1 /*- 2 * Copyright (c) 2003 Silicon Graphics International Corp. 3 * Copyright (c) 2009-2011 Spectra Logic Corporation 4 * Copyright (c) 2012 The FreeBSD Foundation 5 * Copyright (c) 2014-2015 Alexander Motin <mav@FreeBSD.org> 6 * All rights reserved. 7 * 8 * Portions of this software were developed by Edward Tomasz Napierala 9 * under sponsorship from the FreeBSD Foundation. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions, and the following disclaimer, 16 * without modification. 17 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 18 * substantially similar to the "NO WARRANTY" disclaimer below 19 * ("Disclaimer") and any redistribution must be conditioned upon 20 * including a substantially similar Disclaimer requirement for further 21 * binary redistribution. 22 * 23 * NO WARRANTY 24 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 25 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 26 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 27 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 28 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 32 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 33 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 34 * POSSIBILITY OF SUCH DAMAGES. 35 * 36 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 37 */ 38 /* 39 * CAM Target Layer driver backend for block devices. 40 * 41 * Author: Ken Merry <ken@FreeBSD.org> 42 */ 43 #include <sys/cdefs.h> 44 __FBSDID("$FreeBSD$"); 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/kernel.h> 49 #include <sys/types.h> 50 #include <sys/kthread.h> 51 #include <sys/bio.h> 52 #include <sys/fcntl.h> 53 #include <sys/limits.h> 54 #include <sys/lock.h> 55 #include <sys/mutex.h> 56 #include <sys/condvar.h> 57 #include <sys/malloc.h> 58 #include <sys/conf.h> 59 #include <sys/ioccom.h> 60 #include <sys/queue.h> 61 #include <sys/sbuf.h> 62 #include <sys/endian.h> 63 #include <sys/uio.h> 64 #include <sys/buf.h> 65 #include <sys/taskqueue.h> 66 #include <sys/vnode.h> 67 #include <sys/namei.h> 68 #include <sys/mount.h> 69 #include <sys/disk.h> 70 #include <sys/fcntl.h> 71 #include <sys/filedesc.h> 72 #include <sys/filio.h> 73 #include <sys/proc.h> 74 #include <sys/pcpu.h> 75 #include <sys/module.h> 76 #include <sys/sdt.h> 77 #include <sys/devicestat.h> 78 #include <sys/sysctl.h> 79 80 #include <geom/geom.h> 81 82 #include <cam/cam.h> 83 #include <cam/scsi/scsi_all.h> 84 #include <cam/scsi/scsi_da.h> 85 #include <cam/ctl/ctl_io.h> 86 #include <cam/ctl/ctl.h> 87 #include <cam/ctl/ctl_backend.h> 88 #include <cam/ctl/ctl_ioctl.h> 89 #include <cam/ctl/ctl_ha.h> 90 #include <cam/ctl/ctl_scsi_all.h> 91 #include <cam/ctl/ctl_private.h> 92 #include <cam/ctl/ctl_error.h> 93 94 /* 95 * The idea here is that we'll allocate enough S/G space to hold a 1MB 96 * I/O. If we get an I/O larger than that, we'll split it. 97 */ 98 #define CTLBLK_HALF_IO_SIZE (512 * 1024) 99 #define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 100 #define CTLBLK_MAX_SEG MAXPHYS 101 #define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 102 #define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 103 104 #ifdef CTLBLK_DEBUG 105 #define DPRINTF(fmt, args...) \ 106 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 107 #else 108 #define DPRINTF(fmt, args...) do {} while(0) 109 #endif 110 111 #define PRIV(io) \ 112 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 113 #define ARGS(io) \ 114 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 115 116 SDT_PROVIDER_DEFINE(cbb); 117 118 typedef enum { 119 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 120 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 121 CTL_BE_BLOCK_LUN_WAITING = 0x04, 122 } ctl_be_block_lun_flags; 123 124 typedef enum { 125 CTL_BE_BLOCK_NONE, 126 CTL_BE_BLOCK_DEV, 127 CTL_BE_BLOCK_FILE 128 } ctl_be_block_type; 129 130 struct ctl_be_block_filedata { 131 struct ucred *cred; 132 }; 133 134 union ctl_be_block_bedata { 135 struct ctl_be_block_filedata file; 136 }; 137 138 struct ctl_be_block_io; 139 struct ctl_be_block_lun; 140 141 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 142 struct ctl_be_block_io *beio); 143 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 144 const char *attrname); 145 146 /* 147 * Backend LUN structure. There is a 1:1 mapping between a block device 148 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 149 */ 150 struct ctl_be_block_lun { 151 struct ctl_lun_create_params params; 152 char lunname[32]; 153 char *dev_path; 154 ctl_be_block_type dev_type; 155 struct vnode *vn; 156 union ctl_be_block_bedata backend; 157 cbb_dispatch_t dispatch; 158 cbb_dispatch_t lun_flush; 159 cbb_dispatch_t unmap; 160 cbb_dispatch_t get_lba_status; 161 cbb_getattr_t getattr; 162 uma_zone_t lun_zone; 163 uint64_t size_blocks; 164 uint64_t size_bytes; 165 struct ctl_be_block_softc *softc; 166 struct devstat *disk_stats; 167 ctl_be_block_lun_flags flags; 168 STAILQ_ENTRY(ctl_be_block_lun) links; 169 struct ctl_be_lun cbe_lun; 170 struct taskqueue *io_taskqueue; 171 struct task io_task; 172 int num_threads; 173 STAILQ_HEAD(, ctl_io_hdr) input_queue; 174 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 175 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 176 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 177 struct mtx_padalign io_lock; 178 struct mtx_padalign queue_lock; 179 }; 180 181 /* 182 * Overall softc structure for the block backend module. 183 */ 184 struct ctl_be_block_softc { 185 struct mtx lock; 186 int num_luns; 187 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 188 }; 189 190 static struct ctl_be_block_softc backend_block_softc; 191 192 /* 193 * Per-I/O information. 194 */ 195 struct ctl_be_block_io { 196 union ctl_io *io; 197 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 198 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 199 int bio_cmd; 200 int num_segs; 201 int num_bios_sent; 202 int num_bios_done; 203 int send_complete; 204 int num_errors; 205 struct bintime ds_t0; 206 devstat_tag_type ds_tag_type; 207 devstat_trans_flags ds_trans_type; 208 uint64_t io_len; 209 uint64_t io_offset; 210 int io_arg; 211 struct ctl_be_block_softc *softc; 212 struct ctl_be_block_lun *lun; 213 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 214 }; 215 216 extern struct ctl_softc *control_softc; 217 218 static int cbb_num_threads = 14; 219 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 220 "CAM Target Layer Block Backend"); 221 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 222 &cbb_num_threads, 0, "Number of threads per backing file"); 223 224 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 225 static void ctl_free_beio(struct ctl_be_block_io *beio); 226 static void ctl_complete_beio(struct ctl_be_block_io *beio); 227 static int ctl_be_block_move_done(union ctl_io *io); 228 static void ctl_be_block_biodone(struct bio *bio); 229 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 230 struct ctl_be_block_io *beio); 231 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 232 struct ctl_be_block_io *beio); 233 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 234 struct ctl_be_block_io *beio); 235 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 236 const char *attrname); 237 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 238 struct ctl_be_block_io *beio); 239 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 240 struct ctl_be_block_io *beio); 241 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 242 struct ctl_be_block_io *beio); 243 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 244 const char *attrname); 245 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 246 union ctl_io *io); 247 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 248 union ctl_io *io); 249 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 250 union ctl_io *io); 251 static void ctl_be_block_worker(void *context, int pending); 252 static int ctl_be_block_submit(union ctl_io *io); 253 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 254 int flag, struct thread *td); 255 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 256 struct ctl_lun_req *req); 257 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 258 struct ctl_lun_req *req); 259 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 260 static int ctl_be_block_open(struct ctl_be_block_lun *be_lun, 261 struct ctl_lun_req *req); 262 static int ctl_be_block_create(struct ctl_be_block_softc *softc, 263 struct ctl_lun_req *req); 264 static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 265 struct ctl_lun_req *req); 266 static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 267 struct ctl_lun_req *req); 268 static void ctl_be_block_lun_shutdown(void *be_lun); 269 static void ctl_be_block_lun_config_status(void *be_lun, 270 ctl_lun_config_status status); 271 static int ctl_be_block_config_write(union ctl_io *io); 272 static int ctl_be_block_config_read(union ctl_io *io); 273 static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 274 static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 275 int ctl_be_block_init(void); 276 277 static struct ctl_backend_driver ctl_be_block_driver = 278 { 279 .name = "block", 280 .flags = CTL_BE_FLAG_HAS_CONFIG, 281 .init = ctl_be_block_init, 282 .data_submit = ctl_be_block_submit, 283 .data_move_done = ctl_be_block_move_done, 284 .config_read = ctl_be_block_config_read, 285 .config_write = ctl_be_block_config_write, 286 .ioctl = ctl_be_block_ioctl, 287 .lun_info = ctl_be_block_lun_info, 288 .lun_attr = ctl_be_block_lun_attr 289 }; 290 291 MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 292 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 293 294 static uma_zone_t beio_zone; 295 296 static struct ctl_be_block_io * 297 ctl_alloc_beio(struct ctl_be_block_softc *softc) 298 { 299 struct ctl_be_block_io *beio; 300 301 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 302 beio->softc = softc; 303 return (beio); 304 } 305 306 static void 307 ctl_free_beio(struct ctl_be_block_io *beio) 308 { 309 int duplicate_free; 310 int i; 311 312 duplicate_free = 0; 313 314 for (i = 0; i < beio->num_segs; i++) { 315 if (beio->sg_segs[i].addr == NULL) 316 duplicate_free++; 317 318 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 319 beio->sg_segs[i].addr = NULL; 320 321 /* For compare we had two equal S/G lists. */ 322 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 323 uma_zfree(beio->lun->lun_zone, 324 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 325 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 326 } 327 } 328 329 if (duplicate_free > 0) { 330 printf("%s: %d duplicate frees out of %d segments\n", __func__, 331 duplicate_free, beio->num_segs); 332 } 333 334 uma_zfree(beio_zone, beio); 335 } 336 337 static void 338 ctl_complete_beio(struct ctl_be_block_io *beio) 339 { 340 union ctl_io *io = beio->io; 341 342 if (beio->beio_cont != NULL) { 343 beio->beio_cont(beio); 344 } else { 345 ctl_free_beio(beio); 346 ctl_data_submit_done(io); 347 } 348 } 349 350 static size_t 351 cmp(uint8_t *a, uint8_t *b, size_t size) 352 { 353 size_t i; 354 355 for (i = 0; i < size; i++) { 356 if (a[i] != b[i]) 357 break; 358 } 359 return (i); 360 } 361 362 static void 363 ctl_be_block_compare(union ctl_io *io) 364 { 365 struct ctl_be_block_io *beio; 366 uint64_t off, res; 367 int i; 368 uint8_t info[8]; 369 370 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 371 off = 0; 372 for (i = 0; i < beio->num_segs; i++) { 373 res = cmp(beio->sg_segs[i].addr, 374 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 375 beio->sg_segs[i].len); 376 off += res; 377 if (res < beio->sg_segs[i].len) 378 break; 379 } 380 if (i < beio->num_segs) { 381 scsi_u64to8b(off, info); 382 ctl_set_sense(&io->scsiio, /*current_error*/ 1, 383 /*sense_key*/ SSD_KEY_MISCOMPARE, 384 /*asc*/ 0x1D, /*ascq*/ 0x00, 385 /*type*/ SSD_ELEM_INFO, 386 /*size*/ sizeof(info), /*data*/ &info, 387 /*type*/ SSD_ELEM_NONE); 388 } else 389 ctl_set_success(&io->scsiio); 390 } 391 392 static int 393 ctl_be_block_move_done(union ctl_io *io) 394 { 395 struct ctl_be_block_io *beio; 396 struct ctl_be_block_lun *be_lun; 397 struct ctl_lba_len_flags *lbalen; 398 #ifdef CTL_TIME_IO 399 struct bintime cur_bt; 400 #endif 401 402 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 403 be_lun = beio->lun; 404 405 DPRINTF("entered\n"); 406 407 #ifdef CTL_TIME_IO 408 getbinuptime(&cur_bt); 409 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 410 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 411 #endif 412 io->io_hdr.num_dmas++; 413 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 414 415 /* 416 * We set status at this point for read commands, and write 417 * commands with errors. 418 */ 419 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 420 ; 421 } else if ((io->io_hdr.port_status == 0) && 422 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 423 lbalen = ARGS(beio->io); 424 if (lbalen->flags & CTL_LLF_READ) { 425 ctl_set_success(&io->scsiio); 426 } else if (lbalen->flags & CTL_LLF_COMPARE) { 427 /* We have two data blocks ready for comparison. */ 428 ctl_be_block_compare(io); 429 } 430 } else if ((io->io_hdr.port_status != 0) && 431 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 432 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 433 /* 434 * For hardware error sense keys, the sense key 435 * specific value is defined to be a retry count, 436 * but we use it to pass back an internal FETD 437 * error code. XXX KDM Hopefully the FETD is only 438 * using 16 bits for an error code, since that's 439 * all the space we have in the sks field. 440 */ 441 ctl_set_internal_failure(&io->scsiio, 442 /*sks_valid*/ 1, 443 /*retry_count*/ 444 io->io_hdr.port_status); 445 } 446 447 /* 448 * If this is a read, or a write with errors, it is done. 449 */ 450 if ((beio->bio_cmd == BIO_READ) 451 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 452 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 453 ctl_complete_beio(beio); 454 return (0); 455 } 456 457 /* 458 * At this point, we have a write and the DMA completed 459 * successfully. We now have to queue it to the task queue to 460 * execute the backend I/O. That is because we do blocking 461 * memory allocations, and in the file backing case, blocking I/O. 462 * This move done routine is generally called in the SIM's 463 * interrupt context, and therefore we cannot block. 464 */ 465 mtx_lock(&be_lun->queue_lock); 466 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 467 mtx_unlock(&be_lun->queue_lock); 468 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 469 470 return (0); 471 } 472 473 static void 474 ctl_be_block_biodone(struct bio *bio) 475 { 476 struct ctl_be_block_io *beio; 477 struct ctl_be_block_lun *be_lun; 478 union ctl_io *io; 479 int error; 480 481 beio = bio->bio_caller1; 482 be_lun = beio->lun; 483 io = beio->io; 484 485 DPRINTF("entered\n"); 486 487 error = bio->bio_error; 488 mtx_lock(&be_lun->io_lock); 489 if (error != 0) 490 beio->num_errors++; 491 492 beio->num_bios_done++; 493 494 /* 495 * XXX KDM will this cause WITNESS to complain? Holding a lock 496 * during the free might cause it to complain. 497 */ 498 g_destroy_bio(bio); 499 500 /* 501 * If the send complete bit isn't set, or we aren't the last I/O to 502 * complete, then we're done. 503 */ 504 if ((beio->send_complete == 0) 505 || (beio->num_bios_done < beio->num_bios_sent)) { 506 mtx_unlock(&be_lun->io_lock); 507 return; 508 } 509 510 /* 511 * At this point, we've verified that we are the last I/O to 512 * complete, so it's safe to drop the lock. 513 */ 514 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 515 beio->ds_tag_type, beio->ds_trans_type, 516 /*now*/ NULL, /*then*/&beio->ds_t0); 517 mtx_unlock(&be_lun->io_lock); 518 519 /* 520 * If there are any errors from the backing device, we fail the 521 * entire I/O with a medium error. 522 */ 523 if (beio->num_errors > 0) { 524 if (error == EOPNOTSUPP) { 525 ctl_set_invalid_opcode(&io->scsiio); 526 } else if (error == ENOSPC || error == EDQUOT) { 527 ctl_set_space_alloc_fail(&io->scsiio); 528 } else if (error == EROFS || error == EACCES) { 529 ctl_set_hw_write_protected(&io->scsiio); 530 } else if (beio->bio_cmd == BIO_FLUSH) { 531 /* XXX KDM is there is a better error here? */ 532 ctl_set_internal_failure(&io->scsiio, 533 /*sks_valid*/ 1, 534 /*retry_count*/ 0xbad2); 535 } else { 536 ctl_set_medium_error(&io->scsiio, 537 beio->bio_cmd == BIO_READ); 538 } 539 ctl_complete_beio(beio); 540 return; 541 } 542 543 /* 544 * If this is a write, a flush, a delete or verify, we're all done. 545 * If this is a read, we can now send the data to the user. 546 */ 547 if ((beio->bio_cmd == BIO_WRITE) 548 || (beio->bio_cmd == BIO_FLUSH) 549 || (beio->bio_cmd == BIO_DELETE) 550 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 551 ctl_set_success(&io->scsiio); 552 ctl_complete_beio(beio); 553 } else { 554 if ((ARGS(io)->flags & CTL_LLF_READ) && 555 beio->beio_cont == NULL) { 556 ctl_set_success(&io->scsiio); 557 ctl_serseq_done(io); 558 } 559 #ifdef CTL_TIME_IO 560 getbinuptime(&io->io_hdr.dma_start_bt); 561 #endif 562 ctl_datamove(io); 563 } 564 } 565 566 static void 567 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 568 struct ctl_be_block_io *beio) 569 { 570 union ctl_io *io = beio->io; 571 struct mount *mountpoint; 572 int error, lock_flags; 573 574 DPRINTF("entered\n"); 575 576 binuptime(&beio->ds_t0); 577 mtx_lock(&be_lun->io_lock); 578 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 579 mtx_unlock(&be_lun->io_lock); 580 581 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 582 583 if (MNT_SHARED_WRITES(mountpoint) || 584 ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 585 lock_flags = LK_SHARED; 586 else 587 lock_flags = LK_EXCLUSIVE; 588 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 589 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 590 curthread); 591 VOP_UNLOCK(be_lun->vn, 0); 592 593 vn_finished_write(mountpoint); 594 595 mtx_lock(&be_lun->io_lock); 596 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 597 beio->ds_tag_type, beio->ds_trans_type, 598 /*now*/ NULL, /*then*/&beio->ds_t0); 599 mtx_unlock(&be_lun->io_lock); 600 601 if (error == 0) 602 ctl_set_success(&io->scsiio); 603 else { 604 /* XXX KDM is there is a better error here? */ 605 ctl_set_internal_failure(&io->scsiio, 606 /*sks_valid*/ 1, 607 /*retry_count*/ 0xbad1); 608 } 609 610 ctl_complete_beio(beio); 611 } 612 613 SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 614 SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 615 SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 616 SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 617 618 static void 619 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 620 struct ctl_be_block_io *beio) 621 { 622 struct ctl_be_block_filedata *file_data; 623 union ctl_io *io; 624 struct uio xuio; 625 struct iovec *xiovec; 626 size_t s; 627 int error, flags, i; 628 629 DPRINTF("entered\n"); 630 631 file_data = &be_lun->backend.file; 632 io = beio->io; 633 flags = 0; 634 if (ARGS(io)->flags & CTL_LLF_DPO) 635 flags |= IO_DIRECT; 636 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 637 flags |= IO_SYNC; 638 639 bzero(&xuio, sizeof(xuio)); 640 if (beio->bio_cmd == BIO_READ) { 641 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 642 xuio.uio_rw = UIO_READ; 643 } else { 644 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 645 xuio.uio_rw = UIO_WRITE; 646 } 647 xuio.uio_offset = beio->io_offset; 648 xuio.uio_resid = beio->io_len; 649 xuio.uio_segflg = UIO_SYSSPACE; 650 xuio.uio_iov = beio->xiovecs; 651 xuio.uio_iovcnt = beio->num_segs; 652 xuio.uio_td = curthread; 653 654 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 655 xiovec->iov_base = beio->sg_segs[i].addr; 656 xiovec->iov_len = beio->sg_segs[i].len; 657 } 658 659 binuptime(&beio->ds_t0); 660 mtx_lock(&be_lun->io_lock); 661 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 662 mtx_unlock(&be_lun->io_lock); 663 664 if (beio->bio_cmd == BIO_READ) { 665 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 666 667 /* 668 * UFS pays attention to IO_DIRECT for reads. If the 669 * DIRECTIO option is configured into the kernel, it calls 670 * ffs_rawread(). But that only works for single-segment 671 * uios with user space addresses. In our case, with a 672 * kernel uio, it still reads into the buffer cache, but it 673 * will just try to release the buffer from the cache later 674 * on in ffs_read(). 675 * 676 * ZFS does not pay attention to IO_DIRECT for reads. 677 * 678 * UFS does not pay attention to IO_SYNC for reads. 679 * 680 * ZFS pays attention to IO_SYNC (which translates into the 681 * Solaris define FRSYNC for zfs_read()) for reads. It 682 * attempts to sync the file before reading. 683 */ 684 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 685 686 VOP_UNLOCK(be_lun->vn, 0); 687 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 688 if (error == 0 && xuio.uio_resid > 0) { 689 /* 690 * If we red less then requested (EOF), then 691 * we should clean the rest of the buffer. 692 */ 693 s = beio->io_len - xuio.uio_resid; 694 for (i = 0; i < beio->num_segs; i++) { 695 if (s >= beio->sg_segs[i].len) { 696 s -= beio->sg_segs[i].len; 697 continue; 698 } 699 bzero((uint8_t *)beio->sg_segs[i].addr + s, 700 beio->sg_segs[i].len - s); 701 s = 0; 702 } 703 } 704 } else { 705 struct mount *mountpoint; 706 int lock_flags; 707 708 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 709 710 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL) 711 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 712 lock_flags = LK_SHARED; 713 else 714 lock_flags = LK_EXCLUSIVE; 715 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 716 717 /* 718 * UFS pays attention to IO_DIRECT for writes. The write 719 * is done asynchronously. (Normally the write would just 720 * get put into cache. 721 * 722 * UFS pays attention to IO_SYNC for writes. It will 723 * attempt to write the buffer out synchronously if that 724 * flag is set. 725 * 726 * ZFS does not pay attention to IO_DIRECT for writes. 727 * 728 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 729 * for writes. It will flush the transaction from the 730 * cache before returning. 731 */ 732 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 733 VOP_UNLOCK(be_lun->vn, 0); 734 735 vn_finished_write(mountpoint); 736 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 737 } 738 739 mtx_lock(&be_lun->io_lock); 740 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 741 beio->ds_tag_type, beio->ds_trans_type, 742 /*now*/ NULL, /*then*/&beio->ds_t0); 743 mtx_unlock(&be_lun->io_lock); 744 745 /* 746 * If we got an error, set the sense data to "MEDIUM ERROR" and 747 * return the I/O to the user. 748 */ 749 if (error != 0) { 750 if (error == ENOSPC || error == EDQUOT) { 751 ctl_set_space_alloc_fail(&io->scsiio); 752 } else if (error == EROFS || error == EACCES) { 753 ctl_set_hw_write_protected(&io->scsiio); 754 } else { 755 ctl_set_medium_error(&io->scsiio, 756 beio->bio_cmd == BIO_READ); 757 } 758 ctl_complete_beio(beio); 759 return; 760 } 761 762 /* 763 * If this is a write or a verify, we're all done. 764 * If this is a read, we can now send the data to the user. 765 */ 766 if ((beio->bio_cmd == BIO_WRITE) || 767 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 768 ctl_set_success(&io->scsiio); 769 ctl_complete_beio(beio); 770 } else { 771 if ((ARGS(io)->flags & CTL_LLF_READ) && 772 beio->beio_cont == NULL) { 773 ctl_set_success(&io->scsiio); 774 ctl_serseq_done(io); 775 } 776 #ifdef CTL_TIME_IO 777 getbinuptime(&io->io_hdr.dma_start_bt); 778 #endif 779 ctl_datamove(io); 780 } 781 } 782 783 static void 784 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 785 struct ctl_be_block_io *beio) 786 { 787 union ctl_io *io = beio->io; 788 struct ctl_lba_len_flags *lbalen = ARGS(io); 789 struct scsi_get_lba_status_data *data; 790 off_t roff, off; 791 int error, status; 792 793 DPRINTF("entered\n"); 794 795 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 796 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 797 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 798 0, curthread->td_ucred, curthread); 799 if (error == 0 && off > roff) 800 status = 0; /* mapped up to off */ 801 else { 802 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 803 0, curthread->td_ucred, curthread); 804 if (error == 0 && off > roff) 805 status = 1; /* deallocated up to off */ 806 else { 807 status = 0; /* unknown up to the end */ 808 off = be_lun->size_bytes; 809 } 810 } 811 VOP_UNLOCK(be_lun->vn, 0); 812 813 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 814 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 815 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 816 lbalen->lba), data->descr[0].length); 817 data->descr[0].status = status; 818 819 ctl_complete_beio(beio); 820 } 821 822 static uint64_t 823 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 824 { 825 struct vattr vattr; 826 struct statfs statfs; 827 uint64_t val; 828 int error; 829 830 val = UINT64_MAX; 831 if (be_lun->vn == NULL) 832 return (val); 833 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 834 if (strcmp(attrname, "blocksused") == 0) { 835 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 836 if (error == 0) 837 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 838 } 839 if (strcmp(attrname, "blocksavail") == 0 && 840 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 841 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 842 if (error == 0) 843 val = statfs.f_bavail * statfs.f_bsize / 844 be_lun->cbe_lun.blocksize; 845 } 846 VOP_UNLOCK(be_lun->vn, 0); 847 return (val); 848 } 849 850 static void 851 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 852 struct ctl_be_block_io *beio) 853 { 854 union ctl_io *io; 855 struct cdevsw *csw; 856 struct cdev *dev; 857 struct uio xuio; 858 struct iovec *xiovec; 859 int error, flags, i, ref; 860 861 DPRINTF("entered\n"); 862 863 io = beio->io; 864 flags = 0; 865 if (ARGS(io)->flags & CTL_LLF_DPO) 866 flags |= IO_DIRECT; 867 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 868 flags |= IO_SYNC; 869 870 bzero(&xuio, sizeof(xuio)); 871 if (beio->bio_cmd == BIO_READ) { 872 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 873 xuio.uio_rw = UIO_READ; 874 } else { 875 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 876 xuio.uio_rw = UIO_WRITE; 877 } 878 xuio.uio_offset = beio->io_offset; 879 xuio.uio_resid = beio->io_len; 880 xuio.uio_segflg = UIO_SYSSPACE; 881 xuio.uio_iov = beio->xiovecs; 882 xuio.uio_iovcnt = beio->num_segs; 883 xuio.uio_td = curthread; 884 885 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 886 xiovec->iov_base = beio->sg_segs[i].addr; 887 xiovec->iov_len = beio->sg_segs[i].len; 888 } 889 890 binuptime(&beio->ds_t0); 891 mtx_lock(&be_lun->io_lock); 892 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 893 mtx_unlock(&be_lun->io_lock); 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_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 907 else 908 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 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 mtx_lock(&be_lun->io_lock); 1032 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1033 mtx_unlock(&be_lun->io_lock); 1034 1035 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1036 if (csw) { 1037 bio->bio_dev = dev; 1038 csw->d_strategy(bio); 1039 dev_relthread(dev, ref); 1040 } else { 1041 bio->bio_error = ENXIO; 1042 ctl_be_block_biodone(bio); 1043 } 1044 } 1045 1046 static void 1047 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1048 struct ctl_be_block_io *beio, 1049 uint64_t off, uint64_t len, int last) 1050 { 1051 struct bio *bio; 1052 uint64_t maxlen; 1053 struct cdevsw *csw; 1054 struct cdev *dev; 1055 int ref; 1056 1057 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1058 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1059 while (len > 0) { 1060 bio = g_alloc_bio(); 1061 bio->bio_cmd = BIO_DELETE; 1062 bio->bio_dev = dev; 1063 bio->bio_offset = off; 1064 bio->bio_length = MIN(len, maxlen); 1065 bio->bio_data = 0; 1066 bio->bio_done = ctl_be_block_biodone; 1067 bio->bio_caller1 = beio; 1068 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1069 1070 off += bio->bio_length; 1071 len -= bio->bio_length; 1072 1073 mtx_lock(&be_lun->io_lock); 1074 beio->num_bios_sent++; 1075 if (last && len == 0) 1076 beio->send_complete = 1; 1077 mtx_unlock(&be_lun->io_lock); 1078 1079 if (csw) { 1080 csw->d_strategy(bio); 1081 } else { 1082 bio->bio_error = ENXIO; 1083 ctl_be_block_biodone(bio); 1084 } 1085 } 1086 if (csw) 1087 dev_relthread(dev, ref); 1088 } 1089 1090 static void 1091 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1092 struct ctl_be_block_io *beio) 1093 { 1094 union ctl_io *io; 1095 struct ctl_ptr_len_flags *ptrlen; 1096 struct scsi_unmap_desc *buf, *end; 1097 uint64_t len; 1098 1099 io = beio->io; 1100 1101 DPRINTF("entered\n"); 1102 1103 binuptime(&beio->ds_t0); 1104 mtx_lock(&be_lun->io_lock); 1105 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1106 mtx_unlock(&be_lun->io_lock); 1107 1108 if (beio->io_offset == -1) { 1109 beio->io_len = 0; 1110 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1111 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1112 end = buf + ptrlen->len / sizeof(*buf); 1113 for (; buf < end; buf++) { 1114 len = (uint64_t)scsi_4btoul(buf->length) * 1115 be_lun->cbe_lun.blocksize; 1116 beio->io_len += len; 1117 ctl_be_block_unmap_dev_range(be_lun, beio, 1118 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1119 len, (end - buf < 2) ? TRUE : FALSE); 1120 } 1121 } else 1122 ctl_be_block_unmap_dev_range(be_lun, beio, 1123 beio->io_offset, beio->io_len, TRUE); 1124 } 1125 1126 static void 1127 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1128 struct ctl_be_block_io *beio) 1129 { 1130 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1131 struct bio *bio; 1132 struct cdevsw *csw; 1133 struct cdev *dev; 1134 off_t cur_offset; 1135 int i, max_iosize, ref; 1136 1137 DPRINTF("entered\n"); 1138 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1139 1140 /* 1141 * We have to limit our I/O size to the maximum supported by the 1142 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1143 * set it properly, use DFLTPHYS. 1144 */ 1145 if (csw) { 1146 max_iosize = dev->si_iosize_max; 1147 if (max_iosize < PAGE_SIZE) 1148 max_iosize = DFLTPHYS; 1149 } else 1150 max_iosize = DFLTPHYS; 1151 1152 cur_offset = beio->io_offset; 1153 for (i = 0; i < beio->num_segs; i++) { 1154 size_t cur_size; 1155 uint8_t *cur_ptr; 1156 1157 cur_size = beio->sg_segs[i].len; 1158 cur_ptr = beio->sg_segs[i].addr; 1159 1160 while (cur_size > 0) { 1161 /* This can't fail, it's a blocking allocation. */ 1162 bio = g_alloc_bio(); 1163 1164 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1165 1166 bio->bio_cmd = beio->bio_cmd; 1167 bio->bio_dev = dev; 1168 bio->bio_caller1 = beio; 1169 bio->bio_length = min(cur_size, max_iosize); 1170 bio->bio_offset = cur_offset; 1171 bio->bio_data = cur_ptr; 1172 bio->bio_done = ctl_be_block_biodone; 1173 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1174 1175 cur_offset += bio->bio_length; 1176 cur_ptr += bio->bio_length; 1177 cur_size -= bio->bio_length; 1178 1179 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1180 beio->num_bios_sent++; 1181 } 1182 } 1183 binuptime(&beio->ds_t0); 1184 mtx_lock(&be_lun->io_lock); 1185 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1186 beio->send_complete = 1; 1187 mtx_unlock(&be_lun->io_lock); 1188 1189 /* 1190 * Fire off all allocated requests! 1191 */ 1192 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1193 TAILQ_REMOVE(&queue, bio, bio_queue); 1194 if (csw) 1195 csw->d_strategy(bio); 1196 else { 1197 bio->bio_error = ENXIO; 1198 ctl_be_block_biodone(bio); 1199 } 1200 } 1201 if (csw) 1202 dev_relthread(dev, ref); 1203 } 1204 1205 static uint64_t 1206 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1207 { 1208 struct diocgattr_arg arg; 1209 struct cdevsw *csw; 1210 struct cdev *dev; 1211 int error, ref; 1212 1213 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1214 if (csw == NULL) 1215 return (UINT64_MAX); 1216 strlcpy(arg.name, attrname, sizeof(arg.name)); 1217 arg.len = sizeof(arg.value.off); 1218 if (csw->d_ioctl) { 1219 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 1220 curthread); 1221 } else 1222 error = ENODEV; 1223 dev_relthread(dev, ref); 1224 if (error != 0) 1225 return (UINT64_MAX); 1226 return (arg.value.off); 1227 } 1228 1229 static void 1230 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1231 union ctl_io *io) 1232 { 1233 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1234 struct ctl_be_block_io *beio; 1235 struct ctl_lba_len_flags *lbalen; 1236 1237 DPRINTF("entered\n"); 1238 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1239 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1240 1241 beio->io_len = lbalen->len * cbe_lun->blocksize; 1242 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1243 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1244 beio->bio_cmd = BIO_FLUSH; 1245 beio->ds_trans_type = DEVSTAT_NO_DATA; 1246 DPRINTF("SYNC\n"); 1247 be_lun->lun_flush(be_lun, beio); 1248 } 1249 1250 static void 1251 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1252 { 1253 union ctl_io *io; 1254 1255 io = beio->io; 1256 ctl_free_beio(beio); 1257 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1258 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1259 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1260 ctl_config_write_done(io); 1261 return; 1262 } 1263 1264 ctl_be_block_config_write(io); 1265 } 1266 1267 static void 1268 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1269 union ctl_io *io) 1270 { 1271 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1272 struct ctl_be_block_io *beio; 1273 struct ctl_lba_len_flags *lbalen; 1274 uint64_t len_left, lba; 1275 uint32_t pb, pbo, adj; 1276 int i, seglen; 1277 uint8_t *buf, *end; 1278 1279 DPRINTF("entered\n"); 1280 1281 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1282 lbalen = ARGS(beio->io); 1283 1284 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1285 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1286 ctl_free_beio(beio); 1287 ctl_set_invalid_field(&io->scsiio, 1288 /*sks_valid*/ 1, 1289 /*command*/ 1, 1290 /*field*/ 1, 1291 /*bit_valid*/ 0, 1292 /*bit*/ 0); 1293 ctl_config_write_done(io); 1294 return; 1295 } 1296 1297 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1298 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1299 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1300 beio->bio_cmd = BIO_DELETE; 1301 beio->ds_trans_type = DEVSTAT_FREE; 1302 1303 be_lun->unmap(be_lun, beio); 1304 return; 1305 } 1306 1307 beio->bio_cmd = BIO_WRITE; 1308 beio->ds_trans_type = DEVSTAT_WRITE; 1309 1310 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1311 (uintmax_t)lbalen->lba, lbalen->len); 1312 1313 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1314 if (be_lun->cbe_lun.pblockoff > 0) 1315 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1316 else 1317 pbo = 0; 1318 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1319 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1320 1321 /* 1322 * Setup the S/G entry for this chunk. 1323 */ 1324 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1325 if (pb > cbe_lun->blocksize) { 1326 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1327 seglen - pbo) % pb; 1328 if (seglen > adj) 1329 seglen -= adj; 1330 else 1331 seglen -= seglen % cbe_lun->blocksize; 1332 } else 1333 seglen -= seglen % cbe_lun->blocksize; 1334 beio->sg_segs[i].len = seglen; 1335 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1336 1337 DPRINTF("segment %d addr %p len %zd\n", i, 1338 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1339 1340 beio->num_segs++; 1341 len_left -= seglen; 1342 1343 buf = beio->sg_segs[i].addr; 1344 end = buf + seglen; 1345 for (; buf < end; buf += cbe_lun->blocksize) { 1346 if (lbalen->flags & SWS_NDOB) { 1347 memset(buf, 0, cbe_lun->blocksize); 1348 } else { 1349 memcpy(buf, io->scsiio.kern_data_ptr, 1350 cbe_lun->blocksize); 1351 } 1352 if (lbalen->flags & SWS_LBDATA) 1353 scsi_ulto4b(lbalen->lba + lba, buf); 1354 lba++; 1355 } 1356 } 1357 1358 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1359 beio->io_len = lba * cbe_lun->blocksize; 1360 1361 /* We can not do all in one run. Correct and schedule rerun. */ 1362 if (len_left > 0) { 1363 lbalen->lba += lba; 1364 lbalen->len -= lba; 1365 beio->beio_cont = ctl_be_block_cw_done_ws; 1366 } 1367 1368 be_lun->dispatch(be_lun, beio); 1369 } 1370 1371 static void 1372 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1373 union ctl_io *io) 1374 { 1375 struct ctl_be_block_io *beio; 1376 struct ctl_ptr_len_flags *ptrlen; 1377 1378 DPRINTF("entered\n"); 1379 1380 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1381 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1382 1383 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1384 ctl_free_beio(beio); 1385 ctl_set_invalid_field(&io->scsiio, 1386 /*sks_valid*/ 0, 1387 /*command*/ 1, 1388 /*field*/ 0, 1389 /*bit_valid*/ 0, 1390 /*bit*/ 0); 1391 ctl_config_write_done(io); 1392 return; 1393 } 1394 1395 beio->io_len = 0; 1396 beio->io_offset = -1; 1397 beio->bio_cmd = BIO_DELETE; 1398 beio->ds_trans_type = DEVSTAT_FREE; 1399 DPRINTF("UNMAP\n"); 1400 be_lun->unmap(be_lun, beio); 1401 } 1402 1403 static void 1404 ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1405 { 1406 union ctl_io *io; 1407 1408 io = beio->io; 1409 ctl_free_beio(beio); 1410 ctl_config_read_done(io); 1411 } 1412 1413 static void 1414 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1415 union ctl_io *io) 1416 { 1417 struct ctl_be_block_io *beio; 1418 struct ctl_be_block_softc *softc; 1419 1420 DPRINTF("entered\n"); 1421 1422 softc = be_lun->softc; 1423 beio = ctl_alloc_beio(softc); 1424 beio->io = io; 1425 beio->lun = be_lun; 1426 beio->beio_cont = ctl_be_block_cr_done; 1427 PRIV(io)->ptr = (void *)beio; 1428 1429 switch (io->scsiio.cdb[0]) { 1430 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1431 beio->bio_cmd = -1; 1432 beio->ds_trans_type = DEVSTAT_NO_DATA; 1433 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1434 beio->io_len = 0; 1435 if (be_lun->get_lba_status) 1436 be_lun->get_lba_status(be_lun, beio); 1437 else 1438 ctl_be_block_cr_done(beio); 1439 break; 1440 default: 1441 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1442 break; 1443 } 1444 } 1445 1446 static void 1447 ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1448 { 1449 union ctl_io *io; 1450 1451 io = beio->io; 1452 ctl_free_beio(beio); 1453 ctl_config_write_done(io); 1454 } 1455 1456 static void 1457 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1458 union ctl_io *io) 1459 { 1460 struct ctl_be_block_io *beio; 1461 struct ctl_be_block_softc *softc; 1462 1463 DPRINTF("entered\n"); 1464 1465 softc = be_lun->softc; 1466 beio = ctl_alloc_beio(softc); 1467 beio->io = io; 1468 beio->lun = be_lun; 1469 beio->beio_cont = ctl_be_block_cw_done; 1470 switch (io->scsiio.tag_type) { 1471 case CTL_TAG_ORDERED: 1472 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1473 break; 1474 case CTL_TAG_HEAD_OF_QUEUE: 1475 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1476 break; 1477 case CTL_TAG_UNTAGGED: 1478 case CTL_TAG_SIMPLE: 1479 case CTL_TAG_ACA: 1480 default: 1481 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1482 break; 1483 } 1484 PRIV(io)->ptr = (void *)beio; 1485 1486 switch (io->scsiio.cdb[0]) { 1487 case SYNCHRONIZE_CACHE: 1488 case SYNCHRONIZE_CACHE_16: 1489 ctl_be_block_cw_dispatch_sync(be_lun, io); 1490 break; 1491 case WRITE_SAME_10: 1492 case WRITE_SAME_16: 1493 ctl_be_block_cw_dispatch_ws(be_lun, io); 1494 break; 1495 case UNMAP: 1496 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1497 break; 1498 default: 1499 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1500 break; 1501 } 1502 } 1503 1504 SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1505 SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1506 SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1507 SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1508 1509 static void 1510 ctl_be_block_next(struct ctl_be_block_io *beio) 1511 { 1512 struct ctl_be_block_lun *be_lun; 1513 union ctl_io *io; 1514 1515 io = beio->io; 1516 be_lun = beio->lun; 1517 ctl_free_beio(beio); 1518 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1519 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1520 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1521 ctl_data_submit_done(io); 1522 return; 1523 } 1524 1525 io->io_hdr.status &= ~CTL_STATUS_MASK; 1526 io->io_hdr.status |= CTL_STATUS_NONE; 1527 1528 mtx_lock(&be_lun->queue_lock); 1529 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1530 mtx_unlock(&be_lun->queue_lock); 1531 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1532 } 1533 1534 static void 1535 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1536 union ctl_io *io) 1537 { 1538 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1539 struct ctl_be_block_io *beio; 1540 struct ctl_be_block_softc *softc; 1541 struct ctl_lba_len_flags *lbalen; 1542 struct ctl_ptr_len_flags *bptrlen; 1543 uint64_t len_left, lbas; 1544 int i; 1545 1546 softc = be_lun->softc; 1547 1548 DPRINTF("entered\n"); 1549 1550 lbalen = ARGS(io); 1551 if (lbalen->flags & CTL_LLF_WRITE) { 1552 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1553 } else { 1554 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1555 } 1556 1557 beio = ctl_alloc_beio(softc); 1558 beio->io = io; 1559 beio->lun = be_lun; 1560 bptrlen = PRIV(io); 1561 bptrlen->ptr = (void *)beio; 1562 1563 switch (io->scsiio.tag_type) { 1564 case CTL_TAG_ORDERED: 1565 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1566 break; 1567 case CTL_TAG_HEAD_OF_QUEUE: 1568 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1569 break; 1570 case CTL_TAG_UNTAGGED: 1571 case CTL_TAG_SIMPLE: 1572 case CTL_TAG_ACA: 1573 default: 1574 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1575 break; 1576 } 1577 1578 if (lbalen->flags & CTL_LLF_WRITE) { 1579 beio->bio_cmd = BIO_WRITE; 1580 beio->ds_trans_type = DEVSTAT_WRITE; 1581 } else { 1582 beio->bio_cmd = BIO_READ; 1583 beio->ds_trans_type = DEVSTAT_READ; 1584 } 1585 1586 DPRINTF("%s at LBA %jx len %u @%ju\n", 1587 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1588 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1589 if (lbalen->flags & CTL_LLF_COMPARE) 1590 lbas = CTLBLK_HALF_IO_SIZE; 1591 else 1592 lbas = CTLBLK_MAX_IO_SIZE; 1593 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1594 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1595 beio->io_len = lbas * cbe_lun->blocksize; 1596 bptrlen->len += lbas; 1597 1598 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1599 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1600 i, CTLBLK_MAX_SEGS)); 1601 1602 /* 1603 * Setup the S/G entry for this chunk. 1604 */ 1605 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1606 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1607 1608 DPRINTF("segment %d addr %p len %zd\n", i, 1609 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1610 1611 /* Set up second segment for compare operation. */ 1612 if (lbalen->flags & CTL_LLF_COMPARE) { 1613 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1614 beio->sg_segs[i].len; 1615 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1616 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1617 } 1618 1619 beio->num_segs++; 1620 len_left -= beio->sg_segs[i].len; 1621 } 1622 if (bptrlen->len < lbalen->len) 1623 beio->beio_cont = ctl_be_block_next; 1624 io->scsiio.be_move_done = ctl_be_block_move_done; 1625 /* For compare we have separate S/G lists for read and datamove. */ 1626 if (lbalen->flags & CTL_LLF_COMPARE) 1627 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1628 else 1629 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1630 io->scsiio.kern_data_len = beio->io_len; 1631 io->scsiio.kern_data_resid = 0; 1632 io->scsiio.kern_sg_entries = beio->num_segs; 1633 io->io_hdr.flags |= CTL_FLAG_ALLOCATED; 1634 1635 /* 1636 * For the read case, we need to read the data into our buffers and 1637 * then we can send it back to the user. For the write case, we 1638 * need to get the data from the user first. 1639 */ 1640 if (beio->bio_cmd == BIO_READ) { 1641 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1642 be_lun->dispatch(be_lun, beio); 1643 } else { 1644 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1645 #ifdef CTL_TIME_IO 1646 getbinuptime(&io->io_hdr.dma_start_bt); 1647 #endif 1648 ctl_datamove(io); 1649 } 1650 } 1651 1652 static void 1653 ctl_be_block_worker(void *context, int pending) 1654 { 1655 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; 1656 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1657 union ctl_io *io; 1658 struct ctl_be_block_io *beio; 1659 1660 DPRINTF("entered\n"); 1661 /* 1662 * Fetch and process I/Os from all queues. If we detect LUN 1663 * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race, 1664 * so make response maximally opaque to not confuse initiator. 1665 */ 1666 for (;;) { 1667 mtx_lock(&be_lun->queue_lock); 1668 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1669 if (io != NULL) { 1670 DPRINTF("datamove queue\n"); 1671 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1672 ctl_io_hdr, links); 1673 mtx_unlock(&be_lun->queue_lock); 1674 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1675 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1676 ctl_set_busy(&io->scsiio); 1677 ctl_complete_beio(beio); 1678 return; 1679 } 1680 be_lun->dispatch(be_lun, beio); 1681 continue; 1682 } 1683 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1684 if (io != NULL) { 1685 DPRINTF("config write queue\n"); 1686 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1687 ctl_io_hdr, links); 1688 mtx_unlock(&be_lun->queue_lock); 1689 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1690 ctl_set_busy(&io->scsiio); 1691 ctl_config_write_done(io); 1692 return; 1693 } 1694 ctl_be_block_cw_dispatch(be_lun, io); 1695 continue; 1696 } 1697 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1698 if (io != NULL) { 1699 DPRINTF("config read queue\n"); 1700 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1701 ctl_io_hdr, links); 1702 mtx_unlock(&be_lun->queue_lock); 1703 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1704 ctl_set_busy(&io->scsiio); 1705 ctl_config_read_done(io); 1706 return; 1707 } 1708 ctl_be_block_cr_dispatch(be_lun, io); 1709 continue; 1710 } 1711 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1712 if (io != NULL) { 1713 DPRINTF("input queue\n"); 1714 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1715 ctl_io_hdr, links); 1716 mtx_unlock(&be_lun->queue_lock); 1717 if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { 1718 ctl_set_busy(&io->scsiio); 1719 ctl_data_submit_done(io); 1720 return; 1721 } 1722 ctl_be_block_dispatch(be_lun, io); 1723 continue; 1724 } 1725 1726 /* 1727 * If we get here, there is no work left in the queues, so 1728 * just break out and let the task queue go to sleep. 1729 */ 1730 mtx_unlock(&be_lun->queue_lock); 1731 break; 1732 } 1733 } 1734 1735 /* 1736 * Entry point from CTL to the backend for I/O. We queue everything to a 1737 * work thread, so this just puts the I/O on a queue and wakes up the 1738 * thread. 1739 */ 1740 static int 1741 ctl_be_block_submit(union ctl_io *io) 1742 { 1743 struct ctl_be_block_lun *be_lun; 1744 struct ctl_be_lun *cbe_lun; 1745 1746 DPRINTF("entered\n"); 1747 1748 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1749 CTL_PRIV_BACKEND_LUN].ptr; 1750 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 1751 1752 /* 1753 * Make sure we only get SCSI I/O. 1754 */ 1755 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1756 "%#x) encountered", io->io_hdr.io_type)); 1757 1758 PRIV(io)->len = 0; 1759 1760 mtx_lock(&be_lun->queue_lock); 1761 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1762 mtx_unlock(&be_lun->queue_lock); 1763 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1764 1765 return (CTL_RETVAL_COMPLETE); 1766 } 1767 1768 static int 1769 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1770 int flag, struct thread *td) 1771 { 1772 struct ctl_be_block_softc *softc; 1773 int error; 1774 1775 softc = &backend_block_softc; 1776 1777 error = 0; 1778 1779 switch (cmd) { 1780 case CTL_LUN_REQ: { 1781 struct ctl_lun_req *lun_req; 1782 1783 lun_req = (struct ctl_lun_req *)addr; 1784 1785 switch (lun_req->reqtype) { 1786 case CTL_LUNREQ_CREATE: 1787 error = ctl_be_block_create(softc, lun_req); 1788 break; 1789 case CTL_LUNREQ_RM: 1790 error = ctl_be_block_rm(softc, lun_req); 1791 break; 1792 case CTL_LUNREQ_MODIFY: 1793 error = ctl_be_block_modify(softc, lun_req); 1794 break; 1795 default: 1796 lun_req->status = CTL_LUN_ERROR; 1797 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1798 "invalid LUN request type %d", 1799 lun_req->reqtype); 1800 break; 1801 } 1802 break; 1803 } 1804 default: 1805 error = ENOTTY; 1806 break; 1807 } 1808 1809 return (error); 1810 } 1811 1812 static int 1813 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1814 { 1815 struct ctl_be_lun *cbe_lun; 1816 struct ctl_be_block_filedata *file_data; 1817 struct ctl_lun_create_params *params; 1818 char *value; 1819 struct vattr vattr; 1820 off_t ps, pss, po, pos, us, uss, uo, uos; 1821 int error; 1822 1823 cbe_lun = &be_lun->cbe_lun; 1824 file_data = &be_lun->backend.file; 1825 params = &be_lun->params; 1826 1827 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1828 be_lun->dispatch = ctl_be_block_dispatch_file; 1829 be_lun->lun_flush = ctl_be_block_flush_file; 1830 be_lun->get_lba_status = ctl_be_block_gls_file; 1831 be_lun->getattr = ctl_be_block_getattr_file; 1832 be_lun->unmap = NULL; 1833 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1834 1835 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1836 if (error != 0) { 1837 snprintf(req->error_str, sizeof(req->error_str), 1838 "error calling VOP_GETATTR() for file %s", 1839 be_lun->dev_path); 1840 return (error); 1841 } 1842 1843 file_data->cred = crhold(curthread->td_ucred); 1844 if (params->lun_size_bytes != 0) 1845 be_lun->size_bytes = params->lun_size_bytes; 1846 else 1847 be_lun->size_bytes = vattr.va_size; 1848 1849 /* 1850 * For files we can use any logical block size. Prefer 512 bytes 1851 * for compatibility reasons. If file's vattr.va_blocksize 1852 * (preferred I/O block size) is bigger and multiple to chosen 1853 * logical block size -- report it as physical block size. 1854 */ 1855 if (params->blocksize_bytes != 0) 1856 cbe_lun->blocksize = params->blocksize_bytes; 1857 else if (cbe_lun->lun_type == T_CDROM) 1858 cbe_lun->blocksize = 2048; 1859 else 1860 cbe_lun->blocksize = 512; 1861 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1862 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1863 0 : (be_lun->size_blocks - 1); 1864 1865 us = ps = vattr.va_blocksize; 1866 uo = po = 0; 1867 1868 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 1869 if (value != NULL) 1870 ctl_expand_number(value, &ps); 1871 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 1872 if (value != NULL) 1873 ctl_expand_number(value, &po); 1874 pss = ps / cbe_lun->blocksize; 1875 pos = po / cbe_lun->blocksize; 1876 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1877 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1878 cbe_lun->pblockexp = fls(pss) - 1; 1879 cbe_lun->pblockoff = (pss - pos) % pss; 1880 } 1881 1882 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 1883 if (value != NULL) 1884 ctl_expand_number(value, &us); 1885 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 1886 if (value != NULL) 1887 ctl_expand_number(value, &uo); 1888 uss = us / cbe_lun->blocksize; 1889 uos = uo / cbe_lun->blocksize; 1890 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1891 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1892 cbe_lun->ublockexp = fls(uss) - 1; 1893 cbe_lun->ublockoff = (uss - uos) % uss; 1894 } 1895 1896 /* 1897 * Sanity check. The media size has to be at least one 1898 * sector long. 1899 */ 1900 if (be_lun->size_bytes < cbe_lun->blocksize) { 1901 error = EINVAL; 1902 snprintf(req->error_str, sizeof(req->error_str), 1903 "file %s size %ju < block size %u", be_lun->dev_path, 1904 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1905 } 1906 1907 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1908 return (error); 1909 } 1910 1911 static int 1912 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1913 { 1914 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1915 struct ctl_lun_create_params *params; 1916 struct cdevsw *csw; 1917 struct cdev *dev; 1918 char *value; 1919 int error, atomic, maxio, ref, unmap, tmp; 1920 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1921 1922 params = &be_lun->params; 1923 1924 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1925 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1926 if (csw == NULL) 1927 return (ENXIO); 1928 if (strcmp(csw->d_name, "zvol") == 0) { 1929 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1930 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1931 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1932 } else { 1933 be_lun->dispatch = ctl_be_block_dispatch_dev; 1934 be_lun->get_lba_status = NULL; 1935 atomic = 0; 1936 maxio = dev->si_iosize_max; 1937 if (maxio <= 0) 1938 maxio = DFLTPHYS; 1939 if (maxio > CTLBLK_MAX_IO_SIZE) 1940 maxio = CTLBLK_MAX_IO_SIZE; 1941 } 1942 be_lun->lun_flush = ctl_be_block_flush_dev; 1943 be_lun->getattr = ctl_be_block_getattr_dev; 1944 be_lun->unmap = ctl_be_block_unmap_dev; 1945 1946 if (!csw->d_ioctl) { 1947 dev_relthread(dev, ref); 1948 snprintf(req->error_str, sizeof(req->error_str), 1949 "no d_ioctl for device %s!", be_lun->dev_path); 1950 return (ENODEV); 1951 } 1952 1953 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 1954 curthread); 1955 if (error) { 1956 dev_relthread(dev, ref); 1957 snprintf(req->error_str, sizeof(req->error_str), 1958 "error %d returned for DIOCGSECTORSIZE ioctl " 1959 "on %s!", error, be_lun->dev_path); 1960 return (error); 1961 } 1962 1963 /* 1964 * If the user has asked for a blocksize that is greater than the 1965 * backing device's blocksize, we can do it only if the blocksize 1966 * the user is asking for is an even multiple of the underlying 1967 * device's blocksize. 1968 */ 1969 if ((params->blocksize_bytes != 0) && 1970 (params->blocksize_bytes >= tmp)) { 1971 if (params->blocksize_bytes % tmp == 0) { 1972 cbe_lun->blocksize = params->blocksize_bytes; 1973 } else { 1974 dev_relthread(dev, ref); 1975 snprintf(req->error_str, sizeof(req->error_str), 1976 "requested blocksize %u is not an even " 1977 "multiple of backing device blocksize %u", 1978 params->blocksize_bytes, tmp); 1979 return (EINVAL); 1980 } 1981 } else if (params->blocksize_bytes != 0) { 1982 dev_relthread(dev, ref); 1983 snprintf(req->error_str, sizeof(req->error_str), 1984 "requested blocksize %u < backing device " 1985 "blocksize %u", params->blocksize_bytes, tmp); 1986 return (EINVAL); 1987 } else if (cbe_lun->lun_type == T_CDROM) 1988 cbe_lun->blocksize = MAX(tmp, 2048); 1989 else 1990 cbe_lun->blocksize = tmp; 1991 1992 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 1993 curthread); 1994 if (error) { 1995 dev_relthread(dev, ref); 1996 snprintf(req->error_str, sizeof(req->error_str), 1997 "error %d returned for DIOCGMEDIASIZE " 1998 " ioctl on %s!", error, 1999 be_lun->dev_path); 2000 return (error); 2001 } 2002 2003 if (params->lun_size_bytes != 0) { 2004 if (params->lun_size_bytes > otmp) { 2005 dev_relthread(dev, ref); 2006 snprintf(req->error_str, sizeof(req->error_str), 2007 "requested LUN size %ju > backing device " 2008 "size %ju", 2009 (uintmax_t)params->lun_size_bytes, 2010 (uintmax_t)otmp); 2011 return (EINVAL); 2012 } 2013 2014 be_lun->size_bytes = params->lun_size_bytes; 2015 } else 2016 be_lun->size_bytes = otmp; 2017 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2018 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2019 0 : (be_lun->size_blocks - 1); 2020 2021 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, 2022 curthread); 2023 if (error) 2024 ps = po = 0; 2025 else { 2026 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, 2027 FREAD, curthread); 2028 if (error) 2029 po = 0; 2030 } 2031 us = ps; 2032 uo = po; 2033 2034 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 2035 if (value != NULL) 2036 ctl_expand_number(value, &ps); 2037 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 2038 if (value != NULL) 2039 ctl_expand_number(value, &po); 2040 pss = ps / cbe_lun->blocksize; 2041 pos = po / cbe_lun->blocksize; 2042 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 2043 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 2044 cbe_lun->pblockexp = fls(pss) - 1; 2045 cbe_lun->pblockoff = (pss - pos) % pss; 2046 } 2047 2048 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 2049 if (value != NULL) 2050 ctl_expand_number(value, &us); 2051 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 2052 if (value != NULL) 2053 ctl_expand_number(value, &uo); 2054 uss = us / cbe_lun->blocksize; 2055 uos = uo / cbe_lun->blocksize; 2056 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2057 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2058 cbe_lun->ublockexp = fls(uss) - 1; 2059 cbe_lun->ublockoff = (uss - uos) % uss; 2060 } 2061 2062 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2063 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2064 2065 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2066 unmap = 1; 2067 } else { 2068 struct diocgattr_arg arg; 2069 2070 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2071 arg.len = sizeof(arg.value.i); 2072 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 2073 curthread); 2074 unmap = (error == 0) ? arg.value.i : 0; 2075 } 2076 value = ctl_get_opt(&cbe_lun->options, "unmap"); 2077 if (value != NULL) 2078 unmap = (strcmp(value, "on") == 0); 2079 if (unmap) 2080 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2081 else 2082 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2083 2084 dev_relthread(dev, ref); 2085 return (0); 2086 } 2087 2088 static int 2089 ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2090 { 2091 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2092 int flags; 2093 2094 if (be_lun->vn) { 2095 flags = FREAD; 2096 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2097 flags |= FWRITE; 2098 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2099 be_lun->vn = NULL; 2100 2101 switch (be_lun->dev_type) { 2102 case CTL_BE_BLOCK_DEV: 2103 break; 2104 case CTL_BE_BLOCK_FILE: 2105 if (be_lun->backend.file.cred != NULL) { 2106 crfree(be_lun->backend.file.cred); 2107 be_lun->backend.file.cred = NULL; 2108 } 2109 break; 2110 case CTL_BE_BLOCK_NONE: 2111 break; 2112 default: 2113 panic("Unexpected backend type."); 2114 break; 2115 } 2116 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2117 } 2118 return (0); 2119 } 2120 2121 static int 2122 ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2123 { 2124 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2125 struct nameidata nd; 2126 char *value; 2127 int error, flags; 2128 2129 error = 0; 2130 if (rootvnode == NULL) { 2131 snprintf(req->error_str, sizeof(req->error_str), 2132 "Root filesystem is not mounted"); 2133 return (1); 2134 } 2135 pwd_ensure_dirs(); 2136 2137 value = ctl_get_opt(&cbe_lun->options, "file"); 2138 if (value == NULL) { 2139 snprintf(req->error_str, sizeof(req->error_str), 2140 "no file argument specified"); 2141 return (1); 2142 } 2143 free(be_lun->dev_path, M_CTLBLK); 2144 be_lun->dev_path = strdup(value, M_CTLBLK); 2145 2146 flags = FREAD; 2147 value = ctl_get_opt(&cbe_lun->options, "readonly"); 2148 if (value != NULL) { 2149 if (strcmp(value, "on") != 0) 2150 flags |= FWRITE; 2151 } else if (cbe_lun->lun_type == T_DIRECT) 2152 flags |= FWRITE; 2153 2154 again: 2155 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2156 error = vn_open(&nd, &flags, 0, NULL); 2157 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2158 flags &= ~FWRITE; 2159 goto again; 2160 } 2161 if (error) { 2162 /* 2163 * This is the only reasonable guess we can make as far as 2164 * path if the user doesn't give us a fully qualified path. 2165 * If they want to specify a file, they need to specify the 2166 * full path. 2167 */ 2168 if (be_lun->dev_path[0] != '/') { 2169 char *dev_name; 2170 2171 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2172 be_lun->dev_path); 2173 free(be_lun->dev_path, M_CTLBLK); 2174 be_lun->dev_path = dev_name; 2175 goto again; 2176 } 2177 snprintf(req->error_str, sizeof(req->error_str), 2178 "error opening %s: %d", be_lun->dev_path, error); 2179 return (error); 2180 } 2181 if (flags & FWRITE) 2182 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2183 else 2184 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2185 2186 NDFREE(&nd, NDF_ONLY_PNBUF); 2187 be_lun->vn = nd.ni_vp; 2188 2189 /* We only support disks and files. */ 2190 if (vn_isdisk(be_lun->vn, &error)) { 2191 error = ctl_be_block_open_dev(be_lun, req); 2192 } else if (be_lun->vn->v_type == VREG) { 2193 error = ctl_be_block_open_file(be_lun, req); 2194 } else { 2195 error = EINVAL; 2196 snprintf(req->error_str, sizeof(req->error_str), 2197 "%s is not a disk or plain file", be_lun->dev_path); 2198 } 2199 VOP_UNLOCK(be_lun->vn, 0); 2200 2201 if (error != 0) 2202 ctl_be_block_close(be_lun); 2203 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2204 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2205 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2206 value = ctl_get_opt(&cbe_lun->options, "serseq"); 2207 if (value != NULL && strcmp(value, "on") == 0) 2208 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2209 else if (value != NULL && strcmp(value, "read") == 0) 2210 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2211 else if (value != NULL && strcmp(value, "off") == 0) 2212 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2213 return (0); 2214 } 2215 2216 static int 2217 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2218 { 2219 struct ctl_be_lun *cbe_lun; 2220 struct ctl_be_block_lun *be_lun; 2221 struct ctl_lun_create_params *params; 2222 char num_thread_str[16]; 2223 char tmpstr[32]; 2224 char *value; 2225 int retval, num_threads; 2226 int tmp_num_threads; 2227 2228 params = &req->reqdata.create; 2229 retval = 0; 2230 req->status = CTL_LUN_OK; 2231 2232 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2233 cbe_lun = &be_lun->cbe_lun; 2234 cbe_lun->be_lun = be_lun; 2235 be_lun->params = req->reqdata.create; 2236 be_lun->softc = softc; 2237 STAILQ_INIT(&be_lun->input_queue); 2238 STAILQ_INIT(&be_lun->config_read_queue); 2239 STAILQ_INIT(&be_lun->config_write_queue); 2240 STAILQ_INIT(&be_lun->datamove_queue); 2241 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2242 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2243 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2244 ctl_init_opts(&cbe_lun->options, 2245 req->num_be_args, req->kern_be_args); 2246 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2247 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2248 if (be_lun->lun_zone == NULL) { 2249 snprintf(req->error_str, sizeof(req->error_str), 2250 "error allocating UMA zone"); 2251 goto bailout_error; 2252 } 2253 2254 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2255 cbe_lun->lun_type = params->device_type; 2256 else 2257 cbe_lun->lun_type = T_DIRECT; 2258 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2259 cbe_lun->flags = 0; 2260 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2261 if (value != NULL) { 2262 if (strcmp(value, "primary") == 0) 2263 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2264 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2265 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2266 2267 if (cbe_lun->lun_type == T_DIRECT || 2268 cbe_lun->lun_type == T_CDROM) { 2269 be_lun->size_bytes = params->lun_size_bytes; 2270 if (params->blocksize_bytes != 0) 2271 cbe_lun->blocksize = params->blocksize_bytes; 2272 else if (cbe_lun->lun_type == T_CDROM) 2273 cbe_lun->blocksize = 2048; 2274 else 2275 cbe_lun->blocksize = 512; 2276 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2277 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2278 0 : (be_lun->size_blocks - 1); 2279 2280 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2281 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2282 retval = ctl_be_block_open(be_lun, req); 2283 if (retval != 0) { 2284 retval = 0; 2285 req->status = CTL_LUN_WARNING; 2286 } 2287 } 2288 num_threads = cbb_num_threads; 2289 } else { 2290 num_threads = 1; 2291 } 2292 2293 value = ctl_get_opt(&cbe_lun->options, "num_threads"); 2294 if (value != NULL) { 2295 tmp_num_threads = strtol(value, NULL, 0); 2296 2297 /* 2298 * We don't let the user specify less than one 2299 * thread, but hope he's clueful enough not to 2300 * specify 1000 threads. 2301 */ 2302 if (tmp_num_threads < 1) { 2303 snprintf(req->error_str, sizeof(req->error_str), 2304 "invalid number of threads %s", 2305 num_thread_str); 2306 goto bailout_error; 2307 } 2308 num_threads = tmp_num_threads; 2309 } 2310 2311 if (be_lun->vn == NULL) 2312 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2313 /* Tell the user the blocksize we ended up using */ 2314 params->lun_size_bytes = be_lun->size_bytes; 2315 params->blocksize_bytes = cbe_lun->blocksize; 2316 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2317 cbe_lun->req_lun_id = params->req_lun_id; 2318 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2319 } else 2320 cbe_lun->req_lun_id = 0; 2321 2322 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2323 cbe_lun->lun_config_status = ctl_be_block_lun_config_status; 2324 cbe_lun->be = &ctl_be_block_driver; 2325 2326 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2327 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2328 softc->num_luns); 2329 strncpy((char *)cbe_lun->serial_num, tmpstr, 2330 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2331 2332 /* Tell the user what we used for a serial number */ 2333 strncpy((char *)params->serial_num, tmpstr, 2334 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2335 } else { 2336 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2337 MIN(sizeof(cbe_lun->serial_num), 2338 sizeof(params->serial_num))); 2339 } 2340 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2341 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2342 strncpy((char *)cbe_lun->device_id, tmpstr, 2343 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2344 2345 /* Tell the user what we used for a device ID */ 2346 strncpy((char *)params->device_id, tmpstr, 2347 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2348 } else { 2349 strncpy((char *)cbe_lun->device_id, params->device_id, 2350 MIN(sizeof(cbe_lun->device_id), 2351 sizeof(params->device_id))); 2352 } 2353 2354 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2355 2356 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2357 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2358 2359 if (be_lun->io_taskqueue == NULL) { 2360 snprintf(req->error_str, sizeof(req->error_str), 2361 "unable to create taskqueue"); 2362 goto bailout_error; 2363 } 2364 2365 /* 2366 * Note that we start the same number of threads by default for 2367 * both the file case and the block device case. For the file 2368 * case, we need multiple threads to allow concurrency, because the 2369 * vnode interface is designed to be a blocking interface. For the 2370 * block device case, ZFS zvols at least will block the caller's 2371 * context in many instances, and so we need multiple threads to 2372 * overcome that problem. Other block devices don't need as many 2373 * threads, but they shouldn't cause too many problems. 2374 * 2375 * If the user wants to just have a single thread for a block 2376 * device, he can specify that when the LUN is created, or change 2377 * the tunable/sysctl to alter the default number of threads. 2378 */ 2379 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2380 /*num threads*/num_threads, 2381 /*priority*/PWAIT, 2382 /*thread name*/ 2383 "%s taskq", be_lun->lunname); 2384 2385 if (retval != 0) 2386 goto bailout_error; 2387 2388 be_lun->num_threads = num_threads; 2389 2390 mtx_lock(&softc->lock); 2391 softc->num_luns++; 2392 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2393 2394 mtx_unlock(&softc->lock); 2395 2396 retval = ctl_add_lun(&be_lun->cbe_lun); 2397 if (retval != 0) { 2398 mtx_lock(&softc->lock); 2399 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2400 links); 2401 softc->num_luns--; 2402 mtx_unlock(&softc->lock); 2403 snprintf(req->error_str, sizeof(req->error_str), 2404 "ctl_add_lun() returned error %d, see dmesg for " 2405 "details", retval); 2406 retval = 0; 2407 goto bailout_error; 2408 } 2409 2410 mtx_lock(&softc->lock); 2411 2412 /* 2413 * Tell the config_status routine that we're waiting so it won't 2414 * clean up the LUN in the event of an error. 2415 */ 2416 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2417 2418 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2419 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2420 if (retval == EINTR) 2421 break; 2422 } 2423 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2424 2425 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2426 snprintf(req->error_str, sizeof(req->error_str), 2427 "LUN configuration error, see dmesg for details"); 2428 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2429 links); 2430 softc->num_luns--; 2431 mtx_unlock(&softc->lock); 2432 goto bailout_error; 2433 } else { 2434 params->req_lun_id = cbe_lun->lun_id; 2435 } 2436 2437 mtx_unlock(&softc->lock); 2438 2439 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2440 cbe_lun->blocksize, 2441 DEVSTAT_ALL_SUPPORTED, 2442 cbe_lun->lun_type 2443 | DEVSTAT_TYPE_IF_OTHER, 2444 DEVSTAT_PRIORITY_OTHER); 2445 2446 return (retval); 2447 2448 bailout_error: 2449 req->status = CTL_LUN_ERROR; 2450 2451 if (be_lun->io_taskqueue != NULL) 2452 taskqueue_free(be_lun->io_taskqueue); 2453 ctl_be_block_close(be_lun); 2454 if (be_lun->dev_path != NULL) 2455 free(be_lun->dev_path, M_CTLBLK); 2456 if (be_lun->lun_zone != NULL) 2457 uma_zdestroy(be_lun->lun_zone); 2458 ctl_free_opts(&cbe_lun->options); 2459 mtx_destroy(&be_lun->queue_lock); 2460 mtx_destroy(&be_lun->io_lock); 2461 free(be_lun, M_CTLBLK); 2462 2463 return (retval); 2464 } 2465 2466 static int 2467 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2468 { 2469 struct ctl_lun_rm_params *params; 2470 struct ctl_be_block_lun *be_lun; 2471 struct ctl_be_lun *cbe_lun; 2472 int retval; 2473 2474 params = &req->reqdata.rm; 2475 2476 mtx_lock(&softc->lock); 2477 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2478 if (be_lun->cbe_lun.lun_id == params->lun_id) 2479 break; 2480 } 2481 mtx_unlock(&softc->lock); 2482 if (be_lun == NULL) { 2483 snprintf(req->error_str, sizeof(req->error_str), 2484 "LUN %u is not managed by the block backend", 2485 params->lun_id); 2486 goto bailout_error; 2487 } 2488 cbe_lun = &be_lun->cbe_lun; 2489 2490 retval = ctl_disable_lun(cbe_lun); 2491 if (retval != 0) { 2492 snprintf(req->error_str, sizeof(req->error_str), 2493 "error %d returned from ctl_disable_lun() for " 2494 "LUN %d", retval, params->lun_id); 2495 goto bailout_error; 2496 } 2497 2498 if (be_lun->vn != NULL) { 2499 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2500 ctl_lun_no_media(cbe_lun); 2501 taskqueue_drain_all(be_lun->io_taskqueue); 2502 ctl_be_block_close(be_lun); 2503 } 2504 2505 retval = ctl_invalidate_lun(cbe_lun); 2506 if (retval != 0) { 2507 snprintf(req->error_str, sizeof(req->error_str), 2508 "error %d returned from ctl_invalidate_lun() for " 2509 "LUN %d", retval, params->lun_id); 2510 goto bailout_error; 2511 } 2512 2513 mtx_lock(&softc->lock); 2514 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2515 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2516 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2517 if (retval == EINTR) 2518 break; 2519 } 2520 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2521 2522 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2523 snprintf(req->error_str, sizeof(req->error_str), 2524 "interrupted waiting for LUN to be freed"); 2525 mtx_unlock(&softc->lock); 2526 goto bailout_error; 2527 } 2528 2529 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2530 2531 softc->num_luns--; 2532 mtx_unlock(&softc->lock); 2533 2534 taskqueue_drain_all(be_lun->io_taskqueue); 2535 taskqueue_free(be_lun->io_taskqueue); 2536 2537 if (be_lun->disk_stats != NULL) 2538 devstat_remove_entry(be_lun->disk_stats); 2539 2540 uma_zdestroy(be_lun->lun_zone); 2541 2542 ctl_free_opts(&cbe_lun->options); 2543 free(be_lun->dev_path, M_CTLBLK); 2544 mtx_destroy(&be_lun->queue_lock); 2545 mtx_destroy(&be_lun->io_lock); 2546 free(be_lun, M_CTLBLK); 2547 2548 req->status = CTL_LUN_OK; 2549 return (0); 2550 2551 bailout_error: 2552 req->status = CTL_LUN_ERROR; 2553 return (0); 2554 } 2555 2556 static int 2557 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2558 { 2559 struct ctl_lun_modify_params *params; 2560 struct ctl_be_block_lun *be_lun; 2561 struct ctl_be_lun *cbe_lun; 2562 char *value; 2563 uint64_t oldsize; 2564 int error, wasprim; 2565 2566 params = &req->reqdata.modify; 2567 2568 mtx_lock(&softc->lock); 2569 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2570 if (be_lun->cbe_lun.lun_id == params->lun_id) 2571 break; 2572 } 2573 mtx_unlock(&softc->lock); 2574 if (be_lun == NULL) { 2575 snprintf(req->error_str, sizeof(req->error_str), 2576 "LUN %u is not managed by the block backend", 2577 params->lun_id); 2578 goto bailout_error; 2579 } 2580 cbe_lun = &be_lun->cbe_lun; 2581 2582 if (params->lun_size_bytes != 0) 2583 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2584 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); 2585 2586 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2587 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2588 if (value != NULL) { 2589 if (strcmp(value, "primary") == 0) 2590 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2591 else 2592 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2593 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2594 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2595 else 2596 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2597 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2598 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2599 ctl_lun_primary(cbe_lun); 2600 else 2601 ctl_lun_secondary(cbe_lun); 2602 } 2603 2604 oldsize = be_lun->size_blocks; 2605 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2606 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2607 if (be_lun->vn == NULL) 2608 error = ctl_be_block_open(be_lun, req); 2609 else if (vn_isdisk(be_lun->vn, &error)) 2610 error = ctl_be_block_open_dev(be_lun, req); 2611 else if (be_lun->vn->v_type == VREG) { 2612 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2613 error = ctl_be_block_open_file(be_lun, req); 2614 VOP_UNLOCK(be_lun->vn, 0); 2615 } else 2616 error = EINVAL; 2617 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) && 2618 be_lun->vn != NULL) { 2619 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2620 ctl_lun_has_media(cbe_lun); 2621 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 && 2622 be_lun->vn == NULL) { 2623 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2624 ctl_lun_no_media(cbe_lun); 2625 } 2626 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2627 } else { 2628 if (be_lun->vn != NULL) { 2629 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2630 ctl_lun_no_media(cbe_lun); 2631 taskqueue_drain_all(be_lun->io_taskqueue); 2632 error = ctl_be_block_close(be_lun); 2633 } else 2634 error = 0; 2635 } 2636 if (be_lun->size_blocks != oldsize) 2637 ctl_lun_capacity_changed(cbe_lun); 2638 2639 /* Tell the user the exact size we ended up using */ 2640 params->lun_size_bytes = be_lun->size_bytes; 2641 2642 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2643 return (0); 2644 2645 bailout_error: 2646 req->status = CTL_LUN_ERROR; 2647 return (0); 2648 } 2649 2650 static void 2651 ctl_be_block_lun_shutdown(void *be_lun) 2652 { 2653 struct ctl_be_block_lun *lun; 2654 struct ctl_be_block_softc *softc; 2655 2656 lun = (struct ctl_be_block_lun *)be_lun; 2657 softc = lun->softc; 2658 2659 mtx_lock(&softc->lock); 2660 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2661 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2662 wakeup(lun); 2663 mtx_unlock(&softc->lock); 2664 } 2665 2666 static void 2667 ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2668 { 2669 struct ctl_be_block_lun *lun; 2670 struct ctl_be_block_softc *softc; 2671 2672 lun = (struct ctl_be_block_lun *)be_lun; 2673 softc = lun->softc; 2674 2675 if (status == CTL_LUN_CONFIG_OK) { 2676 mtx_lock(&softc->lock); 2677 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2678 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2679 wakeup(lun); 2680 mtx_unlock(&softc->lock); 2681 2682 /* 2683 * We successfully added the LUN, attempt to enable it. 2684 */ 2685 if (ctl_enable_lun(&lun->cbe_lun) != 0) { 2686 printf("%s: ctl_enable_lun() failed!\n", __func__); 2687 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { 2688 printf("%s: ctl_invalidate_lun() failed!\n", 2689 __func__); 2690 } 2691 } 2692 2693 return; 2694 } 2695 2696 2697 mtx_lock(&softc->lock); 2698 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2699 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2700 wakeup(lun); 2701 mtx_unlock(&softc->lock); 2702 } 2703 2704 2705 static int 2706 ctl_be_block_config_write(union ctl_io *io) 2707 { 2708 struct ctl_be_block_lun *be_lun; 2709 struct ctl_be_lun *cbe_lun; 2710 int retval; 2711 2712 DPRINTF("entered\n"); 2713 2714 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2715 CTL_PRIV_BACKEND_LUN].ptr; 2716 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2717 2718 retval = 0; 2719 switch (io->scsiio.cdb[0]) { 2720 case SYNCHRONIZE_CACHE: 2721 case SYNCHRONIZE_CACHE_16: 2722 case WRITE_SAME_10: 2723 case WRITE_SAME_16: 2724 case UNMAP: 2725 /* 2726 * The upper level CTL code will filter out any CDBs with 2727 * the immediate bit set and return the proper error. 2728 * 2729 * We don't really need to worry about what LBA range the 2730 * user asked to be synced out. When they issue a sync 2731 * cache command, we'll sync out the whole thing. 2732 */ 2733 mtx_lock(&be_lun->queue_lock); 2734 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2735 links); 2736 mtx_unlock(&be_lun->queue_lock); 2737 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2738 break; 2739 case START_STOP_UNIT: { 2740 struct scsi_start_stop_unit *cdb; 2741 struct ctl_lun_req req; 2742 2743 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2744 if ((cdb->how & SSS_PC_MASK) != 0) { 2745 ctl_set_success(&io->scsiio); 2746 ctl_config_write_done(io); 2747 break; 2748 } 2749 if (cdb->how & SSS_START) { 2750 if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) { 2751 retval = ctl_be_block_open(be_lun, &req); 2752 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; 2753 if (retval == 0) { 2754 cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; 2755 ctl_lun_has_media(cbe_lun); 2756 } else { 2757 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2758 ctl_lun_no_media(cbe_lun); 2759 } 2760 } 2761 ctl_start_lun(cbe_lun); 2762 } else { 2763 ctl_stop_lun(cbe_lun); 2764 if (cdb->how & SSS_LOEJ) { 2765 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; 2766 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED; 2767 ctl_lun_ejected(cbe_lun); 2768 if (be_lun->vn != NULL) 2769 ctl_be_block_close(be_lun); 2770 } 2771 } 2772 2773 ctl_set_success(&io->scsiio); 2774 ctl_config_write_done(io); 2775 break; 2776 } 2777 case PREVENT_ALLOW: 2778 ctl_set_success(&io->scsiio); 2779 ctl_config_write_done(io); 2780 break; 2781 default: 2782 ctl_set_invalid_opcode(&io->scsiio); 2783 ctl_config_write_done(io); 2784 retval = CTL_RETVAL_COMPLETE; 2785 break; 2786 } 2787 2788 return (retval); 2789 } 2790 2791 static int 2792 ctl_be_block_config_read(union ctl_io *io) 2793 { 2794 struct ctl_be_block_lun *be_lun; 2795 struct ctl_be_lun *cbe_lun; 2796 int retval = 0; 2797 2798 DPRINTF("entered\n"); 2799 2800 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2801 CTL_PRIV_BACKEND_LUN].ptr; 2802 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2803 2804 switch (io->scsiio.cdb[0]) { 2805 case SERVICE_ACTION_IN: 2806 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2807 mtx_lock(&be_lun->queue_lock); 2808 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2809 &io->io_hdr, links); 2810 mtx_unlock(&be_lun->queue_lock); 2811 taskqueue_enqueue(be_lun->io_taskqueue, 2812 &be_lun->io_task); 2813 retval = CTL_RETVAL_QUEUED; 2814 break; 2815 } 2816 ctl_set_invalid_field(&io->scsiio, 2817 /*sks_valid*/ 1, 2818 /*command*/ 1, 2819 /*field*/ 1, 2820 /*bit_valid*/ 1, 2821 /*bit*/ 4); 2822 ctl_config_read_done(io); 2823 retval = CTL_RETVAL_COMPLETE; 2824 break; 2825 default: 2826 ctl_set_invalid_opcode(&io->scsiio); 2827 ctl_config_read_done(io); 2828 retval = CTL_RETVAL_COMPLETE; 2829 break; 2830 } 2831 2832 return (retval); 2833 } 2834 2835 static int 2836 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2837 { 2838 struct ctl_be_block_lun *lun; 2839 int retval; 2840 2841 lun = (struct ctl_be_block_lun *)be_lun; 2842 2843 retval = sbuf_printf(sb, "\t<num_threads>"); 2844 if (retval != 0) 2845 goto bailout; 2846 retval = sbuf_printf(sb, "%d", lun->num_threads); 2847 if (retval != 0) 2848 goto bailout; 2849 retval = sbuf_printf(sb, "</num_threads>\n"); 2850 2851 bailout: 2852 return (retval); 2853 } 2854 2855 static uint64_t 2856 ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2857 { 2858 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2859 2860 if (lun->getattr == NULL) 2861 return (UINT64_MAX); 2862 return (lun->getattr(lun, attrname)); 2863 } 2864 2865 int 2866 ctl_be_block_init(void) 2867 { 2868 struct ctl_be_block_softc *softc; 2869 int retval; 2870 2871 softc = &backend_block_softc; 2872 retval = 0; 2873 2874 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2875 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2876 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2877 STAILQ_INIT(&softc->lun_list); 2878 2879 return (retval); 2880 } 2881