 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2331 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2332 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2333 2334 if (cbe_lun->lun_type == T_DIRECT || 2335 cbe_lun->lun_type == T_CDROM) { 2336 be_lun->size_bytes = params->lun_size_bytes; 2337 if (params->blocksize_bytes != 0) 2338 cbe_lun->blocksize = params->blocksize_bytes; 2339 else if (cbe_lun->lun_type == T_CDROM) 2340 cbe_lun->blocksize = 2048; 2341 else 2342 cbe_lun->blocksize = 512; 2343 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2344 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2345 0 : (be_lun->size_blocks - 1); 2346 2347 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2348 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2349 retval = ctl_be_block_open(be_lun, req); 2350 if (retval != 0) { 2351 retval = 0; 2352 req->status = CTL_LUN_WARNING; 2353 } 2354 } 2355 num_threads = cbb_num_threads; 2356 } else { 2357 num_threads = 1; 2358 } 2359 2360 value = dnvlist_get_string(cbe_lun->options, "num_threads", NULL); 2361 if (value != NULL) { 2362 tmp_num_threads = strtol(value, NULL, 0); 2363 2364 /* 2365 * We don't let the user specify less than one 2366 * thread, but hope he's clueful enough not to 2367 * specify 1000 threads. 2368 */ 2369 if (tmp_num_threads < 1) { 2370 snprintf(req->error_str, sizeof(req->error_str), 2371 "invalid number of threads %s", 2372 num_thread_str); 2373 goto bailout_error; 2374 } 2375 num_threads = tmp_num_threads; 2376 } 2377 2378 if (be_lun->vn == NULL) 2379 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2380 /* Tell the user the blocksize we ended up using */ 2381 params->lun_size_bytes = be_lun->size_bytes; 2382 params->blocksize_bytes = cbe_lun->blocksize; 2383 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2384 cbe_lun->req_lun_id = params->req_lun_id; 2385 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2386 } else 2387 cbe_lun->req_lun_id = 0; 2388 2389 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2390 cbe_lun->be = &ctl_be_block_driver; 2391 2392 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2393 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d", 2394 softc->num_luns); 2395 strncpy((char *)cbe_lun->serial_num, tmpstr, 2396 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2397 2398 /* Tell the user what we used for a serial number */ 2399 strncpy((char *)params->serial_num, tmpstr, 2400 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2401 } else { 2402 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2403 MIN(sizeof(cbe_lun->serial_num), 2404 sizeof(params->serial_num))); 2405 } 2406 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2407 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns); 2408 strncpy((char *)cbe_lun->device_id, tmpstr, 2409 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2410 2411 /* Tell the user what we used for a device ID */ 2412 strncpy((char *)params->device_id, tmpstr, 2413 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2414 } else { 2415 strncpy((char *)cbe_lun->device_id, params->device_id, 2416 MIN(sizeof(cbe_lun->device_id), 2417 sizeof(params->device_id))); 2418 } 2419 2420 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2421 2422 be_lun->io_taskqueue = taskqueue_create("ctlblocktq", M_WAITOK, 2423 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2424 2425 if (be_lun->io_taskqueue == NULL) { 2426 snprintf(req->error_str, sizeof(req->error_str), 2427 "unable to create taskqueue"); 2428 goto bailout_error; 2429 } 2430 2431 /* 2432 * Note that we start the same number of threads by default for 2433 * both the file case and the block device case. For the file 2434 * case, we need multiple threads to allow concurrency, because the 2435 * vnode interface is designed to be a blocking interface. For the 2436 * block device case, ZFS zvols at least will block the caller's 2437 * context in many instances, and so we need multiple threads to 2438 * overcome that problem. Other block devices don't need as many 2439 * threads, but they shouldn't cause too many problems. 2440 * 2441 * If the user wants to just have a single thread for a block 2442 * device, he can specify that when the LUN is created, or change 2443 * the tunable/sysctl to alter the default number of threads. 2444 */ 2445 retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue, 2446 /*num threads*/num_threads, 2447 /*priority*/PUSER, 2448 /*proc*/control_softc->ctl_proc, 2449 /*thread name*/"block"); 2450 2451 if (retval != 0) 2452 goto bailout_error; 2453 2454 be_lun->num_threads = num_threads; 2455 2456 retval = ctl_add_lun(&be_lun->cbe_lun); 2457 if (retval != 0) { 2458 snprintf(req->error_str, sizeof(req->error_str), 2459 "ctl_add_lun() returned error %d, see dmesg for " 2460 "details", retval); 2461 retval = 0; 2462 goto bailout_error; 2463 } 2464 2465 be_lun->disk_stats = devstat_new_entry("cbb", cbe_lun->lun_id, 2466 cbe_lun->blocksize, 2467 DEVSTAT_ALL_SUPPORTED, 2468 cbe_lun->lun_type 2469 | DEVSTAT_TYPE_IF_OTHER, 2470 DEVSTAT_PRIORITY_OTHER); 2471 2472 mtx_lock(&softc->lock); 2473 softc->num_luns++; 2474 SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links); 2475 mtx_unlock(&softc->lock); 2476 2477 params->req_lun_id = cbe_lun->lun_id; 2478 2479 return (retval); 2480 2481 bailout_error: 2482 req->status = CTL_LUN_ERROR; 2483 2484 if (be_lun->io_taskqueue != NULL) 2485 taskqueue_free(be_lun->io_taskqueue); 2486 ctl_be_block_close(be_lun); 2487 if (be_lun->dev_path != NULL) 2488 free(be_lun->dev_path, M_CTLBLK); 2489 nvlist_destroy(cbe_lun->options); 2490 mtx_destroy(&be_lun->queue_lock); 2491 mtx_destroy(&be_lun->io_lock); 2492 free(be_lun, M_CTLBLK); 2493 2494 return (retval); 2495 } 2496 2497 static int 2498 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2499 { 2500 struct ctl_lun_rm_params *params; 2501 struct ctl_be_block_lun *be_lun; 2502 struct ctl_be_lun *cbe_lun; 2503 int retval; 2504 2505 params = &req->reqdata.rm; 2506 2507 sx_xlock(&softc->modify_lock); 2508 mtx_lock(&softc->lock); 2509 SLIST_FOREACH(be_lun, &softc->lun_list, links) { 2510 if (be_lun->cbe_lun.lun_id == params->lun_id) { 2511 SLIST_REMOVE(&softc->lun_list, be_lun, 2512 ctl_be_block_lun, links); 2513 softc->num_luns--; 2514 break; 2515 } 2516 } 2517 mtx_unlock(&softc->lock); 2518 sx_xunlock(&softc->modify_lock); 2519 if (be_lun == NULL) { 2520 snprintf(req->error_str, sizeof(req->error_str), 2521 "LUN %u is not managed by the block backend", 2522 params->lun_id); 2523 goto bailout_error; 2524 } 2525 cbe_lun = &be_lun->cbe_lun; 2526 2527 if (be_lun->vn != NULL) { 2528 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2529 ctl_lun_no_media(cbe_lun); 2530 taskqueue_drain_all(be_lun->io_taskqueue); 2531 ctl_be_block_close(be_lun); 2532 } 2533 2534 mtx_lock(&softc->lock); 2535 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2536 mtx_unlock(&softc->lock); 2537 2538 retval = ctl_remove_lun(cbe_lun); 2539 if (retval != 0) { 2540 snprintf(req->error_str, sizeof(req->error_str), 2541 "error %d returned from ctl_remove_lun() for " 2542 "LUN %d", retval, params->lun_id); 2543 mtx_lock(&softc->lock); 2544 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2545 mtx_unlock(&softc->lock); 2546 goto bailout_error; 2547 } 2548 2549 mtx_lock(&softc->lock); 2550 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2551 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblockrm", 0); 2552 if (retval == EINTR) 2553 break; 2554 } 2555 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2556 if (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2557 mtx_unlock(&softc->lock); 2558 free(be_lun, M_CTLBLK); 2559 } else { 2560 mtx_unlock(&softc->lock); 2561 return (EINTR); 2562 } 2563 2564 req->status = CTL_LUN_OK; 2565 return (0); 2566 2567 bailout_error: 2568 req->status = CTL_LUN_ERROR; 2569 return (0); 2570 } 2571 2572 static int 2573 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2574 { 2575 struct ctl_lun_modify_params *params; 2576 struct ctl_be_block_lun *be_lun; 2577 struct ctl_be_lun *cbe_lun; 2578 const char *value; 2579 uint64_t oldsize; 2580 int error, wasprim; 2581 2582 params = &req->reqdata.modify; 2583 2584 sx_xlock(&softc->modify_lock); 2585 mtx_lock(&softc->lock); 2586 SLIST_FOREACH(be_lun, &softc->lun_list, links) { 2587 if (be_lun->cbe_lun.lun_id == params->lun_id) 2588 break; 2589 } 2590 mtx_unlock(&softc->lock); 2591 if (be_lun == NULL) { 2592 snprintf(req->error_str, sizeof(req->error_str), 2593 "LUN %u is not managed by the block backend", 2594 params->lun_id); 2595 goto bailout_error; 2596 } 2597 cbe_lun = &be_lun->cbe_lun; 2598 2599 if (params->lun_size_bytes != 0) 2600 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2601 2602 if (req->args_nvl != NULL) { 2603 nvlist_destroy(cbe_lun->options); 2604 cbe_lun->options = nvlist_clone(req->args_nvl); 2605 } 2606 2607 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2608 value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL); 2609 if (value != NULL) { 2610 if (strcmp(value, "primary") == 0) 2611 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2612 else 2613 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2614 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2615 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2616 else 2617 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2618 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2619 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2620 ctl_lun_primary(cbe_lun); 2621 else 2622 ctl_lun_secondary(cbe_lun); 2623 } 2624 2625 oldsize = be_lun->size_blocks; 2626 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2627 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2628 if (be_lun->vn == NULL) 2629 error = ctl_be_block_open(be_lun, req); 2630 else if (vn_isdisk_error(be_lun->vn, &error)) 2631 error = ctl_be_block_open_dev(be_lun, req); 2632 else if (be_lun->vn->v_type == VREG) { 2633 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2634 error = ctl_be_block_open_file(be_lun, req); 2635 VOP_UNLOCK(be_lun->vn); 2636 } else 2637 error = EINVAL; 2638 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) && 2639 be_lun->vn != NULL) { 2640 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2641 ctl_lun_has_media(cbe_lun); 2642 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 && 2643 be_lun->vn == NULL) { 2644 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2645 ctl_lun_no_media(cbe_lun); 2646 } 2647 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2648 } else { 2649 if (be_lun->vn != NULL) { 2650 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2651 ctl_lun_no_media(cbe_lun); 2652 taskqueue_drain_all(be_lun->io_taskqueue); 2653 error = ctl_be_block_close(be_lun); 2654 } else 2655 error = 0; 2656 } 2657 if (be_lun->size_blocks != oldsize) 2658 ctl_lun_capacity_changed(cbe_lun); 2659 2660 /* Tell the user the exact size we ended up using */ 2661 params->lun_size_bytes = be_lun->size_bytes; 2662 2663 sx_xunlock(&softc->modify_lock); 2664 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2665 return (0); 2666 2667 bailout_error: 2668 sx_xunlock(&softc->modify_lock); 2669 req->status = CTL_LUN_ERROR; 2670 return (0); 2671 } 2672 2673 static void 2674 ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun) 2675 { 2676 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)cbe_lun; 2677 struct ctl_be_block_softc *softc = be_lun->softc; 2678 2679 taskqueue_drain_all(be_lun->io_taskqueue); 2680 taskqueue_free(be_lun->io_taskqueue); 2681 if (be_lun->disk_stats != NULL) 2682 devstat_remove_entry(be_lun->disk_stats); 2683 nvlist_destroy(be_lun->cbe_lun.options); 2684 free(be_lun->dev_path, M_CTLBLK); 2685 mtx_destroy(&be_lun->queue_lock); 2686 mtx_destroy(&be_lun->io_lock); 2687 2688 mtx_lock(&softc->lock); 2689 be_lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2690 if (be_lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2691 wakeup(be_lun); 2692 else 2693 free(be_lun, M_CTLBLK); 2694 mtx_unlock(&softc->lock); 2695 } 2696 2697 static int 2698 ctl_be_block_config_write(union ctl_io *io) 2699 { 2700 struct ctl_be_block_lun *be_lun; 2701 struct ctl_be_lun *cbe_lun; 2702 int retval; 2703 2704 DPRINTF("entered\n"); 2705 2706 cbe_lun = CTL_BACKEND_LUN(io); 2707 be_lun = (struct ctl_be_block_lun *)cbe_lun; 2708 2709 retval = 0; 2710 switch (io->scsiio.cdb[0]) { 2711 case SYNCHRONIZE_CACHE: 2712 case SYNCHRONIZE_CACHE_16: 2713 case WRITE_SAME_10: 2714 case WRITE_SAME_16: 2715 case UNMAP: 2716 /* 2717 * The upper level CTL code will filter out any CDBs with 2718 * the immediate bit set and return the proper error. 2719 * 2720 * We don't really need to worry about what LBA range the 2721 * user asked to be synced out. When they issue a sync 2722 * cache command, we'll sync out the whole thing. 2723 */ 2724 mtx_lock(&be_lun->queue_lock); 2725 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2726 links); 2727 mtx_unlock(&be_lun->queue_lock); 2728 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2729 break; 2730 case START_STOP_UNIT: { 2731 struct scsi_start_stop_unit *cdb; 2732 struct ctl_lun_req req; 2733 2734 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2735 if ((cdb->how & SSS_PC_MASK) != 0) { 2736 ctl_set_success(&io->scsiio); 2737 ctl_config_write_done(io); 2738 break; 2739 } 2740 if (cdb->how & SSS_START) { 2741 if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) { 2742 retval = ctl_be_block_open(be_lun, &req); 2743 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2744 if (retval == 0) { 2745 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2746 ctl_lun_has_media(cbe_lun); 2747 } else { 2748 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2749 ctl_lun_no_media(cbe_lun); 2750 } 2751 } 2752 ctl_start_lun(cbe_lun); 2753 } else { 2754 ctl_stop_lun(cbe_lun); 2755 if (cdb->how & SSS_LOEJ) { 2756 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2757 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED; 2758 ctl_lun_ejected(cbe_lun); 2759 if (be_lun->vn != NULL) 2760 ctl_be_block_close(be_lun); 2761 } 2762 } 2763 2764 ctl_set_success(&io->scsiio); 2765 ctl_config_write_done(io); 2766 break; 2767 } 2768 case PREVENT_ALLOW: 2769 ctl_set_success(&io->scsiio); 2770 ctl_config_write_done(io); 2771 break; 2772 default: 2773 ctl_set_invalid_opcode(&io->scsiio); 2774 ctl_config_write_done(io); 2775 retval = CTL_RETVAL_COMPLETE; 2776 break; 2777 } 2778 2779 return (retval); 2780 } 2781 2782 static int 2783 ctl_be_block_config_read(union ctl_io *io) 2784 { 2785 struct ctl_be_block_lun *be_lun; 2786 int retval = 0; 2787 2788 DPRINTF("entered\n"); 2789 2790 be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io); 2791 2792 switch (io->scsiio.cdb[0]) { 2793 case SERVICE_ACTION_IN: 2794 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2795 mtx_lock(&be_lun->queue_lock); 2796 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2797 &io->io_hdr, links); 2798 mtx_unlock(&be_lun->queue_lock); 2799 taskqueue_enqueue(be_lun->io_taskqueue, 2800 &be_lun->io_task); 2801 retval = CTL_RETVAL_QUEUED; 2802 break; 2803 } 2804 ctl_set_invalid_field(&io->scsiio, 2805 /*sks_valid*/ 1, 2806 /*command*/ 1, 2807 /*field*/ 1, 2808 /*bit_valid*/ 1, 2809 /*bit*/ 4); 2810 ctl_config_read_done(io); 2811 retval = CTL_RETVAL_COMPLETE; 2812 break; 2813 default: 2814 ctl_set_invalid_opcode(&io->scsiio); 2815 ctl_config_read_done(io); 2816 retval = CTL_RETVAL_COMPLETE; 2817 break; 2818 } 2819 2820 return (retval); 2821 } 2822 2823 static int 2824 ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb) 2825 { 2826 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun; 2827 int retval; 2828 2829 retval = sbuf_cat(sb, "\t<num_threads>"); 2830 if (retval != 0) 2831 goto bailout; 2832 retval = sbuf_printf(sb, "%d", lun->num_threads); 2833 if (retval != 0) 2834 goto bailout; 2835 retval = sbuf_cat(sb, "</num_threads>\n"); 2836 2837 bailout: 2838 return (retval); 2839 } 2840 2841 static uint64_t 2842 ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname) 2843 { 2844 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun; 2845 2846 if (lun->getattr == NULL) 2847 return (UINT64_MAX); 2848 return (lun->getattr(lun, attrname)); 2849 } 2850 2851 static int 2852 ctl_be_block_init(void) 2853 { 2854 struct ctl_be_block_softc *softc = &backend_block_softc; 2855 2856 sx_init(&softc->modify_lock, "ctlblock modify"); 2857 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2858 softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2859 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2860 softc->bufmin_zone = uma_zcreate("ctlblockmin", CTLBLK_MIN_SEG, 2861 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2862 if (CTLBLK_MIN_SEG < CTLBLK_MAX_SEG) 2863 softc->bufmax_zone = uma_zcreate("ctlblockmax", CTLBLK_MAX_SEG, 2864 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2865 SLIST_INIT(&softc->lun_list); 2866 return (0); 2867 } 2868 2869 static int 2870 ctl_be_block_shutdown(void) 2871 { 2872 struct ctl_be_block_softc *softc = &backend_block_softc; 2873 struct ctl_be_block_lun *lun; 2874 2875 mtx_lock(&softc->lock); 2876 while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) { 2877 SLIST_REMOVE_HEAD(&softc->lun_list, links); 2878 softc->num_luns--; 2879 /* 2880 * Drop our lock here. Since ctl_remove_lun() can call 2881 * back into us, this could potentially lead to a recursive 2882 * lock of the same mutex, which would cause a hang. 2883 */ 2884 mtx_unlock(&softc->lock); 2885 ctl_remove_lun(&lun->cbe_lun); 2886 mtx_lock(&softc->lock); 2887 } 2888 mtx_unlock(&softc->lock); 2889 uma_zdestroy(softc->bufmin_zone); 2890 if (CTLBLK_MIN_SEG < CTLBLK_MAX_SEG) 2891 uma_zdestroy(softc->bufmax_zone); 2892 uma_zdestroy(softc->beio_zone); 2893 mtx_destroy(&softc->lock); 2894 sx_destroy(&softc->modify_lock); 2895 return (0); 2896 } 2897