1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2003 Silicon Graphics International Corp. 5 * Copyright (c) 2009-2011 Spectra Logic Corporation 6 * Copyright (c) 2012 The FreeBSD Foundation 7 * Copyright (c) 2014-2015 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 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions, and the following disclaimer, 18 * without modification. 19 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 20 * substantially similar to the "NO WARRANTY" disclaimer below 21 * ("Disclaimer") and any redistribution must be conditioned upon 22 * including a substantially similar Disclaimer requirement for further 23 * binary redistribution. 24 * 25 * NO WARRANTY 26 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 27 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 28 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 29 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 30 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 34 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 35 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGES. 37 * 38 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 39 */ 40 /* 41 * CAM Target Layer driver backend for block devices. 42 * 43 * Author: Ken Merry <ken@FreeBSD.org> 44 */ 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 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 84 #include <geom/geom.h> 85 86 #include <cam/cam.h> 87 #include <cam/scsi/scsi_all.h> 88 #include <cam/scsi/scsi_da.h> 89 #include <cam/ctl/ctl_io.h> 90 #include <cam/ctl/ctl.h> 91 #include <cam/ctl/ctl_backend.h> 92 #include <cam/ctl/ctl_ioctl.h> 93 #include <cam/ctl/ctl_ha.h> 94 #include <cam/ctl/ctl_scsi_all.h> 95 #include <cam/ctl/ctl_private.h> 96 #include <cam/ctl/ctl_error.h> 97 98 /* 99 * The idea here is that we'll allocate enough S/G space to hold a 1MB 100 * I/O. If we get an I/O larger than that, we'll split it. 101 */ 102 #define CTLBLK_HALF_IO_SIZE (512 * 1024) 103 #define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 104 #define CTLBLK_MAX_SEG MAXPHYS 105 #define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 106 #define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 107 108 #ifdef CTLBLK_DEBUG 109 #define DPRINTF(fmt, args...) \ 110 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 111 #else 112 #define DPRINTF(fmt, args...) do {} while(0) 113 #endif 114 115 #define PRIV(io) \ 116 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 117 #define ARGS(io) \ 118 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 119 120 SDT_PROVIDER_DEFINE(cbb); 121 122 typedef enum { 123 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 124 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 125 CTL_BE_BLOCK_LUN_WAITING = 0x04, 126 } ctl_be_block_lun_flags; 127 128 typedef enum { 129 CTL_BE_BLOCK_NONE, 130 CTL_BE_BLOCK_DEV, 131 CTL_BE_BLOCK_FILE 132 } ctl_be_block_type; 133 134 struct ctl_be_block_filedata { 135 struct ucred *cred; 136 }; 137 138 union ctl_be_block_bedata { 139 struct ctl_be_block_filedata file; 140 }; 141 142 struct ctl_be_block_io; 143 struct ctl_be_block_lun; 144 145 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 146 struct ctl_be_block_io *beio); 147 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 148 const char *attrname); 149 150 /* 151 * Backend LUN structure. There is a 1:1 mapping between a block device 152 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 153 */ 154 struct ctl_be_block_lun { 155 struct ctl_lun_create_params params; 156 char lunname[32]; 157 char *dev_path; 158 ctl_be_block_type dev_type; 159 struct vnode *vn; 160 union ctl_be_block_bedata backend; 161 cbb_dispatch_t dispatch; 162 cbb_dispatch_t lun_flush; 163 cbb_dispatch_t unmap; 164 cbb_dispatch_t get_lba_status; 165 cbb_getattr_t getattr; 166 uma_zone_t lun_zone; 167 uint64_t size_blocks; 168 uint64_t size_bytes; 169 struct ctl_be_block_softc *softc; 170 struct devstat *disk_stats; 171 ctl_be_block_lun_flags flags; 172 STAILQ_ENTRY(ctl_be_block_lun) links; 173 struct ctl_be_lun cbe_lun; 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 mtx lock; 190 uma_zone_t beio_zone; 191 int num_luns; 192 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 193 }; 194 195 static struct ctl_be_block_softc backend_block_softc; 196 197 /* 198 * Per-I/O information. 199 */ 200 struct ctl_be_block_io { 201 union ctl_io *io; 202 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 203 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 204 int bio_cmd; 205 int num_segs; 206 int num_bios_sent; 207 int num_bios_done; 208 int send_complete; 209 int first_error; 210 uint64_t first_error_offset; 211 struct bintime ds_t0; 212 devstat_tag_type ds_tag_type; 213 devstat_trans_flags ds_trans_type; 214 uint64_t io_len; 215 uint64_t io_offset; 216 int io_arg; 217 struct ctl_be_block_softc *softc; 218 struct ctl_be_block_lun *lun; 219 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 220 }; 221 222 extern struct ctl_softc *control_softc; 223 224 static int cbb_num_threads = 14; 225 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 226 "CAM Target Layer Block Backend"); 227 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 228 &cbb_num_threads, 0, "Number of threads per backing file"); 229 230 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 231 static void ctl_free_beio(struct ctl_be_block_io *beio); 232 static void ctl_complete_beio(struct ctl_be_block_io *beio); 233 static int ctl_be_block_move_done(union ctl_io *io); 234 static void ctl_be_block_biodone(struct bio *bio); 235 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 236 struct ctl_be_block_io *beio); 237 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 238 struct ctl_be_block_io *beio); 239 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 240 struct ctl_be_block_io *beio); 241 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 242 const char *attrname); 243 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 244 struct ctl_be_block_io *beio); 245 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 246 struct ctl_be_block_io *beio); 247 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 248 struct ctl_be_block_io *beio); 249 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 250 const char *attrname); 251 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 252 union ctl_io *io); 253 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 254 union ctl_io *io); 255 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 256 union ctl_io *io); 257 static void ctl_be_block_worker(void *context, int pending); 258 static int ctl_be_block_submit(union ctl_io *io); 259 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 260 int flag, struct thread *td); 261 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 262 struct ctl_lun_req *req); 263 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 264 struct ctl_lun_req *req); 265 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 266 static int ctl_be_block_open(struct ctl_be_block_lun *be_lun, 267 struct ctl_lun_req *req); 268 static int ctl_be_block_create(struct ctl_be_block_softc *softc, 269 struct ctl_lun_req *req); 270 static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 271 struct ctl_lun_req *req); 272 static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 273 struct ctl_lun_req *req); 274 static void ctl_be_block_lun_shutdown(void *be_lun); 275 static void ctl_be_block_lun_config_status(void *be_lun, 276 ctl_lun_config_status status); 277 static int ctl_be_block_config_write(union ctl_io *io); 278 static int ctl_be_block_config_read(union ctl_io *io); 279 static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 280 static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 281 static int ctl_be_block_init(void); 282 static int ctl_be_block_shutdown(void); 283 284 static struct ctl_backend_driver ctl_be_block_driver = 285 { 286 .name = "block", 287 .flags = CTL_BE_FLAG_HAS_CONFIG, 288 .init = ctl_be_block_init, 289 .shutdown = ctl_be_block_shutdown, 290 .data_submit = ctl_be_block_submit, 291 .data_move_done = ctl_be_block_move_done, 292 .config_read = ctl_be_block_config_read, 293 .config_write = ctl_be_block_config_write, 294 .ioctl = ctl_be_block_ioctl, 295 .lun_info = ctl_be_block_lun_info, 296 .lun_attr = ctl_be_block_lun_attr 297 }; 298 299 MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 300 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 301 302 static struct ctl_be_block_io * 303 ctl_alloc_beio(struct ctl_be_block_softc *softc) 304 { 305 struct ctl_be_block_io *beio; 306 307 beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO); 308 beio->softc = softc; 309 return (beio); 310 } 311 312 static void 313 ctl_free_beio(struct ctl_be_block_io *beio) 314 { 315 int duplicate_free; 316 int i; 317 318 duplicate_free = 0; 319 320 for (i = 0; i < beio->num_segs; i++) { 321 if (beio->sg_segs[i].addr == NULL) 322 duplicate_free++; 323 324 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 325 beio->sg_segs[i].addr = NULL; 326 327 /* For compare we had two equal S/G lists. */ 328 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 329 uma_zfree(beio->lun->lun_zone, 330 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 331 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 332 } 333 } 334 335 if (duplicate_free > 0) { 336 printf("%s: %d duplicate frees out of %d segments\n", __func__, 337 duplicate_free, beio->num_segs); 338 } 339 340 uma_zfree(beio->softc->beio_zone, beio); 341 } 342 343 static void 344 ctl_complete_beio(struct ctl_be_block_io *beio) 345 { 346 union ctl_io *io = beio->io; 347 348 if (beio->beio_cont != NULL) { 349 beio->beio_cont(beio); 350 } else { 351 ctl_free_beio(beio); 352 ctl_data_submit_done(io); 353 } 354 } 355 356 static size_t 357 cmp(uint8_t *a, uint8_t *b, size_t size) 358 { 359 size_t i; 360 361 for (i = 0; i < size; i++) { 362 if (a[i] != b[i]) 363 break; 364 } 365 return (i); 366 } 367 368 static void 369 ctl_be_block_compare(union ctl_io *io) 370 { 371 struct ctl_be_block_io *beio; 372 uint64_t off, res; 373 int i; 374 uint8_t info[8]; 375 376 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 377 off = 0; 378 for (i = 0; i < beio->num_segs; i++) { 379 res = cmp(beio->sg_segs[i].addr, 380 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 381 beio->sg_segs[i].len); 382 off += res; 383 if (res < beio->sg_segs[i].len) 384 break; 385 } 386 if (i < beio->num_segs) { 387 scsi_u64to8b(off, info); 388 ctl_set_sense(&io->scsiio, /*current_error*/ 1, 389 /*sense_key*/ SSD_KEY_MISCOMPARE, 390 /*asc*/ 0x1D, /*ascq*/ 0x00, 391 /*type*/ SSD_ELEM_INFO, 392 /*size*/ sizeof(info), /*data*/ &info, 393 /*type*/ SSD_ELEM_NONE); 394 } else 395 ctl_set_success(&io->scsiio); 396 } 397 398 static int 399 ctl_be_block_move_done(union ctl_io *io) 400 { 401 struct ctl_be_block_io *beio; 402 struct ctl_be_block_lun *be_lun; 403 struct ctl_lba_len_flags *lbalen; 404 #ifdef CTL_TIME_IO 405 struct bintime cur_bt; 406 #endif 407 408 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 409 be_lun = beio->lun; 410 411 DPRINTF("entered\n"); 412 413 #ifdef CTL_TIME_IO 414 getbinuptime(&cur_bt); 415 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 416 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 417 #endif 418 io->io_hdr.num_dmas++; 419 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 420 421 /* 422 * We set status at this point for read commands, and write 423 * commands with errors. 424 */ 425 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 426 ; 427 } else if ((io->io_hdr.port_status != 0) && 428 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 429 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 430 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, 431 /*retry_count*/ io->io_hdr.port_status); 432 } else if (io->scsiio.kern_data_resid != 0 && 433 (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT && 434 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 435 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 436 ctl_set_invalid_field_ciu(&io->scsiio); 437 } else if ((io->io_hdr.port_status == 0) && 438 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 439 lbalen = ARGS(beio->io); 440 if (lbalen->flags & CTL_LLF_READ) { 441 ctl_set_success(&io->scsiio); 442 } else if (lbalen->flags & CTL_LLF_COMPARE) { 443 /* We have two data blocks ready for comparison. */ 444 ctl_be_block_compare(io); 445 } 446 } 447 448 /* 449 * If this is a read, or a write with errors, it is done. 450 */ 451 if ((beio->bio_cmd == BIO_READ) 452 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 453 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 454 ctl_complete_beio(beio); 455 return (0); 456 } 457 458 /* 459 * At this point, we have a write and the DMA completed 460 * successfully. We now have to queue it to the task queue to 461 * execute the backend I/O. That is because we do blocking 462 * memory allocations, and in the file backing case, blocking I/O. 463 * This move done routine is generally called in the SIM's 464 * interrupt context, and therefore we cannot block. 465 */ 466 mtx_lock(&be_lun->queue_lock); 467 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 468 mtx_unlock(&be_lun->queue_lock); 469 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 470 471 return (0); 472 } 473 474 static void 475 ctl_be_block_biodone(struct bio *bio) 476 { 477 struct ctl_be_block_io *beio; 478 struct ctl_be_block_lun *be_lun; 479 union ctl_io *io; 480 int error; 481 482 beio = bio->bio_caller1; 483 be_lun = beio->lun; 484 io = beio->io; 485 486 DPRINTF("entered\n"); 487 488 error = bio->bio_error; 489 mtx_lock(&be_lun->io_lock); 490 if (error != 0 && 491 (beio->first_error == 0 || 492 bio->bio_offset < beio->first_error_offset)) { 493 beio->first_error = error; 494 beio->first_error_offset = bio->bio_offset; 495 } 496 497 beio->num_bios_done++; 498 499 /* 500 * XXX KDM will this cause WITNESS to complain? Holding a lock 501 * during the free might cause it to complain. 502 */ 503 g_destroy_bio(bio); 504 505 /* 506 * If the send complete bit isn't set, or we aren't the last I/O to 507 * complete, then we're done. 508 */ 509 if ((beio->send_complete == 0) 510 || (beio->num_bios_done < beio->num_bios_sent)) { 511 mtx_unlock(&be_lun->io_lock); 512 return; 513 } 514 515 /* 516 * At this point, we've verified that we are the last I/O to 517 * complete, so it's safe to drop the lock. 518 */ 519 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 520 beio->ds_tag_type, beio->ds_trans_type, 521 /*now*/ NULL, /*then*/&beio->ds_t0); 522 mtx_unlock(&be_lun->io_lock); 523 524 /* 525 * If there are any errors from the backing device, we fail the 526 * entire I/O with a medium error. 527 */ 528 error = beio->first_error; 529 if (error != 0) { 530 if (error == EOPNOTSUPP) { 531 ctl_set_invalid_opcode(&io->scsiio); 532 } else if (error == ENOSPC || error == EDQUOT) { 533 ctl_set_space_alloc_fail(&io->scsiio); 534 } else if (error == EROFS || error == EACCES) { 535 ctl_set_hw_write_protected(&io->scsiio); 536 } else if (beio->bio_cmd == BIO_FLUSH) { 537 /* XXX KDM is there is a better error here? */ 538 ctl_set_internal_failure(&io->scsiio, 539 /*sks_valid*/ 1, 540 /*retry_count*/ 0xbad2); 541 } else { 542 ctl_set_medium_error(&io->scsiio, 543 beio->bio_cmd == BIO_READ); 544 } 545 ctl_complete_beio(beio); 546 return; 547 } 548 549 /* 550 * If this is a write, a flush, a delete or verify, we're all done. 551 * If this is a read, we can now send the data to the user. 552 */ 553 if ((beio->bio_cmd == BIO_WRITE) 554 || (beio->bio_cmd == BIO_FLUSH) 555 || (beio->bio_cmd == BIO_DELETE) 556 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 557 ctl_set_success(&io->scsiio); 558 ctl_complete_beio(beio); 559 } else { 560 if ((ARGS(io)->flags & CTL_LLF_READ) && 561 beio->beio_cont == NULL) { 562 ctl_set_success(&io->scsiio); 563 ctl_serseq_done(io); 564 } 565 #ifdef CTL_TIME_IO 566 getbinuptime(&io->io_hdr.dma_start_bt); 567 #endif 568 ctl_datamove(io); 569 } 570 } 571 572 static void 573 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 574 struct ctl_be_block_io *beio) 575 { 576 union ctl_io *io = beio->io; 577 struct mount *mountpoint; 578 int error, lock_flags; 579 580 DPRINTF("entered\n"); 581 582 binuptime(&beio->ds_t0); 583 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 584 585 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 586 587 if (MNT_SHARED_WRITES(mountpoint) || 588 ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 589 lock_flags = LK_SHARED; 590 else 591 lock_flags = LK_EXCLUSIVE; 592 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 593 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 594 curthread); 595 VOP_UNLOCK(be_lun->vn); 596 597 vn_finished_write(mountpoint); 598 599 mtx_lock(&be_lun->io_lock); 600 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 601 beio->ds_tag_type, beio->ds_trans_type, 602 /*now*/ NULL, /*then*/&beio->ds_t0); 603 mtx_unlock(&be_lun->io_lock); 604 605 if (error == 0) 606 ctl_set_success(&io->scsiio); 607 else { 608 /* XXX KDM is there is a better error here? */ 609 ctl_set_internal_failure(&io->scsiio, 610 /*sks_valid*/ 1, 611 /*retry_count*/ 0xbad1); 612 } 613 614 ctl_complete_beio(beio); 615 } 616 617 SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t"); 618 SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t"); 619 SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t"); 620 SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t"); 621 622 static void 623 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 624 struct ctl_be_block_io *beio) 625 { 626 struct ctl_be_block_filedata *file_data; 627 union ctl_io *io; 628 struct uio xuio; 629 struct iovec *xiovec; 630 size_t s; 631 int error, flags, i; 632 633 DPRINTF("entered\n"); 634 635 file_data = &be_lun->backend.file; 636 io = beio->io; 637 flags = 0; 638 if (ARGS(io)->flags & CTL_LLF_DPO) 639 flags |= IO_DIRECT; 640 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 641 flags |= IO_SYNC; 642 643 bzero(&xuio, sizeof(xuio)); 644 if (beio->bio_cmd == BIO_READ) { 645 SDT_PROBE0(cbb, , read, file_start); 646 xuio.uio_rw = UIO_READ; 647 } else { 648 SDT_PROBE0(cbb, , write, file_start); 649 xuio.uio_rw = UIO_WRITE; 650 } 651 xuio.uio_offset = beio->io_offset; 652 xuio.uio_resid = beio->io_len; 653 xuio.uio_segflg = UIO_SYSSPACE; 654 xuio.uio_iov = beio->xiovecs; 655 xuio.uio_iovcnt = beio->num_segs; 656 xuio.uio_td = curthread; 657 658 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 659 xiovec->iov_base = beio->sg_segs[i].addr; 660 xiovec->iov_len = beio->sg_segs[i].len; 661 } 662 663 binuptime(&beio->ds_t0); 664 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 665 666 if (beio->bio_cmd == BIO_READ) { 667 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 668 669 /* 670 * UFS pays attention to IO_DIRECT for reads. If the 671 * DIRECTIO option is configured into the kernel, it calls 672 * ffs_rawread(). But that only works for single-segment 673 * uios with user space addresses. In our case, with a 674 * kernel uio, it still reads into the buffer cache, but it 675 * will just try to release the buffer from the cache later 676 * on in ffs_read(). 677 * 678 * ZFS does not pay attention to IO_DIRECT for reads. 679 * 680 * UFS does not pay attention to IO_SYNC for reads. 681 * 682 * ZFS pays attention to IO_SYNC (which translates into the 683 * Solaris define FRSYNC for zfs_read()) for reads. It 684 * attempts to sync the file before reading. 685 */ 686 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 687 688 VOP_UNLOCK(be_lun->vn); 689 SDT_PROBE0(cbb, , read, file_done); 690 if (error == 0 && xuio.uio_resid > 0) { 691 /* 692 * If we red less then requested (EOF), then 693 * we should clean the rest of the buffer. 694 */ 695 s = beio->io_len - xuio.uio_resid; 696 for (i = 0; i < beio->num_segs; i++) { 697 if (s >= beio->sg_segs[i].len) { 698 s -= beio->sg_segs[i].len; 699 continue; 700 } 701 bzero((uint8_t *)beio->sg_segs[i].addr + s, 702 beio->sg_segs[i].len - s); 703 s = 0; 704 } 705 } 706 } else { 707 struct mount *mountpoint; 708 int lock_flags; 709 710 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 711 712 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL) 713 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 714 lock_flags = LK_SHARED; 715 else 716 lock_flags = LK_EXCLUSIVE; 717 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 718 719 /* 720 * UFS pays attention to IO_DIRECT for writes. The write 721 * is done asynchronously. (Normally the write would just 722 * get put into cache. 723 * 724 * UFS pays attention to IO_SYNC for writes. It will 725 * attempt to write the buffer out synchronously if that 726 * flag is set. 727 * 728 * ZFS does not pay attention to IO_DIRECT for writes. 729 * 730 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 731 * for writes. It will flush the transaction from the 732 * cache before returning. 733 */ 734 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 735 VOP_UNLOCK(be_lun->vn); 736 737 vn_finished_write(mountpoint); 738 SDT_PROBE0(cbb, , write, file_done); 739 } 740 741 mtx_lock(&be_lun->io_lock); 742 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 743 beio->ds_tag_type, beio->ds_trans_type, 744 /*now*/ NULL, /*then*/&beio->ds_t0); 745 mtx_unlock(&be_lun->io_lock); 746 747 /* 748 * If we got an error, set the sense data to "MEDIUM ERROR" and 749 * return the I/O to the user. 750 */ 751 if (error != 0) { 752 if (error == ENOSPC || error == EDQUOT) { 753 ctl_set_space_alloc_fail(&io->scsiio); 754 } else if (error == EROFS || error == EACCES) { 755 ctl_set_hw_write_protected(&io->scsiio); 756 } else { 757 ctl_set_medium_error(&io->scsiio, 758 beio->bio_cmd == BIO_READ); 759 } 760 ctl_complete_beio(beio); 761 return; 762 } 763 764 /* 765 * If this is a write or a verify, we're all done. 766 * If this is a read, we can now send the data to the user. 767 */ 768 if ((beio->bio_cmd == BIO_WRITE) || 769 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 770 ctl_set_success(&io->scsiio); 771 ctl_complete_beio(beio); 772 } else { 773 if ((ARGS(io)->flags & CTL_LLF_READ) && 774 beio->beio_cont == NULL) { 775 ctl_set_success(&io->scsiio); 776 ctl_serseq_done(io); 777 } 778 #ifdef CTL_TIME_IO 779 getbinuptime(&io->io_hdr.dma_start_bt); 780 #endif 781 ctl_datamove(io); 782 } 783 } 784 785 static void 786 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 787 struct ctl_be_block_io *beio) 788 { 789 union ctl_io *io = beio->io; 790 struct ctl_lba_len_flags *lbalen = ARGS(io); 791 struct scsi_get_lba_status_data *data; 792 off_t roff, off; 793 int error, status; 794 795 DPRINTF("entered\n"); 796 797 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 798 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 799 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 800 0, curthread->td_ucred, curthread); 801 if (error == 0 && off > roff) 802 status = 0; /* mapped up to off */ 803 else { 804 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 805 0, curthread->td_ucred, curthread); 806 if (error == 0 && off > roff) 807 status = 1; /* deallocated up to off */ 808 else { 809 status = 0; /* unknown up to the end */ 810 off = be_lun->size_bytes; 811 } 812 } 813 VOP_UNLOCK(be_lun->vn); 814 815 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 816 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 817 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 818 lbalen->lba), data->descr[0].length); 819 data->descr[0].status = status; 820 821 ctl_complete_beio(beio); 822 } 823 824 static uint64_t 825 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 826 { 827 struct vattr vattr; 828 struct statfs statfs; 829 uint64_t val; 830 int error; 831 832 val = UINT64_MAX; 833 if (be_lun->vn == NULL) 834 return (val); 835 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 836 if (strcmp(attrname, "blocksused") == 0) { 837 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 838 if (error == 0) 839 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 840 } 841 if (strcmp(attrname, "blocksavail") == 0 && 842 !VN_IS_DOOMED(be_lun->vn)) { 843 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 844 if (error == 0) 845 val = statfs.f_bavail * statfs.f_bsize / 846 be_lun->cbe_lun.blocksize; 847 } 848 VOP_UNLOCK(be_lun->vn); 849 return (val); 850 } 851 852 static void 853 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 854 struct ctl_be_block_io *beio) 855 { 856 union ctl_io *io; 857 struct cdevsw *csw; 858 struct cdev *dev; 859 struct uio xuio; 860 struct iovec *xiovec; 861 int error, flags, i, ref; 862 863 DPRINTF("entered\n"); 864 865 io = beio->io; 866 flags = 0; 867 if (ARGS(io)->flags & CTL_LLF_DPO) 868 flags |= IO_DIRECT; 869 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 870 flags |= IO_SYNC; 871 872 bzero(&xuio, sizeof(xuio)); 873 if (beio->bio_cmd == BIO_READ) { 874 SDT_PROBE0(cbb, , read, file_start); 875 xuio.uio_rw = UIO_READ; 876 } else { 877 SDT_PROBE0(cbb, , write, file_start); 878 xuio.uio_rw = UIO_WRITE; 879 } 880 xuio.uio_offset = beio->io_offset; 881 xuio.uio_resid = beio->io_len; 882 xuio.uio_segflg = UIO_SYSSPACE; 883 xuio.uio_iov = beio->xiovecs; 884 xuio.uio_iovcnt = beio->num_segs; 885 xuio.uio_td = curthread; 886 887 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 888 xiovec->iov_base = beio->sg_segs[i].addr; 889 xiovec->iov_len = beio->sg_segs[i].len; 890 } 891 892 binuptime(&beio->ds_t0); 893 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 894 895 csw = devvn_refthread(be_lun->vn, &dev, &ref); 896 if (csw) { 897 if (beio->bio_cmd == BIO_READ) 898 error = csw->d_read(dev, &xuio, flags); 899 else 900 error = csw->d_write(dev, &xuio, flags); 901 dev_relthread(dev, ref); 902 } else 903 error = ENXIO; 904 905 if (beio->bio_cmd == BIO_READ) 906 SDT_PROBE0(cbb, , read, file_done); 907 else 908 SDT_PROBE0(cbb, , write, file_done); 909 910 mtx_lock(&be_lun->io_lock); 911 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 912 beio->ds_tag_type, beio->ds_trans_type, 913 /*now*/ NULL, /*then*/&beio->ds_t0); 914 mtx_unlock(&be_lun->io_lock); 915 916 /* 917 * If we got an error, set the sense data to "MEDIUM ERROR" and 918 * return the I/O to the user. 919 */ 920 if (error != 0) { 921 if (error == ENOSPC || error == EDQUOT) { 922 ctl_set_space_alloc_fail(&io->scsiio); 923 } else if (error == EROFS || error == EACCES) { 924 ctl_set_hw_write_protected(&io->scsiio); 925 } else { 926 ctl_set_medium_error(&io->scsiio, 927 beio->bio_cmd == BIO_READ); 928 } 929 ctl_complete_beio(beio); 930 return; 931 } 932 933 /* 934 * If this is a write or a verify, we're all done. 935 * If this is a read, we can now send the data to the user. 936 */ 937 if ((beio->bio_cmd == BIO_WRITE) || 938 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 939 ctl_set_success(&io->scsiio); 940 ctl_complete_beio(beio); 941 } else { 942 if ((ARGS(io)->flags & CTL_LLF_READ) && 943 beio->beio_cont == NULL) { 944 ctl_set_success(&io->scsiio); 945 ctl_serseq_done(io); 946 } 947 #ifdef CTL_TIME_IO 948 getbinuptime(&io->io_hdr.dma_start_bt); 949 #endif 950 ctl_datamove(io); 951 } 952 } 953 954 static void 955 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 956 struct ctl_be_block_io *beio) 957 { 958 union ctl_io *io = beio->io; 959 struct cdevsw *csw; 960 struct cdev *dev; 961 struct ctl_lba_len_flags *lbalen = ARGS(io); 962 struct scsi_get_lba_status_data *data; 963 off_t roff, off; 964 int error, ref, status; 965 966 DPRINTF("entered\n"); 967 968 csw = devvn_refthread(be_lun->vn, &dev, &ref); 969 if (csw == NULL) { 970 status = 0; /* unknown up to the end */ 971 off = be_lun->size_bytes; 972 goto done; 973 } 974 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 975 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD, 976 curthread); 977 if (error == 0 && off > roff) 978 status = 0; /* mapped up to off */ 979 else { 980 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD, 981 curthread); 982 if (error == 0 && off > roff) 983 status = 1; /* deallocated up to off */ 984 else { 985 status = 0; /* unknown up to the end */ 986 off = be_lun->size_bytes; 987 } 988 } 989 dev_relthread(dev, ref); 990 991 done: 992 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 993 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 994 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 995 lbalen->lba), data->descr[0].length); 996 data->descr[0].status = status; 997 998 ctl_complete_beio(beio); 999 } 1000 1001 static void 1002 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 1003 struct ctl_be_block_io *beio) 1004 { 1005 struct bio *bio; 1006 struct cdevsw *csw; 1007 struct cdev *dev; 1008 int ref; 1009 1010 DPRINTF("entered\n"); 1011 1012 /* This can't fail, it's a blocking allocation. */ 1013 bio = g_alloc_bio(); 1014 1015 bio->bio_cmd = BIO_FLUSH; 1016 bio->bio_offset = 0; 1017 bio->bio_data = 0; 1018 bio->bio_done = ctl_be_block_biodone; 1019 bio->bio_caller1 = beio; 1020 bio->bio_pblkno = 0; 1021 1022 /* 1023 * We don't need to acquire the LUN lock here, because we are only 1024 * sending one bio, and so there is no other context to synchronize 1025 * with. 1026 */ 1027 beio->num_bios_sent = 1; 1028 beio->send_complete = 1; 1029 1030 binuptime(&beio->ds_t0); 1031 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1032 1033 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1034 if (csw) { 1035 bio->bio_dev = dev; 1036 csw->d_strategy(bio); 1037 dev_relthread(dev, ref); 1038 } else { 1039 bio->bio_error = ENXIO; 1040 ctl_be_block_biodone(bio); 1041 } 1042 } 1043 1044 static void 1045 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1046 struct ctl_be_block_io *beio, 1047 uint64_t off, uint64_t len, int last) 1048 { 1049 struct bio *bio; 1050 uint64_t maxlen; 1051 struct cdevsw *csw; 1052 struct cdev *dev; 1053 int ref; 1054 1055 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1056 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1057 while (len > 0) { 1058 bio = g_alloc_bio(); 1059 bio->bio_cmd = BIO_DELETE; 1060 bio->bio_dev = dev; 1061 bio->bio_offset = off; 1062 bio->bio_length = MIN(len, maxlen); 1063 bio->bio_data = 0; 1064 bio->bio_done = ctl_be_block_biodone; 1065 bio->bio_caller1 = beio; 1066 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1067 1068 off += bio->bio_length; 1069 len -= bio->bio_length; 1070 1071 mtx_lock(&be_lun->io_lock); 1072 beio->num_bios_sent++; 1073 if (last && len == 0) 1074 beio->send_complete = 1; 1075 mtx_unlock(&be_lun->io_lock); 1076 1077 if (csw) { 1078 csw->d_strategy(bio); 1079 } else { 1080 bio->bio_error = ENXIO; 1081 ctl_be_block_biodone(bio); 1082 } 1083 } 1084 if (csw) 1085 dev_relthread(dev, ref); 1086 } 1087 1088 static void 1089 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1090 struct ctl_be_block_io *beio) 1091 { 1092 union ctl_io *io; 1093 struct ctl_ptr_len_flags *ptrlen; 1094 struct scsi_unmap_desc *buf, *end; 1095 uint64_t len; 1096 1097 io = beio->io; 1098 1099 DPRINTF("entered\n"); 1100 1101 binuptime(&beio->ds_t0); 1102 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1103 1104 if (beio->io_offset == -1) { 1105 beio->io_len = 0; 1106 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1107 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1108 end = buf + ptrlen->len / sizeof(*buf); 1109 for (; buf < end; buf++) { 1110 len = (uint64_t)scsi_4btoul(buf->length) * 1111 be_lun->cbe_lun.blocksize; 1112 beio->io_len += len; 1113 ctl_be_block_unmap_dev_range(be_lun, beio, 1114 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1115 len, (end - buf < 2) ? TRUE : FALSE); 1116 } 1117 } else 1118 ctl_be_block_unmap_dev_range(be_lun, beio, 1119 beio->io_offset, beio->io_len, TRUE); 1120 } 1121 1122 static void 1123 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1124 struct ctl_be_block_io *beio) 1125 { 1126 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1127 struct bio *bio; 1128 struct cdevsw *csw; 1129 struct cdev *dev; 1130 off_t cur_offset; 1131 int i, max_iosize, ref; 1132 1133 DPRINTF("entered\n"); 1134 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1135 1136 /* 1137 * We have to limit our I/O size to the maximum supported by the 1138 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1139 * set it properly, use DFLTPHYS. 1140 */ 1141 if (csw) { 1142 max_iosize = dev->si_iosize_max; 1143 if (max_iosize < PAGE_SIZE) 1144 max_iosize = DFLTPHYS; 1145 } else 1146 max_iosize = DFLTPHYS; 1147 1148 cur_offset = beio->io_offset; 1149 for (i = 0; i < beio->num_segs; i++) { 1150 size_t cur_size; 1151 uint8_t *cur_ptr; 1152 1153 cur_size = beio->sg_segs[i].len; 1154 cur_ptr = beio->sg_segs[i].addr; 1155 1156 while (cur_size > 0) { 1157 /* This can't fail, it's a blocking allocation. */ 1158 bio = g_alloc_bio(); 1159 1160 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1161 1162 bio->bio_cmd = beio->bio_cmd; 1163 bio->bio_dev = dev; 1164 bio->bio_caller1 = beio; 1165 bio->bio_length = min(cur_size, max_iosize); 1166 bio->bio_offset = cur_offset; 1167 bio->bio_data = cur_ptr; 1168 bio->bio_done = ctl_be_block_biodone; 1169 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1170 1171 cur_offset += bio->bio_length; 1172 cur_ptr += bio->bio_length; 1173 cur_size -= bio->bio_length; 1174 1175 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1176 beio->num_bios_sent++; 1177 } 1178 } 1179 beio->send_complete = 1; 1180 binuptime(&beio->ds_t0); 1181 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1182 1183 /* 1184 * Fire off all allocated requests! 1185 */ 1186 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1187 TAILQ_REMOVE(&queue, bio, bio_queue); 1188 if (csw) 1189 csw->d_strategy(bio); 1190 else { 1191 bio->bio_error = ENXIO; 1192 ctl_be_block_biodone(bio); 1193 } 1194 } 1195 if (csw) 1196 dev_relthread(dev, ref); 1197 } 1198 1199 static uint64_t 1200 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1201 { 1202 struct diocgattr_arg arg; 1203 struct cdevsw *csw; 1204 struct cdev *dev; 1205 int error, ref; 1206 1207 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1208 if (csw == NULL) 1209 return (UINT64_MAX); 1210 strlcpy(arg.name, attrname, sizeof(arg.name)); 1211 arg.len = sizeof(arg.value.off); 1212 if (csw->d_ioctl) { 1213 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 1214 curthread); 1215 } else 1216 error = ENODEV; 1217 dev_relthread(dev, ref); 1218 if (error != 0) 1219 return (UINT64_MAX); 1220 return (arg.value.off); 1221 } 1222 1223 static void 1224 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1225 union ctl_io *io) 1226 { 1227 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1228 struct ctl_be_block_io *beio; 1229 struct ctl_lba_len_flags *lbalen; 1230 1231 DPRINTF("entered\n"); 1232 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1233 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1234 1235 beio->io_len = lbalen->len * cbe_lun->blocksize; 1236 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1237 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1238 beio->bio_cmd = BIO_FLUSH; 1239 beio->ds_trans_type = DEVSTAT_NO_DATA; 1240 DPRINTF("SYNC\n"); 1241 be_lun->lun_flush(be_lun, beio); 1242 } 1243 1244 static void 1245 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1246 { 1247 union ctl_io *io; 1248 1249 io = beio->io; 1250 ctl_free_beio(beio); 1251 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1252 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1253 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1254 ctl_config_write_done(io); 1255 return; 1256 } 1257 1258 ctl_be_block_config_write(io); 1259 } 1260 1261 static void 1262 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1263 union ctl_io *io) 1264 { 1265 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1266 struct ctl_be_block_io *beio; 1267 struct ctl_lba_len_flags *lbalen; 1268 uint64_t len_left, lba; 1269 uint32_t pb, pbo, adj; 1270 int i, seglen; 1271 uint8_t *buf, *end; 1272 1273 DPRINTF("entered\n"); 1274 1275 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1276 lbalen = ARGS(beio->io); 1277 1278 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1279 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1280 ctl_free_beio(beio); 1281 ctl_set_invalid_field(&io->scsiio, 1282 /*sks_valid*/ 1, 1283 /*command*/ 1, 1284 /*field*/ 1, 1285 /*bit_valid*/ 0, 1286 /*bit*/ 0); 1287 ctl_config_write_done(io); 1288 return; 1289 } 1290 1291 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1292 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1293 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1294 beio->bio_cmd = BIO_DELETE; 1295 beio->ds_trans_type = DEVSTAT_FREE; 1296 1297 be_lun->unmap(be_lun, beio); 1298 return; 1299 } 1300 1301 beio->bio_cmd = BIO_WRITE; 1302 beio->ds_trans_type = DEVSTAT_WRITE; 1303 1304 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1305 (uintmax_t)lbalen->lba, lbalen->len); 1306 1307 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1308 if (be_lun->cbe_lun.pblockoff > 0) 1309 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1310 else 1311 pbo = 0; 1312 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1313 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1314 1315 /* 1316 * Setup the S/G entry for this chunk. 1317 */ 1318 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1319 if (pb > cbe_lun->blocksize) { 1320 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1321 seglen - pbo) % pb; 1322 if (seglen > adj) 1323 seglen -= adj; 1324 else 1325 seglen -= seglen % cbe_lun->blocksize; 1326 } else 1327 seglen -= seglen % cbe_lun->blocksize; 1328 beio->sg_segs[i].len = seglen; 1329 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1330 1331 DPRINTF("segment %d addr %p len %zd\n", i, 1332 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1333 1334 beio->num_segs++; 1335 len_left -= seglen; 1336 1337 buf = beio->sg_segs[i].addr; 1338 end = buf + seglen; 1339 for (; buf < end; buf += cbe_lun->blocksize) { 1340 if (lbalen->flags & SWS_NDOB) { 1341 memset(buf, 0, cbe_lun->blocksize); 1342 } else { 1343 memcpy(buf, io->scsiio.kern_data_ptr, 1344 cbe_lun->blocksize); 1345 } 1346 if (lbalen->flags & SWS_LBDATA) 1347 scsi_ulto4b(lbalen->lba + lba, buf); 1348 lba++; 1349 } 1350 } 1351 1352 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1353 beio->io_len = lba * cbe_lun->blocksize; 1354 1355 /* We can not do all in one run. Correct and schedule rerun. */ 1356 if (len_left > 0) { 1357 lbalen->lba += lba; 1358 lbalen->len -= lba; 1359 beio->beio_cont = ctl_be_block_cw_done_ws; 1360 } 1361 1362 be_lun->dispatch(be_lun, beio); 1363 } 1364 1365 static void 1366 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1367 union ctl_io *io) 1368 { 1369 struct ctl_be_block_io *beio; 1370 struct ctl_ptr_len_flags *ptrlen; 1371 1372 DPRINTF("entered\n"); 1373 1374 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1375 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1376 1377 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1378 ctl_free_beio(beio); 1379 ctl_set_invalid_field(&io->scsiio, 1380 /*sks_valid*/ 0, 1381 /*command*/ 1, 1382 /*field*/ 0, 1383 /*bit_valid*/ 0, 1384 /*bit*/ 0); 1385 ctl_config_write_done(io); 1386 return; 1387 } 1388 1389 beio->io_len = 0; 1390 beio->io_offset = -1; 1391 beio->bio_cmd = BIO_DELETE; 1392 beio->ds_trans_type = DEVSTAT_FREE; 1393 DPRINTF("UNMAP\n"); 1394 be_lun->unmap(be_lun, beio); 1395 } 1396 1397 static void 1398 ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1399 { 1400 union ctl_io *io; 1401 1402 io = beio->io; 1403 ctl_free_beio(beio); 1404 ctl_config_read_done(io); 1405 } 1406 1407 static void 1408 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1409 union ctl_io *io) 1410 { 1411 struct ctl_be_block_io *beio; 1412 struct ctl_be_block_softc *softc; 1413 1414 DPRINTF("entered\n"); 1415 1416 softc = be_lun->softc; 1417 beio = ctl_alloc_beio(softc); 1418 beio->io = io; 1419 beio->lun = be_lun; 1420 beio->beio_cont = ctl_be_block_cr_done; 1421 PRIV(io)->ptr = (void *)beio; 1422 1423 switch (io->scsiio.cdb[0]) { 1424 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1425 beio->bio_cmd = -1; 1426 beio->ds_trans_type = DEVSTAT_NO_DATA; 1427 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1428 beio->io_len = 0; 1429 if (be_lun->get_lba_status) 1430 be_lun->get_lba_status(be_lun, beio); 1431 else 1432 ctl_be_block_cr_done(beio); 1433 break; 1434 default: 1435 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1436 break; 1437 } 1438 } 1439 1440 static void 1441 ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1442 { 1443 union ctl_io *io; 1444 1445 io = beio->io; 1446 ctl_free_beio(beio); 1447 ctl_config_write_done(io); 1448 } 1449 1450 static void 1451 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1452 union ctl_io *io) 1453 { 1454 struct ctl_be_block_io *beio; 1455 struct ctl_be_block_softc *softc; 1456 1457 DPRINTF("entered\n"); 1458 1459 softc = be_lun->softc; 1460 beio = ctl_alloc_beio(softc); 1461 beio->io = io; 1462 beio->lun = be_lun; 1463 beio->beio_cont = ctl_be_block_cw_done; 1464 switch (io->scsiio.tag_type) { 1465 case CTL_TAG_ORDERED: 1466 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1467 break; 1468 case CTL_TAG_HEAD_OF_QUEUE: 1469 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1470 break; 1471 case CTL_TAG_UNTAGGED: 1472 case CTL_TAG_SIMPLE: 1473 case CTL_TAG_ACA: 1474 default: 1475 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1476 break; 1477 } 1478 PRIV(io)->ptr = (void *)beio; 1479 1480 switch (io->scsiio.cdb[0]) { 1481 case SYNCHRONIZE_CACHE: 1482 case SYNCHRONIZE_CACHE_16: 1483 ctl_be_block_cw_dispatch_sync(be_lun, io); 1484 break; 1485 case WRITE_SAME_10: 1486 case WRITE_SAME_16: 1487 ctl_be_block_cw_dispatch_ws(be_lun, io); 1488 break; 1489 case UNMAP: 1490 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1491 break; 1492 default: 1493 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1494 break; 1495 } 1496 } 1497 1498 SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t"); 1499 SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t"); 1500 SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t"); 1501 SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t"); 1502 1503 static void 1504 ctl_be_block_next(struct ctl_be_block_io *beio) 1505 { 1506 struct ctl_be_block_lun *be_lun; 1507 union ctl_io *io; 1508 1509 io = beio->io; 1510 be_lun = beio->lun; 1511 ctl_free_beio(beio); 1512 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1513 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1514 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1515 ctl_data_submit_done(io); 1516 return; 1517 } 1518 1519 io->io_hdr.status &= ~CTL_STATUS_MASK; 1520 io->io_hdr.status |= CTL_STATUS_NONE; 1521 1522 mtx_lock(&be_lun->queue_lock); 1523 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1524 mtx_unlock(&be_lun->queue_lock); 1525 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1526 } 1527 1528 static void 1529 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1530 union ctl_io *io) 1531 { 1532 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1533 struct ctl_be_block_io *beio; 1534 struct ctl_be_block_softc *softc; 1535 struct ctl_lba_len_flags *lbalen; 1536 struct ctl_ptr_len_flags *bptrlen; 1537 uint64_t len_left, lbas; 1538 int i; 1539 1540 softc = be_lun->softc; 1541 1542 DPRINTF("entered\n"); 1543 1544 lbalen = ARGS(io); 1545 if (lbalen->flags & CTL_LLF_WRITE) { 1546 SDT_PROBE0(cbb, , write, start); 1547 } else { 1548 SDT_PROBE0(cbb, , read, start); 1549 } 1550 1551 beio = ctl_alloc_beio(softc); 1552 beio->io = io; 1553 beio->lun = be_lun; 1554 bptrlen = PRIV(io); 1555 bptrlen->ptr = (void *)beio; 1556 1557 switch (io->scsiio.tag_type) { 1558 case CTL_TAG_ORDERED: 1559 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1560 break; 1561 case CTL_TAG_HEAD_OF_QUEUE: 1562 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1563 break; 1564 case CTL_TAG_UNTAGGED: 1565 case CTL_TAG_SIMPLE: 1566 case CTL_TAG_ACA: 1567 default: 1568 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1569 break; 1570 } 1571 1572 if (lbalen->flags & CTL_LLF_WRITE) { 1573 beio->bio_cmd = BIO_WRITE; 1574 beio->ds_trans_type = DEVSTAT_WRITE; 1575 } else { 1576 beio->bio_cmd = BIO_READ; 1577 beio->ds_trans_type = DEVSTAT_READ; 1578 } 1579 1580 DPRINTF("%s at LBA %jx len %u @%ju\n", 1581 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1582 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1583 if (lbalen->flags & CTL_LLF_COMPARE) 1584 lbas = CTLBLK_HALF_IO_SIZE; 1585 else 1586 lbas = CTLBLK_MAX_IO_SIZE; 1587 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1588 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1589 beio->io_len = lbas * cbe_lun->blocksize; 1590 bptrlen->len += lbas; 1591 1592 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1593 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1594 i, CTLBLK_MAX_SEGS)); 1595 1596 /* 1597 * Setup the S/G entry for this chunk. 1598 */ 1599 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1600 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1601 1602 DPRINTF("segment %d addr %p len %zd\n", i, 1603 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1604 1605 /* Set up second segment for compare operation. */ 1606 if (lbalen->flags & CTL_LLF_COMPARE) { 1607 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1608 beio->sg_segs[i].len; 1609 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1610 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1611 } 1612 1613 beio->num_segs++; 1614 len_left -= beio->sg_segs[i].len; 1615 } 1616 if (bptrlen->len < lbalen->len) 1617 beio->beio_cont = ctl_be_block_next; 1618 io->scsiio.be_move_done = ctl_be_block_move_done; 1619 /* For compare we have separate S/G lists for read and datamove. */ 1620 if (lbalen->flags & CTL_LLF_COMPARE) 1621 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1622 else 1623 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1624 io->scsiio.kern_data_len = beio->io_len; 1625 io->scsiio.kern_sg_entries = beio->num_segs; 1626 io->io_hdr.flags |= CTL_FLAG_ALLOCATED; 1627 1628 /* 1629 * For the read case, we need to read the data into our buffers and 1630 * then we can send it back to the user. For the write case, we 1631 * need to get the data from the user first. 1632 */ 1633 if (beio->bio_cmd == BIO_READ) { 1634 SDT_PROBE0(cbb, , read, alloc_done); 1635 be_lun->dispatch(be_lun, beio); 1636 } else { 1637 SDT_PROBE0(cbb, , write, alloc_done); 1638 #ifdef CTL_TIME_IO 1639 getbinuptime(&io->io_hdr.dma_start_bt); 1640 #endif 1641 ctl_datamove(io); 1642 } 1643 } 1644 1645 static void 1646 ctl_be_block_worker(void *context, int pending) 1647 { 1648 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; 1649 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1650 union ctl_io *io; 1651 struct ctl_be_block_io *beio; 1652 1653 DPRINTF("entered\n"); 1654 /* 1655 * Fetch and process I/Os from all queues. If we detect LUN 1656 * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race, 1657 * so make response maximally opaque to not confuse initiator. 1658 */ 1659 for (;;) { 1660 mtx_lock(&be_lun->queue_lock); 1661 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1662 if (io != NULL) { 1663 DPRINTF("datamove queue\n"); 1664 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1665 ctl_io_hdr, links); 1666 mtx_unlock(&be_lun->queue_lock); 1667 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1668 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1669 ctl_set_busy(&io->scsiio); 1670 ctl_complete_beio(beio); 1671 return; 1672 } 1673 be_lun->dispatch(be_lun, beio); 1674 continue; 1675 } 1676 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1677 if (io != NULL) { 1678 DPRINTF("config write queue\n"); 1679 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1680 ctl_io_hdr, links); 1681 mtx_unlock(&be_lun->queue_lock); 1682 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1683 ctl_set_busy(&io->scsiio); 1684 ctl_config_write_done(io); 1685 return; 1686 } 1687 ctl_be_block_cw_dispatch(be_lun, io); 1688 continue; 1689 } 1690 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1691 if (io != NULL) { 1692 DPRINTF("config read queue\n"); 1693 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1694 ctl_io_hdr, links); 1695 mtx_unlock(&be_lun->queue_lock); 1696 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1697 ctl_set_busy(&io->scsiio); 1698 ctl_config_read_done(io); 1699 return; 1700 } 1701 ctl_be_block_cr_dispatch(be_lun, io); 1702 continue; 1703 } 1704 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1705 if (io != NULL) { 1706 DPRINTF("input queue\n"); 1707 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1708 ctl_io_hdr, links); 1709 mtx_unlock(&be_lun->queue_lock); 1710 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1711 ctl_set_busy(&io->scsiio); 1712 ctl_data_submit_done(io); 1713 return; 1714 } 1715 ctl_be_block_dispatch(be_lun, io); 1716 continue; 1717 } 1718 1719 /* 1720 * If we get here, there is no work left in the queues, so 1721 * just break out and let the task queue go to sleep. 1722 */ 1723 mtx_unlock(&be_lun->queue_lock); 1724 break; 1725 } 1726 } 1727 1728 /* 1729 * Entry point from CTL to the backend for I/O. We queue everything to a 1730 * work thread, so this just puts the I/O on a queue and wakes up the 1731 * thread. 1732 */ 1733 static int 1734 ctl_be_block_submit(union ctl_io *io) 1735 { 1736 struct ctl_be_block_lun *be_lun; 1737 struct ctl_be_lun *cbe_lun; 1738 1739 DPRINTF("entered\n"); 1740 1741 cbe_lun = CTL_BACKEND_LUN(io); 1742 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 1743 1744 /* 1745 * Make sure we only get SCSI I/O. 1746 */ 1747 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1748 "%#x) encountered", io->io_hdr.io_type)); 1749 1750 PRIV(io)->len = 0; 1751 1752 mtx_lock(&be_lun->queue_lock); 1753 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1754 mtx_unlock(&be_lun->queue_lock); 1755 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1756 1757 return (CTL_RETVAL_COMPLETE); 1758 } 1759 1760 static int 1761 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1762 int flag, struct thread *td) 1763 { 1764 struct ctl_be_block_softc *softc; 1765 int error; 1766 1767 softc = &backend_block_softc; 1768 1769 error = 0; 1770 1771 switch (cmd) { 1772 case CTL_LUN_REQ: { 1773 struct ctl_lun_req *lun_req; 1774 1775 lun_req = (struct ctl_lun_req *)addr; 1776 1777 switch (lun_req->reqtype) { 1778 case CTL_LUNREQ_CREATE: 1779 error = ctl_be_block_create(softc, lun_req); 1780 break; 1781 case CTL_LUNREQ_RM: 1782 error = ctl_be_block_rm(softc, lun_req); 1783 break; 1784 case CTL_LUNREQ_MODIFY: 1785 error = ctl_be_block_modify(softc, lun_req); 1786 break; 1787 default: 1788 lun_req->status = CTL_LUN_ERROR; 1789 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1790 "invalid LUN request type %d", 1791 lun_req->reqtype); 1792 break; 1793 } 1794 break; 1795 } 1796 default: 1797 error = ENOTTY; 1798 break; 1799 } 1800 1801 return (error); 1802 } 1803 1804 static int 1805 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1806 { 1807 struct ctl_be_lun *cbe_lun; 1808 struct ctl_be_block_filedata *file_data; 1809 struct ctl_lun_create_params *params; 1810 const char *value; 1811 struct vattr vattr; 1812 off_t ps, pss, po, pos, us, uss, uo, uos; 1813 int error; 1814 1815 cbe_lun = &be_lun->cbe_lun; 1816 file_data = &be_lun->backend.file; 1817 params = &be_lun->params; 1818 1819 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1820 be_lun->dispatch = ctl_be_block_dispatch_file; 1821 be_lun->lun_flush = ctl_be_block_flush_file; 1822 be_lun->get_lba_status = ctl_be_block_gls_file; 1823 be_lun->getattr = ctl_be_block_getattr_file; 1824 be_lun->unmap = NULL; 1825 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1826 1827 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1828 if (error != 0) { 1829 snprintf(req->error_str, sizeof(req->error_str), 1830 "error calling VOP_GETATTR() for file %s", 1831 be_lun->dev_path); 1832 return (error); 1833 } 1834 1835 file_data->cred = crhold(curthread->td_ucred); 1836 if (params->lun_size_bytes != 0) 1837 be_lun->size_bytes = params->lun_size_bytes; 1838 else 1839 be_lun->size_bytes = vattr.va_size; 1840 1841 /* 1842 * For files we can use any logical block size. Prefer 512 bytes 1843 * for compatibility reasons. If file's vattr.va_blocksize 1844 * (preferred I/O block size) is bigger and multiple to chosen 1845 * logical block size -- report it as physical block size. 1846 */ 1847 if (params->blocksize_bytes != 0) 1848 cbe_lun->blocksize = params->blocksize_bytes; 1849 else if (cbe_lun->lun_type == T_CDROM) 1850 cbe_lun->blocksize = 2048; 1851 else 1852 cbe_lun->blocksize = 512; 1853 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1854 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1855 0 : (be_lun->size_blocks - 1); 1856 1857 us = ps = vattr.va_blocksize; 1858 uo = po = 0; 1859 1860 value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL); 1861 if (value != NULL) 1862 ctl_expand_number(value, &ps); 1863 value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL); 1864 if (value != NULL) 1865 ctl_expand_number(value, &po); 1866 pss = ps / cbe_lun->blocksize; 1867 pos = po / cbe_lun->blocksize; 1868 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1869 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1870 cbe_lun->pblockexp = fls(pss) - 1; 1871 cbe_lun->pblockoff = (pss - pos) % pss; 1872 } 1873 1874 value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL); 1875 if (value != NULL) 1876 ctl_expand_number(value, &us); 1877 value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL); 1878 if (value != NULL) 1879 ctl_expand_number(value, &uo); 1880 uss = us / cbe_lun->blocksize; 1881 uos = uo / cbe_lun->blocksize; 1882 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1883 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1884 cbe_lun->ublockexp = fls(uss) - 1; 1885 cbe_lun->ublockoff = (uss - uos) % uss; 1886 } 1887 1888 /* 1889 * Sanity check. The media size has to be at least one 1890 * sector long. 1891 */ 1892 if (be_lun->size_bytes < cbe_lun->blocksize) { 1893 error = EINVAL; 1894 snprintf(req->error_str, sizeof(req->error_str), 1895 "file %s size %ju < block size %u", be_lun->dev_path, 1896 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1897 } 1898 1899 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1900 return (error); 1901 } 1902 1903 static int 1904 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1905 { 1906 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1907 struct ctl_lun_create_params *params; 1908 struct cdevsw *csw; 1909 struct cdev *dev; 1910 const char *value; 1911 int error, atomic, maxio, ref, unmap, tmp; 1912 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1913 1914 params = &be_lun->params; 1915 1916 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1917 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1918 if (csw == NULL) 1919 return (ENXIO); 1920 if (strcmp(csw->d_name, "zvol") == 0) { 1921 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1922 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1923 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1924 } else { 1925 be_lun->dispatch = ctl_be_block_dispatch_dev; 1926 be_lun->get_lba_status = NULL; 1927 atomic = 0; 1928 maxio = dev->si_iosize_max; 1929 if (maxio <= 0) 1930 maxio = DFLTPHYS; 1931 if (maxio > CTLBLK_MAX_IO_SIZE) 1932 maxio = CTLBLK_MAX_IO_SIZE; 1933 } 1934 be_lun->lun_flush = ctl_be_block_flush_dev; 1935 be_lun->getattr = ctl_be_block_getattr_dev; 1936 be_lun->unmap = ctl_be_block_unmap_dev; 1937 1938 if (!csw->d_ioctl) { 1939 dev_relthread(dev, ref); 1940 snprintf(req->error_str, sizeof(req->error_str), 1941 "no d_ioctl for device %s!", be_lun->dev_path); 1942 return (ENODEV); 1943 } 1944 1945 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 1946 curthread); 1947 if (error) { 1948 dev_relthread(dev, ref); 1949 snprintf(req->error_str, sizeof(req->error_str), 1950 "error %d returned for DIOCGSECTORSIZE ioctl " 1951 "on %s!", error, be_lun->dev_path); 1952 return (error); 1953 } 1954 1955 /* 1956 * If the user has asked for a blocksize that is greater than the 1957 * backing device's blocksize, we can do it only if the blocksize 1958 * the user is asking for is an even multiple of the underlying 1959 * device's blocksize. 1960 */ 1961 if ((params->blocksize_bytes != 0) && 1962 (params->blocksize_bytes >= tmp)) { 1963 if (params->blocksize_bytes % tmp == 0) { 1964 cbe_lun->blocksize = params->blocksize_bytes; 1965 } else { 1966 dev_relthread(dev, ref); 1967 snprintf(req->error_str, sizeof(req->error_str), 1968 "requested blocksize %u is not an even " 1969 "multiple of backing device blocksize %u", 1970 params->blocksize_bytes, tmp); 1971 return (EINVAL); 1972 } 1973 } else if (params->blocksize_bytes != 0) { 1974 dev_relthread(dev, ref); 1975 snprintf(req->error_str, sizeof(req->error_str), 1976 "requested blocksize %u < backing device " 1977 "blocksize %u", params->blocksize_bytes, tmp); 1978 return (EINVAL); 1979 } else if (cbe_lun->lun_type == T_CDROM) 1980 cbe_lun->blocksize = MAX(tmp, 2048); 1981 else 1982 cbe_lun->blocksize = tmp; 1983 1984 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 1985 curthread); 1986 if (error) { 1987 dev_relthread(dev, ref); 1988 snprintf(req->error_str, sizeof(req->error_str), 1989 "error %d returned for DIOCGMEDIASIZE " 1990 " ioctl on %s!", error, 1991 be_lun->dev_path); 1992 return (error); 1993 } 1994 1995 if (params->lun_size_bytes != 0) { 1996 if (params->lun_size_bytes > otmp) { 1997 dev_relthread(dev, ref); 1998 snprintf(req->error_str, sizeof(req->error_str), 1999 "requested LUN size %ju > backing device " 2000 "size %ju", 2001 (uintmax_t)params->lun_size_bytes, 2002 (uintmax_t)otmp); 2003 return (EINVAL); 2004 } 2005 2006 be_lun->size_bytes = params->lun_size_bytes; 2007 } else 2008 be_lun->size_bytes = otmp; 2009 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2010 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2011 0 : (be_lun->size_blocks - 1); 2012 2013 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, 2014 curthread); 2015 if (error) 2016 ps = po = 0; 2017 else { 2018 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, 2019 FREAD, curthread); 2020 if (error) 2021 po = 0; 2022 } 2023 us = ps; 2024 uo = po; 2025 2026 value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL); 2027 if (value != NULL) 2028 ctl_expand_number(value, &ps); 2029 value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL); 2030 if (value != NULL) 2031 ctl_expand_number(value, &po); 2032 pss = ps / cbe_lun->blocksize; 2033 pos = po / cbe_lun->blocksize; 2034 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 2035 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 2036 cbe_lun->pblockexp = fls(pss) - 1; 2037 cbe_lun->pblockoff = (pss - pos) % pss; 2038 } 2039 2040 value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL); 2041 if (value != NULL) 2042 ctl_expand_number(value, &us); 2043 value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL); 2044 if (value != NULL) 2045 ctl_expand_number(value, &uo); 2046 uss = us / cbe_lun->blocksize; 2047 uos = uo / cbe_lun->blocksize; 2048 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2049 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2050 cbe_lun->ublockexp = fls(uss) - 1; 2051 cbe_lun->ublockoff = (uss - uos) % uss; 2052 } 2053 2054 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2055 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2056 2057 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2058 unmap = 1; 2059 } else { 2060 struct diocgattr_arg arg; 2061 2062 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2063 arg.len = sizeof(arg.value.i); 2064 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 2065 curthread); 2066 unmap = (error == 0) ? arg.value.i : 0; 2067 } 2068 value = dnvlist_get_string(cbe_lun->options, "unmap", NULL); 2069 if (value != NULL) 2070 unmap = (strcmp(value, "on") == 0); 2071 if (unmap) 2072 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2073 else 2074 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2075 2076 dev_relthread(dev, ref); 2077 return (0); 2078 } 2079 2080 static int 2081 ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2082 { 2083 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2084 int flags; 2085 2086 if (be_lun->vn) { 2087 flags = FREAD; 2088 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2089 flags |= FWRITE; 2090 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2091 be_lun->vn = NULL; 2092 2093 switch (be_lun->dev_type) { 2094 case CTL_BE_BLOCK_DEV: 2095 break; 2096 case CTL_BE_BLOCK_FILE: 2097 if (be_lun->backend.file.cred != NULL) { 2098 crfree(be_lun->backend.file.cred); 2099 be_lun->backend.file.cred = NULL; 2100 } 2101 break; 2102 case CTL_BE_BLOCK_NONE: 2103 break; 2104 default: 2105 panic("Unexpected backend type %d", be_lun->dev_type); 2106 break; 2107 } 2108 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2109 } 2110 return (0); 2111 } 2112 2113 static int 2114 ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2115 { 2116 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2117 struct nameidata nd; 2118 const char *value; 2119 int error, flags; 2120 2121 error = 0; 2122 if (rootvnode == NULL) { 2123 snprintf(req->error_str, sizeof(req->error_str), 2124 "Root filesystem is not mounted"); 2125 return (1); 2126 } 2127 pwd_ensure_dirs(); 2128 2129 value = dnvlist_get_string(cbe_lun->options, "file", NULL); 2130 if (value == NULL) { 2131 snprintf(req->error_str, sizeof(req->error_str), 2132 "no file argument specified"); 2133 return (1); 2134 } 2135 free(be_lun->dev_path, M_CTLBLK); 2136 be_lun->dev_path = strdup(value, M_CTLBLK); 2137 2138 flags = FREAD; 2139 value = dnvlist_get_string(cbe_lun->options, "readonly", NULL); 2140 if (value != NULL) { 2141 if (strcmp(value, "on") != 0) 2142 flags |= FWRITE; 2143 } else if (cbe_lun->lun_type == T_DIRECT) 2144 flags |= FWRITE; 2145 2146 again: 2147 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2148 error = vn_open(&nd, &flags, 0, NULL); 2149 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2150 flags &= ~FWRITE; 2151 goto again; 2152 } 2153 if (error) { 2154 /* 2155 * This is the only reasonable guess we can make as far as 2156 * path if the user doesn't give us a fully qualified path. 2157 * If they want to specify a file, they need to specify the 2158 * full path. 2159 */ 2160 if (be_lun->dev_path[0] != '/') { 2161 char *dev_name; 2162 2163 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2164 be_lun->dev_path); 2165 free(be_lun->dev_path, M_CTLBLK); 2166 be_lun->dev_path = dev_name; 2167 goto again; 2168 } 2169 snprintf(req->error_str, sizeof(req->error_str), 2170 "error opening %s: %d", be_lun->dev_path, error); 2171 return (error); 2172 } 2173 if (flags & FWRITE) 2174 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2175 else 2176 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2177 2178 NDFREE(&nd, NDF_ONLY_PNBUF); 2179 be_lun->vn = nd.ni_vp; 2180 2181 /* We only support disks and files. */ 2182 if (vn_isdisk(be_lun->vn, &error)) { 2183 error = ctl_be_block_open_dev(be_lun, req); 2184 } else if (be_lun->vn->v_type == VREG) { 2185 error = ctl_be_block_open_file(be_lun, req); 2186 } else { 2187 error = EINVAL; 2188 snprintf(req->error_str, sizeof(req->error_str), 2189 "%s is not a disk or plain file", be_lun->dev_path); 2190 } 2191 VOP_UNLOCK(be_lun->vn); 2192 2193 if (error != 0) 2194 ctl_be_block_close(be_lun); 2195 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2196 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2197 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2198 value = dnvlist_get_string(cbe_lun->options, "serseq", NULL); 2199 if (value != NULL && strcmp(value, "on") == 0) 2200 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2201 else if (value != NULL && strcmp(value, "read") == 0) 2202 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2203 else if (value != NULL && strcmp(value, "off") == 0) 2204 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2205 return (0); 2206 } 2207 2208 static int 2209 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2210 { 2211 struct ctl_be_lun *cbe_lun; 2212 struct ctl_be_block_lun *be_lun; 2213 struct ctl_lun_create_params *params; 2214 char num_thread_str[16]; 2215 char tmpstr[32]; 2216 const char *value; 2217 int retval, num_threads; 2218 int tmp_num_threads; 2219 2220 params = &req->reqdata.create; 2221 retval = 0; 2222 req->status = CTL_LUN_OK; 2223 2224 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2225 cbe_lun = &be_lun->cbe_lun; 2226 cbe_lun->be_lun = be_lun; 2227 be_lun->params = req->reqdata.create; 2228 be_lun->softc = softc; 2229 STAILQ_INIT(&be_lun->input_queue); 2230 STAILQ_INIT(&be_lun->config_read_queue); 2231 STAILQ_INIT(&be_lun->config_write_queue); 2232 STAILQ_INIT(&be_lun->datamove_queue); 2233 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2234 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2235 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2236 cbe_lun->options = nvlist_clone(req->args_nvl); 2237 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2238 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2239 if (be_lun->lun_zone == NULL) { 2240 snprintf(req->error_str, sizeof(req->error_str), 2241 "error allocating UMA zone"); 2242 goto bailout_error; 2243 } 2244 2245 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2246 cbe_lun->lun_type = params->device_type; 2247 else 2248 cbe_lun->lun_type = T_DIRECT; 2249 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2250 cbe_lun->flags = 0; 2251 value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL); 2252 if (value != NULL) { 2253 if (strcmp(value, "primary") == 0) 2254 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2255 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2256 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2257 2258 if (cbe_lun->lun_type == T_DIRECT || 2259 cbe_lun->lun_type == T_CDROM) { 2260 be_lun->size_bytes = params->lun_size_bytes; 2261 if (params->blocksize_bytes != 0) 2262 cbe_lun->blocksize = params->blocksize_bytes; 2263 else if (cbe_lun->lun_type == T_CDROM) 2264 cbe_lun->blocksize = 2048; 2265 else 2266 cbe_lun->blocksize = 512; 2267 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2268 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2269 0 : (be_lun->size_blocks - 1); 2270 2271 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2272 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2273 retval = ctl_be_block_open(be_lun, req); 2274 if (retval != 0) { 2275 retval = 0; 2276 req->status = CTL_LUN_WARNING; 2277 } 2278 } 2279 num_threads = cbb_num_threads; 2280 } else { 2281 num_threads = 1; 2282 } 2283 2284 value = dnvlist_get_string(cbe_lun->options, "num_threads", NULL); 2285 if (value != NULL) { 2286 tmp_num_threads = strtol(value, NULL, 0); 2287 2288 /* 2289 * We don't let the user specify less than one 2290 * thread, but hope he's clueful enough not to 2291 * specify 1000 threads. 2292 */ 2293 if (tmp_num_threads < 1) { 2294 snprintf(req->error_str, sizeof(req->error_str), 2295 "invalid number of threads %s", 2296 num_thread_str); 2297 goto bailout_error; 2298 } 2299 num_threads = tmp_num_threads; 2300 } 2301 2302 if (be_lun->vn == NULL) 2303 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2304 /* Tell the user the blocksize we ended up using */ 2305 params->lun_size_bytes = be_lun->size_bytes; 2306 params->blocksize_bytes = cbe_lun->blocksize; 2307 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2308 cbe_lun->req_lun_id = params->req_lun_id; 2309 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2310 } else 2311 cbe_lun->req_lun_id = 0; 2312 2313 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2314 cbe_lun->lun_config_status = ctl_be_block_lun_config_status; 2315 cbe_lun->be = &ctl_be_block_driver; 2316 2317 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2318 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d", 2319 softc->num_luns); 2320 strncpy((char *)cbe_lun->serial_num, tmpstr, 2321 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2322 2323 /* Tell the user what we used for a serial number */ 2324 strncpy((char *)params->serial_num, tmpstr, 2325 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2326 } else { 2327 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2328 MIN(sizeof(cbe_lun->serial_num), 2329 sizeof(params->serial_num))); 2330 } 2331 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2332 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns); 2333 strncpy((char *)cbe_lun->device_id, tmpstr, 2334 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2335 2336 /* Tell the user what we used for a device ID */ 2337 strncpy((char *)params->device_id, tmpstr, 2338 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2339 } else { 2340 strncpy((char *)cbe_lun->device_id, params->device_id, 2341 MIN(sizeof(cbe_lun->device_id), 2342 sizeof(params->device_id))); 2343 } 2344 2345 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2346 2347 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2348 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2349 2350 if (be_lun->io_taskqueue == NULL) { 2351 snprintf(req->error_str, sizeof(req->error_str), 2352 "unable to create taskqueue"); 2353 goto bailout_error; 2354 } 2355 2356 /* 2357 * Note that we start the same number of threads by default for 2358 * both the file case and the block device case. For the file 2359 * case, we need multiple threads to allow concurrency, because the 2360 * vnode interface is designed to be a blocking interface. For the 2361 * block device case, ZFS zvols at least will block the caller's 2362 * context in many instances, and so we need multiple threads to 2363 * overcome that problem. Other block devices don't need as many 2364 * threads, but they shouldn't cause too many problems. 2365 * 2366 * If the user wants to just have a single thread for a block 2367 * device, he can specify that when the LUN is created, or change 2368 * the tunable/sysctl to alter the default number of threads. 2369 */ 2370 retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue, 2371 /*num threads*/num_threads, 2372 /*priority*/PUSER, 2373 /*proc*/control_softc->ctl_proc, 2374 /*thread name*/ 2375 "%s taskq", be_lun->lunname); 2376 2377 if (retval != 0) 2378 goto bailout_error; 2379 2380 be_lun->num_threads = num_threads; 2381 2382 mtx_lock(&softc->lock); 2383 softc->num_luns++; 2384 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2385 2386 mtx_unlock(&softc->lock); 2387 2388 retval = ctl_add_lun(&be_lun->cbe_lun); 2389 if (retval != 0) { 2390 mtx_lock(&softc->lock); 2391 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2392 links); 2393 softc->num_luns--; 2394 mtx_unlock(&softc->lock); 2395 snprintf(req->error_str, sizeof(req->error_str), 2396 "ctl_add_lun() returned error %d, see dmesg for " 2397 "details", retval); 2398 retval = 0; 2399 goto bailout_error; 2400 } 2401 2402 mtx_lock(&softc->lock); 2403 2404 /* 2405 * Tell the config_status routine that we're waiting so it won't 2406 * clean up the LUN in the event of an error. 2407 */ 2408 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2409 2410 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2411 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2412 if (retval == EINTR) 2413 break; 2414 } 2415 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2416 2417 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2418 snprintf(req->error_str, sizeof(req->error_str), 2419 "LUN configuration error, see dmesg for details"); 2420 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2421 links); 2422 softc->num_luns--; 2423 mtx_unlock(&softc->lock); 2424 goto bailout_error; 2425 } else { 2426 params->req_lun_id = cbe_lun->lun_id; 2427 } 2428 2429 mtx_unlock(&softc->lock); 2430 2431 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2432 cbe_lun->blocksize, 2433 DEVSTAT_ALL_SUPPORTED, 2434 cbe_lun->lun_type 2435 | DEVSTAT_TYPE_IF_OTHER, 2436 DEVSTAT_PRIORITY_OTHER); 2437 2438 return (retval); 2439 2440 bailout_error: 2441 req->status = CTL_LUN_ERROR; 2442 2443 if (be_lun->io_taskqueue != NULL) 2444 taskqueue_free(be_lun->io_taskqueue); 2445 ctl_be_block_close(be_lun); 2446 if (be_lun->dev_path != NULL) 2447 free(be_lun->dev_path, M_CTLBLK); 2448 if (be_lun->lun_zone != NULL) 2449 uma_zdestroy(be_lun->lun_zone); 2450 nvlist_destroy(cbe_lun->options); 2451 mtx_destroy(&be_lun->queue_lock); 2452 mtx_destroy(&be_lun->io_lock); 2453 free(be_lun, M_CTLBLK); 2454 2455 return (retval); 2456 } 2457 2458 static int 2459 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2460 { 2461 struct ctl_lun_rm_params *params; 2462 struct ctl_be_block_lun *be_lun; 2463 struct ctl_be_lun *cbe_lun; 2464 int retval; 2465 2466 params = &req->reqdata.rm; 2467 2468 mtx_lock(&softc->lock); 2469 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2470 if (be_lun->cbe_lun.lun_id == params->lun_id) 2471 break; 2472 } 2473 mtx_unlock(&softc->lock); 2474 if (be_lun == NULL) { 2475 snprintf(req->error_str, sizeof(req->error_str), 2476 "LUN %u is not managed by the block backend", 2477 params->lun_id); 2478 goto bailout_error; 2479 } 2480 cbe_lun = &be_lun->cbe_lun; 2481 2482 retval = ctl_disable_lun(cbe_lun); 2483 if (retval != 0) { 2484 snprintf(req->error_str, sizeof(req->error_str), 2485 "error %d returned from ctl_disable_lun() for " 2486 "LUN %d", retval, params->lun_id); 2487 goto bailout_error; 2488 } 2489 2490 if (be_lun->vn != NULL) { 2491 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2492 ctl_lun_no_media(cbe_lun); 2493 taskqueue_drain_all(be_lun->io_taskqueue); 2494 ctl_be_block_close(be_lun); 2495 } 2496 2497 retval = ctl_invalidate_lun(cbe_lun); 2498 if (retval != 0) { 2499 snprintf(req->error_str, sizeof(req->error_str), 2500 "error %d returned from ctl_invalidate_lun() for " 2501 "LUN %d", retval, params->lun_id); 2502 goto bailout_error; 2503 } 2504 2505 mtx_lock(&softc->lock); 2506 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2507 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2508 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2509 if (retval == EINTR) 2510 break; 2511 } 2512 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2513 2514 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2515 snprintf(req->error_str, sizeof(req->error_str), 2516 "interrupted waiting for LUN to be freed"); 2517 mtx_unlock(&softc->lock); 2518 goto bailout_error; 2519 } 2520 2521 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2522 2523 softc->num_luns--; 2524 mtx_unlock(&softc->lock); 2525 2526 taskqueue_drain_all(be_lun->io_taskqueue); 2527 taskqueue_free(be_lun->io_taskqueue); 2528 2529 if (be_lun->disk_stats != NULL) 2530 devstat_remove_entry(be_lun->disk_stats); 2531 2532 uma_zdestroy(be_lun->lun_zone); 2533 2534 nvlist_destroy(cbe_lun->options); 2535 free(be_lun->dev_path, M_CTLBLK); 2536 mtx_destroy(&be_lun->queue_lock); 2537 mtx_destroy(&be_lun->io_lock); 2538 free(be_lun, M_CTLBLK); 2539 2540 req->status = CTL_LUN_OK; 2541 return (0); 2542 2543 bailout_error: 2544 req->status = CTL_LUN_ERROR; 2545 return (0); 2546 } 2547 2548 static int 2549 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2550 { 2551 struct ctl_lun_modify_params *params; 2552 struct ctl_be_block_lun *be_lun; 2553 struct ctl_be_lun *cbe_lun; 2554 const char *value; 2555 uint64_t oldsize; 2556 int error, wasprim; 2557 2558 params = &req->reqdata.modify; 2559 2560 mtx_lock(&softc->lock); 2561 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2562 if (be_lun->cbe_lun.lun_id == params->lun_id) 2563 break; 2564 } 2565 mtx_unlock(&softc->lock); 2566 if (be_lun == NULL) { 2567 snprintf(req->error_str, sizeof(req->error_str), 2568 "LUN %u is not managed by the block backend", 2569 params->lun_id); 2570 goto bailout_error; 2571 } 2572 cbe_lun = &be_lun->cbe_lun; 2573 2574 if (params->lun_size_bytes != 0) 2575 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2576 2577 nvlist_destroy(cbe_lun->options); 2578 cbe_lun->options = nvlist_clone(req->args_nvl); 2579 2580 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2581 value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL); 2582 if (value != NULL) { 2583 if (strcmp(value, "primary") == 0) 2584 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2585 else 2586 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2587 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2588 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2589 else 2590 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2591 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2592 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2593 ctl_lun_primary(cbe_lun); 2594 else 2595 ctl_lun_secondary(cbe_lun); 2596 } 2597 2598 oldsize = be_lun->size_blocks; 2599 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2600 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2601 if (be_lun->vn == NULL) 2602 error = ctl_be_block_open(be_lun, req); 2603 else if (vn_isdisk(be_lun->vn, &error)) 2604 error = ctl_be_block_open_dev(be_lun, req); 2605 else if (be_lun->vn->v_type == VREG) { 2606 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2607 error = ctl_be_block_open_file(be_lun, req); 2608 VOP_UNLOCK(be_lun->vn); 2609 } else 2610 error = EINVAL; 2611 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) && 2612 be_lun->vn != NULL) { 2613 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2614 ctl_lun_has_media(cbe_lun); 2615 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 && 2616 be_lun->vn == NULL) { 2617 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2618 ctl_lun_no_media(cbe_lun); 2619 } 2620 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2621 } else { 2622 if (be_lun->vn != NULL) { 2623 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2624 ctl_lun_no_media(cbe_lun); 2625 taskqueue_drain_all(be_lun->io_taskqueue); 2626 error = ctl_be_block_close(be_lun); 2627 } else 2628 error = 0; 2629 } 2630 if (be_lun->size_blocks != oldsize) 2631 ctl_lun_capacity_changed(cbe_lun); 2632 2633 /* Tell the user the exact size we ended up using */ 2634 params->lun_size_bytes = be_lun->size_bytes; 2635 2636 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2637 return (0); 2638 2639 bailout_error: 2640 req->status = CTL_LUN_ERROR; 2641 return (0); 2642 } 2643 2644 static void 2645 ctl_be_block_lun_shutdown(void *be_lun) 2646 { 2647 struct ctl_be_block_lun *lun = be_lun; 2648 struct ctl_be_block_softc *softc = lun->softc; 2649 2650 mtx_lock(&softc->lock); 2651 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2652 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2653 wakeup(lun); 2654 mtx_unlock(&softc->lock); 2655 } 2656 2657 static void 2658 ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2659 { 2660 struct ctl_be_block_lun *lun; 2661 struct ctl_be_block_softc *softc; 2662 2663 lun = (struct ctl_be_block_lun *)be_lun; 2664 softc = lun->softc; 2665 2666 if (status == CTL_LUN_CONFIG_OK) { 2667 mtx_lock(&softc->lock); 2668 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2669 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2670 wakeup(lun); 2671 mtx_unlock(&softc->lock); 2672 2673 /* 2674 * We successfully added the LUN, attempt to enable it. 2675 */ 2676 if (ctl_enable_lun(&lun->cbe_lun) != 0) { 2677 printf("%s: ctl_enable_lun() failed!\n", __func__); 2678 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { 2679 printf("%s: ctl_invalidate_lun() failed!\n", 2680 __func__); 2681 } 2682 } 2683 2684 return; 2685 } 2686 2687 2688 mtx_lock(&softc->lock); 2689 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2690 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2691 wakeup(lun); 2692 mtx_unlock(&softc->lock); 2693 } 2694 2695 2696 static int 2697 ctl_be_block_config_write(union ctl_io *io) 2698 { 2699 struct ctl_be_block_lun *be_lun; 2700 struct ctl_be_lun *cbe_lun; 2701 int retval; 2702 2703 DPRINTF("entered\n"); 2704 2705 cbe_lun = CTL_BACKEND_LUN(io); 2706 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2707 2708 retval = 0; 2709 switch (io->scsiio.cdb[0]) { 2710 case SYNCHRONIZE_CACHE: 2711 case SYNCHRONIZE_CACHE_16: 2712 case WRITE_SAME_10: 2713 case WRITE_SAME_16: 2714 case UNMAP: 2715 /* 2716 * The upper level CTL code will filter out any CDBs with 2717 * the immediate bit set and return the proper error. 2718 * 2719 * We don't really need to worry about what LBA range the 2720 * user asked to be synced out. When they issue a sync 2721 * cache command, we'll sync out the whole thing. 2722 */ 2723 mtx_lock(&be_lun->queue_lock); 2724 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2725 links); 2726 mtx_unlock(&be_lun->queue_lock); 2727 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2728 break; 2729 case START_STOP_UNIT: { 2730 struct scsi_start_stop_unit *cdb; 2731 struct ctl_lun_req req; 2732 2733 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2734 if ((cdb->how & SSS_PC_MASK) != 0) { 2735 ctl_set_success(&io->scsiio); 2736 ctl_config_write_done(io); 2737 break; 2738 } 2739 if (cdb->how & SSS_START) { 2740 if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) { 2741 retval = ctl_be_block_open(be_lun, &req); 2742 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2743 if (retval == 0) { 2744 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2745 ctl_lun_has_media(cbe_lun); 2746 } else { 2747 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2748 ctl_lun_no_media(cbe_lun); 2749 } 2750 } 2751 ctl_start_lun(cbe_lun); 2752 } else { 2753 ctl_stop_lun(cbe_lun); 2754 if (cdb->how & SSS_LOEJ) { 2755 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2756 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED; 2757 ctl_lun_ejected(cbe_lun); 2758 if (be_lun->vn != NULL) 2759 ctl_be_block_close(be_lun); 2760 } 2761 } 2762 2763 ctl_set_success(&io->scsiio); 2764 ctl_config_write_done(io); 2765 break; 2766 } 2767 case PREVENT_ALLOW: 2768 ctl_set_success(&io->scsiio); 2769 ctl_config_write_done(io); 2770 break; 2771 default: 2772 ctl_set_invalid_opcode(&io->scsiio); 2773 ctl_config_write_done(io); 2774 retval = CTL_RETVAL_COMPLETE; 2775 break; 2776 } 2777 2778 return (retval); 2779 } 2780 2781 static int 2782 ctl_be_block_config_read(union ctl_io *io) 2783 { 2784 struct ctl_be_block_lun *be_lun; 2785 struct ctl_be_lun *cbe_lun; 2786 int retval = 0; 2787 2788 DPRINTF("entered\n"); 2789 2790 cbe_lun = CTL_BACKEND_LUN(io); 2791 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2792 2793 switch (io->scsiio.cdb[0]) { 2794 case SERVICE_ACTION_IN: 2795 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2796 mtx_lock(&be_lun->queue_lock); 2797 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2798 &io->io_hdr, links); 2799 mtx_unlock(&be_lun->queue_lock); 2800 taskqueue_enqueue(be_lun->io_taskqueue, 2801 &be_lun->io_task); 2802 retval = CTL_RETVAL_QUEUED; 2803 break; 2804 } 2805 ctl_set_invalid_field(&io->scsiio, 2806 /*sks_valid*/ 1, 2807 /*command*/ 1, 2808 /*field*/ 1, 2809 /*bit_valid*/ 1, 2810 /*bit*/ 4); 2811 ctl_config_read_done(io); 2812 retval = CTL_RETVAL_COMPLETE; 2813 break; 2814 default: 2815 ctl_set_invalid_opcode(&io->scsiio); 2816 ctl_config_read_done(io); 2817 retval = CTL_RETVAL_COMPLETE; 2818 break; 2819 } 2820 2821 return (retval); 2822 } 2823 2824 static int 2825 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2826 { 2827 struct ctl_be_block_lun *lun; 2828 int retval; 2829 2830 lun = (struct ctl_be_block_lun *)be_lun; 2831 2832 retval = sbuf_printf(sb, "\t<num_threads>"); 2833 if (retval != 0) 2834 goto bailout; 2835 retval = sbuf_printf(sb, "%d", lun->num_threads); 2836 if (retval != 0) 2837 goto bailout; 2838 retval = sbuf_printf(sb, "</num_threads>\n"); 2839 2840 bailout: 2841 return (retval); 2842 } 2843 2844 static uint64_t 2845 ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2846 { 2847 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2848 2849 if (lun->getattr == NULL) 2850 return (UINT64_MAX); 2851 return (lun->getattr(lun, attrname)); 2852 } 2853 2854 static int 2855 ctl_be_block_init(void) 2856 { 2857 struct ctl_be_block_softc *softc = &backend_block_softc; 2858 2859 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2860 softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2861 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2862 STAILQ_INIT(&softc->lun_list); 2863 return (0); 2864 } 2865 2866 2867 static int 2868 ctl_be_block_shutdown(void) 2869 { 2870 struct ctl_be_block_softc *softc = &backend_block_softc; 2871 struct ctl_be_block_lun *lun, *next_lun; 2872 2873 mtx_lock(&softc->lock); 2874 STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun) { 2875 /* 2876 * Drop our lock here. Since ctl_invalidate_lun() can call 2877 * back into us, this could potentially lead to a recursive 2878 * lock of the same mutex, which would cause a hang. 2879 */ 2880 mtx_unlock(&softc->lock); 2881 ctl_disable_lun(&lun->cbe_lun); 2882 ctl_invalidate_lun(&lun->cbe_lun); 2883 mtx_lock(&softc->lock); 2884 } 2885 mtx_unlock(&softc->lock); 2886 2887 uma_zdestroy(softc->beio_zone); 2888 mtx_destroy(&softc->lock); 2889 return (0); 2890 } 2891