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 * All rights reserved. 6 * 7 * Portions of this software were developed by Edward Tomasz Napierala 8 * under sponsorship from the FreeBSD Foundation. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions, and the following disclaimer, 15 * without modification. 16 * 2. Redistributions in binary form must reproduce at minimum a disclaimer 17 * substantially similar to the "NO WARRANTY" disclaimer below 18 * ("Disclaimer") and any redistribution must be conditioned upon 19 * including a substantially similar Disclaimer requirement for further 20 * binary redistribution. 21 * 22 * NO WARRANTY 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR 26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 27 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGES. 34 * 35 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ 36 */ 37 /* 38 * CAM Target Layer driver backend for block devices. 39 * 40 * Author: Ken Merry <ken@FreeBSD.org> 41 */ 42 #include <sys/cdefs.h> 43 __FBSDID("$FreeBSD$"); 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/kernel.h> 48 #include <sys/types.h> 49 #include <sys/kthread.h> 50 #include <sys/bio.h> 51 #include <sys/fcntl.h> 52 #include <sys/limits.h> 53 #include <sys/lock.h> 54 #include <sys/mutex.h> 55 #include <sys/condvar.h> 56 #include <sys/malloc.h> 57 #include <sys/conf.h> 58 #include <sys/ioccom.h> 59 #include <sys/queue.h> 60 #include <sys/sbuf.h> 61 #include <sys/endian.h> 62 #include <sys/uio.h> 63 #include <sys/buf.h> 64 #include <sys/taskqueue.h> 65 #include <sys/vnode.h> 66 #include <sys/namei.h> 67 #include <sys/mount.h> 68 #include <sys/disk.h> 69 #include <sys/fcntl.h> 70 #include <sys/filedesc.h> 71 #include <sys/filio.h> 72 #include <sys/proc.h> 73 #include <sys/pcpu.h> 74 #include <sys/module.h> 75 #include <sys/sdt.h> 76 #include <sys/devicestat.h> 77 #include <sys/sysctl.h> 78 79 #include <geom/geom.h> 80 81 #include <cam/cam.h> 82 #include <cam/scsi/scsi_all.h> 83 #include <cam/scsi/scsi_da.h> 84 #include <cam/ctl/ctl_io.h> 85 #include <cam/ctl/ctl.h> 86 #include <cam/ctl/ctl_backend.h> 87 #include <cam/ctl/ctl_frontend_internal.h> 88 #include <cam/ctl/ctl_ioctl.h> 89 #include <cam/ctl/ctl_scsi_all.h> 90 #include <cam/ctl/ctl_error.h> 91 92 /* 93 * The idea here is that we'll allocate enough S/G space to hold a 1MB 94 * I/O. If we get an I/O larger than that, we'll split it. 95 */ 96 #define CTLBLK_HALF_IO_SIZE (512 * 1024) 97 #define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 98 #define CTLBLK_MAX_SEG MAXPHYS 99 #define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 100 #define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 101 102 #ifdef CTLBLK_DEBUG 103 #define DPRINTF(fmt, args...) \ 104 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 105 #else 106 #define DPRINTF(fmt, args...) do {} while(0) 107 #endif 108 109 #define PRIV(io) \ 110 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 111 #define ARGS(io) \ 112 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 113 114 SDT_PROVIDER_DEFINE(cbb); 115 116 typedef enum { 117 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 118 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 119 CTL_BE_BLOCK_LUN_WAITING = 0x04, 120 CTL_BE_BLOCK_LUN_MULTI_THREAD = 0x08 121 } ctl_be_block_lun_flags; 122 123 typedef enum { 124 CTL_BE_BLOCK_NONE, 125 CTL_BE_BLOCK_DEV, 126 CTL_BE_BLOCK_FILE 127 } ctl_be_block_type; 128 129 struct ctl_be_block_devdata { 130 struct cdev *cdev; 131 struct cdevsw *csw; 132 int dev_ref; 133 }; 134 135 struct ctl_be_block_filedata { 136 struct ucred *cred; 137 }; 138 139 union ctl_be_block_bedata { 140 struct ctl_be_block_devdata dev; 141 struct ctl_be_block_filedata file; 142 }; 143 144 struct ctl_be_block_io; 145 struct ctl_be_block_lun; 146 147 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 148 struct ctl_be_block_io *beio); 149 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 150 const char *attrname); 151 152 /* 153 * Backend LUN structure. There is a 1:1 mapping between a block device 154 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 155 */ 156 struct ctl_be_block_lun { 157 struct ctl_lun_create_params params; 158 struct ctl_block_disk *disk; 159 char lunname[32]; 160 char *dev_path; 161 ctl_be_block_type dev_type; 162 struct vnode *vn; 163 union ctl_be_block_bedata backend; 164 cbb_dispatch_t dispatch; 165 cbb_dispatch_t lun_flush; 166 cbb_dispatch_t unmap; 167 cbb_dispatch_t get_lba_status; 168 cbb_getattr_t getattr; 169 uma_zone_t lun_zone; 170 uint64_t size_blocks; 171 uint64_t size_bytes; 172 uint32_t blocksize; 173 int blocksize_shift; 174 uint16_t pblockexp; 175 uint16_t pblockoff; 176 struct ctl_be_block_softc *softc; 177 struct devstat *disk_stats; 178 ctl_be_block_lun_flags flags; 179 STAILQ_ENTRY(ctl_be_block_lun) links; 180 struct ctl_be_lun ctl_be_lun; 181 struct taskqueue *io_taskqueue; 182 struct task io_task; 183 int num_threads; 184 STAILQ_HEAD(, ctl_io_hdr) input_queue; 185 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 186 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 187 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 188 struct mtx_padalign io_lock; 189 struct mtx_padalign queue_lock; 190 }; 191 192 /* 193 * Overall softc structure for the block backend module. 194 */ 195 struct ctl_be_block_softc { 196 struct mtx lock; 197 int num_disks; 198 STAILQ_HEAD(, ctl_block_disk) disk_list; 199 int num_luns; 200 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 201 }; 202 203 static struct ctl_be_block_softc backend_block_softc; 204 205 /* 206 * Per-I/O information. 207 */ 208 struct ctl_be_block_io { 209 union ctl_io *io; 210 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 211 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 212 int bio_cmd; 213 int num_segs; 214 int num_bios_sent; 215 int num_bios_done; 216 int send_complete; 217 int num_errors; 218 struct bintime ds_t0; 219 devstat_tag_type ds_tag_type; 220 devstat_trans_flags ds_trans_type; 221 uint64_t io_len; 222 uint64_t io_offset; 223 struct ctl_be_block_softc *softc; 224 struct ctl_be_block_lun *lun; 225 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 226 }; 227 228 static int cbb_num_threads = 14; 229 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 230 "CAM Target Layer Block Backend"); 231 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 232 &cbb_num_threads, 0, "Number of threads per backing file"); 233 234 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 235 static void ctl_free_beio(struct ctl_be_block_io *beio); 236 static void ctl_complete_beio(struct ctl_be_block_io *beio); 237 static int ctl_be_block_move_done(union ctl_io *io); 238 static void ctl_be_block_biodone(struct bio *bio); 239 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 240 struct ctl_be_block_io *beio); 241 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 242 struct ctl_be_block_io *beio); 243 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 244 struct ctl_be_block_io *beio); 245 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 246 const char *attrname); 247 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 248 struct ctl_be_block_io *beio); 249 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 250 struct ctl_be_block_io *beio); 251 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 252 struct ctl_be_block_io *beio); 253 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 254 const char *attrname); 255 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 256 union ctl_io *io); 257 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 258 union ctl_io *io); 259 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 260 union ctl_io *io); 261 static void ctl_be_block_worker(void *context, int pending); 262 static int ctl_be_block_submit(union ctl_io *io); 263 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 264 int flag, struct thread *td); 265 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 266 struct ctl_lun_req *req); 267 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 268 struct ctl_lun_req *req); 269 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 270 static int ctl_be_block_open(struct ctl_be_block_softc *softc, 271 struct ctl_be_block_lun *be_lun, 272 struct ctl_lun_req *req); 273 static int ctl_be_block_create(struct ctl_be_block_softc *softc, 274 struct ctl_lun_req *req); 275 static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 276 struct ctl_lun_req *req); 277 static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 278 struct ctl_lun_req *req); 279 static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 280 struct ctl_lun_req *req); 281 static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 282 struct ctl_lun_req *req); 283 static void ctl_be_block_lun_shutdown(void *be_lun); 284 static void ctl_be_block_lun_config_status(void *be_lun, 285 ctl_lun_config_status status); 286 static int ctl_be_block_config_write(union ctl_io *io); 287 static int ctl_be_block_config_read(union ctl_io *io); 288 static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 289 static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 290 int ctl_be_block_init(void); 291 292 static struct ctl_backend_driver ctl_be_block_driver = 293 { 294 .name = "block", 295 .flags = CTL_BE_FLAG_HAS_CONFIG, 296 .init = ctl_be_block_init, 297 .data_submit = ctl_be_block_submit, 298 .data_move_done = ctl_be_block_move_done, 299 .config_read = ctl_be_block_config_read, 300 .config_write = ctl_be_block_config_write, 301 .ioctl = ctl_be_block_ioctl, 302 .lun_info = ctl_be_block_lun_info, 303 .lun_attr = ctl_be_block_lun_attr 304 }; 305 306 MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 307 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 308 309 static uma_zone_t beio_zone; 310 311 static struct ctl_be_block_io * 312 ctl_alloc_beio(struct ctl_be_block_softc *softc) 313 { 314 struct ctl_be_block_io *beio; 315 316 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 317 beio->softc = softc; 318 return (beio); 319 } 320 321 static void 322 ctl_free_beio(struct ctl_be_block_io *beio) 323 { 324 int duplicate_free; 325 int i; 326 327 duplicate_free = 0; 328 329 for (i = 0; i < beio->num_segs; i++) { 330 if (beio->sg_segs[i].addr == NULL) 331 duplicate_free++; 332 333 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 334 beio->sg_segs[i].addr = NULL; 335 336 /* For compare we had two equal S/G lists. */ 337 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 338 uma_zfree(beio->lun->lun_zone, 339 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 340 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 341 } 342 } 343 344 if (duplicate_free > 0) { 345 printf("%s: %d duplicate frees out of %d segments\n", __func__, 346 duplicate_free, beio->num_segs); 347 } 348 349 uma_zfree(beio_zone, beio); 350 } 351 352 static void 353 ctl_complete_beio(struct ctl_be_block_io *beio) 354 { 355 union ctl_io *io = beio->io; 356 357 if (beio->beio_cont != NULL) { 358 beio->beio_cont(beio); 359 } else { 360 ctl_free_beio(beio); 361 ctl_data_submit_done(io); 362 } 363 } 364 365 static int 366 ctl_be_block_move_done(union ctl_io *io) 367 { 368 struct ctl_be_block_io *beio; 369 struct ctl_be_block_lun *be_lun; 370 struct ctl_lba_len_flags *lbalen; 371 #ifdef CTL_TIME_IO 372 struct bintime cur_bt; 373 #endif 374 int i; 375 376 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 377 be_lun = beio->lun; 378 379 DPRINTF("entered\n"); 380 381 #ifdef CTL_TIME_IO 382 getbintime(&cur_bt); 383 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 384 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 385 io->io_hdr.num_dmas++; 386 #endif 387 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 388 389 /* 390 * We set status at this point for read commands, and write 391 * commands with errors. 392 */ 393 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 394 ; 395 } else if ((io->io_hdr.port_status == 0) && 396 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 397 lbalen = ARGS(beio->io); 398 if (lbalen->flags & CTL_LLF_READ) { 399 ctl_set_success(&io->scsiio); 400 } else if (lbalen->flags & CTL_LLF_COMPARE) { 401 /* We have two data blocks ready for comparison. */ 402 for (i = 0; i < beio->num_segs; i++) { 403 if (memcmp(beio->sg_segs[i].addr, 404 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 405 beio->sg_segs[i].len) != 0) 406 break; 407 } 408 if (i < beio->num_segs) 409 ctl_set_sense(&io->scsiio, 410 /*current_error*/ 1, 411 /*sense_key*/ SSD_KEY_MISCOMPARE, 412 /*asc*/ 0x1D, 413 /*ascq*/ 0x00, 414 SSD_ELEM_NONE); 415 else 416 ctl_set_success(&io->scsiio); 417 } 418 } else if ((io->io_hdr.port_status != 0) && 419 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 420 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 421 /* 422 * For hardware error sense keys, the sense key 423 * specific value is defined to be a retry count, 424 * but we use it to pass back an internal FETD 425 * error code. XXX KDM Hopefully the FETD is only 426 * using 16 bits for an error code, since that's 427 * all the space we have in the sks field. 428 */ 429 ctl_set_internal_failure(&io->scsiio, 430 /*sks_valid*/ 1, 431 /*retry_count*/ 432 io->io_hdr.port_status); 433 } 434 435 /* 436 * If this is a read, or a write with errors, it is done. 437 */ 438 if ((beio->bio_cmd == BIO_READ) 439 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 440 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 441 ctl_complete_beio(beio); 442 return (0); 443 } 444 445 /* 446 * At this point, we have a write and the DMA completed 447 * successfully. We now have to queue it to the task queue to 448 * execute the backend I/O. That is because we do blocking 449 * memory allocations, and in the file backing case, blocking I/O. 450 * This move done routine is generally called in the SIM's 451 * interrupt context, and therefore we cannot block. 452 */ 453 mtx_lock(&be_lun->queue_lock); 454 /* 455 * XXX KDM make sure that links is okay to use at this point. 456 * Otherwise, we either need to add another field to ctl_io_hdr, 457 * or deal with resource allocation here. 458 */ 459 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 460 mtx_unlock(&be_lun->queue_lock); 461 462 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 463 464 return (0); 465 } 466 467 static void 468 ctl_be_block_biodone(struct bio *bio) 469 { 470 struct ctl_be_block_io *beio; 471 struct ctl_be_block_lun *be_lun; 472 union ctl_io *io; 473 int error; 474 475 beio = bio->bio_caller1; 476 be_lun = beio->lun; 477 io = beio->io; 478 479 DPRINTF("entered\n"); 480 481 error = bio->bio_error; 482 mtx_lock(&be_lun->io_lock); 483 if (error != 0) 484 beio->num_errors++; 485 486 beio->num_bios_done++; 487 488 /* 489 * XXX KDM will this cause WITNESS to complain? Holding a lock 490 * during the free might cause it to complain. 491 */ 492 g_destroy_bio(bio); 493 494 /* 495 * If the send complete bit isn't set, or we aren't the last I/O to 496 * complete, then we're done. 497 */ 498 if ((beio->send_complete == 0) 499 || (beio->num_bios_done < beio->num_bios_sent)) { 500 mtx_unlock(&be_lun->io_lock); 501 return; 502 } 503 504 /* 505 * At this point, we've verified that we are the last I/O to 506 * complete, so it's safe to drop the lock. 507 */ 508 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 509 beio->ds_tag_type, beio->ds_trans_type, 510 /*now*/ NULL, /*then*/&beio->ds_t0); 511 mtx_unlock(&be_lun->io_lock); 512 513 /* 514 * If there are any errors from the backing device, we fail the 515 * entire I/O with a medium error. 516 */ 517 if (beio->num_errors > 0) { 518 if (error == EOPNOTSUPP) { 519 ctl_set_invalid_opcode(&io->scsiio); 520 } else if (error == ENOSPC) { 521 ctl_set_space_alloc_fail(&io->scsiio); 522 } else if (beio->bio_cmd == BIO_FLUSH) { 523 /* XXX KDM is there is a better error here? */ 524 ctl_set_internal_failure(&io->scsiio, 525 /*sks_valid*/ 1, 526 /*retry_count*/ 0xbad2); 527 } else 528 ctl_set_medium_error(&io->scsiio); 529 ctl_complete_beio(beio); 530 return; 531 } 532 533 /* 534 * If this is a write, a flush, a delete or verify, we're all done. 535 * If this is a read, we can now send the data to the user. 536 */ 537 if ((beio->bio_cmd == BIO_WRITE) 538 || (beio->bio_cmd == BIO_FLUSH) 539 || (beio->bio_cmd == BIO_DELETE) 540 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 541 ctl_set_success(&io->scsiio); 542 ctl_complete_beio(beio); 543 } else { 544 if ((ARGS(io)->flags & CTL_LLF_READ) && 545 beio->beio_cont == NULL) 546 ctl_set_success(&io->scsiio); 547 #ifdef CTL_TIME_IO 548 getbintime(&io->io_hdr.dma_start_bt); 549 #endif 550 ctl_datamove(io); 551 } 552 } 553 554 static void 555 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 556 struct ctl_be_block_io *beio) 557 { 558 union ctl_io *io = beio->io; 559 struct mount *mountpoint; 560 int error, lock_flags; 561 562 DPRINTF("entered\n"); 563 564 binuptime(&beio->ds_t0); 565 mtx_lock(&be_lun->io_lock); 566 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 567 mtx_unlock(&be_lun->io_lock); 568 569 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 570 571 if (MNT_SHARED_WRITES(mountpoint) 572 || ((mountpoint == NULL) 573 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 574 lock_flags = LK_SHARED; 575 else 576 lock_flags = LK_EXCLUSIVE; 577 578 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 579 580 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 581 VOP_UNLOCK(be_lun->vn, 0); 582 583 vn_finished_write(mountpoint); 584 585 mtx_lock(&be_lun->io_lock); 586 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 587 beio->ds_tag_type, beio->ds_trans_type, 588 /*now*/ NULL, /*then*/&beio->ds_t0); 589 mtx_unlock(&be_lun->io_lock); 590 591 if (error == 0) 592 ctl_set_success(&io->scsiio); 593 else { 594 /* XXX KDM is there is a better error here? */ 595 ctl_set_internal_failure(&io->scsiio, 596 /*sks_valid*/ 1, 597 /*retry_count*/ 0xbad1); 598 } 599 600 ctl_complete_beio(beio); 601 } 602 603 SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 604 SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 605 SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 606 SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 607 608 static void 609 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 610 struct ctl_be_block_io *beio) 611 { 612 struct ctl_be_block_filedata *file_data; 613 union ctl_io *io; 614 struct uio xuio; 615 struct iovec *xiovec; 616 int flags; 617 int error, i; 618 619 DPRINTF("entered\n"); 620 621 file_data = &be_lun->backend.file; 622 io = beio->io; 623 flags = 0; 624 if (ARGS(io)->flags & CTL_LLF_DPO) 625 flags |= IO_DIRECT; 626 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 627 flags |= IO_SYNC; 628 629 bzero(&xuio, sizeof(xuio)); 630 if (beio->bio_cmd == BIO_READ) { 631 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 632 xuio.uio_rw = UIO_READ; 633 } else { 634 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 635 xuio.uio_rw = UIO_WRITE; 636 } 637 xuio.uio_offset = beio->io_offset; 638 xuio.uio_resid = beio->io_len; 639 xuio.uio_segflg = UIO_SYSSPACE; 640 xuio.uio_iov = beio->xiovecs; 641 xuio.uio_iovcnt = beio->num_segs; 642 xuio.uio_td = curthread; 643 644 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 645 xiovec->iov_base = beio->sg_segs[i].addr; 646 xiovec->iov_len = beio->sg_segs[i].len; 647 } 648 649 binuptime(&beio->ds_t0); 650 mtx_lock(&be_lun->io_lock); 651 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 652 mtx_unlock(&be_lun->io_lock); 653 654 if (beio->bio_cmd == BIO_READ) { 655 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 656 657 /* 658 * UFS pays attention to IO_DIRECT for reads. If the 659 * DIRECTIO option is configured into the kernel, it calls 660 * ffs_rawread(). But that only works for single-segment 661 * uios with user space addresses. In our case, with a 662 * kernel uio, it still reads into the buffer cache, but it 663 * will just try to release the buffer from the cache later 664 * on in ffs_read(). 665 * 666 * ZFS does not pay attention to IO_DIRECT for reads. 667 * 668 * UFS does not pay attention to IO_SYNC for reads. 669 * 670 * ZFS pays attention to IO_SYNC (which translates into the 671 * Solaris define FRSYNC for zfs_read()) for reads. It 672 * attempts to sync the file before reading. 673 * 674 * So, to attempt to provide some barrier semantics in the 675 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 676 */ 677 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 678 679 VOP_UNLOCK(be_lun->vn, 0); 680 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 681 } else { 682 struct mount *mountpoint; 683 int lock_flags; 684 685 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 686 687 if (MNT_SHARED_WRITES(mountpoint) 688 || ((mountpoint == NULL) 689 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 690 lock_flags = LK_SHARED; 691 else 692 lock_flags = LK_EXCLUSIVE; 693 694 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 695 696 /* 697 * UFS pays attention to IO_DIRECT for writes. The write 698 * is done asynchronously. (Normally the write would just 699 * get put into cache. 700 * 701 * UFS pays attention to IO_SYNC for writes. It will 702 * attempt to write the buffer out synchronously if that 703 * flag is set. 704 * 705 * ZFS does not pay attention to IO_DIRECT for writes. 706 * 707 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 708 * for writes. It will flush the transaction from the 709 * cache before returning. 710 * 711 * So if we've got the BIO_ORDERED flag set, we want 712 * IO_SYNC in either the UFS or ZFS case. 713 */ 714 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 715 VOP_UNLOCK(be_lun->vn, 0); 716 717 vn_finished_write(mountpoint); 718 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 719 } 720 721 mtx_lock(&be_lun->io_lock); 722 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 723 beio->ds_tag_type, beio->ds_trans_type, 724 /*now*/ NULL, /*then*/&beio->ds_t0); 725 mtx_unlock(&be_lun->io_lock); 726 727 /* 728 * If we got an error, set the sense data to "MEDIUM ERROR" and 729 * return the I/O to the user. 730 */ 731 if (error != 0) { 732 char path_str[32]; 733 734 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 735 printf("%s%s command returned errno %d\n", path_str, 736 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 737 if (error == ENOSPC) { 738 ctl_set_space_alloc_fail(&io->scsiio); 739 } else 740 ctl_set_medium_error(&io->scsiio); 741 ctl_complete_beio(beio); 742 return; 743 } 744 745 /* 746 * If this is a write or a verify, we're all done. 747 * If this is a read, we can now send the data to the user. 748 */ 749 if ((beio->bio_cmd == BIO_WRITE) || 750 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 751 ctl_set_success(&io->scsiio); 752 ctl_complete_beio(beio); 753 } else { 754 if ((ARGS(io)->flags & CTL_LLF_READ) && 755 beio->beio_cont == NULL) 756 ctl_set_success(&io->scsiio); 757 #ifdef CTL_TIME_IO 758 getbintime(&io->io_hdr.dma_start_bt); 759 #endif 760 ctl_datamove(io); 761 } 762 } 763 764 static void 765 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 766 struct ctl_be_block_io *beio) 767 { 768 union ctl_io *io = beio->io; 769 struct ctl_lba_len_flags *lbalen = ARGS(io); 770 struct scsi_get_lba_status_data *data; 771 off_t roff, off; 772 int error, status; 773 774 DPRINTF("entered\n"); 775 776 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 777 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 778 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 779 0, curthread->td_ucred, curthread); 780 if (error == 0 && off > roff) 781 status = 0; /* mapped up to off */ 782 else { 783 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 784 0, curthread->td_ucred, curthread); 785 if (error == 0 && off > roff) 786 status = 1; /* deallocated up to off */ 787 else { 788 status = 0; /* unknown up to the end */ 789 off = be_lun->size_bytes; 790 } 791 } 792 VOP_UNLOCK(be_lun->vn, 0); 793 794 off >>= be_lun->blocksize_shift; 795 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 796 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 797 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 798 data->descr[0].length); 799 data->descr[0].status = status; 800 801 ctl_complete_beio(beio); 802 } 803 804 static uint64_t 805 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 806 { 807 struct vattr vattr; 808 struct statfs statfs; 809 int error; 810 811 if (be_lun->vn == NULL) 812 return (UINT64_MAX); 813 if (strcmp(attrname, "blocksused") == 0) { 814 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 815 if (error != 0) 816 return (UINT64_MAX); 817 return (vattr.va_bytes >> be_lun->blocksize_shift); 818 } 819 if (strcmp(attrname, "blocksavail") == 0) { 820 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 821 if (error != 0) 822 return (UINT64_MAX); 823 return ((statfs.f_bavail * statfs.f_bsize) >> 824 be_lun->blocksize_shift); 825 } 826 return (UINT64_MAX); 827 } 828 829 static void 830 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 831 struct ctl_be_block_io *beio) 832 { 833 struct ctl_be_block_devdata *dev_data; 834 union ctl_io *io; 835 struct uio xuio; 836 struct iovec *xiovec; 837 int flags; 838 int error, i; 839 840 DPRINTF("entered\n"); 841 842 dev_data = &be_lun->backend.dev; 843 io = beio->io; 844 flags = 0; 845 if (ARGS(io)->flags & CTL_LLF_DPO) 846 flags |= IO_DIRECT; 847 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 848 flags |= IO_SYNC; 849 850 bzero(&xuio, sizeof(xuio)); 851 if (beio->bio_cmd == BIO_READ) { 852 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 853 xuio.uio_rw = UIO_READ; 854 } else { 855 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 856 xuio.uio_rw = UIO_WRITE; 857 } 858 xuio.uio_offset = beio->io_offset; 859 xuio.uio_resid = beio->io_len; 860 xuio.uio_segflg = UIO_SYSSPACE; 861 xuio.uio_iov = beio->xiovecs; 862 xuio.uio_iovcnt = beio->num_segs; 863 xuio.uio_td = curthread; 864 865 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 866 xiovec->iov_base = beio->sg_segs[i].addr; 867 xiovec->iov_len = beio->sg_segs[i].len; 868 } 869 870 binuptime(&beio->ds_t0); 871 mtx_lock(&be_lun->io_lock); 872 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 873 mtx_unlock(&be_lun->io_lock); 874 875 if (beio->bio_cmd == BIO_READ) { 876 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 877 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 878 } else { 879 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 880 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 881 } 882 883 mtx_lock(&be_lun->io_lock); 884 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 885 beio->ds_tag_type, beio->ds_trans_type, 886 /*now*/ NULL, /*then*/&beio->ds_t0); 887 mtx_unlock(&be_lun->io_lock); 888 889 /* 890 * If we got an error, set the sense data to "MEDIUM ERROR" and 891 * return the I/O to the user. 892 */ 893 if (error != 0) { 894 if (error == ENOSPC) { 895 ctl_set_space_alloc_fail(&io->scsiio); 896 } else 897 ctl_set_medium_error(&io->scsiio); 898 ctl_complete_beio(beio); 899 return; 900 } 901 902 /* 903 * If this is a write or a verify, we're all done. 904 * If this is a read, we can now send the data to the user. 905 */ 906 if ((beio->bio_cmd == BIO_WRITE) || 907 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 908 ctl_set_success(&io->scsiio); 909 ctl_complete_beio(beio); 910 } else { 911 if ((ARGS(io)->flags & CTL_LLF_READ) && 912 beio->beio_cont == NULL) 913 ctl_set_success(&io->scsiio); 914 #ifdef CTL_TIME_IO 915 getbintime(&io->io_hdr.dma_start_bt); 916 #endif 917 ctl_datamove(io); 918 } 919 } 920 921 static void 922 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 923 struct ctl_be_block_io *beio) 924 { 925 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 926 union ctl_io *io = beio->io; 927 struct ctl_lba_len_flags *lbalen = ARGS(io); 928 struct scsi_get_lba_status_data *data; 929 off_t roff, off; 930 int error, status; 931 932 DPRINTF("entered\n"); 933 934 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 935 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKHOLE, 936 (caddr_t)&off, FREAD, curthread); 937 if (error == 0 && off > roff) 938 status = 0; /* mapped up to off */ 939 else { 940 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKDATA, 941 (caddr_t)&off, FREAD, curthread); 942 if (error == 0 && off > roff) 943 status = 1; /* deallocated up to off */ 944 else { 945 status = 0; /* unknown up to the end */ 946 off = be_lun->size_bytes; 947 } 948 } 949 950 off >>= be_lun->blocksize_shift; 951 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 952 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 953 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 954 data->descr[0].length); 955 data->descr[0].status = status; 956 957 ctl_complete_beio(beio); 958 } 959 960 static void 961 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 962 struct ctl_be_block_io *beio) 963 { 964 struct bio *bio; 965 union ctl_io *io; 966 struct ctl_be_block_devdata *dev_data; 967 968 dev_data = &be_lun->backend.dev; 969 io = beio->io; 970 971 DPRINTF("entered\n"); 972 973 /* This can't fail, it's a blocking allocation. */ 974 bio = g_alloc_bio(); 975 976 bio->bio_cmd = BIO_FLUSH; 977 bio->bio_flags |= BIO_ORDERED; 978 bio->bio_dev = dev_data->cdev; 979 bio->bio_offset = 0; 980 bio->bio_data = 0; 981 bio->bio_done = ctl_be_block_biodone; 982 bio->bio_caller1 = beio; 983 bio->bio_pblkno = 0; 984 985 /* 986 * We don't need to acquire the LUN lock here, because we are only 987 * sending one bio, and so there is no other context to synchronize 988 * with. 989 */ 990 beio->num_bios_sent = 1; 991 beio->send_complete = 1; 992 993 binuptime(&beio->ds_t0); 994 mtx_lock(&be_lun->io_lock); 995 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 996 mtx_unlock(&be_lun->io_lock); 997 998 (*dev_data->csw->d_strategy)(bio); 999 } 1000 1001 static void 1002 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1003 struct ctl_be_block_io *beio, 1004 uint64_t off, uint64_t len, int last) 1005 { 1006 struct bio *bio; 1007 struct ctl_be_block_devdata *dev_data; 1008 uint64_t maxlen; 1009 1010 dev_data = &be_lun->backend.dev; 1011 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 1012 while (len > 0) { 1013 bio = g_alloc_bio(); 1014 bio->bio_cmd = BIO_DELETE; 1015 bio->bio_dev = dev_data->cdev; 1016 bio->bio_offset = off; 1017 bio->bio_length = MIN(len, maxlen); 1018 bio->bio_data = 0; 1019 bio->bio_done = ctl_be_block_biodone; 1020 bio->bio_caller1 = beio; 1021 bio->bio_pblkno = off / be_lun->blocksize; 1022 1023 off += bio->bio_length; 1024 len -= bio->bio_length; 1025 1026 mtx_lock(&be_lun->io_lock); 1027 beio->num_bios_sent++; 1028 if (last && len == 0) 1029 beio->send_complete = 1; 1030 mtx_unlock(&be_lun->io_lock); 1031 1032 (*dev_data->csw->d_strategy)(bio); 1033 } 1034 } 1035 1036 static void 1037 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1038 struct ctl_be_block_io *beio) 1039 { 1040 union ctl_io *io; 1041 struct ctl_be_block_devdata *dev_data; 1042 struct ctl_ptr_len_flags *ptrlen; 1043 struct scsi_unmap_desc *buf, *end; 1044 uint64_t len; 1045 1046 dev_data = &be_lun->backend.dev; 1047 io = beio->io; 1048 1049 DPRINTF("entered\n"); 1050 1051 binuptime(&beio->ds_t0); 1052 mtx_lock(&be_lun->io_lock); 1053 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1054 mtx_unlock(&be_lun->io_lock); 1055 1056 if (beio->io_offset == -1) { 1057 beio->io_len = 0; 1058 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1059 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1060 end = buf + ptrlen->len / sizeof(*buf); 1061 for (; buf < end; buf++) { 1062 len = (uint64_t)scsi_4btoul(buf->length) * 1063 be_lun->blocksize; 1064 beio->io_len += len; 1065 ctl_be_block_unmap_dev_range(be_lun, beio, 1066 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 1067 (end - buf < 2) ? TRUE : FALSE); 1068 } 1069 } else 1070 ctl_be_block_unmap_dev_range(be_lun, beio, 1071 beio->io_offset, beio->io_len, TRUE); 1072 } 1073 1074 static void 1075 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1076 struct ctl_be_block_io *beio) 1077 { 1078 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1079 int i; 1080 struct bio *bio; 1081 struct ctl_be_block_devdata *dev_data; 1082 off_t cur_offset; 1083 int max_iosize; 1084 1085 DPRINTF("entered\n"); 1086 1087 dev_data = &be_lun->backend.dev; 1088 1089 /* 1090 * We have to limit our I/O size to the maximum supported by the 1091 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1092 * set it properly, use DFLTPHYS. 1093 */ 1094 max_iosize = dev_data->cdev->si_iosize_max; 1095 if (max_iosize < PAGE_SIZE) 1096 max_iosize = DFLTPHYS; 1097 1098 cur_offset = beio->io_offset; 1099 for (i = 0; i < beio->num_segs; i++) { 1100 size_t cur_size; 1101 uint8_t *cur_ptr; 1102 1103 cur_size = beio->sg_segs[i].len; 1104 cur_ptr = beio->sg_segs[i].addr; 1105 1106 while (cur_size > 0) { 1107 /* This can't fail, it's a blocking allocation. */ 1108 bio = g_alloc_bio(); 1109 1110 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1111 1112 bio->bio_cmd = beio->bio_cmd; 1113 bio->bio_dev = dev_data->cdev; 1114 bio->bio_caller1 = beio; 1115 bio->bio_length = min(cur_size, max_iosize); 1116 bio->bio_offset = cur_offset; 1117 bio->bio_data = cur_ptr; 1118 bio->bio_done = ctl_be_block_biodone; 1119 bio->bio_pblkno = cur_offset / be_lun->blocksize; 1120 1121 cur_offset += bio->bio_length; 1122 cur_ptr += bio->bio_length; 1123 cur_size -= bio->bio_length; 1124 1125 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1126 beio->num_bios_sent++; 1127 } 1128 } 1129 binuptime(&beio->ds_t0); 1130 mtx_lock(&be_lun->io_lock); 1131 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1132 beio->send_complete = 1; 1133 mtx_unlock(&be_lun->io_lock); 1134 1135 /* 1136 * Fire off all allocated requests! 1137 */ 1138 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1139 TAILQ_REMOVE(&queue, bio, bio_queue); 1140 (*dev_data->csw->d_strategy)(bio); 1141 } 1142 } 1143 1144 static uint64_t 1145 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1146 { 1147 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 1148 struct diocgattr_arg arg; 1149 int error; 1150 1151 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) 1152 return (UINT64_MAX); 1153 strlcpy(arg.name, attrname, sizeof(arg.name)); 1154 arg.len = sizeof(arg.value.off); 1155 error = dev_data->csw->d_ioctl(dev_data->cdev, 1156 DIOCGATTR, (caddr_t)&arg, FREAD, curthread); 1157 if (error != 0) 1158 return (UINT64_MAX); 1159 return (arg.value.off); 1160 } 1161 1162 static void 1163 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1164 { 1165 union ctl_io *io; 1166 1167 io = beio->io; 1168 ctl_free_beio(beio); 1169 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1170 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1171 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1172 ctl_config_write_done(io); 1173 return; 1174 } 1175 1176 ctl_be_block_config_write(io); 1177 } 1178 1179 static void 1180 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1181 union ctl_io *io) 1182 { 1183 struct ctl_be_block_io *beio; 1184 struct ctl_be_block_softc *softc; 1185 struct ctl_lba_len_flags *lbalen; 1186 uint64_t len_left, lba; 1187 int i, seglen; 1188 uint8_t *buf, *end; 1189 1190 DPRINTF("entered\n"); 1191 1192 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1193 softc = be_lun->softc; 1194 lbalen = ARGS(beio->io); 1195 1196 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1197 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1198 ctl_free_beio(beio); 1199 ctl_set_invalid_field(&io->scsiio, 1200 /*sks_valid*/ 1, 1201 /*command*/ 1, 1202 /*field*/ 1, 1203 /*bit_valid*/ 0, 1204 /*bit*/ 0); 1205 ctl_config_write_done(io); 1206 return; 1207 } 1208 1209 switch (io->scsiio.tag_type) { 1210 case CTL_TAG_ORDERED: 1211 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1212 break; 1213 case CTL_TAG_HEAD_OF_QUEUE: 1214 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1215 break; 1216 case CTL_TAG_UNTAGGED: 1217 case CTL_TAG_SIMPLE: 1218 case CTL_TAG_ACA: 1219 default: 1220 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1221 break; 1222 } 1223 1224 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1225 beio->io_offset = lbalen->lba * be_lun->blocksize; 1226 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1227 beio->bio_cmd = BIO_DELETE; 1228 beio->ds_trans_type = DEVSTAT_FREE; 1229 1230 be_lun->unmap(be_lun, beio); 1231 return; 1232 } 1233 1234 beio->bio_cmd = BIO_WRITE; 1235 beio->ds_trans_type = DEVSTAT_WRITE; 1236 1237 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1238 (uintmax_t)lbalen->lba, lbalen->len); 1239 1240 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1241 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1242 1243 /* 1244 * Setup the S/G entry for this chunk. 1245 */ 1246 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1247 seglen -= seglen % be_lun->blocksize; 1248 beio->sg_segs[i].len = seglen; 1249 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1250 1251 DPRINTF("segment %d addr %p len %zd\n", i, 1252 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1253 1254 beio->num_segs++; 1255 len_left -= seglen; 1256 1257 buf = beio->sg_segs[i].addr; 1258 end = buf + seglen; 1259 for (; buf < end; buf += be_lun->blocksize) { 1260 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1261 if (lbalen->flags & SWS_LBDATA) 1262 scsi_ulto4b(lbalen->lba + lba, buf); 1263 lba++; 1264 } 1265 } 1266 1267 beio->io_offset = lbalen->lba * be_lun->blocksize; 1268 beio->io_len = lba * be_lun->blocksize; 1269 1270 /* We can not do all in one run. Correct and schedule rerun. */ 1271 if (len_left > 0) { 1272 lbalen->lba += lba; 1273 lbalen->len -= lba; 1274 beio->beio_cont = ctl_be_block_cw_done_ws; 1275 } 1276 1277 be_lun->dispatch(be_lun, beio); 1278 } 1279 1280 static void 1281 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1282 union ctl_io *io) 1283 { 1284 struct ctl_be_block_io *beio; 1285 struct ctl_be_block_softc *softc; 1286 struct ctl_ptr_len_flags *ptrlen; 1287 1288 DPRINTF("entered\n"); 1289 1290 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1291 softc = be_lun->softc; 1292 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1293 1294 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1295 ctl_free_beio(beio); 1296 ctl_set_invalid_field(&io->scsiio, 1297 /*sks_valid*/ 0, 1298 /*command*/ 1, 1299 /*field*/ 0, 1300 /*bit_valid*/ 0, 1301 /*bit*/ 0); 1302 ctl_config_write_done(io); 1303 return; 1304 } 1305 1306 switch (io->scsiio.tag_type) { 1307 case CTL_TAG_ORDERED: 1308 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1309 break; 1310 case CTL_TAG_HEAD_OF_QUEUE: 1311 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1312 break; 1313 case CTL_TAG_UNTAGGED: 1314 case CTL_TAG_SIMPLE: 1315 case CTL_TAG_ACA: 1316 default: 1317 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1318 break; 1319 } 1320 1321 beio->io_len = 0; 1322 beio->io_offset = -1; 1323 1324 beio->bio_cmd = BIO_DELETE; 1325 beio->ds_trans_type = DEVSTAT_FREE; 1326 1327 DPRINTF("UNMAP\n"); 1328 1329 be_lun->unmap(be_lun, beio); 1330 } 1331 1332 static void 1333 ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1334 { 1335 union ctl_io *io; 1336 1337 io = beio->io; 1338 ctl_free_beio(beio); 1339 ctl_config_read_done(io); 1340 } 1341 1342 static void 1343 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1344 union ctl_io *io) 1345 { 1346 struct ctl_be_block_io *beio; 1347 struct ctl_be_block_softc *softc; 1348 1349 DPRINTF("entered\n"); 1350 1351 softc = be_lun->softc; 1352 beio = ctl_alloc_beio(softc); 1353 beio->io = io; 1354 beio->lun = be_lun; 1355 beio->beio_cont = ctl_be_block_cr_done; 1356 PRIV(io)->ptr = (void *)beio; 1357 1358 switch (io->scsiio.cdb[0]) { 1359 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1360 beio->bio_cmd = -1; 1361 beio->ds_trans_type = DEVSTAT_NO_DATA; 1362 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1363 beio->io_len = 0; 1364 if (be_lun->get_lba_status) 1365 be_lun->get_lba_status(be_lun, beio); 1366 else 1367 ctl_be_block_cr_done(beio); 1368 break; 1369 default: 1370 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1371 break; 1372 } 1373 } 1374 1375 static void 1376 ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1377 { 1378 union ctl_io *io; 1379 1380 io = beio->io; 1381 ctl_free_beio(beio); 1382 ctl_config_write_done(io); 1383 } 1384 1385 static void 1386 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1387 union ctl_io *io) 1388 { 1389 struct ctl_be_block_io *beio; 1390 struct ctl_be_block_softc *softc; 1391 1392 DPRINTF("entered\n"); 1393 1394 softc = be_lun->softc; 1395 beio = ctl_alloc_beio(softc); 1396 beio->io = io; 1397 beio->lun = be_lun; 1398 beio->beio_cont = ctl_be_block_cw_done; 1399 PRIV(io)->ptr = (void *)beio; 1400 1401 switch (io->scsiio.cdb[0]) { 1402 case SYNCHRONIZE_CACHE: 1403 case SYNCHRONIZE_CACHE_16: 1404 beio->bio_cmd = BIO_FLUSH; 1405 beio->ds_trans_type = DEVSTAT_NO_DATA; 1406 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1407 beio->io_len = 0; 1408 be_lun->lun_flush(be_lun, beio); 1409 break; 1410 case WRITE_SAME_10: 1411 case WRITE_SAME_16: 1412 ctl_be_block_cw_dispatch_ws(be_lun, io); 1413 break; 1414 case UNMAP: 1415 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1416 break; 1417 default: 1418 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1419 break; 1420 } 1421 } 1422 1423 SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1424 SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1425 SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1426 SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1427 1428 static void 1429 ctl_be_block_next(struct ctl_be_block_io *beio) 1430 { 1431 struct ctl_be_block_lun *be_lun; 1432 union ctl_io *io; 1433 1434 io = beio->io; 1435 be_lun = beio->lun; 1436 ctl_free_beio(beio); 1437 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1438 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1439 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1440 ctl_data_submit_done(io); 1441 return; 1442 } 1443 1444 io->io_hdr.status &= ~CTL_STATUS_MASK; 1445 io->io_hdr.status |= CTL_STATUS_NONE; 1446 1447 mtx_lock(&be_lun->queue_lock); 1448 /* 1449 * XXX KDM make sure that links is okay to use at this point. 1450 * Otherwise, we either need to add another field to ctl_io_hdr, 1451 * or deal with resource allocation here. 1452 */ 1453 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1454 mtx_unlock(&be_lun->queue_lock); 1455 1456 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1457 } 1458 1459 static void 1460 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1461 union ctl_io *io) 1462 { 1463 struct ctl_be_block_io *beio; 1464 struct ctl_be_block_softc *softc; 1465 struct ctl_lba_len_flags *lbalen; 1466 struct ctl_ptr_len_flags *bptrlen; 1467 uint64_t len_left, lbas; 1468 int i; 1469 1470 softc = be_lun->softc; 1471 1472 DPRINTF("entered\n"); 1473 1474 lbalen = ARGS(io); 1475 if (lbalen->flags & CTL_LLF_WRITE) { 1476 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1477 } else { 1478 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1479 } 1480 1481 beio = ctl_alloc_beio(softc); 1482 beio->io = io; 1483 beio->lun = be_lun; 1484 bptrlen = PRIV(io); 1485 bptrlen->ptr = (void *)beio; 1486 1487 switch (io->scsiio.tag_type) { 1488 case CTL_TAG_ORDERED: 1489 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1490 break; 1491 case CTL_TAG_HEAD_OF_QUEUE: 1492 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1493 break; 1494 case CTL_TAG_UNTAGGED: 1495 case CTL_TAG_SIMPLE: 1496 case CTL_TAG_ACA: 1497 default: 1498 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1499 break; 1500 } 1501 1502 if (lbalen->flags & CTL_LLF_WRITE) { 1503 beio->bio_cmd = BIO_WRITE; 1504 beio->ds_trans_type = DEVSTAT_WRITE; 1505 } else { 1506 beio->bio_cmd = BIO_READ; 1507 beio->ds_trans_type = DEVSTAT_READ; 1508 } 1509 1510 DPRINTF("%s at LBA %jx len %u @%ju\n", 1511 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1512 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1513 if (lbalen->flags & CTL_LLF_COMPARE) 1514 lbas = CTLBLK_HALF_IO_SIZE; 1515 else 1516 lbas = CTLBLK_MAX_IO_SIZE; 1517 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1518 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1519 beio->io_len = lbas * be_lun->blocksize; 1520 bptrlen->len += lbas; 1521 1522 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1523 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1524 i, CTLBLK_MAX_SEGS)); 1525 1526 /* 1527 * Setup the S/G entry for this chunk. 1528 */ 1529 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1530 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1531 1532 DPRINTF("segment %d addr %p len %zd\n", i, 1533 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1534 1535 /* Set up second segment for compare operation. */ 1536 if (lbalen->flags & CTL_LLF_COMPARE) { 1537 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1538 beio->sg_segs[i].len; 1539 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1540 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1541 } 1542 1543 beio->num_segs++; 1544 len_left -= beio->sg_segs[i].len; 1545 } 1546 if (bptrlen->len < lbalen->len) 1547 beio->beio_cont = ctl_be_block_next; 1548 io->scsiio.be_move_done = ctl_be_block_move_done; 1549 /* For compare we have separate S/G lists for read and datamove. */ 1550 if (lbalen->flags & CTL_LLF_COMPARE) 1551 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1552 else 1553 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1554 io->scsiio.kern_data_len = beio->io_len; 1555 io->scsiio.kern_data_resid = 0; 1556 io->scsiio.kern_sg_entries = beio->num_segs; 1557 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1558 1559 /* 1560 * For the read case, we need to read the data into our buffers and 1561 * then we can send it back to the user. For the write case, we 1562 * need to get the data from the user first. 1563 */ 1564 if (beio->bio_cmd == BIO_READ) { 1565 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1566 be_lun->dispatch(be_lun, beio); 1567 } else { 1568 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1569 #ifdef CTL_TIME_IO 1570 getbintime(&io->io_hdr.dma_start_bt); 1571 #endif 1572 ctl_datamove(io); 1573 } 1574 } 1575 1576 static void 1577 ctl_be_block_worker(void *context, int pending) 1578 { 1579 struct ctl_be_block_lun *be_lun; 1580 struct ctl_be_block_softc *softc; 1581 union ctl_io *io; 1582 1583 be_lun = (struct ctl_be_block_lun *)context; 1584 softc = be_lun->softc; 1585 1586 DPRINTF("entered\n"); 1587 1588 mtx_lock(&be_lun->queue_lock); 1589 for (;;) { 1590 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1591 if (io != NULL) { 1592 struct ctl_be_block_io *beio; 1593 1594 DPRINTF("datamove queue\n"); 1595 1596 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1597 ctl_io_hdr, links); 1598 1599 mtx_unlock(&be_lun->queue_lock); 1600 1601 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1602 1603 be_lun->dispatch(be_lun, beio); 1604 1605 mtx_lock(&be_lun->queue_lock); 1606 continue; 1607 } 1608 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1609 if (io != NULL) { 1610 DPRINTF("config write queue\n"); 1611 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1612 ctl_io_hdr, links); 1613 mtx_unlock(&be_lun->queue_lock); 1614 ctl_be_block_cw_dispatch(be_lun, io); 1615 mtx_lock(&be_lun->queue_lock); 1616 continue; 1617 } 1618 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1619 if (io != NULL) { 1620 DPRINTF("config read queue\n"); 1621 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1622 ctl_io_hdr, links); 1623 mtx_unlock(&be_lun->queue_lock); 1624 ctl_be_block_cr_dispatch(be_lun, io); 1625 mtx_lock(&be_lun->queue_lock); 1626 continue; 1627 } 1628 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1629 if (io != NULL) { 1630 DPRINTF("input queue\n"); 1631 1632 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1633 ctl_io_hdr, links); 1634 mtx_unlock(&be_lun->queue_lock); 1635 1636 /* 1637 * We must drop the lock, since this routine and 1638 * its children may sleep. 1639 */ 1640 ctl_be_block_dispatch(be_lun, io); 1641 1642 mtx_lock(&be_lun->queue_lock); 1643 continue; 1644 } 1645 1646 /* 1647 * If we get here, there is no work left in the queues, so 1648 * just break out and let the task queue go to sleep. 1649 */ 1650 break; 1651 } 1652 mtx_unlock(&be_lun->queue_lock); 1653 } 1654 1655 /* 1656 * Entry point from CTL to the backend for I/O. We queue everything to a 1657 * work thread, so this just puts the I/O on a queue and wakes up the 1658 * thread. 1659 */ 1660 static int 1661 ctl_be_block_submit(union ctl_io *io) 1662 { 1663 struct ctl_be_block_lun *be_lun; 1664 struct ctl_be_lun *ctl_be_lun; 1665 1666 DPRINTF("entered\n"); 1667 1668 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1669 CTL_PRIV_BACKEND_LUN].ptr; 1670 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1671 1672 /* 1673 * Make sure we only get SCSI I/O. 1674 */ 1675 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1676 "%#x) encountered", io->io_hdr.io_type)); 1677 1678 PRIV(io)->len = 0; 1679 1680 mtx_lock(&be_lun->queue_lock); 1681 /* 1682 * XXX KDM make sure that links is okay to use at this point. 1683 * Otherwise, we either need to add another field to ctl_io_hdr, 1684 * or deal with resource allocation here. 1685 */ 1686 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1687 mtx_unlock(&be_lun->queue_lock); 1688 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1689 1690 return (CTL_RETVAL_COMPLETE); 1691 } 1692 1693 static int 1694 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1695 int flag, struct thread *td) 1696 { 1697 struct ctl_be_block_softc *softc; 1698 int error; 1699 1700 softc = &backend_block_softc; 1701 1702 error = 0; 1703 1704 switch (cmd) { 1705 case CTL_LUN_REQ: { 1706 struct ctl_lun_req *lun_req; 1707 1708 lun_req = (struct ctl_lun_req *)addr; 1709 1710 switch (lun_req->reqtype) { 1711 case CTL_LUNREQ_CREATE: 1712 error = ctl_be_block_create(softc, lun_req); 1713 break; 1714 case CTL_LUNREQ_RM: 1715 error = ctl_be_block_rm(softc, lun_req); 1716 break; 1717 case CTL_LUNREQ_MODIFY: 1718 error = ctl_be_block_modify(softc, lun_req); 1719 break; 1720 default: 1721 lun_req->status = CTL_LUN_ERROR; 1722 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1723 "invalid LUN request type %d", 1724 lun_req->reqtype); 1725 break; 1726 } 1727 break; 1728 } 1729 default: 1730 error = ENOTTY; 1731 break; 1732 } 1733 1734 return (error); 1735 } 1736 1737 static int 1738 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1739 { 1740 struct ctl_be_block_filedata *file_data; 1741 struct ctl_lun_create_params *params; 1742 struct vattr vattr; 1743 off_t pss; 1744 int error; 1745 1746 error = 0; 1747 file_data = &be_lun->backend.file; 1748 params = &be_lun->params; 1749 1750 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1751 be_lun->dispatch = ctl_be_block_dispatch_file; 1752 be_lun->lun_flush = ctl_be_block_flush_file; 1753 be_lun->get_lba_status = ctl_be_block_gls_file; 1754 be_lun->getattr = ctl_be_block_getattr_file; 1755 1756 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1757 if (error != 0) { 1758 snprintf(req->error_str, sizeof(req->error_str), 1759 "error calling VOP_GETATTR() for file %s", 1760 be_lun->dev_path); 1761 return (error); 1762 } 1763 1764 /* 1765 * Verify that we have the ability to upgrade to exclusive 1766 * access on this file so we can trap errors at open instead 1767 * of reporting them during first access. 1768 */ 1769 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1770 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1771 if (be_lun->vn->v_iflag & VI_DOOMED) { 1772 error = EBADF; 1773 snprintf(req->error_str, sizeof(req->error_str), 1774 "error locking file %s", be_lun->dev_path); 1775 return (error); 1776 } 1777 } 1778 1779 1780 file_data->cred = crhold(curthread->td_ucred); 1781 if (params->lun_size_bytes != 0) 1782 be_lun->size_bytes = params->lun_size_bytes; 1783 else 1784 be_lun->size_bytes = vattr.va_size; 1785 /* 1786 * We set the multi thread flag for file operations because all 1787 * filesystems (in theory) are capable of allowing multiple readers 1788 * of a file at once. So we want to get the maximum possible 1789 * concurrency. 1790 */ 1791 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1792 1793 /* 1794 * For files we can use any logical block size. Prefer 512 bytes 1795 * for compatibility reasons. If file's vattr.va_blocksize 1796 * (preferred I/O block size) is bigger and multiple to chosen 1797 * logical block size -- report it as physical block size. 1798 */ 1799 if (params->blocksize_bytes != 0) 1800 be_lun->blocksize = params->blocksize_bytes; 1801 else 1802 be_lun->blocksize = 512; 1803 pss = vattr.va_blocksize / be_lun->blocksize; 1804 if ((pss > 0) && (pss * be_lun->blocksize == vattr.va_blocksize) && 1805 ((pss & (pss - 1)) == 0)) { 1806 be_lun->pblockexp = fls(pss) - 1; 1807 be_lun->pblockoff = 0; 1808 } 1809 1810 /* 1811 * Sanity check. The media size has to be at least one 1812 * sector long. 1813 */ 1814 if (be_lun->size_bytes < be_lun->blocksize) { 1815 error = EINVAL; 1816 snprintf(req->error_str, sizeof(req->error_str), 1817 "file %s size %ju < block size %u", be_lun->dev_path, 1818 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1819 } 1820 return (error); 1821 } 1822 1823 static int 1824 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1825 { 1826 struct ctl_lun_create_params *params; 1827 struct vattr vattr; 1828 struct cdev *dev; 1829 struct cdevsw *devsw; 1830 int error; 1831 off_t ps, pss, po, pos; 1832 1833 params = &be_lun->params; 1834 1835 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1836 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1837 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1838 &be_lun->backend.dev.dev_ref); 1839 if (be_lun->backend.dev.csw == NULL) 1840 panic("Unable to retrieve device switch"); 1841 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) { 1842 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1843 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1844 } else 1845 be_lun->dispatch = ctl_be_block_dispatch_dev; 1846 be_lun->lun_flush = ctl_be_block_flush_dev; 1847 be_lun->unmap = ctl_be_block_unmap_dev; 1848 be_lun->getattr = ctl_be_block_getattr_dev; 1849 1850 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1851 if (error) { 1852 snprintf(req->error_str, sizeof(req->error_str), 1853 "error getting vnode attributes for device %s", 1854 be_lun->dev_path); 1855 return (error); 1856 } 1857 1858 dev = be_lun->vn->v_rdev; 1859 devsw = dev->si_devsw; 1860 if (!devsw->d_ioctl) { 1861 snprintf(req->error_str, sizeof(req->error_str), 1862 "no d_ioctl for device %s!", 1863 be_lun->dev_path); 1864 return (ENODEV); 1865 } 1866 1867 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1868 (caddr_t)&be_lun->blocksize, FREAD, 1869 curthread); 1870 if (error) { 1871 snprintf(req->error_str, sizeof(req->error_str), 1872 "error %d returned for DIOCGSECTORSIZE ioctl " 1873 "on %s!", error, be_lun->dev_path); 1874 return (error); 1875 } 1876 1877 /* 1878 * If the user has asked for a blocksize that is greater than the 1879 * backing device's blocksize, we can do it only if the blocksize 1880 * the user is asking for is an even multiple of the underlying 1881 * device's blocksize. 1882 */ 1883 if ((params->blocksize_bytes != 0) 1884 && (params->blocksize_bytes > be_lun->blocksize)) { 1885 uint32_t bs_multiple, tmp_blocksize; 1886 1887 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1888 1889 tmp_blocksize = bs_multiple * be_lun->blocksize; 1890 1891 if (tmp_blocksize == params->blocksize_bytes) { 1892 be_lun->blocksize = params->blocksize_bytes; 1893 } else { 1894 snprintf(req->error_str, sizeof(req->error_str), 1895 "requested blocksize %u is not an even " 1896 "multiple of backing device blocksize %u", 1897 params->blocksize_bytes, 1898 be_lun->blocksize); 1899 return (EINVAL); 1900 1901 } 1902 } else if ((params->blocksize_bytes != 0) 1903 && (params->blocksize_bytes != be_lun->blocksize)) { 1904 snprintf(req->error_str, sizeof(req->error_str), 1905 "requested blocksize %u < backing device " 1906 "blocksize %u", params->blocksize_bytes, 1907 be_lun->blocksize); 1908 return (EINVAL); 1909 } 1910 1911 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1912 (caddr_t)&be_lun->size_bytes, FREAD, 1913 curthread); 1914 if (error) { 1915 snprintf(req->error_str, sizeof(req->error_str), 1916 "error %d returned for DIOCGMEDIASIZE " 1917 " ioctl on %s!", error, 1918 be_lun->dev_path); 1919 return (error); 1920 } 1921 1922 if (params->lun_size_bytes != 0) { 1923 if (params->lun_size_bytes > be_lun->size_bytes) { 1924 snprintf(req->error_str, sizeof(req->error_str), 1925 "requested LUN size %ju > backing device " 1926 "size %ju", 1927 (uintmax_t)params->lun_size_bytes, 1928 (uintmax_t)be_lun->size_bytes); 1929 return (EINVAL); 1930 } 1931 1932 be_lun->size_bytes = params->lun_size_bytes; 1933 } 1934 1935 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1936 (caddr_t)&ps, FREAD, curthread); 1937 if (error) 1938 ps = po = 0; 1939 else { 1940 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1941 (caddr_t)&po, FREAD, curthread); 1942 if (error) 1943 po = 0; 1944 } 1945 pss = ps / be_lun->blocksize; 1946 pos = po / be_lun->blocksize; 1947 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1948 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1949 be_lun->pblockexp = fls(pss) - 1; 1950 be_lun->pblockoff = (pss - pos) % pss; 1951 } 1952 1953 return (0); 1954 } 1955 1956 static int 1957 ctl_be_block_close(struct ctl_be_block_lun *be_lun) 1958 { 1959 DROP_GIANT(); 1960 if (be_lun->vn) { 1961 int flags = FREAD | FWRITE; 1962 1963 switch (be_lun->dev_type) { 1964 case CTL_BE_BLOCK_DEV: 1965 if (be_lun->backend.dev.csw) { 1966 dev_relthread(be_lun->backend.dev.cdev, 1967 be_lun->backend.dev.dev_ref); 1968 be_lun->backend.dev.csw = NULL; 1969 be_lun->backend.dev.cdev = NULL; 1970 } 1971 break; 1972 case CTL_BE_BLOCK_FILE: 1973 break; 1974 case CTL_BE_BLOCK_NONE: 1975 break; 1976 default: 1977 panic("Unexpected backend type."); 1978 break; 1979 } 1980 1981 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 1982 be_lun->vn = NULL; 1983 1984 switch (be_lun->dev_type) { 1985 case CTL_BE_BLOCK_DEV: 1986 break; 1987 case CTL_BE_BLOCK_FILE: 1988 if (be_lun->backend.file.cred != NULL) { 1989 crfree(be_lun->backend.file.cred); 1990 be_lun->backend.file.cred = NULL; 1991 } 1992 break; 1993 case CTL_BE_BLOCK_NONE: 1994 break; 1995 default: 1996 panic("Unexpected backend type."); 1997 break; 1998 } 1999 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2000 } 2001 PICKUP_GIANT(); 2002 2003 return (0); 2004 } 2005 2006 static int 2007 ctl_be_block_open(struct ctl_be_block_softc *softc, 2008 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2009 { 2010 struct nameidata nd; 2011 int flags; 2012 int error; 2013 2014 /* 2015 * XXX KDM allow a read-only option? 2016 */ 2017 flags = FREAD | FWRITE; 2018 error = 0; 2019 2020 if (rootvnode == NULL) { 2021 snprintf(req->error_str, sizeof(req->error_str), 2022 "Root filesystem is not mounted"); 2023 return (1); 2024 } 2025 2026 if (!curthread->td_proc->p_fd->fd_cdir) { 2027 curthread->td_proc->p_fd->fd_cdir = rootvnode; 2028 VREF(rootvnode); 2029 } 2030 if (!curthread->td_proc->p_fd->fd_rdir) { 2031 curthread->td_proc->p_fd->fd_rdir = rootvnode; 2032 VREF(rootvnode); 2033 } 2034 if (!curthread->td_proc->p_fd->fd_jdir) { 2035 curthread->td_proc->p_fd->fd_jdir = rootvnode; 2036 VREF(rootvnode); 2037 } 2038 2039 again: 2040 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2041 error = vn_open(&nd, &flags, 0, NULL); 2042 if (error) { 2043 /* 2044 * This is the only reasonable guess we can make as far as 2045 * path if the user doesn't give us a fully qualified path. 2046 * If they want to specify a file, they need to specify the 2047 * full path. 2048 */ 2049 if (be_lun->dev_path[0] != '/') { 2050 char *dev_path = "/dev/"; 2051 char *dev_name; 2052 2053 /* Try adding device path at beginning of name */ 2054 dev_name = malloc(strlen(be_lun->dev_path) 2055 + strlen(dev_path) + 1, 2056 M_CTLBLK, M_WAITOK); 2057 if (dev_name) { 2058 sprintf(dev_name, "%s%s", dev_path, 2059 be_lun->dev_path); 2060 free(be_lun->dev_path, M_CTLBLK); 2061 be_lun->dev_path = dev_name; 2062 goto again; 2063 } 2064 } 2065 snprintf(req->error_str, sizeof(req->error_str), 2066 "error opening %s: %d", be_lun->dev_path, error); 2067 return (error); 2068 } 2069 2070 NDFREE(&nd, NDF_ONLY_PNBUF); 2071 2072 be_lun->vn = nd.ni_vp; 2073 2074 /* We only support disks and files. */ 2075 if (vn_isdisk(be_lun->vn, &error)) { 2076 error = ctl_be_block_open_dev(be_lun, req); 2077 } else if (be_lun->vn->v_type == VREG) { 2078 error = ctl_be_block_open_file(be_lun, req); 2079 } else { 2080 error = EINVAL; 2081 snprintf(req->error_str, sizeof(req->error_str), 2082 "%s is not a disk or plain file", be_lun->dev_path); 2083 } 2084 VOP_UNLOCK(be_lun->vn, 0); 2085 2086 if (error != 0) { 2087 ctl_be_block_close(be_lun); 2088 return (error); 2089 } 2090 2091 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2092 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2093 2094 return (0); 2095 } 2096 2097 static int 2098 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2099 { 2100 struct ctl_be_block_lun *be_lun; 2101 struct ctl_lun_create_params *params; 2102 char num_thread_str[16]; 2103 char tmpstr[32]; 2104 char *value; 2105 int retval, num_threads, unmap; 2106 int tmp_num_threads; 2107 2108 params = &req->reqdata.create; 2109 retval = 0; 2110 req->status = CTL_LUN_OK; 2111 2112 num_threads = cbb_num_threads; 2113 2114 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2115 2116 be_lun->params = req->reqdata.create; 2117 be_lun->softc = softc; 2118 STAILQ_INIT(&be_lun->input_queue); 2119 STAILQ_INIT(&be_lun->config_read_queue); 2120 STAILQ_INIT(&be_lun->config_write_queue); 2121 STAILQ_INIT(&be_lun->datamove_queue); 2122 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2123 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2124 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2125 ctl_init_opts(&be_lun->ctl_be_lun.options, 2126 req->num_be_args, req->kern_be_args); 2127 2128 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2129 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2130 2131 if (be_lun->lun_zone == NULL) { 2132 snprintf(req->error_str, sizeof(req->error_str), 2133 "error allocating UMA zone"); 2134 goto bailout_error; 2135 } 2136 2137 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2138 be_lun->ctl_be_lun.lun_type = params->device_type; 2139 else 2140 be_lun->ctl_be_lun.lun_type = T_DIRECT; 2141 2142 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 2143 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 2144 if (value == NULL) { 2145 snprintf(req->error_str, sizeof(req->error_str), 2146 "no file argument specified"); 2147 goto bailout_error; 2148 } 2149 be_lun->dev_path = strdup(value, M_CTLBLK); 2150 be_lun->blocksize = 512; 2151 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2152 2153 retval = ctl_be_block_open(softc, be_lun, req); 2154 if (retval != 0) { 2155 retval = 0; 2156 req->status = CTL_LUN_WARNING; 2157 } 2158 } else { 2159 /* 2160 * For processor devices, we don't have any size. 2161 */ 2162 be_lun->blocksize = 0; 2163 be_lun->pblockexp = 0; 2164 be_lun->pblockoff = 0; 2165 be_lun->size_blocks = 0; 2166 be_lun->size_bytes = 0; 2167 be_lun->ctl_be_lun.maxlba = 0; 2168 2169 /* 2170 * Default to just 1 thread for processor devices. 2171 */ 2172 num_threads = 1; 2173 } 2174 2175 /* 2176 * XXX This searching loop might be refactored to be combined with 2177 * the loop above, 2178 */ 2179 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2180 if (value != NULL) { 2181 tmp_num_threads = strtol(value, NULL, 0); 2182 2183 /* 2184 * We don't let the user specify less than one 2185 * thread, but hope he's clueful enough not to 2186 * specify 1000 threads. 2187 */ 2188 if (tmp_num_threads < 1) { 2189 snprintf(req->error_str, sizeof(req->error_str), 2190 "invalid number of threads %s", 2191 num_thread_str); 2192 goto bailout_error; 2193 } 2194 num_threads = tmp_num_threads; 2195 } 2196 unmap = (be_lun->dispatch == ctl_be_block_dispatch_zvol); 2197 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2198 if (value != NULL) 2199 unmap = (strcmp(value, "on") == 0); 2200 2201 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2202 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2203 if (be_lun->vn == NULL) 2204 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; 2205 if (unmap) 2206 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2207 be_lun->ctl_be_lun.be_lun = be_lun; 2208 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2209 0 : (be_lun->size_blocks - 1); 2210 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2211 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2212 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2213 if (be_lun->dispatch == ctl_be_block_dispatch_zvol && 2214 be_lun->blocksize != 0) 2215 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE / 2216 be_lun->blocksize; 2217 /* Tell the user the blocksize we ended up using */ 2218 params->lun_size_bytes = be_lun->size_bytes; 2219 params->blocksize_bytes = be_lun->blocksize; 2220 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2221 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2222 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2223 } else 2224 be_lun->ctl_be_lun.req_lun_id = 0; 2225 2226 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2227 be_lun->ctl_be_lun.lun_config_status = 2228 ctl_be_block_lun_config_status; 2229 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2230 2231 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2232 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2233 softc->num_luns); 2234 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2235 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2236 sizeof(tmpstr))); 2237 2238 /* Tell the user what we used for a serial number */ 2239 strncpy((char *)params->serial_num, tmpstr, 2240 ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); 2241 } else { 2242 strncpy((char *)be_lun->ctl_be_lun.serial_num, 2243 params->serial_num, 2244 ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), 2245 sizeof(params->serial_num))); 2246 } 2247 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2248 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2249 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2250 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2251 sizeof(tmpstr))); 2252 2253 /* Tell the user what we used for a device ID */ 2254 strncpy((char *)params->device_id, tmpstr, 2255 ctl_min(sizeof(params->device_id), sizeof(tmpstr))); 2256 } else { 2257 strncpy((char *)be_lun->ctl_be_lun.device_id, 2258 params->device_id, 2259 ctl_min(sizeof(be_lun->ctl_be_lun.device_id), 2260 sizeof(params->device_id))); 2261 } 2262 2263 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2264 2265 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2266 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2267 2268 if (be_lun->io_taskqueue == NULL) { 2269 snprintf(req->error_str, sizeof(req->error_str), 2270 "unable to create taskqueue"); 2271 goto bailout_error; 2272 } 2273 2274 /* 2275 * Note that we start the same number of threads by default for 2276 * both the file case and the block device case. For the file 2277 * case, we need multiple threads to allow concurrency, because the 2278 * vnode interface is designed to be a blocking interface. For the 2279 * block device case, ZFS zvols at least will block the caller's 2280 * context in many instances, and so we need multiple threads to 2281 * overcome that problem. Other block devices don't need as many 2282 * threads, but they shouldn't cause too many problems. 2283 * 2284 * If the user wants to just have a single thread for a block 2285 * device, he can specify that when the LUN is created, or change 2286 * the tunable/sysctl to alter the default number of threads. 2287 */ 2288 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2289 /*num threads*/num_threads, 2290 /*priority*/PWAIT, 2291 /*thread name*/ 2292 "%s taskq", be_lun->lunname); 2293 2294 if (retval != 0) 2295 goto bailout_error; 2296 2297 be_lun->num_threads = num_threads; 2298 2299 mtx_lock(&softc->lock); 2300 softc->num_luns++; 2301 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2302 2303 mtx_unlock(&softc->lock); 2304 2305 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2306 if (retval != 0) { 2307 mtx_lock(&softc->lock); 2308 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2309 links); 2310 softc->num_luns--; 2311 mtx_unlock(&softc->lock); 2312 snprintf(req->error_str, sizeof(req->error_str), 2313 "ctl_add_lun() returned error %d, see dmesg for " 2314 "details", retval); 2315 retval = 0; 2316 goto bailout_error; 2317 } 2318 2319 mtx_lock(&softc->lock); 2320 2321 /* 2322 * Tell the config_status routine that we're waiting so it won't 2323 * clean up the LUN in the event of an error. 2324 */ 2325 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2326 2327 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2328 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2329 if (retval == EINTR) 2330 break; 2331 } 2332 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2333 2334 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2335 snprintf(req->error_str, sizeof(req->error_str), 2336 "LUN configuration error, see dmesg for details"); 2337 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2338 links); 2339 softc->num_luns--; 2340 mtx_unlock(&softc->lock); 2341 goto bailout_error; 2342 } else { 2343 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2344 } 2345 2346 mtx_unlock(&softc->lock); 2347 2348 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2349 be_lun->blocksize, 2350 DEVSTAT_ALL_SUPPORTED, 2351 be_lun->ctl_be_lun.lun_type 2352 | DEVSTAT_TYPE_IF_OTHER, 2353 DEVSTAT_PRIORITY_OTHER); 2354 2355 return (retval); 2356 2357 bailout_error: 2358 req->status = CTL_LUN_ERROR; 2359 2360 if (be_lun->io_taskqueue != NULL) 2361 taskqueue_free(be_lun->io_taskqueue); 2362 ctl_be_block_close(be_lun); 2363 if (be_lun->dev_path != NULL) 2364 free(be_lun->dev_path, M_CTLBLK); 2365 if (be_lun->lun_zone != NULL) 2366 uma_zdestroy(be_lun->lun_zone); 2367 ctl_free_opts(&be_lun->ctl_be_lun.options); 2368 mtx_destroy(&be_lun->queue_lock); 2369 mtx_destroy(&be_lun->io_lock); 2370 free(be_lun, M_CTLBLK); 2371 2372 return (retval); 2373 } 2374 2375 static int 2376 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2377 { 2378 struct ctl_lun_rm_params *params; 2379 struct ctl_be_block_lun *be_lun; 2380 int retval; 2381 2382 params = &req->reqdata.rm; 2383 2384 mtx_lock(&softc->lock); 2385 2386 be_lun = NULL; 2387 2388 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2389 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2390 break; 2391 } 2392 mtx_unlock(&softc->lock); 2393 2394 if (be_lun == NULL) { 2395 snprintf(req->error_str, sizeof(req->error_str), 2396 "LUN %u is not managed by the block backend", 2397 params->lun_id); 2398 goto bailout_error; 2399 } 2400 2401 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2402 2403 if (retval != 0) { 2404 snprintf(req->error_str, sizeof(req->error_str), 2405 "error %d returned from ctl_disable_lun() for " 2406 "LUN %d", retval, params->lun_id); 2407 goto bailout_error; 2408 2409 } 2410 2411 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2412 if (retval != 0) { 2413 snprintf(req->error_str, sizeof(req->error_str), 2414 "error %d returned from ctl_invalidate_lun() for " 2415 "LUN %d", retval, params->lun_id); 2416 goto bailout_error; 2417 } 2418 2419 mtx_lock(&softc->lock); 2420 2421 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2422 2423 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2424 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2425 if (retval == EINTR) 2426 break; 2427 } 2428 2429 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2430 2431 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2432 snprintf(req->error_str, sizeof(req->error_str), 2433 "interrupted waiting for LUN to be freed"); 2434 mtx_unlock(&softc->lock); 2435 goto bailout_error; 2436 } 2437 2438 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2439 2440 softc->num_luns--; 2441 mtx_unlock(&softc->lock); 2442 2443 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2444 2445 taskqueue_free(be_lun->io_taskqueue); 2446 2447 ctl_be_block_close(be_lun); 2448 2449 if (be_lun->disk_stats != NULL) 2450 devstat_remove_entry(be_lun->disk_stats); 2451 2452 uma_zdestroy(be_lun->lun_zone); 2453 2454 ctl_free_opts(&be_lun->ctl_be_lun.options); 2455 free(be_lun->dev_path, M_CTLBLK); 2456 mtx_destroy(&be_lun->queue_lock); 2457 mtx_destroy(&be_lun->io_lock); 2458 free(be_lun, M_CTLBLK); 2459 2460 req->status = CTL_LUN_OK; 2461 2462 return (0); 2463 2464 bailout_error: 2465 2466 req->status = CTL_LUN_ERROR; 2467 2468 return (0); 2469 } 2470 2471 static int 2472 ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2473 struct ctl_lun_req *req) 2474 { 2475 struct vattr vattr; 2476 int error; 2477 struct ctl_lun_create_params *params = &be_lun->params; 2478 2479 if (params->lun_size_bytes != 0) { 2480 be_lun->size_bytes = params->lun_size_bytes; 2481 } else { 2482 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2483 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2484 VOP_UNLOCK(be_lun->vn, 0); 2485 if (error != 0) { 2486 snprintf(req->error_str, sizeof(req->error_str), 2487 "error calling VOP_GETATTR() for file %s", 2488 be_lun->dev_path); 2489 return (error); 2490 } 2491 2492 be_lun->size_bytes = vattr.va_size; 2493 } 2494 2495 return (0); 2496 } 2497 2498 static int 2499 ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2500 struct ctl_lun_req *req) 2501 { 2502 struct ctl_be_block_devdata *dev_data; 2503 int error; 2504 struct ctl_lun_create_params *params = &be_lun->params; 2505 uint64_t size_bytes; 2506 2507 dev_data = &be_lun->backend.dev; 2508 if (!dev_data->csw->d_ioctl) { 2509 snprintf(req->error_str, sizeof(req->error_str), 2510 "no d_ioctl for device %s!", be_lun->dev_path); 2511 return (ENODEV); 2512 } 2513 2514 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2515 (caddr_t)&size_bytes, FREAD, 2516 curthread); 2517 if (error) { 2518 snprintf(req->error_str, sizeof(req->error_str), 2519 "error %d returned for DIOCGMEDIASIZE ioctl " 2520 "on %s!", error, be_lun->dev_path); 2521 return (error); 2522 } 2523 2524 if (params->lun_size_bytes != 0) { 2525 if (params->lun_size_bytes > size_bytes) { 2526 snprintf(req->error_str, sizeof(req->error_str), 2527 "requested LUN size %ju > backing device " 2528 "size %ju", 2529 (uintmax_t)params->lun_size_bytes, 2530 (uintmax_t)size_bytes); 2531 return (EINVAL); 2532 } 2533 2534 be_lun->size_bytes = params->lun_size_bytes; 2535 } else { 2536 be_lun->size_bytes = size_bytes; 2537 } 2538 2539 return (0); 2540 } 2541 2542 static int 2543 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2544 { 2545 struct ctl_lun_modify_params *params; 2546 struct ctl_be_block_lun *be_lun; 2547 uint64_t oldsize; 2548 int error; 2549 2550 params = &req->reqdata.modify; 2551 2552 mtx_lock(&softc->lock); 2553 be_lun = NULL; 2554 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2555 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2556 break; 2557 } 2558 mtx_unlock(&softc->lock); 2559 2560 if (be_lun == NULL) { 2561 snprintf(req->error_str, sizeof(req->error_str), 2562 "LUN %u is not managed by the block backend", 2563 params->lun_id); 2564 goto bailout_error; 2565 } 2566 2567 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2568 2569 oldsize = be_lun->size_bytes; 2570 if (be_lun->vn == NULL) 2571 error = ctl_be_block_open(softc, be_lun, req); 2572 else if (be_lun->vn->v_type == VREG) 2573 error = ctl_be_block_modify_file(be_lun, req); 2574 else 2575 error = ctl_be_block_modify_dev(be_lun, req); 2576 2577 if (error == 0 && be_lun->size_bytes != oldsize) { 2578 be_lun->size_blocks = be_lun->size_bytes >> 2579 be_lun->blocksize_shift; 2580 2581 /* 2582 * The maximum LBA is the size - 1. 2583 * 2584 * XXX: Note that this field is being updated without locking, 2585 * which might cause problems on 32-bit architectures. 2586 */ 2587 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2588 0 : (be_lun->size_blocks - 1); 2589 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2590 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2591 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2592 if (be_lun->dispatch == ctl_be_block_dispatch_zvol && 2593 be_lun->blocksize != 0) 2594 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE / 2595 be_lun->blocksize; 2596 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2597 if (oldsize == 0 && be_lun->size_blocks != 0) 2598 ctl_lun_online(&be_lun->ctl_be_lun); 2599 } 2600 2601 /* Tell the user the exact size we ended up using */ 2602 params->lun_size_bytes = be_lun->size_bytes; 2603 2604 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2605 2606 return (0); 2607 2608 bailout_error: 2609 req->status = CTL_LUN_ERROR; 2610 2611 return (0); 2612 } 2613 2614 static void 2615 ctl_be_block_lun_shutdown(void *be_lun) 2616 { 2617 struct ctl_be_block_lun *lun; 2618 struct ctl_be_block_softc *softc; 2619 2620 lun = (struct ctl_be_block_lun *)be_lun; 2621 2622 softc = lun->softc; 2623 2624 mtx_lock(&softc->lock); 2625 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2626 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2627 wakeup(lun); 2628 mtx_unlock(&softc->lock); 2629 2630 } 2631 2632 static void 2633 ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2634 { 2635 struct ctl_be_block_lun *lun; 2636 struct ctl_be_block_softc *softc; 2637 2638 lun = (struct ctl_be_block_lun *)be_lun; 2639 softc = lun->softc; 2640 2641 if (status == CTL_LUN_CONFIG_OK) { 2642 mtx_lock(&softc->lock); 2643 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2644 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2645 wakeup(lun); 2646 mtx_unlock(&softc->lock); 2647 2648 /* 2649 * We successfully added the LUN, attempt to enable it. 2650 */ 2651 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2652 printf("%s: ctl_enable_lun() failed!\n", __func__); 2653 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2654 printf("%s: ctl_invalidate_lun() failed!\n", 2655 __func__); 2656 } 2657 } 2658 2659 return; 2660 } 2661 2662 2663 mtx_lock(&softc->lock); 2664 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2665 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2666 wakeup(lun); 2667 mtx_unlock(&softc->lock); 2668 } 2669 2670 2671 static int 2672 ctl_be_block_config_write(union ctl_io *io) 2673 { 2674 struct ctl_be_block_lun *be_lun; 2675 struct ctl_be_lun *ctl_be_lun; 2676 int retval; 2677 2678 retval = 0; 2679 2680 DPRINTF("entered\n"); 2681 2682 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2683 CTL_PRIV_BACKEND_LUN].ptr; 2684 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2685 2686 switch (io->scsiio.cdb[0]) { 2687 case SYNCHRONIZE_CACHE: 2688 case SYNCHRONIZE_CACHE_16: 2689 case WRITE_SAME_10: 2690 case WRITE_SAME_16: 2691 case UNMAP: 2692 /* 2693 * The upper level CTL code will filter out any CDBs with 2694 * the immediate bit set and return the proper error. 2695 * 2696 * We don't really need to worry about what LBA range the 2697 * user asked to be synced out. When they issue a sync 2698 * cache command, we'll sync out the whole thing. 2699 */ 2700 mtx_lock(&be_lun->queue_lock); 2701 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2702 links); 2703 mtx_unlock(&be_lun->queue_lock); 2704 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2705 break; 2706 case START_STOP_UNIT: { 2707 struct scsi_start_stop_unit *cdb; 2708 2709 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2710 2711 if (cdb->how & SSS_START) 2712 retval = ctl_start_lun(ctl_be_lun); 2713 else { 2714 retval = ctl_stop_lun(ctl_be_lun); 2715 /* 2716 * XXX KDM Copan-specific offline behavior. 2717 * Figure out a reasonable way to port this? 2718 */ 2719 #ifdef NEEDTOPORT 2720 if ((retval == 0) 2721 && (cdb->byte2 & SSS_ONOFFLINE)) 2722 retval = ctl_lun_offline(ctl_be_lun); 2723 #endif 2724 } 2725 2726 /* 2727 * In general, the above routines should not fail. They 2728 * just set state for the LUN. So we've got something 2729 * pretty wrong here if we can't start or stop the LUN. 2730 */ 2731 if (retval != 0) { 2732 ctl_set_internal_failure(&io->scsiio, 2733 /*sks_valid*/ 1, 2734 /*retry_count*/ 0xf051); 2735 retval = CTL_RETVAL_COMPLETE; 2736 } else { 2737 ctl_set_success(&io->scsiio); 2738 } 2739 ctl_config_write_done(io); 2740 break; 2741 } 2742 default: 2743 ctl_set_invalid_opcode(&io->scsiio); 2744 ctl_config_write_done(io); 2745 retval = CTL_RETVAL_COMPLETE; 2746 break; 2747 } 2748 2749 return (retval); 2750 } 2751 2752 static int 2753 ctl_be_block_config_read(union ctl_io *io) 2754 { 2755 struct ctl_be_block_lun *be_lun; 2756 struct ctl_be_lun *ctl_be_lun; 2757 int retval = 0; 2758 2759 DPRINTF("entered\n"); 2760 2761 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2762 CTL_PRIV_BACKEND_LUN].ptr; 2763 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2764 2765 switch (io->scsiio.cdb[0]) { 2766 case SERVICE_ACTION_IN: 2767 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2768 mtx_lock(&be_lun->queue_lock); 2769 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2770 &io->io_hdr, links); 2771 mtx_unlock(&be_lun->queue_lock); 2772 taskqueue_enqueue(be_lun->io_taskqueue, 2773 &be_lun->io_task); 2774 retval = CTL_RETVAL_QUEUED; 2775 break; 2776 } 2777 ctl_set_invalid_field(&io->scsiio, 2778 /*sks_valid*/ 1, 2779 /*command*/ 1, 2780 /*field*/ 1, 2781 /*bit_valid*/ 1, 2782 /*bit*/ 4); 2783 ctl_config_read_done(io); 2784 retval = CTL_RETVAL_COMPLETE; 2785 break; 2786 default: 2787 ctl_set_invalid_opcode(&io->scsiio); 2788 ctl_config_read_done(io); 2789 retval = CTL_RETVAL_COMPLETE; 2790 break; 2791 } 2792 2793 return (retval); 2794 } 2795 2796 static int 2797 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2798 { 2799 struct ctl_be_block_lun *lun; 2800 int retval; 2801 2802 lun = (struct ctl_be_block_lun *)be_lun; 2803 retval = 0; 2804 2805 retval = sbuf_printf(sb, "\t<num_threads>"); 2806 2807 if (retval != 0) 2808 goto bailout; 2809 2810 retval = sbuf_printf(sb, "%d", lun->num_threads); 2811 2812 if (retval != 0) 2813 goto bailout; 2814 2815 retval = sbuf_printf(sb, "</num_threads>\n"); 2816 2817 bailout: 2818 2819 return (retval); 2820 } 2821 2822 static uint64_t 2823 ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2824 { 2825 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2826 2827 if (lun->getattr == NULL) 2828 return (UINT64_MAX); 2829 return (lun->getattr(lun, attrname)); 2830 } 2831 2832 int 2833 ctl_be_block_init(void) 2834 { 2835 struct ctl_be_block_softc *softc; 2836 int retval; 2837 2838 softc = &backend_block_softc; 2839 retval = 0; 2840 2841 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2842 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2843 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2844 STAILQ_INIT(&softc->disk_list); 2845 STAILQ_INIT(&softc->lun_list); 2846 2847 return (retval); 2848 } 2849