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_ioctl.h> 88 #include <cam/ctl/ctl_ha.h> 89 #include <cam/ctl/ctl_scsi_all.h> 90 #include <cam/ctl/ctl_private.h> 91 #include <cam/ctl/ctl_error.h> 92 93 /* 94 * The idea here is that we'll allocate enough S/G space to hold a 1MB 95 * I/O. If we get an I/O larger than that, we'll split it. 96 */ 97 #define CTLBLK_HALF_IO_SIZE (512 * 1024) 98 #define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) 99 #define CTLBLK_MAX_SEG MAXPHYS 100 #define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) 101 #define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) 102 103 #ifdef CTLBLK_DEBUG 104 #define DPRINTF(fmt, args...) \ 105 printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) 106 #else 107 #define DPRINTF(fmt, args...) do {} while(0) 108 #endif 109 110 #define PRIV(io) \ 111 ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) 112 #define ARGS(io) \ 113 ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) 114 115 SDT_PROVIDER_DEFINE(cbb); 116 117 typedef enum { 118 CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, 119 CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, 120 CTL_BE_BLOCK_LUN_WAITING = 0x04, 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_filedata { 130 struct ucred *cred; 131 }; 132 133 union ctl_be_block_bedata { 134 struct ctl_be_block_filedata file; 135 }; 136 137 struct ctl_be_block_io; 138 struct ctl_be_block_lun; 139 140 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, 141 struct ctl_be_block_io *beio); 142 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, 143 const char *attrname); 144 145 /* 146 * Backend LUN structure. There is a 1:1 mapping between a block device 147 * and a backend block LUN, and between a backend block LUN and a CTL LUN. 148 */ 149 struct ctl_be_block_lun { 150 struct ctl_lun_create_params params; 151 char lunname[32]; 152 char *dev_path; 153 ctl_be_block_type dev_type; 154 struct vnode *vn; 155 union ctl_be_block_bedata backend; 156 cbb_dispatch_t dispatch; 157 cbb_dispatch_t lun_flush; 158 cbb_dispatch_t unmap; 159 cbb_dispatch_t get_lba_status; 160 cbb_getattr_t getattr; 161 uma_zone_t lun_zone; 162 uint64_t size_blocks; 163 uint64_t size_bytes; 164 struct ctl_be_block_softc *softc; 165 struct devstat *disk_stats; 166 ctl_be_block_lun_flags flags; 167 STAILQ_ENTRY(ctl_be_block_lun) links; 168 struct ctl_be_lun cbe_lun; 169 struct taskqueue *io_taskqueue; 170 struct task io_task; 171 int num_threads; 172 STAILQ_HEAD(, ctl_io_hdr) input_queue; 173 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 174 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 175 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 176 struct mtx_padalign io_lock; 177 struct mtx_padalign queue_lock; 178 }; 179 180 /* 181 * Overall softc structure for the block backend module. 182 */ 183 struct ctl_be_block_softc { 184 struct mtx lock; 185 int num_luns; 186 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 187 }; 188 189 static struct ctl_be_block_softc backend_block_softc; 190 191 /* 192 * Per-I/O information. 193 */ 194 struct ctl_be_block_io { 195 union ctl_io *io; 196 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 197 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 198 int bio_cmd; 199 int num_segs; 200 int num_bios_sent; 201 int num_bios_done; 202 int send_complete; 203 int num_errors; 204 struct bintime ds_t0; 205 devstat_tag_type ds_tag_type; 206 devstat_trans_flags ds_trans_type; 207 uint64_t io_len; 208 uint64_t io_offset; 209 int io_arg; 210 struct ctl_be_block_softc *softc; 211 struct ctl_be_block_lun *lun; 212 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 213 }; 214 215 extern struct ctl_softc *control_softc; 216 217 static int cbb_num_threads = 14; 218 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 219 "CAM Target Layer Block Backend"); 220 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 221 &cbb_num_threads, 0, "Number of threads per backing file"); 222 223 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 224 static void ctl_free_beio(struct ctl_be_block_io *beio); 225 static void ctl_complete_beio(struct ctl_be_block_io *beio); 226 static int ctl_be_block_move_done(union ctl_io *io); 227 static void ctl_be_block_biodone(struct bio *bio); 228 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 229 struct ctl_be_block_io *beio); 230 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 231 struct ctl_be_block_io *beio); 232 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 233 struct ctl_be_block_io *beio); 234 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 235 const char *attrname); 236 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 237 struct ctl_be_block_io *beio); 238 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 239 struct ctl_be_block_io *beio); 240 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 241 struct ctl_be_block_io *beio); 242 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 243 const char *attrname); 244 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 245 union ctl_io *io); 246 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 247 union ctl_io *io); 248 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 249 union ctl_io *io); 250 static void ctl_be_block_worker(void *context, int pending); 251 static int ctl_be_block_submit(union ctl_io *io); 252 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 253 int flag, struct thread *td); 254 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 255 struct ctl_lun_req *req); 256 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 257 struct ctl_lun_req *req); 258 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 259 static int ctl_be_block_open(struct ctl_be_block_softc *softc, 260 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 getbintime(&cur_bt); 409 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 410 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 411 io->io_hdr.num_dmas++; 412 #endif 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 /* 467 * XXX KDM make sure that links is okay to use at this point. 468 * Otherwise, we either need to add another field to ctl_io_hdr, 469 * or deal with resource allocation here. 470 */ 471 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 472 mtx_unlock(&be_lun->queue_lock); 473 474 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 475 476 return (0); 477 } 478 479 static void 480 ctl_be_block_biodone(struct bio *bio) 481 { 482 struct ctl_be_block_io *beio; 483 struct ctl_be_block_lun *be_lun; 484 union ctl_io *io; 485 int error; 486 487 beio = bio->bio_caller1; 488 be_lun = beio->lun; 489 io = beio->io; 490 491 DPRINTF("entered\n"); 492 493 error = bio->bio_error; 494 mtx_lock(&be_lun->io_lock); 495 if (error != 0) 496 beio->num_errors++; 497 498 beio->num_bios_done++; 499 500 /* 501 * XXX KDM will this cause WITNESS to complain? Holding a lock 502 * during the free might cause it to complain. 503 */ 504 g_destroy_bio(bio); 505 506 /* 507 * If the send complete bit isn't set, or we aren't the last I/O to 508 * complete, then we're done. 509 */ 510 if ((beio->send_complete == 0) 511 || (beio->num_bios_done < beio->num_bios_sent)) { 512 mtx_unlock(&be_lun->io_lock); 513 return; 514 } 515 516 /* 517 * At this point, we've verified that we are the last I/O to 518 * complete, so it's safe to drop the lock. 519 */ 520 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 521 beio->ds_tag_type, beio->ds_trans_type, 522 /*now*/ NULL, /*then*/&beio->ds_t0); 523 mtx_unlock(&be_lun->io_lock); 524 525 /* 526 * If there are any errors from the backing device, we fail the 527 * entire I/O with a medium error. 528 */ 529 if (beio->num_errors > 0) { 530 if (error == EOPNOTSUPP) { 531 ctl_set_invalid_opcode(&io->scsiio); 532 } else if (error == ENOSPC || error == EDQUOT) { 533 ctl_set_space_alloc_fail(&io->scsiio); 534 } else if (error == EROFS || error == EACCES) { 535 ctl_set_hw_write_protected(&io->scsiio); 536 } else if (beio->bio_cmd == BIO_FLUSH) { 537 /* XXX KDM is there is a better error here? */ 538 ctl_set_internal_failure(&io->scsiio, 539 /*sks_valid*/ 1, 540 /*retry_count*/ 0xbad2); 541 } else { 542 ctl_set_medium_error(&io->scsiio, 543 beio->bio_cmd == BIO_READ); 544 } 545 ctl_complete_beio(beio); 546 return; 547 } 548 549 /* 550 * If this is a write, a flush, a delete or verify, we're all done. 551 * If this is a read, we can now send the data to the user. 552 */ 553 if ((beio->bio_cmd == BIO_WRITE) 554 || (beio->bio_cmd == BIO_FLUSH) 555 || (beio->bio_cmd == BIO_DELETE) 556 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 557 ctl_set_success(&io->scsiio); 558 ctl_complete_beio(beio); 559 } else { 560 if ((ARGS(io)->flags & CTL_LLF_READ) && 561 beio->beio_cont == NULL) { 562 ctl_set_success(&io->scsiio); 563 ctl_serseq_done(io); 564 } 565 #ifdef CTL_TIME_IO 566 getbintime(&io->io_hdr.dma_start_bt); 567 #endif 568 ctl_datamove(io); 569 } 570 } 571 572 static void 573 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 574 struct ctl_be_block_io *beio) 575 { 576 union ctl_io *io = beio->io; 577 struct mount *mountpoint; 578 int error, lock_flags; 579 580 DPRINTF("entered\n"); 581 582 binuptime(&beio->ds_t0); 583 mtx_lock(&be_lun->io_lock); 584 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 585 mtx_unlock(&be_lun->io_lock); 586 587 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 588 589 if (MNT_SHARED_WRITES(mountpoint) 590 || ((mountpoint == NULL) 591 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 592 lock_flags = LK_SHARED; 593 else 594 lock_flags = LK_EXCLUSIVE; 595 596 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 597 598 error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, 599 curthread); 600 VOP_UNLOCK(be_lun->vn, 0); 601 602 vn_finished_write(mountpoint); 603 604 mtx_lock(&be_lun->io_lock); 605 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 606 beio->ds_tag_type, beio->ds_trans_type, 607 /*now*/ NULL, /*then*/&beio->ds_t0); 608 mtx_unlock(&be_lun->io_lock); 609 610 if (error == 0) 611 ctl_set_success(&io->scsiio); 612 else { 613 /* XXX KDM is there is a better error here? */ 614 ctl_set_internal_failure(&io->scsiio, 615 /*sks_valid*/ 1, 616 /*retry_count*/ 0xbad1); 617 } 618 619 ctl_complete_beio(beio); 620 } 621 622 SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 623 SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 624 SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 625 SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 626 627 static void 628 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 629 struct ctl_be_block_io *beio) 630 { 631 struct ctl_be_block_filedata *file_data; 632 union ctl_io *io; 633 struct uio xuio; 634 struct iovec *xiovec; 635 size_t s; 636 int error, flags, i; 637 638 DPRINTF("entered\n"); 639 640 file_data = &be_lun->backend.file; 641 io = beio->io; 642 flags = 0; 643 if (ARGS(io)->flags & CTL_LLF_DPO) 644 flags |= IO_DIRECT; 645 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 646 flags |= IO_SYNC; 647 648 bzero(&xuio, sizeof(xuio)); 649 if (beio->bio_cmd == BIO_READ) { 650 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 651 xuio.uio_rw = UIO_READ; 652 } else { 653 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 654 xuio.uio_rw = UIO_WRITE; 655 } 656 xuio.uio_offset = beio->io_offset; 657 xuio.uio_resid = beio->io_len; 658 xuio.uio_segflg = UIO_SYSSPACE; 659 xuio.uio_iov = beio->xiovecs; 660 xuio.uio_iovcnt = beio->num_segs; 661 xuio.uio_td = curthread; 662 663 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 664 xiovec->iov_base = beio->sg_segs[i].addr; 665 xiovec->iov_len = beio->sg_segs[i].len; 666 } 667 668 binuptime(&beio->ds_t0); 669 mtx_lock(&be_lun->io_lock); 670 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 671 mtx_unlock(&be_lun->io_lock); 672 673 if (beio->bio_cmd == BIO_READ) { 674 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 675 676 /* 677 * UFS pays attention to IO_DIRECT for reads. If the 678 * DIRECTIO option is configured into the kernel, it calls 679 * ffs_rawread(). But that only works for single-segment 680 * uios with user space addresses. In our case, with a 681 * kernel uio, it still reads into the buffer cache, but it 682 * will just try to release the buffer from the cache later 683 * on in ffs_read(). 684 * 685 * ZFS does not pay attention to IO_DIRECT for reads. 686 * 687 * UFS does not pay attention to IO_SYNC for reads. 688 * 689 * ZFS pays attention to IO_SYNC (which translates into the 690 * Solaris define FRSYNC for zfs_read()) for reads. It 691 * attempts to sync the file before reading. 692 */ 693 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 694 695 VOP_UNLOCK(be_lun->vn, 0); 696 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 697 if (error == 0 && xuio.uio_resid > 0) { 698 /* 699 * If we red less then requested (EOF), then 700 * we should clean the rest of the buffer. 701 */ 702 s = beio->io_len - xuio.uio_resid; 703 for (i = 0; i < beio->num_segs; i++) { 704 if (s >= beio->sg_segs[i].len) { 705 s -= beio->sg_segs[i].len; 706 continue; 707 } 708 bzero((uint8_t *)beio->sg_segs[i].addr + s, 709 beio->sg_segs[i].len - s); 710 s = 0; 711 } 712 } 713 } else { 714 struct mount *mountpoint; 715 int lock_flags; 716 717 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 718 719 if (MNT_SHARED_WRITES(mountpoint) 720 || ((mountpoint == NULL) 721 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 722 lock_flags = LK_SHARED; 723 else 724 lock_flags = LK_EXCLUSIVE; 725 726 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 727 728 /* 729 * UFS pays attention to IO_DIRECT for writes. The write 730 * is done asynchronously. (Normally the write would just 731 * get put into cache. 732 * 733 * UFS pays attention to IO_SYNC for writes. It will 734 * attempt to write the buffer out synchronously if that 735 * flag is set. 736 * 737 * ZFS does not pay attention to IO_DIRECT for writes. 738 * 739 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 740 * for writes. It will flush the transaction from the 741 * cache before returning. 742 */ 743 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 744 VOP_UNLOCK(be_lun->vn, 0); 745 746 vn_finished_write(mountpoint); 747 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 748 } 749 750 mtx_lock(&be_lun->io_lock); 751 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 752 beio->ds_tag_type, beio->ds_trans_type, 753 /*now*/ NULL, /*then*/&beio->ds_t0); 754 mtx_unlock(&be_lun->io_lock); 755 756 /* 757 * If we got an error, set the sense data to "MEDIUM ERROR" and 758 * return the I/O to the user. 759 */ 760 if (error != 0) { 761 if (error == ENOSPC || error == EDQUOT) { 762 ctl_set_space_alloc_fail(&io->scsiio); 763 } else if (error == EROFS || error == EACCES) { 764 ctl_set_hw_write_protected(&io->scsiio); 765 } else { 766 ctl_set_medium_error(&io->scsiio, 767 beio->bio_cmd == BIO_READ); 768 } 769 ctl_complete_beio(beio); 770 return; 771 } 772 773 /* 774 * If this is a write or a verify, we're all done. 775 * If this is a read, we can now send the data to the user. 776 */ 777 if ((beio->bio_cmd == BIO_WRITE) || 778 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 779 ctl_set_success(&io->scsiio); 780 ctl_complete_beio(beio); 781 } else { 782 if ((ARGS(io)->flags & CTL_LLF_READ) && 783 beio->beio_cont == NULL) { 784 ctl_set_success(&io->scsiio); 785 ctl_serseq_done(io); 786 } 787 #ifdef CTL_TIME_IO 788 getbintime(&io->io_hdr.dma_start_bt); 789 #endif 790 ctl_datamove(io); 791 } 792 } 793 794 static void 795 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 796 struct ctl_be_block_io *beio) 797 { 798 union ctl_io *io = beio->io; 799 struct ctl_lba_len_flags *lbalen = ARGS(io); 800 struct scsi_get_lba_status_data *data; 801 off_t roff, off; 802 int error, status; 803 804 DPRINTF("entered\n"); 805 806 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 807 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 808 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 809 0, curthread->td_ucred, curthread); 810 if (error == 0 && off > roff) 811 status = 0; /* mapped up to off */ 812 else { 813 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 814 0, curthread->td_ucred, curthread); 815 if (error == 0 && off > roff) 816 status = 1; /* deallocated up to off */ 817 else { 818 status = 0; /* unknown up to the end */ 819 off = be_lun->size_bytes; 820 } 821 } 822 VOP_UNLOCK(be_lun->vn, 0); 823 824 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 825 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 826 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 827 lbalen->lba), data->descr[0].length); 828 data->descr[0].status = status; 829 830 ctl_complete_beio(beio); 831 } 832 833 static uint64_t 834 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 835 { 836 struct vattr vattr; 837 struct statfs statfs; 838 uint64_t val; 839 int error; 840 841 val = UINT64_MAX; 842 if (be_lun->vn == NULL) 843 return (val); 844 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 845 if (strcmp(attrname, "blocksused") == 0) { 846 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 847 if (error == 0) 848 val = vattr.va_bytes / be_lun->cbe_lun.blocksize; 849 } 850 if (strcmp(attrname, "blocksavail") == 0 && 851 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 852 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 853 if (error == 0) 854 val = statfs.f_bavail * statfs.f_bsize / 855 be_lun->cbe_lun.blocksize; 856 } 857 VOP_UNLOCK(be_lun->vn, 0); 858 return (val); 859 } 860 861 static void 862 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 863 struct ctl_be_block_io *beio) 864 { 865 union ctl_io *io; 866 struct cdevsw *csw; 867 struct cdev *dev; 868 struct uio xuio; 869 struct iovec *xiovec; 870 int error, flags, i, ref; 871 872 DPRINTF("entered\n"); 873 874 io = beio->io; 875 flags = 0; 876 if (ARGS(io)->flags & CTL_LLF_DPO) 877 flags |= IO_DIRECT; 878 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 879 flags |= IO_SYNC; 880 881 bzero(&xuio, sizeof(xuio)); 882 if (beio->bio_cmd == BIO_READ) { 883 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 884 xuio.uio_rw = UIO_READ; 885 } else { 886 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 887 xuio.uio_rw = UIO_WRITE; 888 } 889 xuio.uio_offset = beio->io_offset; 890 xuio.uio_resid = beio->io_len; 891 xuio.uio_segflg = UIO_SYSSPACE; 892 xuio.uio_iov = beio->xiovecs; 893 xuio.uio_iovcnt = beio->num_segs; 894 xuio.uio_td = curthread; 895 896 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 897 xiovec->iov_base = beio->sg_segs[i].addr; 898 xiovec->iov_len = beio->sg_segs[i].len; 899 } 900 901 binuptime(&beio->ds_t0); 902 mtx_lock(&be_lun->io_lock); 903 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 904 mtx_unlock(&be_lun->io_lock); 905 906 csw = devvn_refthread(be_lun->vn, &dev, &ref); 907 if (csw) { 908 if (beio->bio_cmd == BIO_READ) 909 error = csw->d_read(dev, &xuio, flags); 910 else 911 error = csw->d_write(dev, &xuio, flags); 912 dev_relthread(dev, ref); 913 } else 914 error = ENXIO; 915 916 if (beio->bio_cmd == BIO_READ) 917 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 918 else 919 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 920 921 mtx_lock(&be_lun->io_lock); 922 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 923 beio->ds_tag_type, beio->ds_trans_type, 924 /*now*/ NULL, /*then*/&beio->ds_t0); 925 mtx_unlock(&be_lun->io_lock); 926 927 /* 928 * If we got an error, set the sense data to "MEDIUM ERROR" and 929 * return the I/O to the user. 930 */ 931 if (error != 0) { 932 if (error == ENOSPC || error == EDQUOT) { 933 ctl_set_space_alloc_fail(&io->scsiio); 934 } else if (error == EROFS || error == EACCES) { 935 ctl_set_hw_write_protected(&io->scsiio); 936 } else { 937 ctl_set_medium_error(&io->scsiio, 938 beio->bio_cmd == BIO_READ); 939 } 940 ctl_complete_beio(beio); 941 return; 942 } 943 944 /* 945 * If this is a write or a verify, we're all done. 946 * If this is a read, we can now send the data to the user. 947 */ 948 if ((beio->bio_cmd == BIO_WRITE) || 949 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 950 ctl_set_success(&io->scsiio); 951 ctl_complete_beio(beio); 952 } else { 953 if ((ARGS(io)->flags & CTL_LLF_READ) && 954 beio->beio_cont == NULL) { 955 ctl_set_success(&io->scsiio); 956 ctl_serseq_done(io); 957 } 958 #ifdef CTL_TIME_IO 959 getbintime(&io->io_hdr.dma_start_bt); 960 #endif 961 ctl_datamove(io); 962 } 963 } 964 965 static void 966 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 967 struct ctl_be_block_io *beio) 968 { 969 union ctl_io *io = beio->io; 970 struct cdevsw *csw; 971 struct cdev *dev; 972 struct ctl_lba_len_flags *lbalen = ARGS(io); 973 struct scsi_get_lba_status_data *data; 974 off_t roff, off; 975 int error, ref, status; 976 977 DPRINTF("entered\n"); 978 979 csw = devvn_refthread(be_lun->vn, &dev, &ref); 980 if (csw == NULL) { 981 status = 0; /* unknown up to the end */ 982 off = be_lun->size_bytes; 983 goto done; 984 } 985 off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; 986 error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD, 987 curthread); 988 if (error == 0 && off > roff) 989 status = 0; /* mapped up to off */ 990 else { 991 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD, 992 curthread); 993 if (error == 0 && off > roff) 994 status = 1; /* deallocated up to off */ 995 else { 996 status = 0; /* unknown up to the end */ 997 off = be_lun->size_bytes; 998 } 999 } 1000 dev_relthread(dev, ref); 1001 1002 done: 1003 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 1004 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 1005 scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - 1006 lbalen->lba), data->descr[0].length); 1007 data->descr[0].status = status; 1008 1009 ctl_complete_beio(beio); 1010 } 1011 1012 static void 1013 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 1014 struct ctl_be_block_io *beio) 1015 { 1016 struct bio *bio; 1017 union ctl_io *io; 1018 struct cdevsw *csw; 1019 struct cdev *dev; 1020 int ref; 1021 1022 io = beio->io; 1023 1024 DPRINTF("entered\n"); 1025 1026 /* This can't fail, it's a blocking allocation. */ 1027 bio = g_alloc_bio(); 1028 1029 bio->bio_cmd = BIO_FLUSH; 1030 bio->bio_offset = 0; 1031 bio->bio_data = 0; 1032 bio->bio_done = ctl_be_block_biodone; 1033 bio->bio_caller1 = beio; 1034 bio->bio_pblkno = 0; 1035 1036 /* 1037 * We don't need to acquire the LUN lock here, because we are only 1038 * sending one bio, and so there is no other context to synchronize 1039 * with. 1040 */ 1041 beio->num_bios_sent = 1; 1042 beio->send_complete = 1; 1043 1044 binuptime(&beio->ds_t0); 1045 mtx_lock(&be_lun->io_lock); 1046 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1047 mtx_unlock(&be_lun->io_lock); 1048 1049 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1050 if (csw) { 1051 bio->bio_dev = dev; 1052 csw->d_strategy(bio); 1053 dev_relthread(dev, ref); 1054 } else { 1055 bio->bio_error = ENXIO; 1056 ctl_be_block_biodone(bio); 1057 } 1058 } 1059 1060 static void 1061 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1062 struct ctl_be_block_io *beio, 1063 uint64_t off, uint64_t len, int last) 1064 { 1065 struct bio *bio; 1066 uint64_t maxlen; 1067 struct cdevsw *csw; 1068 struct cdev *dev; 1069 int ref; 1070 1071 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1072 maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); 1073 while (len > 0) { 1074 bio = g_alloc_bio(); 1075 bio->bio_cmd = BIO_DELETE; 1076 bio->bio_dev = dev; 1077 bio->bio_offset = off; 1078 bio->bio_length = MIN(len, maxlen); 1079 bio->bio_data = 0; 1080 bio->bio_done = ctl_be_block_biodone; 1081 bio->bio_caller1 = beio; 1082 bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; 1083 1084 off += bio->bio_length; 1085 len -= bio->bio_length; 1086 1087 mtx_lock(&be_lun->io_lock); 1088 beio->num_bios_sent++; 1089 if (last && len == 0) 1090 beio->send_complete = 1; 1091 mtx_unlock(&be_lun->io_lock); 1092 1093 if (csw) { 1094 csw->d_strategy(bio); 1095 } else { 1096 bio->bio_error = ENXIO; 1097 ctl_be_block_biodone(bio); 1098 } 1099 } 1100 if (csw) 1101 dev_relthread(dev, ref); 1102 } 1103 1104 static void 1105 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1106 struct ctl_be_block_io *beio) 1107 { 1108 union ctl_io *io; 1109 struct ctl_ptr_len_flags *ptrlen; 1110 struct scsi_unmap_desc *buf, *end; 1111 uint64_t len; 1112 1113 io = beio->io; 1114 1115 DPRINTF("entered\n"); 1116 1117 binuptime(&beio->ds_t0); 1118 mtx_lock(&be_lun->io_lock); 1119 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1120 mtx_unlock(&be_lun->io_lock); 1121 1122 if (beio->io_offset == -1) { 1123 beio->io_len = 0; 1124 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1125 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1126 end = buf + ptrlen->len / sizeof(*buf); 1127 for (; buf < end; buf++) { 1128 len = (uint64_t)scsi_4btoul(buf->length) * 1129 be_lun->cbe_lun.blocksize; 1130 beio->io_len += len; 1131 ctl_be_block_unmap_dev_range(be_lun, beio, 1132 scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, 1133 len, (end - buf < 2) ? TRUE : FALSE); 1134 } 1135 } else 1136 ctl_be_block_unmap_dev_range(be_lun, beio, 1137 beio->io_offset, beio->io_len, TRUE); 1138 } 1139 1140 static void 1141 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1142 struct ctl_be_block_io *beio) 1143 { 1144 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1145 struct bio *bio; 1146 struct cdevsw *csw; 1147 struct cdev *dev; 1148 off_t cur_offset; 1149 int i, max_iosize, ref; 1150 1151 DPRINTF("entered\n"); 1152 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1153 1154 /* 1155 * We have to limit our I/O size to the maximum supported by the 1156 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1157 * set it properly, use DFLTPHYS. 1158 */ 1159 if (csw) { 1160 max_iosize = dev->si_iosize_max; 1161 if (max_iosize < PAGE_SIZE) 1162 max_iosize = DFLTPHYS; 1163 } else 1164 max_iosize = DFLTPHYS; 1165 1166 cur_offset = beio->io_offset; 1167 for (i = 0; i < beio->num_segs; i++) { 1168 size_t cur_size; 1169 uint8_t *cur_ptr; 1170 1171 cur_size = beio->sg_segs[i].len; 1172 cur_ptr = beio->sg_segs[i].addr; 1173 1174 while (cur_size > 0) { 1175 /* This can't fail, it's a blocking allocation. */ 1176 bio = g_alloc_bio(); 1177 1178 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1179 1180 bio->bio_cmd = beio->bio_cmd; 1181 bio->bio_dev = dev; 1182 bio->bio_caller1 = beio; 1183 bio->bio_length = min(cur_size, max_iosize); 1184 bio->bio_offset = cur_offset; 1185 bio->bio_data = cur_ptr; 1186 bio->bio_done = ctl_be_block_biodone; 1187 bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; 1188 1189 cur_offset += bio->bio_length; 1190 cur_ptr += bio->bio_length; 1191 cur_size -= bio->bio_length; 1192 1193 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1194 beio->num_bios_sent++; 1195 } 1196 } 1197 binuptime(&beio->ds_t0); 1198 mtx_lock(&be_lun->io_lock); 1199 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1200 beio->send_complete = 1; 1201 mtx_unlock(&be_lun->io_lock); 1202 1203 /* 1204 * Fire off all allocated requests! 1205 */ 1206 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1207 TAILQ_REMOVE(&queue, bio, bio_queue); 1208 if (csw) 1209 csw->d_strategy(bio); 1210 else { 1211 bio->bio_error = ENXIO; 1212 ctl_be_block_biodone(bio); 1213 } 1214 } 1215 if (csw) 1216 dev_relthread(dev, ref); 1217 } 1218 1219 static uint64_t 1220 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1221 { 1222 struct diocgattr_arg arg; 1223 struct cdevsw *csw; 1224 struct cdev *dev; 1225 int error, ref; 1226 1227 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1228 if (csw == NULL) 1229 return (UINT64_MAX); 1230 strlcpy(arg.name, attrname, sizeof(arg.name)); 1231 arg.len = sizeof(arg.value.off); 1232 if (csw->d_ioctl) { 1233 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 1234 curthread); 1235 } else 1236 error = ENODEV; 1237 dev_relthread(dev, ref); 1238 if (error != 0) 1239 return (UINT64_MAX); 1240 return (arg.value.off); 1241 } 1242 1243 static void 1244 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, 1245 union ctl_io *io) 1246 { 1247 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1248 struct ctl_be_block_io *beio; 1249 struct ctl_lba_len_flags *lbalen; 1250 1251 DPRINTF("entered\n"); 1252 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1253 lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1254 1255 beio->io_len = lbalen->len * cbe_lun->blocksize; 1256 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1257 beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; 1258 beio->bio_cmd = BIO_FLUSH; 1259 beio->ds_trans_type = DEVSTAT_NO_DATA; 1260 DPRINTF("SYNC\n"); 1261 be_lun->lun_flush(be_lun, beio); 1262 } 1263 1264 static void 1265 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1266 { 1267 union ctl_io *io; 1268 1269 io = beio->io; 1270 ctl_free_beio(beio); 1271 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1272 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1273 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1274 ctl_config_write_done(io); 1275 return; 1276 } 1277 1278 ctl_be_block_config_write(io); 1279 } 1280 1281 static void 1282 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1283 union ctl_io *io) 1284 { 1285 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1286 struct ctl_be_block_io *beio; 1287 struct ctl_lba_len_flags *lbalen; 1288 uint64_t len_left, lba; 1289 uint32_t pb, pbo, adj; 1290 int i, seglen; 1291 uint8_t *buf, *end; 1292 1293 DPRINTF("entered\n"); 1294 1295 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1296 lbalen = ARGS(beio->io); 1297 1298 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1299 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1300 ctl_free_beio(beio); 1301 ctl_set_invalid_field(&io->scsiio, 1302 /*sks_valid*/ 1, 1303 /*command*/ 1, 1304 /*field*/ 1, 1305 /*bit_valid*/ 0, 1306 /*bit*/ 0); 1307 ctl_config_write_done(io); 1308 return; 1309 } 1310 1311 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1312 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1313 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; 1314 beio->bio_cmd = BIO_DELETE; 1315 beio->ds_trans_type = DEVSTAT_FREE; 1316 1317 be_lun->unmap(be_lun, beio); 1318 return; 1319 } 1320 1321 beio->bio_cmd = BIO_WRITE; 1322 beio->ds_trans_type = DEVSTAT_WRITE; 1323 1324 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1325 (uintmax_t)lbalen->lba, lbalen->len); 1326 1327 pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; 1328 if (be_lun->cbe_lun.pblockoff > 0) 1329 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; 1330 else 1331 pbo = 0; 1332 len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; 1333 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1334 1335 /* 1336 * Setup the S/G entry for this chunk. 1337 */ 1338 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1339 if (pb > cbe_lun->blocksize) { 1340 adj = ((lbalen->lba + lba) * cbe_lun->blocksize + 1341 seglen - pbo) % pb; 1342 if (seglen > adj) 1343 seglen -= adj; 1344 else 1345 seglen -= seglen % cbe_lun->blocksize; 1346 } else 1347 seglen -= seglen % cbe_lun->blocksize; 1348 beio->sg_segs[i].len = seglen; 1349 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1350 1351 DPRINTF("segment %d addr %p len %zd\n", i, 1352 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1353 1354 beio->num_segs++; 1355 len_left -= seglen; 1356 1357 buf = beio->sg_segs[i].addr; 1358 end = buf + seglen; 1359 for (; buf < end; buf += cbe_lun->blocksize) { 1360 memcpy(buf, io->scsiio.kern_data_ptr, cbe_lun->blocksize); 1361 if (lbalen->flags & SWS_LBDATA) 1362 scsi_ulto4b(lbalen->lba + lba, buf); 1363 lba++; 1364 } 1365 } 1366 1367 beio->io_offset = lbalen->lba * cbe_lun->blocksize; 1368 beio->io_len = lba * cbe_lun->blocksize; 1369 1370 /* We can not do all in one run. Correct and schedule rerun. */ 1371 if (len_left > 0) { 1372 lbalen->lba += lba; 1373 lbalen->len -= lba; 1374 beio->beio_cont = ctl_be_block_cw_done_ws; 1375 } 1376 1377 be_lun->dispatch(be_lun, beio); 1378 } 1379 1380 static void 1381 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1382 union ctl_io *io) 1383 { 1384 struct ctl_be_block_io *beio; 1385 struct ctl_ptr_len_flags *ptrlen; 1386 1387 DPRINTF("entered\n"); 1388 1389 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1390 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1391 1392 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1393 ctl_free_beio(beio); 1394 ctl_set_invalid_field(&io->scsiio, 1395 /*sks_valid*/ 0, 1396 /*command*/ 1, 1397 /*field*/ 0, 1398 /*bit_valid*/ 0, 1399 /*bit*/ 0); 1400 ctl_config_write_done(io); 1401 return; 1402 } 1403 1404 beio->io_len = 0; 1405 beio->io_offset = -1; 1406 beio->bio_cmd = BIO_DELETE; 1407 beio->ds_trans_type = DEVSTAT_FREE; 1408 DPRINTF("UNMAP\n"); 1409 be_lun->unmap(be_lun, beio); 1410 } 1411 1412 static void 1413 ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1414 { 1415 union ctl_io *io; 1416 1417 io = beio->io; 1418 ctl_free_beio(beio); 1419 ctl_config_read_done(io); 1420 } 1421 1422 static void 1423 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1424 union ctl_io *io) 1425 { 1426 struct ctl_be_block_io *beio; 1427 struct ctl_be_block_softc *softc; 1428 1429 DPRINTF("entered\n"); 1430 1431 softc = be_lun->softc; 1432 beio = ctl_alloc_beio(softc); 1433 beio->io = io; 1434 beio->lun = be_lun; 1435 beio->beio_cont = ctl_be_block_cr_done; 1436 PRIV(io)->ptr = (void *)beio; 1437 1438 switch (io->scsiio.cdb[0]) { 1439 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1440 beio->bio_cmd = -1; 1441 beio->ds_trans_type = DEVSTAT_NO_DATA; 1442 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1443 beio->io_len = 0; 1444 if (be_lun->get_lba_status) 1445 be_lun->get_lba_status(be_lun, beio); 1446 else 1447 ctl_be_block_cr_done(beio); 1448 break; 1449 default: 1450 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1451 break; 1452 } 1453 } 1454 1455 static void 1456 ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1457 { 1458 union ctl_io *io; 1459 1460 io = beio->io; 1461 ctl_free_beio(beio); 1462 ctl_config_write_done(io); 1463 } 1464 1465 static void 1466 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1467 union ctl_io *io) 1468 { 1469 struct ctl_be_block_io *beio; 1470 struct ctl_be_block_softc *softc; 1471 1472 DPRINTF("entered\n"); 1473 1474 softc = be_lun->softc; 1475 beio = ctl_alloc_beio(softc); 1476 beio->io = io; 1477 beio->lun = be_lun; 1478 beio->beio_cont = ctl_be_block_cw_done; 1479 switch (io->scsiio.tag_type) { 1480 case CTL_TAG_ORDERED: 1481 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1482 break; 1483 case CTL_TAG_HEAD_OF_QUEUE: 1484 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1485 break; 1486 case CTL_TAG_UNTAGGED: 1487 case CTL_TAG_SIMPLE: 1488 case CTL_TAG_ACA: 1489 default: 1490 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1491 break; 1492 } 1493 PRIV(io)->ptr = (void *)beio; 1494 1495 switch (io->scsiio.cdb[0]) { 1496 case SYNCHRONIZE_CACHE: 1497 case SYNCHRONIZE_CACHE_16: 1498 ctl_be_block_cw_dispatch_sync(be_lun, io); 1499 break; 1500 case WRITE_SAME_10: 1501 case WRITE_SAME_16: 1502 ctl_be_block_cw_dispatch_ws(be_lun, io); 1503 break; 1504 case UNMAP: 1505 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1506 break; 1507 default: 1508 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1509 break; 1510 } 1511 } 1512 1513 SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1514 SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1515 SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1516 SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1517 1518 static void 1519 ctl_be_block_next(struct ctl_be_block_io *beio) 1520 { 1521 struct ctl_be_block_lun *be_lun; 1522 union ctl_io *io; 1523 1524 io = beio->io; 1525 be_lun = beio->lun; 1526 ctl_free_beio(beio); 1527 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1528 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1529 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1530 ctl_data_submit_done(io); 1531 return; 1532 } 1533 1534 io->io_hdr.status &= ~CTL_STATUS_MASK; 1535 io->io_hdr.status |= CTL_STATUS_NONE; 1536 1537 mtx_lock(&be_lun->queue_lock); 1538 /* 1539 * XXX KDM make sure that links is okay to use at this point. 1540 * Otherwise, we either need to add another field to ctl_io_hdr, 1541 * or deal with resource allocation here. 1542 */ 1543 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1544 mtx_unlock(&be_lun->queue_lock); 1545 1546 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1547 } 1548 1549 static void 1550 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1551 union ctl_io *io) 1552 { 1553 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1554 struct ctl_be_block_io *beio; 1555 struct ctl_be_block_softc *softc; 1556 struct ctl_lba_len_flags *lbalen; 1557 struct ctl_ptr_len_flags *bptrlen; 1558 uint64_t len_left, lbas; 1559 int i; 1560 1561 softc = be_lun->softc; 1562 1563 DPRINTF("entered\n"); 1564 1565 lbalen = ARGS(io); 1566 if (lbalen->flags & CTL_LLF_WRITE) { 1567 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1568 } else { 1569 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1570 } 1571 1572 beio = ctl_alloc_beio(softc); 1573 beio->io = io; 1574 beio->lun = be_lun; 1575 bptrlen = PRIV(io); 1576 bptrlen->ptr = (void *)beio; 1577 1578 switch (io->scsiio.tag_type) { 1579 case CTL_TAG_ORDERED: 1580 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1581 break; 1582 case CTL_TAG_HEAD_OF_QUEUE: 1583 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1584 break; 1585 case CTL_TAG_UNTAGGED: 1586 case CTL_TAG_SIMPLE: 1587 case CTL_TAG_ACA: 1588 default: 1589 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1590 break; 1591 } 1592 1593 if (lbalen->flags & CTL_LLF_WRITE) { 1594 beio->bio_cmd = BIO_WRITE; 1595 beio->ds_trans_type = DEVSTAT_WRITE; 1596 } else { 1597 beio->bio_cmd = BIO_READ; 1598 beio->ds_trans_type = DEVSTAT_READ; 1599 } 1600 1601 DPRINTF("%s at LBA %jx len %u @%ju\n", 1602 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1603 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1604 if (lbalen->flags & CTL_LLF_COMPARE) 1605 lbas = CTLBLK_HALF_IO_SIZE; 1606 else 1607 lbas = CTLBLK_MAX_IO_SIZE; 1608 lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); 1609 beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; 1610 beio->io_len = lbas * cbe_lun->blocksize; 1611 bptrlen->len += lbas; 1612 1613 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1614 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1615 i, CTLBLK_MAX_SEGS)); 1616 1617 /* 1618 * Setup the S/G entry for this chunk. 1619 */ 1620 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1621 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1622 1623 DPRINTF("segment %d addr %p len %zd\n", i, 1624 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1625 1626 /* Set up second segment for compare operation. */ 1627 if (lbalen->flags & CTL_LLF_COMPARE) { 1628 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1629 beio->sg_segs[i].len; 1630 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1631 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1632 } 1633 1634 beio->num_segs++; 1635 len_left -= beio->sg_segs[i].len; 1636 } 1637 if (bptrlen->len < lbalen->len) 1638 beio->beio_cont = ctl_be_block_next; 1639 io->scsiio.be_move_done = ctl_be_block_move_done; 1640 /* For compare we have separate S/G lists for read and datamove. */ 1641 if (lbalen->flags & CTL_LLF_COMPARE) 1642 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1643 else 1644 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1645 io->scsiio.kern_data_len = beio->io_len; 1646 io->scsiio.kern_data_resid = 0; 1647 io->scsiio.kern_sg_entries = beio->num_segs; 1648 io->io_hdr.flags |= CTL_FLAG_ALLOCATED; 1649 1650 /* 1651 * For the read case, we need to read the data into our buffers and 1652 * then we can send it back to the user. For the write case, we 1653 * need to get the data from the user first. 1654 */ 1655 if (beio->bio_cmd == BIO_READ) { 1656 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1657 be_lun->dispatch(be_lun, beio); 1658 } else { 1659 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1660 #ifdef CTL_TIME_IO 1661 getbintime(&io->io_hdr.dma_start_bt); 1662 #endif 1663 ctl_datamove(io); 1664 } 1665 } 1666 1667 static void 1668 ctl_be_block_worker(void *context, int pending) 1669 { 1670 struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; 1671 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1672 union ctl_io *io; 1673 struct ctl_be_block_io *beio; 1674 1675 DPRINTF("entered\n"); 1676 /* 1677 * Fetch and process I/Os from all queues. If we detect LUN 1678 * CTL_LUN_FLAG_OFFLINE status here -- it is result of a race, 1679 * so make response maximally opaque to not confuse initiator. 1680 */ 1681 for (;;) { 1682 mtx_lock(&be_lun->queue_lock); 1683 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1684 if (io != NULL) { 1685 DPRINTF("datamove queue\n"); 1686 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1687 ctl_io_hdr, links); 1688 mtx_unlock(&be_lun->queue_lock); 1689 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1690 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) { 1691 ctl_set_busy(&io->scsiio); 1692 ctl_complete_beio(beio); 1693 return; 1694 } 1695 be_lun->dispatch(be_lun, beio); 1696 continue; 1697 } 1698 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1699 if (io != NULL) { 1700 DPRINTF("config write queue\n"); 1701 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1702 ctl_io_hdr, links); 1703 mtx_unlock(&be_lun->queue_lock); 1704 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) { 1705 ctl_set_busy(&io->scsiio); 1706 ctl_config_write_done(io); 1707 return; 1708 } 1709 ctl_be_block_cw_dispatch(be_lun, io); 1710 continue; 1711 } 1712 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1713 if (io != NULL) { 1714 DPRINTF("config read queue\n"); 1715 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1716 ctl_io_hdr, links); 1717 mtx_unlock(&be_lun->queue_lock); 1718 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) { 1719 ctl_set_busy(&io->scsiio); 1720 ctl_config_read_done(io); 1721 return; 1722 } 1723 ctl_be_block_cr_dispatch(be_lun, io); 1724 continue; 1725 } 1726 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1727 if (io != NULL) { 1728 DPRINTF("input queue\n"); 1729 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1730 ctl_io_hdr, links); 1731 mtx_unlock(&be_lun->queue_lock); 1732 if (cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) { 1733 ctl_set_busy(&io->scsiio); 1734 ctl_data_submit_done(io); 1735 return; 1736 } 1737 ctl_be_block_dispatch(be_lun, io); 1738 continue; 1739 } 1740 1741 /* 1742 * If we get here, there is no work left in the queues, so 1743 * just break out and let the task queue go to sleep. 1744 */ 1745 mtx_unlock(&be_lun->queue_lock); 1746 break; 1747 } 1748 } 1749 1750 /* 1751 * Entry point from CTL to the backend for I/O. We queue everything to a 1752 * work thread, so this just puts the I/O on a queue and wakes up the 1753 * thread. 1754 */ 1755 static int 1756 ctl_be_block_submit(union ctl_io *io) 1757 { 1758 struct ctl_be_block_lun *be_lun; 1759 struct ctl_be_lun *cbe_lun; 1760 1761 DPRINTF("entered\n"); 1762 1763 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1764 CTL_PRIV_BACKEND_LUN].ptr; 1765 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 1766 1767 /* 1768 * Make sure we only get SCSI I/O. 1769 */ 1770 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1771 "%#x) encountered", io->io_hdr.io_type)); 1772 1773 PRIV(io)->len = 0; 1774 1775 mtx_lock(&be_lun->queue_lock); 1776 /* 1777 * XXX KDM make sure that links is okay to use at this point. 1778 * Otherwise, we either need to add another field to ctl_io_hdr, 1779 * or deal with resource allocation here. 1780 */ 1781 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1782 mtx_unlock(&be_lun->queue_lock); 1783 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1784 1785 return (CTL_RETVAL_COMPLETE); 1786 } 1787 1788 static int 1789 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1790 int flag, struct thread *td) 1791 { 1792 struct ctl_be_block_softc *softc; 1793 int error; 1794 1795 softc = &backend_block_softc; 1796 1797 error = 0; 1798 1799 switch (cmd) { 1800 case CTL_LUN_REQ: { 1801 struct ctl_lun_req *lun_req; 1802 1803 lun_req = (struct ctl_lun_req *)addr; 1804 1805 switch (lun_req->reqtype) { 1806 case CTL_LUNREQ_CREATE: 1807 error = ctl_be_block_create(softc, lun_req); 1808 break; 1809 case CTL_LUNREQ_RM: 1810 error = ctl_be_block_rm(softc, lun_req); 1811 break; 1812 case CTL_LUNREQ_MODIFY: 1813 error = ctl_be_block_modify(softc, lun_req); 1814 break; 1815 default: 1816 lun_req->status = CTL_LUN_ERROR; 1817 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1818 "invalid LUN request type %d", 1819 lun_req->reqtype); 1820 break; 1821 } 1822 break; 1823 } 1824 default: 1825 error = ENOTTY; 1826 break; 1827 } 1828 1829 return (error); 1830 } 1831 1832 static int 1833 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1834 { 1835 struct ctl_be_lun *cbe_lun; 1836 struct ctl_be_block_filedata *file_data; 1837 struct ctl_lun_create_params *params; 1838 char *value; 1839 struct vattr vattr; 1840 off_t ps, pss, po, pos, us, uss, uo, uos; 1841 int error; 1842 1843 error = 0; 1844 cbe_lun = &be_lun->cbe_lun; 1845 file_data = &be_lun->backend.file; 1846 params = &be_lun->params; 1847 1848 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1849 be_lun->dispatch = ctl_be_block_dispatch_file; 1850 be_lun->lun_flush = ctl_be_block_flush_file; 1851 be_lun->get_lba_status = ctl_be_block_gls_file; 1852 be_lun->getattr = ctl_be_block_getattr_file; 1853 be_lun->unmap = NULL; 1854 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 1855 1856 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1857 if (error != 0) { 1858 snprintf(req->error_str, sizeof(req->error_str), 1859 "error calling VOP_GETATTR() for file %s", 1860 be_lun->dev_path); 1861 return (error); 1862 } 1863 1864 /* 1865 * Verify that we have the ability to upgrade to exclusive 1866 * access on this file so we can trap errors at open instead 1867 * of reporting them during first access. 1868 */ 1869 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1870 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1871 if (be_lun->vn->v_iflag & VI_DOOMED) { 1872 error = EBADF; 1873 snprintf(req->error_str, sizeof(req->error_str), 1874 "error locking file %s", be_lun->dev_path); 1875 return (error); 1876 } 1877 } 1878 1879 file_data->cred = crhold(curthread->td_ucred); 1880 if (params->lun_size_bytes != 0) 1881 be_lun->size_bytes = params->lun_size_bytes; 1882 else 1883 be_lun->size_bytes = vattr.va_size; 1884 1885 /* 1886 * For files we can use any logical block size. Prefer 512 bytes 1887 * for compatibility reasons. If file's vattr.va_blocksize 1888 * (preferred I/O block size) is bigger and multiple to chosen 1889 * logical block size -- report it as physical block size. 1890 */ 1891 if (params->blocksize_bytes != 0) 1892 cbe_lun->blocksize = params->blocksize_bytes; 1893 else 1894 cbe_lun->blocksize = 512; 1895 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 1896 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 1897 0 : (be_lun->size_blocks - 1); 1898 1899 us = ps = vattr.va_blocksize; 1900 uo = po = 0; 1901 1902 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 1903 if (value != NULL) 1904 ctl_expand_number(value, &ps); 1905 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 1906 if (value != NULL) 1907 ctl_expand_number(value, &po); 1908 pss = ps / cbe_lun->blocksize; 1909 pos = po / cbe_lun->blocksize; 1910 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 1911 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 1912 cbe_lun->pblockexp = fls(pss) - 1; 1913 cbe_lun->pblockoff = (pss - pos) % pss; 1914 } 1915 1916 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 1917 if (value != NULL) 1918 ctl_expand_number(value, &us); 1919 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 1920 if (value != NULL) 1921 ctl_expand_number(value, &uo); 1922 uss = us / cbe_lun->blocksize; 1923 uos = uo / cbe_lun->blocksize; 1924 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 1925 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 1926 cbe_lun->ublockexp = fls(uss) - 1; 1927 cbe_lun->ublockoff = (uss - uos) % uss; 1928 } 1929 1930 /* 1931 * Sanity check. The media size has to be at least one 1932 * sector long. 1933 */ 1934 if (be_lun->size_bytes < cbe_lun->blocksize) { 1935 error = EINVAL; 1936 snprintf(req->error_str, sizeof(req->error_str), 1937 "file %s size %ju < block size %u", be_lun->dev_path, 1938 (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); 1939 } 1940 1941 cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; 1942 return (error); 1943 } 1944 1945 static int 1946 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1947 { 1948 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 1949 struct ctl_lun_create_params *params; 1950 struct cdevsw *csw; 1951 struct cdev *dev; 1952 char *value; 1953 int error, atomic, maxio, ref, unmap, tmp; 1954 off_t ps, pss, po, pos, us, uss, uo, uos, otmp; 1955 1956 params = &be_lun->params; 1957 1958 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1959 csw = devvn_refthread(be_lun->vn, &dev, &ref); 1960 if (csw == NULL) 1961 return (ENXIO); 1962 if (strcmp(csw->d_name, "zvol") == 0) { 1963 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1964 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1965 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1966 } else { 1967 be_lun->dispatch = ctl_be_block_dispatch_dev; 1968 be_lun->get_lba_status = NULL; 1969 atomic = 0; 1970 maxio = dev->si_iosize_max; 1971 if (maxio <= 0) 1972 maxio = DFLTPHYS; 1973 if (maxio > CTLBLK_MAX_IO_SIZE) 1974 maxio = CTLBLK_MAX_IO_SIZE; 1975 } 1976 be_lun->lun_flush = ctl_be_block_flush_dev; 1977 be_lun->getattr = ctl_be_block_getattr_dev; 1978 be_lun->unmap = ctl_be_block_unmap_dev; 1979 1980 if (!csw->d_ioctl) { 1981 dev_relthread(dev, ref); 1982 snprintf(req->error_str, sizeof(req->error_str), 1983 "no d_ioctl for device %s!", be_lun->dev_path); 1984 return (ENODEV); 1985 } 1986 1987 error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, 1988 curthread); 1989 if (error) { 1990 dev_relthread(dev, ref); 1991 snprintf(req->error_str, sizeof(req->error_str), 1992 "error %d returned for DIOCGSECTORSIZE ioctl " 1993 "on %s!", error, be_lun->dev_path); 1994 return (error); 1995 } 1996 1997 /* 1998 * If the user has asked for a blocksize that is greater than the 1999 * backing device's blocksize, we can do it only if the blocksize 2000 * the user is asking for is an even multiple of the underlying 2001 * device's blocksize. 2002 */ 2003 if ((params->blocksize_bytes != 0) && 2004 (params->blocksize_bytes >= tmp)) { 2005 if (params->blocksize_bytes % tmp == 0) { 2006 cbe_lun->blocksize = params->blocksize_bytes; 2007 } else { 2008 dev_relthread(dev, ref); 2009 snprintf(req->error_str, sizeof(req->error_str), 2010 "requested blocksize %u is not an even " 2011 "multiple of backing device blocksize %u", 2012 params->blocksize_bytes, tmp); 2013 return (EINVAL); 2014 } 2015 } else if (params->blocksize_bytes != 0) { 2016 dev_relthread(dev, ref); 2017 snprintf(req->error_str, sizeof(req->error_str), 2018 "requested blocksize %u < backing device " 2019 "blocksize %u", params->blocksize_bytes, tmp); 2020 return (EINVAL); 2021 } else 2022 cbe_lun->blocksize = tmp; 2023 2024 error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, 2025 curthread); 2026 if (error) { 2027 dev_relthread(dev, ref); 2028 snprintf(req->error_str, sizeof(req->error_str), 2029 "error %d returned for DIOCGMEDIASIZE " 2030 " ioctl on %s!", error, 2031 be_lun->dev_path); 2032 return (error); 2033 } 2034 2035 if (params->lun_size_bytes != 0) { 2036 if (params->lun_size_bytes > otmp) { 2037 dev_relthread(dev, ref); 2038 snprintf(req->error_str, sizeof(req->error_str), 2039 "requested LUN size %ju > backing device " 2040 "size %ju", 2041 (uintmax_t)params->lun_size_bytes, 2042 (uintmax_t)otmp); 2043 return (EINVAL); 2044 } 2045 2046 be_lun->size_bytes = params->lun_size_bytes; 2047 } else 2048 be_lun->size_bytes = otmp; 2049 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2050 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2051 0 : (be_lun->size_blocks - 1); 2052 2053 error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, 2054 curthread); 2055 if (error) 2056 ps = po = 0; 2057 else { 2058 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, 2059 FREAD, curthread); 2060 if (error) 2061 po = 0; 2062 } 2063 us = ps; 2064 uo = po; 2065 2066 value = ctl_get_opt(&cbe_lun->options, "pblocksize"); 2067 if (value != NULL) 2068 ctl_expand_number(value, &ps); 2069 value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); 2070 if (value != NULL) 2071 ctl_expand_number(value, &po); 2072 pss = ps / cbe_lun->blocksize; 2073 pos = po / cbe_lun->blocksize; 2074 if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && 2075 ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { 2076 cbe_lun->pblockexp = fls(pss) - 1; 2077 cbe_lun->pblockoff = (pss - pos) % pss; 2078 } 2079 2080 value = ctl_get_opt(&cbe_lun->options, "ublocksize"); 2081 if (value != NULL) 2082 ctl_expand_number(value, &us); 2083 value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); 2084 if (value != NULL) 2085 ctl_expand_number(value, &uo); 2086 uss = us / cbe_lun->blocksize; 2087 uos = uo / cbe_lun->blocksize; 2088 if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && 2089 ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { 2090 cbe_lun->ublockexp = fls(uss) - 1; 2091 cbe_lun->ublockoff = (uss - uos) % uss; 2092 } 2093 2094 cbe_lun->atomicblock = atomic / cbe_lun->blocksize; 2095 cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; 2096 2097 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2098 unmap = 1; 2099 } else { 2100 struct diocgattr_arg arg; 2101 2102 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2103 arg.len = sizeof(arg.value.i); 2104 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, 2105 curthread); 2106 unmap = (error == 0) ? arg.value.i : 0; 2107 } 2108 value = ctl_get_opt(&cbe_lun->options, "unmap"); 2109 if (value != NULL) 2110 unmap = (strcmp(value, "on") == 0); 2111 if (unmap) 2112 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; 2113 else 2114 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; 2115 2116 dev_relthread(dev, ref); 2117 return (0); 2118 } 2119 2120 static int 2121 ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2122 { 2123 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2124 int flags; 2125 2126 if (be_lun->vn) { 2127 flags = FREAD; 2128 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) 2129 flags |= FWRITE; 2130 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2131 be_lun->vn = NULL; 2132 2133 switch (be_lun->dev_type) { 2134 case CTL_BE_BLOCK_DEV: 2135 break; 2136 case CTL_BE_BLOCK_FILE: 2137 if (be_lun->backend.file.cred != NULL) { 2138 crfree(be_lun->backend.file.cred); 2139 be_lun->backend.file.cred = NULL; 2140 } 2141 break; 2142 case CTL_BE_BLOCK_NONE: 2143 break; 2144 default: 2145 panic("Unexpected backend type."); 2146 break; 2147 } 2148 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2149 } 2150 return (0); 2151 } 2152 2153 static int 2154 ctl_be_block_open(struct ctl_be_block_softc *softc, 2155 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2156 { 2157 struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; 2158 struct nameidata nd; 2159 char *value; 2160 int error, flags; 2161 2162 error = 0; 2163 if (rootvnode == NULL) { 2164 snprintf(req->error_str, sizeof(req->error_str), 2165 "Root filesystem is not mounted"); 2166 return (1); 2167 } 2168 pwd_ensure_dirs(); 2169 2170 value = ctl_get_opt(&cbe_lun->options, "file"); 2171 if (value == NULL) { 2172 snprintf(req->error_str, sizeof(req->error_str), 2173 "no file argument specified"); 2174 return (1); 2175 } 2176 free(be_lun->dev_path, M_CTLBLK); 2177 be_lun->dev_path = strdup(value, M_CTLBLK); 2178 2179 flags = FREAD; 2180 value = ctl_get_opt(&cbe_lun->options, "readonly"); 2181 if (value == NULL || strcmp(value, "on") != 0) 2182 flags |= FWRITE; 2183 2184 again: 2185 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2186 error = vn_open(&nd, &flags, 0, NULL); 2187 if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { 2188 flags &= ~FWRITE; 2189 goto again; 2190 } 2191 if (error) { 2192 /* 2193 * This is the only reasonable guess we can make as far as 2194 * path if the user doesn't give us a fully qualified path. 2195 * If they want to specify a file, they need to specify the 2196 * full path. 2197 */ 2198 if (be_lun->dev_path[0] != '/') { 2199 char *dev_name; 2200 2201 asprintf(&dev_name, M_CTLBLK, "/dev/%s", 2202 be_lun->dev_path); 2203 free(be_lun->dev_path, M_CTLBLK); 2204 be_lun->dev_path = dev_name; 2205 goto again; 2206 } 2207 snprintf(req->error_str, sizeof(req->error_str), 2208 "error opening %s: %d", be_lun->dev_path, error); 2209 return (error); 2210 } 2211 if (flags & FWRITE) 2212 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; 2213 else 2214 cbe_lun->flags |= CTL_LUN_FLAG_READONLY; 2215 2216 NDFREE(&nd, NDF_ONLY_PNBUF); 2217 be_lun->vn = nd.ni_vp; 2218 2219 /* We only support disks and files. */ 2220 if (vn_isdisk(be_lun->vn, &error)) { 2221 error = ctl_be_block_open_dev(be_lun, req); 2222 } else if (be_lun->vn->v_type == VREG) { 2223 error = ctl_be_block_open_file(be_lun, req); 2224 } else { 2225 error = EINVAL; 2226 snprintf(req->error_str, sizeof(req->error_str), 2227 "%s is not a disk or plain file", be_lun->dev_path); 2228 } 2229 VOP_UNLOCK(be_lun->vn, 0); 2230 2231 if (error != 0) 2232 ctl_be_block_close(be_lun); 2233 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2234 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2235 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2236 value = ctl_get_opt(&cbe_lun->options, "serseq"); 2237 if (value != NULL && strcmp(value, "on") == 0) 2238 cbe_lun->serseq = CTL_LUN_SERSEQ_ON; 2239 else if (value != NULL && strcmp(value, "read") == 0) 2240 cbe_lun->serseq = CTL_LUN_SERSEQ_READ; 2241 else if (value != NULL && strcmp(value, "off") == 0) 2242 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; 2243 return (0); 2244 } 2245 2246 static int 2247 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2248 { 2249 struct ctl_be_lun *cbe_lun; 2250 struct ctl_be_block_lun *be_lun; 2251 struct ctl_lun_create_params *params; 2252 char num_thread_str[16]; 2253 char tmpstr[32]; 2254 char *value; 2255 int retval, num_threads; 2256 int tmp_num_threads; 2257 2258 params = &req->reqdata.create; 2259 retval = 0; 2260 req->status = CTL_LUN_OK; 2261 2262 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2263 cbe_lun = &be_lun->cbe_lun; 2264 cbe_lun->be_lun = be_lun; 2265 be_lun->params = req->reqdata.create; 2266 be_lun->softc = softc; 2267 STAILQ_INIT(&be_lun->input_queue); 2268 STAILQ_INIT(&be_lun->config_read_queue); 2269 STAILQ_INIT(&be_lun->config_write_queue); 2270 STAILQ_INIT(&be_lun->datamove_queue); 2271 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2272 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2273 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2274 ctl_init_opts(&cbe_lun->options, 2275 req->num_be_args, req->kern_be_args); 2276 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2277 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2278 if (be_lun->lun_zone == NULL) { 2279 snprintf(req->error_str, sizeof(req->error_str), 2280 "error allocating UMA zone"); 2281 goto bailout_error; 2282 } 2283 2284 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2285 cbe_lun->lun_type = params->device_type; 2286 else 2287 cbe_lun->lun_type = T_DIRECT; 2288 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2289 cbe_lun->flags = 0; 2290 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2291 if (value != NULL) { 2292 if (strcmp(value, "primary") == 0) 2293 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2294 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2295 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2296 2297 if (cbe_lun->lun_type == T_DIRECT) { 2298 be_lun->size_bytes = params->lun_size_bytes; 2299 if (params->blocksize_bytes != 0) 2300 cbe_lun->blocksize = params->blocksize_bytes; 2301 else 2302 cbe_lun->blocksize = 512; 2303 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; 2304 cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 2305 0 : (be_lun->size_blocks - 1); 2306 2307 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2308 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2309 retval = ctl_be_block_open(softc, be_lun, req); 2310 if (retval != 0) { 2311 retval = 0; 2312 req->status = CTL_LUN_WARNING; 2313 } 2314 } 2315 num_threads = cbb_num_threads; 2316 } else { 2317 num_threads = 1; 2318 } 2319 2320 /* 2321 * XXX This searching loop might be refactored to be combined with 2322 * the loop above, 2323 */ 2324 value = ctl_get_opt(&cbe_lun->options, "num_threads"); 2325 if (value != NULL) { 2326 tmp_num_threads = strtol(value, NULL, 0); 2327 2328 /* 2329 * We don't let the user specify less than one 2330 * thread, but hope he's clueful enough not to 2331 * specify 1000 threads. 2332 */ 2333 if (tmp_num_threads < 1) { 2334 snprintf(req->error_str, sizeof(req->error_str), 2335 "invalid number of threads %s", 2336 num_thread_str); 2337 goto bailout_error; 2338 } 2339 num_threads = tmp_num_threads; 2340 } 2341 2342 if (be_lun->vn == NULL) 2343 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE; 2344 /* Tell the user the blocksize we ended up using */ 2345 params->lun_size_bytes = be_lun->size_bytes; 2346 params->blocksize_bytes = cbe_lun->blocksize; 2347 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2348 cbe_lun->req_lun_id = params->req_lun_id; 2349 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; 2350 } else 2351 cbe_lun->req_lun_id = 0; 2352 2353 cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; 2354 cbe_lun->lun_config_status = ctl_be_block_lun_config_status; 2355 cbe_lun->be = &ctl_be_block_driver; 2356 2357 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2358 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2359 softc->num_luns); 2360 strncpy((char *)cbe_lun->serial_num, tmpstr, 2361 MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); 2362 2363 /* Tell the user what we used for a serial number */ 2364 strncpy((char *)params->serial_num, tmpstr, 2365 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2366 } else { 2367 strncpy((char *)cbe_lun->serial_num, params->serial_num, 2368 MIN(sizeof(cbe_lun->serial_num), 2369 sizeof(params->serial_num))); 2370 } 2371 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2372 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2373 strncpy((char *)cbe_lun->device_id, tmpstr, 2374 MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); 2375 2376 /* Tell the user what we used for a device ID */ 2377 strncpy((char *)params->device_id, tmpstr, 2378 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2379 } else { 2380 strncpy((char *)cbe_lun->device_id, params->device_id, 2381 MIN(sizeof(cbe_lun->device_id), 2382 sizeof(params->device_id))); 2383 } 2384 2385 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2386 2387 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2388 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2389 2390 if (be_lun->io_taskqueue == NULL) { 2391 snprintf(req->error_str, sizeof(req->error_str), 2392 "unable to create taskqueue"); 2393 goto bailout_error; 2394 } 2395 2396 /* 2397 * Note that we start the same number of threads by default for 2398 * both the file case and the block device case. For the file 2399 * case, we need multiple threads to allow concurrency, because the 2400 * vnode interface is designed to be a blocking interface. For the 2401 * block device case, ZFS zvols at least will block the caller's 2402 * context in many instances, and so we need multiple threads to 2403 * overcome that problem. Other block devices don't need as many 2404 * threads, but they shouldn't cause too many problems. 2405 * 2406 * If the user wants to just have a single thread for a block 2407 * device, he can specify that when the LUN is created, or change 2408 * the tunable/sysctl to alter the default number of threads. 2409 */ 2410 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2411 /*num threads*/num_threads, 2412 /*priority*/PWAIT, 2413 /*thread name*/ 2414 "%s taskq", be_lun->lunname); 2415 2416 if (retval != 0) 2417 goto bailout_error; 2418 2419 be_lun->num_threads = num_threads; 2420 2421 mtx_lock(&softc->lock); 2422 softc->num_luns++; 2423 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2424 2425 mtx_unlock(&softc->lock); 2426 2427 retval = ctl_add_lun(&be_lun->cbe_lun); 2428 if (retval != 0) { 2429 mtx_lock(&softc->lock); 2430 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2431 links); 2432 softc->num_luns--; 2433 mtx_unlock(&softc->lock); 2434 snprintf(req->error_str, sizeof(req->error_str), 2435 "ctl_add_lun() returned error %d, see dmesg for " 2436 "details", retval); 2437 retval = 0; 2438 goto bailout_error; 2439 } 2440 2441 mtx_lock(&softc->lock); 2442 2443 /* 2444 * Tell the config_status routine that we're waiting so it won't 2445 * clean up the LUN in the event of an error. 2446 */ 2447 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2448 2449 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2450 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2451 if (retval == EINTR) 2452 break; 2453 } 2454 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2455 2456 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2457 snprintf(req->error_str, sizeof(req->error_str), 2458 "LUN configuration error, see dmesg for details"); 2459 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2460 links); 2461 softc->num_luns--; 2462 mtx_unlock(&softc->lock); 2463 goto bailout_error; 2464 } else { 2465 params->req_lun_id = cbe_lun->lun_id; 2466 } 2467 2468 mtx_unlock(&softc->lock); 2469 2470 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2471 cbe_lun->blocksize, 2472 DEVSTAT_ALL_SUPPORTED, 2473 cbe_lun->lun_type 2474 | DEVSTAT_TYPE_IF_OTHER, 2475 DEVSTAT_PRIORITY_OTHER); 2476 2477 return (retval); 2478 2479 bailout_error: 2480 req->status = CTL_LUN_ERROR; 2481 2482 if (be_lun->io_taskqueue != NULL) 2483 taskqueue_free(be_lun->io_taskqueue); 2484 ctl_be_block_close(be_lun); 2485 if (be_lun->dev_path != NULL) 2486 free(be_lun->dev_path, M_CTLBLK); 2487 if (be_lun->lun_zone != NULL) 2488 uma_zdestroy(be_lun->lun_zone); 2489 ctl_free_opts(&cbe_lun->options); 2490 mtx_destroy(&be_lun->queue_lock); 2491 mtx_destroy(&be_lun->io_lock); 2492 free(be_lun, M_CTLBLK); 2493 2494 return (retval); 2495 } 2496 2497 static int 2498 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2499 { 2500 struct ctl_lun_rm_params *params; 2501 struct ctl_be_block_lun *be_lun; 2502 struct ctl_be_lun *cbe_lun; 2503 int retval; 2504 2505 params = &req->reqdata.rm; 2506 2507 mtx_lock(&softc->lock); 2508 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2509 if (be_lun->cbe_lun.lun_id == params->lun_id) 2510 break; 2511 } 2512 mtx_unlock(&softc->lock); 2513 2514 if (be_lun == NULL) { 2515 snprintf(req->error_str, sizeof(req->error_str), 2516 "LUN %u is not managed by the block backend", 2517 params->lun_id); 2518 goto bailout_error; 2519 } 2520 cbe_lun = &be_lun->cbe_lun; 2521 2522 retval = ctl_disable_lun(cbe_lun); 2523 if (retval != 0) { 2524 snprintf(req->error_str, sizeof(req->error_str), 2525 "error %d returned from ctl_disable_lun() for " 2526 "LUN %d", retval, params->lun_id); 2527 goto bailout_error; 2528 } 2529 2530 if (be_lun->vn != NULL) { 2531 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE; 2532 ctl_lun_offline(cbe_lun); 2533 taskqueue_drain_all(be_lun->io_taskqueue); 2534 ctl_be_block_close(be_lun); 2535 } 2536 2537 retval = ctl_invalidate_lun(cbe_lun); 2538 if (retval != 0) { 2539 snprintf(req->error_str, sizeof(req->error_str), 2540 "error %d returned from ctl_invalidate_lun() for " 2541 "LUN %d", retval, params->lun_id); 2542 goto bailout_error; 2543 } 2544 2545 mtx_lock(&softc->lock); 2546 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2547 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2548 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2549 if (retval == EINTR) 2550 break; 2551 } 2552 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2553 2554 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2555 snprintf(req->error_str, sizeof(req->error_str), 2556 "interrupted waiting for LUN to be freed"); 2557 mtx_unlock(&softc->lock); 2558 goto bailout_error; 2559 } 2560 2561 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2562 2563 softc->num_luns--; 2564 mtx_unlock(&softc->lock); 2565 2566 taskqueue_drain_all(be_lun->io_taskqueue); 2567 taskqueue_free(be_lun->io_taskqueue); 2568 2569 if (be_lun->disk_stats != NULL) 2570 devstat_remove_entry(be_lun->disk_stats); 2571 2572 uma_zdestroy(be_lun->lun_zone); 2573 2574 ctl_free_opts(&cbe_lun->options); 2575 free(be_lun->dev_path, M_CTLBLK); 2576 mtx_destroy(&be_lun->queue_lock); 2577 mtx_destroy(&be_lun->io_lock); 2578 free(be_lun, M_CTLBLK); 2579 2580 req->status = CTL_LUN_OK; 2581 2582 return (0); 2583 2584 bailout_error: 2585 2586 req->status = CTL_LUN_ERROR; 2587 2588 return (0); 2589 } 2590 2591 static int 2592 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2593 { 2594 struct ctl_lun_modify_params *params; 2595 struct ctl_be_block_lun *be_lun; 2596 struct ctl_be_lun *cbe_lun; 2597 char *value; 2598 uint64_t oldsize; 2599 int error, wasprim; 2600 2601 params = &req->reqdata.modify; 2602 2603 mtx_lock(&softc->lock); 2604 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2605 if (be_lun->cbe_lun.lun_id == params->lun_id) 2606 break; 2607 } 2608 mtx_unlock(&softc->lock); 2609 2610 if (be_lun == NULL) { 2611 snprintf(req->error_str, sizeof(req->error_str), 2612 "LUN %u is not managed by the block backend", 2613 params->lun_id); 2614 goto bailout_error; 2615 } 2616 cbe_lun = &be_lun->cbe_lun; 2617 2618 if (params->lun_size_bytes != 0) 2619 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2620 ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); 2621 2622 wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); 2623 value = ctl_get_opt(&cbe_lun->options, "ha_role"); 2624 if (value != NULL) { 2625 if (strcmp(value, "primary") == 0) 2626 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2627 else 2628 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2629 } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) 2630 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; 2631 else 2632 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; 2633 if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { 2634 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) 2635 ctl_lun_primary(cbe_lun); 2636 else 2637 ctl_lun_secondary(cbe_lun); 2638 } 2639 2640 oldsize = be_lun->size_blocks; 2641 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || 2642 control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { 2643 if (be_lun->vn == NULL) 2644 error = ctl_be_block_open(softc, be_lun, req); 2645 else if (vn_isdisk(be_lun->vn, &error)) 2646 error = ctl_be_block_open_dev(be_lun, req); 2647 else if (be_lun->vn->v_type == VREG) 2648 error = ctl_be_block_open_file(be_lun, req); 2649 else 2650 error = EINVAL; 2651 if ((cbe_lun->flags & CTL_LUN_FLAG_OFFLINE) && 2652 be_lun->vn != NULL) { 2653 cbe_lun->flags &= ~CTL_LUN_FLAG_OFFLINE; 2654 ctl_lun_online(cbe_lun); 2655 } 2656 } else { 2657 if (be_lun->vn != NULL) { 2658 cbe_lun->flags |= CTL_LUN_FLAG_OFFLINE; 2659 ctl_lun_offline(cbe_lun); 2660 taskqueue_drain_all(be_lun->io_taskqueue); 2661 error = ctl_be_block_close(be_lun); 2662 } else 2663 error = 0; 2664 } 2665 if (be_lun->size_blocks != oldsize) 2666 ctl_lun_capacity_changed(cbe_lun); 2667 2668 /* Tell the user the exact size we ended up using */ 2669 params->lun_size_bytes = be_lun->size_bytes; 2670 2671 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2672 return (0); 2673 2674 bailout_error: 2675 req->status = CTL_LUN_ERROR; 2676 return (0); 2677 } 2678 2679 static void 2680 ctl_be_block_lun_shutdown(void *be_lun) 2681 { 2682 struct ctl_be_block_lun *lun; 2683 struct ctl_be_block_softc *softc; 2684 2685 lun = (struct ctl_be_block_lun *)be_lun; 2686 2687 softc = lun->softc; 2688 2689 mtx_lock(&softc->lock); 2690 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2691 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2692 wakeup(lun); 2693 mtx_unlock(&softc->lock); 2694 2695 } 2696 2697 static void 2698 ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2699 { 2700 struct ctl_be_block_lun *lun; 2701 struct ctl_be_block_softc *softc; 2702 2703 lun = (struct ctl_be_block_lun *)be_lun; 2704 softc = lun->softc; 2705 2706 if (status == CTL_LUN_CONFIG_OK) { 2707 mtx_lock(&softc->lock); 2708 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2709 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2710 wakeup(lun); 2711 mtx_unlock(&softc->lock); 2712 2713 /* 2714 * We successfully added the LUN, attempt to enable it. 2715 */ 2716 if (ctl_enable_lun(&lun->cbe_lun) != 0) { 2717 printf("%s: ctl_enable_lun() failed!\n", __func__); 2718 if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { 2719 printf("%s: ctl_invalidate_lun() failed!\n", 2720 __func__); 2721 } 2722 } 2723 2724 return; 2725 } 2726 2727 2728 mtx_lock(&softc->lock); 2729 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2730 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2731 wakeup(lun); 2732 mtx_unlock(&softc->lock); 2733 } 2734 2735 2736 static int 2737 ctl_be_block_config_write(union ctl_io *io) 2738 { 2739 struct ctl_be_block_lun *be_lun; 2740 struct ctl_be_lun *cbe_lun; 2741 int retval; 2742 2743 retval = 0; 2744 2745 DPRINTF("entered\n"); 2746 2747 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2748 CTL_PRIV_BACKEND_LUN].ptr; 2749 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2750 2751 switch (io->scsiio.cdb[0]) { 2752 case SYNCHRONIZE_CACHE: 2753 case SYNCHRONIZE_CACHE_16: 2754 case WRITE_SAME_10: 2755 case WRITE_SAME_16: 2756 case UNMAP: 2757 /* 2758 * The upper level CTL code will filter out any CDBs with 2759 * the immediate bit set and return the proper error. 2760 * 2761 * We don't really need to worry about what LBA range the 2762 * user asked to be synced out. When they issue a sync 2763 * cache command, we'll sync out the whole thing. 2764 */ 2765 mtx_lock(&be_lun->queue_lock); 2766 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2767 links); 2768 mtx_unlock(&be_lun->queue_lock); 2769 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2770 break; 2771 case START_STOP_UNIT: { 2772 struct scsi_start_stop_unit *cdb; 2773 2774 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2775 2776 if (cdb->how & SSS_START) 2777 retval = ctl_start_lun(cbe_lun); 2778 else { 2779 retval = ctl_stop_lun(cbe_lun); 2780 /* 2781 * XXX KDM Copan-specific offline behavior. 2782 * Figure out a reasonable way to port this? 2783 */ 2784 #ifdef NEEDTOPORT 2785 if ((retval == 0) 2786 && (cdb->byte2 & SSS_ONOFFLINE)) 2787 retval = ctl_lun_offline(cbe_lun); 2788 #endif 2789 } 2790 2791 /* 2792 * In general, the above routines should not fail. They 2793 * just set state for the LUN. So we've got something 2794 * pretty wrong here if we can't start or stop the LUN. 2795 */ 2796 if (retval != 0) { 2797 ctl_set_internal_failure(&io->scsiio, 2798 /*sks_valid*/ 1, 2799 /*retry_count*/ 0xf051); 2800 retval = CTL_RETVAL_COMPLETE; 2801 } else { 2802 ctl_set_success(&io->scsiio); 2803 } 2804 ctl_config_write_done(io); 2805 break; 2806 } 2807 default: 2808 ctl_set_invalid_opcode(&io->scsiio); 2809 ctl_config_write_done(io); 2810 retval = CTL_RETVAL_COMPLETE; 2811 break; 2812 } 2813 2814 return (retval); 2815 } 2816 2817 static int 2818 ctl_be_block_config_read(union ctl_io *io) 2819 { 2820 struct ctl_be_block_lun *be_lun; 2821 struct ctl_be_lun *cbe_lun; 2822 int retval = 0; 2823 2824 DPRINTF("entered\n"); 2825 2826 cbe_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2827 CTL_PRIV_BACKEND_LUN].ptr; 2828 be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; 2829 2830 switch (io->scsiio.cdb[0]) { 2831 case SERVICE_ACTION_IN: 2832 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2833 mtx_lock(&be_lun->queue_lock); 2834 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2835 &io->io_hdr, links); 2836 mtx_unlock(&be_lun->queue_lock); 2837 taskqueue_enqueue(be_lun->io_taskqueue, 2838 &be_lun->io_task); 2839 retval = CTL_RETVAL_QUEUED; 2840 break; 2841 } 2842 ctl_set_invalid_field(&io->scsiio, 2843 /*sks_valid*/ 1, 2844 /*command*/ 1, 2845 /*field*/ 1, 2846 /*bit_valid*/ 1, 2847 /*bit*/ 4); 2848 ctl_config_read_done(io); 2849 retval = CTL_RETVAL_COMPLETE; 2850 break; 2851 default: 2852 ctl_set_invalid_opcode(&io->scsiio); 2853 ctl_config_read_done(io); 2854 retval = CTL_RETVAL_COMPLETE; 2855 break; 2856 } 2857 2858 return (retval); 2859 } 2860 2861 static int 2862 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2863 { 2864 struct ctl_be_block_lun *lun; 2865 int retval; 2866 2867 lun = (struct ctl_be_block_lun *)be_lun; 2868 retval = 0; 2869 2870 retval = sbuf_printf(sb, "\t<num_threads>"); 2871 2872 if (retval != 0) 2873 goto bailout; 2874 2875 retval = sbuf_printf(sb, "%d", lun->num_threads); 2876 2877 if (retval != 0) 2878 goto bailout; 2879 2880 retval = sbuf_printf(sb, "</num_threads>\n"); 2881 2882 bailout: 2883 2884 return (retval); 2885 } 2886 2887 static uint64_t 2888 ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2889 { 2890 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2891 2892 if (lun->getattr == NULL) 2893 return (UINT64_MAX); 2894 return (lun->getattr(lun, attrname)); 2895 } 2896 2897 int 2898 ctl_be_block_init(void) 2899 { 2900 struct ctl_be_block_softc *softc; 2901 int retval; 2902 2903 softc = &backend_block_softc; 2904 retval = 0; 2905 2906 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2907 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2908 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2909 STAILQ_INIT(&softc->lun_list); 2910 2911 return (retval); 2912 } 2913