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 uint16_t ublockexp; 177 uint16_t ublockoff; 178 uint32_t atomicblock; 179 uint32_t opttxferlen; 180 struct ctl_be_block_softc *softc; 181 struct devstat *disk_stats; 182 ctl_be_block_lun_flags flags; 183 STAILQ_ENTRY(ctl_be_block_lun) links; 184 struct ctl_be_lun ctl_be_lun; 185 struct taskqueue *io_taskqueue; 186 struct task io_task; 187 int num_threads; 188 STAILQ_HEAD(, ctl_io_hdr) input_queue; 189 STAILQ_HEAD(, ctl_io_hdr) config_read_queue; 190 STAILQ_HEAD(, ctl_io_hdr) config_write_queue; 191 STAILQ_HEAD(, ctl_io_hdr) datamove_queue; 192 struct mtx_padalign io_lock; 193 struct mtx_padalign queue_lock; 194 }; 195 196 /* 197 * Overall softc structure for the block backend module. 198 */ 199 struct ctl_be_block_softc { 200 struct mtx lock; 201 int num_disks; 202 STAILQ_HEAD(, ctl_block_disk) disk_list; 203 int num_luns; 204 STAILQ_HEAD(, ctl_be_block_lun) lun_list; 205 }; 206 207 static struct ctl_be_block_softc backend_block_softc; 208 209 /* 210 * Per-I/O information. 211 */ 212 struct ctl_be_block_io { 213 union ctl_io *io; 214 struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; 215 struct iovec xiovecs[CTLBLK_MAX_SEGS]; 216 int bio_cmd; 217 int num_segs; 218 int num_bios_sent; 219 int num_bios_done; 220 int send_complete; 221 int num_errors; 222 struct bintime ds_t0; 223 devstat_tag_type ds_tag_type; 224 devstat_trans_flags ds_trans_type; 225 uint64_t io_len; 226 uint64_t io_offset; 227 struct ctl_be_block_softc *softc; 228 struct ctl_be_block_lun *lun; 229 void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ 230 }; 231 232 static int cbb_num_threads = 14; 233 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, 234 "CAM Target Layer Block Backend"); 235 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, 236 &cbb_num_threads, 0, "Number of threads per backing file"); 237 238 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); 239 static void ctl_free_beio(struct ctl_be_block_io *beio); 240 static void ctl_complete_beio(struct ctl_be_block_io *beio); 241 static int ctl_be_block_move_done(union ctl_io *io); 242 static void ctl_be_block_biodone(struct bio *bio); 243 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 244 struct ctl_be_block_io *beio); 245 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 246 struct ctl_be_block_io *beio); 247 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 248 struct ctl_be_block_io *beio); 249 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, 250 const char *attrname); 251 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 252 struct ctl_be_block_io *beio); 253 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 254 struct ctl_be_block_io *beio); 255 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 256 struct ctl_be_block_io *beio); 257 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, 258 const char *attrname); 259 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 260 union ctl_io *io); 261 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 262 union ctl_io *io); 263 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 264 union ctl_io *io); 265 static void ctl_be_block_worker(void *context, int pending); 266 static int ctl_be_block_submit(union ctl_io *io); 267 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 268 int flag, struct thread *td); 269 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, 270 struct ctl_lun_req *req); 271 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, 272 struct ctl_lun_req *req); 273 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); 274 static int ctl_be_block_open(struct ctl_be_block_softc *softc, 275 struct ctl_be_block_lun *be_lun, 276 struct ctl_lun_req *req); 277 static int ctl_be_block_create(struct ctl_be_block_softc *softc, 278 struct ctl_lun_req *req); 279 static int ctl_be_block_rm(struct ctl_be_block_softc *softc, 280 struct ctl_lun_req *req); 281 static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 282 struct ctl_lun_req *req); 283 static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 284 struct ctl_lun_req *req); 285 static int ctl_be_block_modify(struct ctl_be_block_softc *softc, 286 struct ctl_lun_req *req); 287 static void ctl_be_block_lun_shutdown(void *be_lun); 288 static void ctl_be_block_lun_config_status(void *be_lun, 289 ctl_lun_config_status status); 290 static int ctl_be_block_config_write(union ctl_io *io); 291 static int ctl_be_block_config_read(union ctl_io *io); 292 static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); 293 static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); 294 int ctl_be_block_init(void); 295 296 static struct ctl_backend_driver ctl_be_block_driver = 297 { 298 .name = "block", 299 .flags = CTL_BE_FLAG_HAS_CONFIG, 300 .init = ctl_be_block_init, 301 .data_submit = ctl_be_block_submit, 302 .data_move_done = ctl_be_block_move_done, 303 .config_read = ctl_be_block_config_read, 304 .config_write = ctl_be_block_config_write, 305 .ioctl = ctl_be_block_ioctl, 306 .lun_info = ctl_be_block_lun_info, 307 .lun_attr = ctl_be_block_lun_attr 308 }; 309 310 MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); 311 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); 312 313 static uma_zone_t beio_zone; 314 315 static struct ctl_be_block_io * 316 ctl_alloc_beio(struct ctl_be_block_softc *softc) 317 { 318 struct ctl_be_block_io *beio; 319 320 beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO); 321 beio->softc = softc; 322 return (beio); 323 } 324 325 static void 326 ctl_free_beio(struct ctl_be_block_io *beio) 327 { 328 int duplicate_free; 329 int i; 330 331 duplicate_free = 0; 332 333 for (i = 0; i < beio->num_segs; i++) { 334 if (beio->sg_segs[i].addr == NULL) 335 duplicate_free++; 336 337 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); 338 beio->sg_segs[i].addr = NULL; 339 340 /* For compare we had two equal S/G lists. */ 341 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { 342 uma_zfree(beio->lun->lun_zone, 343 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); 344 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; 345 } 346 } 347 348 if (duplicate_free > 0) { 349 printf("%s: %d duplicate frees out of %d segments\n", __func__, 350 duplicate_free, beio->num_segs); 351 } 352 353 uma_zfree(beio_zone, beio); 354 } 355 356 static void 357 ctl_complete_beio(struct ctl_be_block_io *beio) 358 { 359 union ctl_io *io = beio->io; 360 361 if (beio->beio_cont != NULL) { 362 beio->beio_cont(beio); 363 } else { 364 ctl_free_beio(beio); 365 ctl_data_submit_done(io); 366 } 367 } 368 369 static int 370 ctl_be_block_move_done(union ctl_io *io) 371 { 372 struct ctl_be_block_io *beio; 373 struct ctl_be_block_lun *be_lun; 374 struct ctl_lba_len_flags *lbalen; 375 #ifdef CTL_TIME_IO 376 struct bintime cur_bt; 377 #endif 378 int i; 379 380 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 381 be_lun = beio->lun; 382 383 DPRINTF("entered\n"); 384 385 #ifdef CTL_TIME_IO 386 getbintime(&cur_bt); 387 bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); 388 bintime_add(&io->io_hdr.dma_bt, &cur_bt); 389 io->io_hdr.num_dmas++; 390 #endif 391 io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; 392 393 /* 394 * We set status at this point for read commands, and write 395 * commands with errors. 396 */ 397 if (io->io_hdr.flags & CTL_FLAG_ABORT) { 398 ; 399 } else if ((io->io_hdr.port_status == 0) && 400 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { 401 lbalen = ARGS(beio->io); 402 if (lbalen->flags & CTL_LLF_READ) { 403 ctl_set_success(&io->scsiio); 404 } else if (lbalen->flags & CTL_LLF_COMPARE) { 405 /* We have two data blocks ready for comparison. */ 406 for (i = 0; i < beio->num_segs; i++) { 407 if (memcmp(beio->sg_segs[i].addr, 408 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, 409 beio->sg_segs[i].len) != 0) 410 break; 411 } 412 if (i < beio->num_segs) 413 ctl_set_sense(&io->scsiio, 414 /*current_error*/ 1, 415 /*sense_key*/ SSD_KEY_MISCOMPARE, 416 /*asc*/ 0x1D, 417 /*ascq*/ 0x00, 418 SSD_ELEM_NONE); 419 else 420 ctl_set_success(&io->scsiio); 421 } 422 } else if ((io->io_hdr.port_status != 0) && 423 ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || 424 (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { 425 /* 426 * For hardware error sense keys, the sense key 427 * specific value is defined to be a retry count, 428 * but we use it to pass back an internal FETD 429 * error code. XXX KDM Hopefully the FETD is only 430 * using 16 bits for an error code, since that's 431 * all the space we have in the sks field. 432 */ 433 ctl_set_internal_failure(&io->scsiio, 434 /*sks_valid*/ 1, 435 /*retry_count*/ 436 io->io_hdr.port_status); 437 } 438 439 /* 440 * If this is a read, or a write with errors, it is done. 441 */ 442 if ((beio->bio_cmd == BIO_READ) 443 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) 444 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { 445 ctl_complete_beio(beio); 446 return (0); 447 } 448 449 /* 450 * At this point, we have a write and the DMA completed 451 * successfully. We now have to queue it to the task queue to 452 * execute the backend I/O. That is because we do blocking 453 * memory allocations, and in the file backing case, blocking I/O. 454 * This move done routine is generally called in the SIM's 455 * interrupt context, and therefore we cannot block. 456 */ 457 mtx_lock(&be_lun->queue_lock); 458 /* 459 * XXX KDM make sure that links is okay to use at this point. 460 * Otherwise, we either need to add another field to ctl_io_hdr, 461 * or deal with resource allocation here. 462 */ 463 STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); 464 mtx_unlock(&be_lun->queue_lock); 465 466 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 467 468 return (0); 469 } 470 471 static void 472 ctl_be_block_biodone(struct bio *bio) 473 { 474 struct ctl_be_block_io *beio; 475 struct ctl_be_block_lun *be_lun; 476 union ctl_io *io; 477 int error; 478 479 beio = bio->bio_caller1; 480 be_lun = beio->lun; 481 io = beio->io; 482 483 DPRINTF("entered\n"); 484 485 error = bio->bio_error; 486 mtx_lock(&be_lun->io_lock); 487 if (error != 0) 488 beio->num_errors++; 489 490 beio->num_bios_done++; 491 492 /* 493 * XXX KDM will this cause WITNESS to complain? Holding a lock 494 * during the free might cause it to complain. 495 */ 496 g_destroy_bio(bio); 497 498 /* 499 * If the send complete bit isn't set, or we aren't the last I/O to 500 * complete, then we're done. 501 */ 502 if ((beio->send_complete == 0) 503 || (beio->num_bios_done < beio->num_bios_sent)) { 504 mtx_unlock(&be_lun->io_lock); 505 return; 506 } 507 508 /* 509 * At this point, we've verified that we are the last I/O to 510 * complete, so it's safe to drop the lock. 511 */ 512 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 513 beio->ds_tag_type, beio->ds_trans_type, 514 /*now*/ NULL, /*then*/&beio->ds_t0); 515 mtx_unlock(&be_lun->io_lock); 516 517 /* 518 * If there are any errors from the backing device, we fail the 519 * entire I/O with a medium error. 520 */ 521 if (beio->num_errors > 0) { 522 if (error == EOPNOTSUPP) { 523 ctl_set_invalid_opcode(&io->scsiio); 524 } else if (error == ENOSPC || error == EDQUOT) { 525 ctl_set_space_alloc_fail(&io->scsiio); 526 } else if (beio->bio_cmd == BIO_FLUSH) { 527 /* XXX KDM is there is a better error here? */ 528 ctl_set_internal_failure(&io->scsiio, 529 /*sks_valid*/ 1, 530 /*retry_count*/ 0xbad2); 531 } else 532 ctl_set_medium_error(&io->scsiio); 533 ctl_complete_beio(beio); 534 return; 535 } 536 537 /* 538 * If this is a write, a flush, a delete or verify, we're all done. 539 * If this is a read, we can now send the data to the user. 540 */ 541 if ((beio->bio_cmd == BIO_WRITE) 542 || (beio->bio_cmd == BIO_FLUSH) 543 || (beio->bio_cmd == BIO_DELETE) 544 || (ARGS(io)->flags & CTL_LLF_VERIFY)) { 545 ctl_set_success(&io->scsiio); 546 ctl_complete_beio(beio); 547 } else { 548 if ((ARGS(io)->flags & CTL_LLF_READ) && 549 beio->beio_cont == NULL) 550 ctl_set_success(&io->scsiio); 551 #ifdef CTL_TIME_IO 552 getbintime(&io->io_hdr.dma_start_bt); 553 #endif 554 ctl_datamove(io); 555 } 556 } 557 558 static void 559 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, 560 struct ctl_be_block_io *beio) 561 { 562 union ctl_io *io = beio->io; 563 struct mount *mountpoint; 564 int error, lock_flags; 565 566 DPRINTF("entered\n"); 567 568 binuptime(&beio->ds_t0); 569 mtx_lock(&be_lun->io_lock); 570 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 571 mtx_unlock(&be_lun->io_lock); 572 573 (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 574 575 if (MNT_SHARED_WRITES(mountpoint) 576 || ((mountpoint == NULL) 577 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 578 lock_flags = LK_SHARED; 579 else 580 lock_flags = LK_EXCLUSIVE; 581 582 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 583 584 error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread); 585 VOP_UNLOCK(be_lun->vn, 0); 586 587 vn_finished_write(mountpoint); 588 589 mtx_lock(&be_lun->io_lock); 590 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 591 beio->ds_tag_type, beio->ds_trans_type, 592 /*now*/ NULL, /*then*/&beio->ds_t0); 593 mtx_unlock(&be_lun->io_lock); 594 595 if (error == 0) 596 ctl_set_success(&io->scsiio); 597 else { 598 /* XXX KDM is there is a better error here? */ 599 ctl_set_internal_failure(&io->scsiio, 600 /*sks_valid*/ 1, 601 /*retry_count*/ 0xbad1); 602 } 603 604 ctl_complete_beio(beio); 605 } 606 607 SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t"); 608 SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); 609 SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); 610 SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t"); 611 612 static void 613 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, 614 struct ctl_be_block_io *beio) 615 { 616 struct ctl_be_block_filedata *file_data; 617 union ctl_io *io; 618 struct uio xuio; 619 struct iovec *xiovec; 620 int flags; 621 int error, i; 622 623 DPRINTF("entered\n"); 624 625 file_data = &be_lun->backend.file; 626 io = beio->io; 627 flags = 0; 628 if (ARGS(io)->flags & CTL_LLF_DPO) 629 flags |= IO_DIRECT; 630 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 631 flags |= IO_SYNC; 632 633 bzero(&xuio, sizeof(xuio)); 634 if (beio->bio_cmd == BIO_READ) { 635 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 636 xuio.uio_rw = UIO_READ; 637 } else { 638 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 639 xuio.uio_rw = UIO_WRITE; 640 } 641 xuio.uio_offset = beio->io_offset; 642 xuio.uio_resid = beio->io_len; 643 xuio.uio_segflg = UIO_SYSSPACE; 644 xuio.uio_iov = beio->xiovecs; 645 xuio.uio_iovcnt = beio->num_segs; 646 xuio.uio_td = curthread; 647 648 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 649 xiovec->iov_base = beio->sg_segs[i].addr; 650 xiovec->iov_len = beio->sg_segs[i].len; 651 } 652 653 binuptime(&beio->ds_t0); 654 mtx_lock(&be_lun->io_lock); 655 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 656 mtx_unlock(&be_lun->io_lock); 657 658 if (beio->bio_cmd == BIO_READ) { 659 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 660 661 /* 662 * UFS pays attention to IO_DIRECT for reads. If the 663 * DIRECTIO option is configured into the kernel, it calls 664 * ffs_rawread(). But that only works for single-segment 665 * uios with user space addresses. In our case, with a 666 * kernel uio, it still reads into the buffer cache, but it 667 * will just try to release the buffer from the cache later 668 * on in ffs_read(). 669 * 670 * ZFS does not pay attention to IO_DIRECT for reads. 671 * 672 * UFS does not pay attention to IO_SYNC for reads. 673 * 674 * ZFS pays attention to IO_SYNC (which translates into the 675 * Solaris define FRSYNC for zfs_read()) for reads. It 676 * attempts to sync the file before reading. 677 * 678 * So, to attempt to provide some barrier semantics in the 679 * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. 680 */ 681 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); 682 683 VOP_UNLOCK(be_lun->vn, 0); 684 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 685 } else { 686 struct mount *mountpoint; 687 int lock_flags; 688 689 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); 690 691 if (MNT_SHARED_WRITES(mountpoint) 692 || ((mountpoint == NULL) 693 && MNT_SHARED_WRITES(be_lun->vn->v_mount))) 694 lock_flags = LK_SHARED; 695 else 696 lock_flags = LK_EXCLUSIVE; 697 698 vn_lock(be_lun->vn, lock_flags | LK_RETRY); 699 700 /* 701 * UFS pays attention to IO_DIRECT for writes. The write 702 * is done asynchronously. (Normally the write would just 703 * get put into cache. 704 * 705 * UFS pays attention to IO_SYNC for writes. It will 706 * attempt to write the buffer out synchronously if that 707 * flag is set. 708 * 709 * ZFS does not pay attention to IO_DIRECT for writes. 710 * 711 * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) 712 * for writes. It will flush the transaction from the 713 * cache before returning. 714 * 715 * So if we've got the BIO_ORDERED flag set, we want 716 * IO_SYNC in either the UFS or ZFS case. 717 */ 718 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); 719 VOP_UNLOCK(be_lun->vn, 0); 720 721 vn_finished_write(mountpoint); 722 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 723 } 724 725 mtx_lock(&be_lun->io_lock); 726 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 727 beio->ds_tag_type, beio->ds_trans_type, 728 /*now*/ NULL, /*then*/&beio->ds_t0); 729 mtx_unlock(&be_lun->io_lock); 730 731 /* 732 * If we got an error, set the sense data to "MEDIUM ERROR" and 733 * return the I/O to the user. 734 */ 735 if (error != 0) { 736 char path_str[32]; 737 738 ctl_scsi_path_string(io, path_str, sizeof(path_str)); 739 printf("%s%s command returned errno %d\n", path_str, 740 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); 741 if (error == ENOSPC || error == EDQUOT) { 742 ctl_set_space_alloc_fail(&io->scsiio); 743 } else 744 ctl_set_medium_error(&io->scsiio); 745 ctl_complete_beio(beio); 746 return; 747 } 748 749 /* 750 * If this is a write or a verify, we're all done. 751 * If this is a read, we can now send the data to the user. 752 */ 753 if ((beio->bio_cmd == BIO_WRITE) || 754 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 755 ctl_set_success(&io->scsiio); 756 ctl_complete_beio(beio); 757 } else { 758 if ((ARGS(io)->flags & CTL_LLF_READ) && 759 beio->beio_cont == NULL) 760 ctl_set_success(&io->scsiio); 761 #ifdef CTL_TIME_IO 762 getbintime(&io->io_hdr.dma_start_bt); 763 #endif 764 ctl_datamove(io); 765 } 766 } 767 768 static void 769 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, 770 struct ctl_be_block_io *beio) 771 { 772 union ctl_io *io = beio->io; 773 struct ctl_lba_len_flags *lbalen = ARGS(io); 774 struct scsi_get_lba_status_data *data; 775 off_t roff, off; 776 int error, status; 777 778 DPRINTF("entered\n"); 779 780 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 781 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 782 error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 783 0, curthread->td_ucred, curthread); 784 if (error == 0 && off > roff) 785 status = 0; /* mapped up to off */ 786 else { 787 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 788 0, curthread->td_ucred, curthread); 789 if (error == 0 && off > roff) 790 status = 1; /* deallocated up to off */ 791 else { 792 status = 0; /* unknown up to the end */ 793 off = be_lun->size_bytes; 794 } 795 } 796 VOP_UNLOCK(be_lun->vn, 0); 797 798 off >>= be_lun->blocksize_shift; 799 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 800 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 801 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 802 data->descr[0].length); 803 data->descr[0].status = status; 804 805 ctl_complete_beio(beio); 806 } 807 808 static uint64_t 809 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) 810 { 811 struct vattr vattr; 812 struct statfs statfs; 813 uint64_t val; 814 int error; 815 816 val = UINT64_MAX; 817 if (be_lun->vn == NULL) 818 return (val); 819 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 820 if (strcmp(attrname, "blocksused") == 0) { 821 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 822 if (error == 0) 823 val = vattr.va_bytes >> be_lun->blocksize_shift; 824 } 825 if (strcmp(attrname, "blocksavail") == 0 && 826 (be_lun->vn->v_iflag & VI_DOOMED) == 0) { 827 error = VFS_STATFS(be_lun->vn->v_mount, &statfs); 828 if (error == 0) 829 val = (statfs.f_bavail * statfs.f_bsize) >> 830 be_lun->blocksize_shift; 831 } 832 VOP_UNLOCK(be_lun->vn, 0); 833 return (val); 834 } 835 836 static void 837 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, 838 struct ctl_be_block_io *beio) 839 { 840 struct ctl_be_block_devdata *dev_data; 841 union ctl_io *io; 842 struct uio xuio; 843 struct iovec *xiovec; 844 int flags; 845 int error, i; 846 847 DPRINTF("entered\n"); 848 849 dev_data = &be_lun->backend.dev; 850 io = beio->io; 851 flags = 0; 852 if (ARGS(io)->flags & CTL_LLF_DPO) 853 flags |= IO_DIRECT; 854 if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) 855 flags |= IO_SYNC; 856 857 bzero(&xuio, sizeof(xuio)); 858 if (beio->bio_cmd == BIO_READ) { 859 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); 860 xuio.uio_rw = UIO_READ; 861 } else { 862 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 863 xuio.uio_rw = UIO_WRITE; 864 } 865 xuio.uio_offset = beio->io_offset; 866 xuio.uio_resid = beio->io_len; 867 xuio.uio_segflg = UIO_SYSSPACE; 868 xuio.uio_iov = beio->xiovecs; 869 xuio.uio_iovcnt = beio->num_segs; 870 xuio.uio_td = curthread; 871 872 for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { 873 xiovec->iov_base = beio->sg_segs[i].addr; 874 xiovec->iov_len = beio->sg_segs[i].len; 875 } 876 877 binuptime(&beio->ds_t0); 878 mtx_lock(&be_lun->io_lock); 879 devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); 880 mtx_unlock(&be_lun->io_lock); 881 882 if (beio->bio_cmd == BIO_READ) { 883 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); 884 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); 885 } else { 886 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); 887 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); 888 } 889 890 mtx_lock(&be_lun->io_lock); 891 devstat_end_transaction(beio->lun->disk_stats, beio->io_len, 892 beio->ds_tag_type, beio->ds_trans_type, 893 /*now*/ NULL, /*then*/&beio->ds_t0); 894 mtx_unlock(&be_lun->io_lock); 895 896 /* 897 * If we got an error, set the sense data to "MEDIUM ERROR" and 898 * return the I/O to the user. 899 */ 900 if (error != 0) { 901 if (error == ENOSPC || error == EDQUOT) { 902 ctl_set_space_alloc_fail(&io->scsiio); 903 } else 904 ctl_set_medium_error(&io->scsiio); 905 ctl_complete_beio(beio); 906 return; 907 } 908 909 /* 910 * If this is a write or a verify, we're all done. 911 * If this is a read, we can now send the data to the user. 912 */ 913 if ((beio->bio_cmd == BIO_WRITE) || 914 (ARGS(io)->flags & CTL_LLF_VERIFY)) { 915 ctl_set_success(&io->scsiio); 916 ctl_complete_beio(beio); 917 } else { 918 if ((ARGS(io)->flags & CTL_LLF_READ) && 919 beio->beio_cont == NULL) 920 ctl_set_success(&io->scsiio); 921 #ifdef CTL_TIME_IO 922 getbintime(&io->io_hdr.dma_start_bt); 923 #endif 924 ctl_datamove(io); 925 } 926 } 927 928 static void 929 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, 930 struct ctl_be_block_io *beio) 931 { 932 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 933 union ctl_io *io = beio->io; 934 struct ctl_lba_len_flags *lbalen = ARGS(io); 935 struct scsi_get_lba_status_data *data; 936 off_t roff, off; 937 int error, status; 938 939 DPRINTF("entered\n"); 940 941 off = roff = ((off_t)lbalen->lba) << be_lun->blocksize_shift; 942 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKHOLE, 943 (caddr_t)&off, FREAD, curthread); 944 if (error == 0 && off > roff) 945 status = 0; /* mapped up to off */ 946 else { 947 error = (*dev_data->csw->d_ioctl)(dev_data->cdev, FIOSEEKDATA, 948 (caddr_t)&off, FREAD, curthread); 949 if (error == 0 && off > roff) 950 status = 1; /* deallocated up to off */ 951 else { 952 status = 0; /* unknown up to the end */ 953 off = be_lun->size_bytes; 954 } 955 } 956 957 off >>= be_lun->blocksize_shift; 958 data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; 959 scsi_u64to8b(lbalen->lba, data->descr[0].addr); 960 scsi_ulto4b(MIN(UINT32_MAX, off - lbalen->lba), 961 data->descr[0].length); 962 data->descr[0].status = status; 963 964 ctl_complete_beio(beio); 965 } 966 967 static void 968 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, 969 struct ctl_be_block_io *beio) 970 { 971 struct bio *bio; 972 union ctl_io *io; 973 struct ctl_be_block_devdata *dev_data; 974 975 dev_data = &be_lun->backend.dev; 976 io = beio->io; 977 978 DPRINTF("entered\n"); 979 980 /* This can't fail, it's a blocking allocation. */ 981 bio = g_alloc_bio(); 982 983 bio->bio_cmd = BIO_FLUSH; 984 bio->bio_flags |= BIO_ORDERED; 985 bio->bio_dev = dev_data->cdev; 986 bio->bio_offset = 0; 987 bio->bio_data = 0; 988 bio->bio_done = ctl_be_block_biodone; 989 bio->bio_caller1 = beio; 990 bio->bio_pblkno = 0; 991 992 /* 993 * We don't need to acquire the LUN lock here, because we are only 994 * sending one bio, and so there is no other context to synchronize 995 * with. 996 */ 997 beio->num_bios_sent = 1; 998 beio->send_complete = 1; 999 1000 binuptime(&beio->ds_t0); 1001 mtx_lock(&be_lun->io_lock); 1002 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1003 mtx_unlock(&be_lun->io_lock); 1004 1005 (*dev_data->csw->d_strategy)(bio); 1006 } 1007 1008 static void 1009 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, 1010 struct ctl_be_block_io *beio, 1011 uint64_t off, uint64_t len, int last) 1012 { 1013 struct bio *bio; 1014 struct ctl_be_block_devdata *dev_data; 1015 uint64_t maxlen; 1016 1017 dev_data = &be_lun->backend.dev; 1018 maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); 1019 while (len > 0) { 1020 bio = g_alloc_bio(); 1021 bio->bio_cmd = BIO_DELETE; 1022 bio->bio_dev = dev_data->cdev; 1023 bio->bio_offset = off; 1024 bio->bio_length = MIN(len, maxlen); 1025 bio->bio_data = 0; 1026 bio->bio_done = ctl_be_block_biodone; 1027 bio->bio_caller1 = beio; 1028 bio->bio_pblkno = off / be_lun->blocksize; 1029 1030 off += bio->bio_length; 1031 len -= bio->bio_length; 1032 1033 mtx_lock(&be_lun->io_lock); 1034 beio->num_bios_sent++; 1035 if (last && len == 0) 1036 beio->send_complete = 1; 1037 mtx_unlock(&be_lun->io_lock); 1038 1039 (*dev_data->csw->d_strategy)(bio); 1040 } 1041 } 1042 1043 static void 1044 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, 1045 struct ctl_be_block_io *beio) 1046 { 1047 union ctl_io *io; 1048 struct ctl_be_block_devdata *dev_data; 1049 struct ctl_ptr_len_flags *ptrlen; 1050 struct scsi_unmap_desc *buf, *end; 1051 uint64_t len; 1052 1053 dev_data = &be_lun->backend.dev; 1054 io = beio->io; 1055 1056 DPRINTF("entered\n"); 1057 1058 binuptime(&beio->ds_t0); 1059 mtx_lock(&be_lun->io_lock); 1060 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1061 mtx_unlock(&be_lun->io_lock); 1062 1063 if (beio->io_offset == -1) { 1064 beio->io_len = 0; 1065 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1066 buf = (struct scsi_unmap_desc *)ptrlen->ptr; 1067 end = buf + ptrlen->len / sizeof(*buf); 1068 for (; buf < end; buf++) { 1069 len = (uint64_t)scsi_4btoul(buf->length) * 1070 be_lun->blocksize; 1071 beio->io_len += len; 1072 ctl_be_block_unmap_dev_range(be_lun, beio, 1073 scsi_8btou64(buf->lba) * be_lun->blocksize, len, 1074 (end - buf < 2) ? TRUE : FALSE); 1075 } 1076 } else 1077 ctl_be_block_unmap_dev_range(be_lun, beio, 1078 beio->io_offset, beio->io_len, TRUE); 1079 } 1080 1081 static void 1082 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, 1083 struct ctl_be_block_io *beio) 1084 { 1085 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 1086 int i; 1087 struct bio *bio; 1088 struct ctl_be_block_devdata *dev_data; 1089 off_t cur_offset; 1090 int max_iosize; 1091 1092 DPRINTF("entered\n"); 1093 1094 dev_data = &be_lun->backend.dev; 1095 1096 /* 1097 * We have to limit our I/O size to the maximum supported by the 1098 * backend device. Hopefully it is MAXPHYS. If the driver doesn't 1099 * set it properly, use DFLTPHYS. 1100 */ 1101 max_iosize = dev_data->cdev->si_iosize_max; 1102 if (max_iosize < PAGE_SIZE) 1103 max_iosize = DFLTPHYS; 1104 1105 cur_offset = beio->io_offset; 1106 for (i = 0; i < beio->num_segs; i++) { 1107 size_t cur_size; 1108 uint8_t *cur_ptr; 1109 1110 cur_size = beio->sg_segs[i].len; 1111 cur_ptr = beio->sg_segs[i].addr; 1112 1113 while (cur_size > 0) { 1114 /* This can't fail, it's a blocking allocation. */ 1115 bio = g_alloc_bio(); 1116 1117 KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); 1118 1119 bio->bio_cmd = beio->bio_cmd; 1120 bio->bio_dev = dev_data->cdev; 1121 bio->bio_caller1 = beio; 1122 bio->bio_length = min(cur_size, max_iosize); 1123 bio->bio_offset = cur_offset; 1124 bio->bio_data = cur_ptr; 1125 bio->bio_done = ctl_be_block_biodone; 1126 bio->bio_pblkno = cur_offset / be_lun->blocksize; 1127 1128 cur_offset += bio->bio_length; 1129 cur_ptr += bio->bio_length; 1130 cur_size -= bio->bio_length; 1131 1132 TAILQ_INSERT_TAIL(&queue, bio, bio_queue); 1133 beio->num_bios_sent++; 1134 } 1135 } 1136 binuptime(&beio->ds_t0); 1137 mtx_lock(&be_lun->io_lock); 1138 devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); 1139 beio->send_complete = 1; 1140 mtx_unlock(&be_lun->io_lock); 1141 1142 /* 1143 * Fire off all allocated requests! 1144 */ 1145 while ((bio = TAILQ_FIRST(&queue)) != NULL) { 1146 TAILQ_REMOVE(&queue, bio, bio_queue); 1147 (*dev_data->csw->d_strategy)(bio); 1148 } 1149 } 1150 1151 static uint64_t 1152 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) 1153 { 1154 struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; 1155 struct diocgattr_arg arg; 1156 int error; 1157 1158 if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) 1159 return (UINT64_MAX); 1160 strlcpy(arg.name, attrname, sizeof(arg.name)); 1161 arg.len = sizeof(arg.value.off); 1162 error = dev_data->csw->d_ioctl(dev_data->cdev, 1163 DIOCGATTR, (caddr_t)&arg, FREAD, curthread); 1164 if (error != 0) 1165 return (UINT64_MAX); 1166 return (arg.value.off); 1167 } 1168 1169 static void 1170 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) 1171 { 1172 union ctl_io *io; 1173 1174 io = beio->io; 1175 ctl_free_beio(beio); 1176 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1177 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1178 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1179 ctl_config_write_done(io); 1180 return; 1181 } 1182 1183 ctl_be_block_config_write(io); 1184 } 1185 1186 static void 1187 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, 1188 union ctl_io *io) 1189 { 1190 struct ctl_be_block_io *beio; 1191 struct ctl_be_block_softc *softc; 1192 struct ctl_lba_len_flags *lbalen; 1193 uint64_t len_left, lba; 1194 uint32_t pb, pbo, adj; 1195 int i, seglen; 1196 uint8_t *buf, *end; 1197 1198 DPRINTF("entered\n"); 1199 1200 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1201 softc = be_lun->softc; 1202 lbalen = ARGS(beio->io); 1203 1204 if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || 1205 (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { 1206 ctl_free_beio(beio); 1207 ctl_set_invalid_field(&io->scsiio, 1208 /*sks_valid*/ 1, 1209 /*command*/ 1, 1210 /*field*/ 1, 1211 /*bit_valid*/ 0, 1212 /*bit*/ 0); 1213 ctl_config_write_done(io); 1214 return; 1215 } 1216 1217 switch (io->scsiio.tag_type) { 1218 case CTL_TAG_ORDERED: 1219 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1220 break; 1221 case CTL_TAG_HEAD_OF_QUEUE: 1222 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1223 break; 1224 case CTL_TAG_UNTAGGED: 1225 case CTL_TAG_SIMPLE: 1226 case CTL_TAG_ACA: 1227 default: 1228 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1229 break; 1230 } 1231 1232 if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { 1233 beio->io_offset = lbalen->lba * be_lun->blocksize; 1234 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize; 1235 beio->bio_cmd = BIO_DELETE; 1236 beio->ds_trans_type = DEVSTAT_FREE; 1237 1238 be_lun->unmap(be_lun, beio); 1239 return; 1240 } 1241 1242 beio->bio_cmd = BIO_WRITE; 1243 beio->ds_trans_type = DEVSTAT_WRITE; 1244 1245 DPRINTF("WRITE SAME at LBA %jx len %u\n", 1246 (uintmax_t)lbalen->lba, lbalen->len); 1247 1248 pb = be_lun->blocksize << be_lun->pblockexp; 1249 if (be_lun->pblockoff > 0) 1250 pbo = pb - be_lun->blocksize * be_lun->pblockoff; 1251 else 1252 pbo = 0; 1253 len_left = (uint64_t)lbalen->len * be_lun->blocksize; 1254 for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { 1255 1256 /* 1257 * Setup the S/G entry for this chunk. 1258 */ 1259 seglen = MIN(CTLBLK_MAX_SEG, len_left); 1260 if (pb > be_lun->blocksize) { 1261 adj = ((lbalen->lba + lba) * be_lun->blocksize + 1262 seglen - pbo) % pb; 1263 if (seglen > adj) 1264 seglen -= adj; 1265 else 1266 seglen -= seglen % be_lun->blocksize; 1267 } else 1268 seglen -= seglen % be_lun->blocksize; 1269 beio->sg_segs[i].len = seglen; 1270 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1271 1272 DPRINTF("segment %d addr %p len %zd\n", i, 1273 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1274 1275 beio->num_segs++; 1276 len_left -= seglen; 1277 1278 buf = beio->sg_segs[i].addr; 1279 end = buf + seglen; 1280 for (; buf < end; buf += be_lun->blocksize) { 1281 memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); 1282 if (lbalen->flags & SWS_LBDATA) 1283 scsi_ulto4b(lbalen->lba + lba, buf); 1284 lba++; 1285 } 1286 } 1287 1288 beio->io_offset = lbalen->lba * be_lun->blocksize; 1289 beio->io_len = lba * be_lun->blocksize; 1290 1291 /* We can not do all in one run. Correct and schedule rerun. */ 1292 if (len_left > 0) { 1293 lbalen->lba += lba; 1294 lbalen->len -= lba; 1295 beio->beio_cont = ctl_be_block_cw_done_ws; 1296 } 1297 1298 be_lun->dispatch(be_lun, beio); 1299 } 1300 1301 static void 1302 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, 1303 union ctl_io *io) 1304 { 1305 struct ctl_be_block_io *beio; 1306 struct ctl_be_block_softc *softc; 1307 struct ctl_ptr_len_flags *ptrlen; 1308 1309 DPRINTF("entered\n"); 1310 1311 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1312 softc = be_lun->softc; 1313 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; 1314 1315 if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { 1316 ctl_free_beio(beio); 1317 ctl_set_invalid_field(&io->scsiio, 1318 /*sks_valid*/ 0, 1319 /*command*/ 1, 1320 /*field*/ 0, 1321 /*bit_valid*/ 0, 1322 /*bit*/ 0); 1323 ctl_config_write_done(io); 1324 return; 1325 } 1326 1327 switch (io->scsiio.tag_type) { 1328 case CTL_TAG_ORDERED: 1329 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1330 break; 1331 case CTL_TAG_HEAD_OF_QUEUE: 1332 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1333 break; 1334 case CTL_TAG_UNTAGGED: 1335 case CTL_TAG_SIMPLE: 1336 case CTL_TAG_ACA: 1337 default: 1338 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1339 break; 1340 } 1341 1342 beio->io_len = 0; 1343 beio->io_offset = -1; 1344 1345 beio->bio_cmd = BIO_DELETE; 1346 beio->ds_trans_type = DEVSTAT_FREE; 1347 1348 DPRINTF("UNMAP\n"); 1349 1350 be_lun->unmap(be_lun, beio); 1351 } 1352 1353 static void 1354 ctl_be_block_cr_done(struct ctl_be_block_io *beio) 1355 { 1356 union ctl_io *io; 1357 1358 io = beio->io; 1359 ctl_free_beio(beio); 1360 ctl_config_read_done(io); 1361 } 1362 1363 static void 1364 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, 1365 union ctl_io *io) 1366 { 1367 struct ctl_be_block_io *beio; 1368 struct ctl_be_block_softc *softc; 1369 1370 DPRINTF("entered\n"); 1371 1372 softc = be_lun->softc; 1373 beio = ctl_alloc_beio(softc); 1374 beio->io = io; 1375 beio->lun = be_lun; 1376 beio->beio_cont = ctl_be_block_cr_done; 1377 PRIV(io)->ptr = (void *)beio; 1378 1379 switch (io->scsiio.cdb[0]) { 1380 case SERVICE_ACTION_IN: /* GET LBA STATUS */ 1381 beio->bio_cmd = -1; 1382 beio->ds_trans_type = DEVSTAT_NO_DATA; 1383 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1384 beio->io_len = 0; 1385 if (be_lun->get_lba_status) 1386 be_lun->get_lba_status(be_lun, beio); 1387 else 1388 ctl_be_block_cr_done(beio); 1389 break; 1390 default: 1391 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1392 break; 1393 } 1394 } 1395 1396 static void 1397 ctl_be_block_cw_done(struct ctl_be_block_io *beio) 1398 { 1399 union ctl_io *io; 1400 1401 io = beio->io; 1402 ctl_free_beio(beio); 1403 ctl_config_write_done(io); 1404 } 1405 1406 static void 1407 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, 1408 union ctl_io *io) 1409 { 1410 struct ctl_be_block_io *beio; 1411 struct ctl_be_block_softc *softc; 1412 1413 DPRINTF("entered\n"); 1414 1415 softc = be_lun->softc; 1416 beio = ctl_alloc_beio(softc); 1417 beio->io = io; 1418 beio->lun = be_lun; 1419 beio->beio_cont = ctl_be_block_cw_done; 1420 PRIV(io)->ptr = (void *)beio; 1421 1422 switch (io->scsiio.cdb[0]) { 1423 case SYNCHRONIZE_CACHE: 1424 case SYNCHRONIZE_CACHE_16: 1425 beio->bio_cmd = BIO_FLUSH; 1426 beio->ds_trans_type = DEVSTAT_NO_DATA; 1427 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1428 beio->io_len = 0; 1429 be_lun->lun_flush(be_lun, beio); 1430 break; 1431 case WRITE_SAME_10: 1432 case WRITE_SAME_16: 1433 ctl_be_block_cw_dispatch_ws(be_lun, io); 1434 break; 1435 case UNMAP: 1436 ctl_be_block_cw_dispatch_unmap(be_lun, io); 1437 break; 1438 default: 1439 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); 1440 break; 1441 } 1442 } 1443 1444 SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t"); 1445 SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); 1446 SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); 1447 SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t"); 1448 1449 static void 1450 ctl_be_block_next(struct ctl_be_block_io *beio) 1451 { 1452 struct ctl_be_block_lun *be_lun; 1453 union ctl_io *io; 1454 1455 io = beio->io; 1456 be_lun = beio->lun; 1457 ctl_free_beio(beio); 1458 if ((io->io_hdr.flags & CTL_FLAG_ABORT) || 1459 ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && 1460 (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { 1461 ctl_data_submit_done(io); 1462 return; 1463 } 1464 1465 io->io_hdr.status &= ~CTL_STATUS_MASK; 1466 io->io_hdr.status |= CTL_STATUS_NONE; 1467 1468 mtx_lock(&be_lun->queue_lock); 1469 /* 1470 * XXX KDM make sure that links is okay to use at this point. 1471 * Otherwise, we either need to add another field to ctl_io_hdr, 1472 * or deal with resource allocation here. 1473 */ 1474 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1475 mtx_unlock(&be_lun->queue_lock); 1476 1477 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1478 } 1479 1480 static void 1481 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, 1482 union ctl_io *io) 1483 { 1484 struct ctl_be_block_io *beio; 1485 struct ctl_be_block_softc *softc; 1486 struct ctl_lba_len_flags *lbalen; 1487 struct ctl_ptr_len_flags *bptrlen; 1488 uint64_t len_left, lbas; 1489 int i; 1490 1491 softc = be_lun->softc; 1492 1493 DPRINTF("entered\n"); 1494 1495 lbalen = ARGS(io); 1496 if (lbalen->flags & CTL_LLF_WRITE) { 1497 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); 1498 } else { 1499 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); 1500 } 1501 1502 beio = ctl_alloc_beio(softc); 1503 beio->io = io; 1504 beio->lun = be_lun; 1505 bptrlen = PRIV(io); 1506 bptrlen->ptr = (void *)beio; 1507 1508 switch (io->scsiio.tag_type) { 1509 case CTL_TAG_ORDERED: 1510 beio->ds_tag_type = DEVSTAT_TAG_ORDERED; 1511 break; 1512 case CTL_TAG_HEAD_OF_QUEUE: 1513 beio->ds_tag_type = DEVSTAT_TAG_HEAD; 1514 break; 1515 case CTL_TAG_UNTAGGED: 1516 case CTL_TAG_SIMPLE: 1517 case CTL_TAG_ACA: 1518 default: 1519 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; 1520 break; 1521 } 1522 1523 if (lbalen->flags & CTL_LLF_WRITE) { 1524 beio->bio_cmd = BIO_WRITE; 1525 beio->ds_trans_type = DEVSTAT_WRITE; 1526 } else { 1527 beio->bio_cmd = BIO_READ; 1528 beio->ds_trans_type = DEVSTAT_READ; 1529 } 1530 1531 DPRINTF("%s at LBA %jx len %u @%ju\n", 1532 (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", 1533 (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); 1534 if (lbalen->flags & CTL_LLF_COMPARE) 1535 lbas = CTLBLK_HALF_IO_SIZE; 1536 else 1537 lbas = CTLBLK_MAX_IO_SIZE; 1538 lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize); 1539 beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; 1540 beio->io_len = lbas * be_lun->blocksize; 1541 bptrlen->len += lbas; 1542 1543 for (i = 0, len_left = beio->io_len; len_left > 0; i++) { 1544 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", 1545 i, CTLBLK_MAX_SEGS)); 1546 1547 /* 1548 * Setup the S/G entry for this chunk. 1549 */ 1550 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); 1551 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); 1552 1553 DPRINTF("segment %d addr %p len %zd\n", i, 1554 beio->sg_segs[i].addr, beio->sg_segs[i].len); 1555 1556 /* Set up second segment for compare operation. */ 1557 if (lbalen->flags & CTL_LLF_COMPARE) { 1558 beio->sg_segs[i + CTLBLK_HALF_SEGS].len = 1559 beio->sg_segs[i].len; 1560 beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = 1561 uma_zalloc(be_lun->lun_zone, M_WAITOK); 1562 } 1563 1564 beio->num_segs++; 1565 len_left -= beio->sg_segs[i].len; 1566 } 1567 if (bptrlen->len < lbalen->len) 1568 beio->beio_cont = ctl_be_block_next; 1569 io->scsiio.be_move_done = ctl_be_block_move_done; 1570 /* For compare we have separate S/G lists for read and datamove. */ 1571 if (lbalen->flags & CTL_LLF_COMPARE) 1572 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; 1573 else 1574 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; 1575 io->scsiio.kern_data_len = beio->io_len; 1576 io->scsiio.kern_data_resid = 0; 1577 io->scsiio.kern_sg_entries = beio->num_segs; 1578 io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; 1579 1580 /* 1581 * For the read case, we need to read the data into our buffers and 1582 * then we can send it back to the user. For the write case, we 1583 * need to get the data from the user first. 1584 */ 1585 if (beio->bio_cmd == BIO_READ) { 1586 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); 1587 be_lun->dispatch(be_lun, beio); 1588 } else { 1589 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); 1590 #ifdef CTL_TIME_IO 1591 getbintime(&io->io_hdr.dma_start_bt); 1592 #endif 1593 ctl_datamove(io); 1594 } 1595 } 1596 1597 static void 1598 ctl_be_block_worker(void *context, int pending) 1599 { 1600 struct ctl_be_block_lun *be_lun; 1601 struct ctl_be_block_softc *softc; 1602 union ctl_io *io; 1603 1604 be_lun = (struct ctl_be_block_lun *)context; 1605 softc = be_lun->softc; 1606 1607 DPRINTF("entered\n"); 1608 1609 mtx_lock(&be_lun->queue_lock); 1610 for (;;) { 1611 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); 1612 if (io != NULL) { 1613 struct ctl_be_block_io *beio; 1614 1615 DPRINTF("datamove queue\n"); 1616 1617 STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, 1618 ctl_io_hdr, links); 1619 1620 mtx_unlock(&be_lun->queue_lock); 1621 1622 beio = (struct ctl_be_block_io *)PRIV(io)->ptr; 1623 1624 be_lun->dispatch(be_lun, beio); 1625 1626 mtx_lock(&be_lun->queue_lock); 1627 continue; 1628 } 1629 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); 1630 if (io != NULL) { 1631 DPRINTF("config write queue\n"); 1632 STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, 1633 ctl_io_hdr, links); 1634 mtx_unlock(&be_lun->queue_lock); 1635 ctl_be_block_cw_dispatch(be_lun, io); 1636 mtx_lock(&be_lun->queue_lock); 1637 continue; 1638 } 1639 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); 1640 if (io != NULL) { 1641 DPRINTF("config read queue\n"); 1642 STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, 1643 ctl_io_hdr, links); 1644 mtx_unlock(&be_lun->queue_lock); 1645 ctl_be_block_cr_dispatch(be_lun, io); 1646 mtx_lock(&be_lun->queue_lock); 1647 continue; 1648 } 1649 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); 1650 if (io != NULL) { 1651 DPRINTF("input queue\n"); 1652 1653 STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, 1654 ctl_io_hdr, links); 1655 mtx_unlock(&be_lun->queue_lock); 1656 1657 /* 1658 * We must drop the lock, since this routine and 1659 * its children may sleep. 1660 */ 1661 ctl_be_block_dispatch(be_lun, io); 1662 1663 mtx_lock(&be_lun->queue_lock); 1664 continue; 1665 } 1666 1667 /* 1668 * If we get here, there is no work left in the queues, so 1669 * just break out and let the task queue go to sleep. 1670 */ 1671 break; 1672 } 1673 mtx_unlock(&be_lun->queue_lock); 1674 } 1675 1676 /* 1677 * Entry point from CTL to the backend for I/O. We queue everything to a 1678 * work thread, so this just puts the I/O on a queue and wakes up the 1679 * thread. 1680 */ 1681 static int 1682 ctl_be_block_submit(union ctl_io *io) 1683 { 1684 struct ctl_be_block_lun *be_lun; 1685 struct ctl_be_lun *ctl_be_lun; 1686 1687 DPRINTF("entered\n"); 1688 1689 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 1690 CTL_PRIV_BACKEND_LUN].ptr; 1691 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 1692 1693 /* 1694 * Make sure we only get SCSI I/O. 1695 */ 1696 KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " 1697 "%#x) encountered", io->io_hdr.io_type)); 1698 1699 PRIV(io)->len = 0; 1700 1701 mtx_lock(&be_lun->queue_lock); 1702 /* 1703 * XXX KDM make sure that links is okay to use at this point. 1704 * Otherwise, we either need to add another field to ctl_io_hdr, 1705 * or deal with resource allocation here. 1706 */ 1707 STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); 1708 mtx_unlock(&be_lun->queue_lock); 1709 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 1710 1711 return (CTL_RETVAL_COMPLETE); 1712 } 1713 1714 static int 1715 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, 1716 int flag, struct thread *td) 1717 { 1718 struct ctl_be_block_softc *softc; 1719 int error; 1720 1721 softc = &backend_block_softc; 1722 1723 error = 0; 1724 1725 switch (cmd) { 1726 case CTL_LUN_REQ: { 1727 struct ctl_lun_req *lun_req; 1728 1729 lun_req = (struct ctl_lun_req *)addr; 1730 1731 switch (lun_req->reqtype) { 1732 case CTL_LUNREQ_CREATE: 1733 error = ctl_be_block_create(softc, lun_req); 1734 break; 1735 case CTL_LUNREQ_RM: 1736 error = ctl_be_block_rm(softc, lun_req); 1737 break; 1738 case CTL_LUNREQ_MODIFY: 1739 error = ctl_be_block_modify(softc, lun_req); 1740 break; 1741 default: 1742 lun_req->status = CTL_LUN_ERROR; 1743 snprintf(lun_req->error_str, sizeof(lun_req->error_str), 1744 "invalid LUN request type %d", 1745 lun_req->reqtype); 1746 break; 1747 } 1748 break; 1749 } 1750 default: 1751 error = ENOTTY; 1752 break; 1753 } 1754 1755 return (error); 1756 } 1757 1758 static int 1759 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1760 { 1761 struct ctl_be_block_filedata *file_data; 1762 struct ctl_lun_create_params *params; 1763 char *value; 1764 struct vattr vattr; 1765 off_t ps, pss, po, pos, us, uss, uo, uos; 1766 int error; 1767 1768 error = 0; 1769 file_data = &be_lun->backend.file; 1770 params = &be_lun->params; 1771 1772 be_lun->dev_type = CTL_BE_BLOCK_FILE; 1773 be_lun->dispatch = ctl_be_block_dispatch_file; 1774 be_lun->lun_flush = ctl_be_block_flush_file; 1775 be_lun->get_lba_status = ctl_be_block_gls_file; 1776 be_lun->getattr = ctl_be_block_getattr_file; 1777 1778 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 1779 if (error != 0) { 1780 snprintf(req->error_str, sizeof(req->error_str), 1781 "error calling VOP_GETATTR() for file %s", 1782 be_lun->dev_path); 1783 return (error); 1784 } 1785 1786 /* 1787 * Verify that we have the ability to upgrade to exclusive 1788 * access on this file so we can trap errors at open instead 1789 * of reporting them during first access. 1790 */ 1791 if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { 1792 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); 1793 if (be_lun->vn->v_iflag & VI_DOOMED) { 1794 error = EBADF; 1795 snprintf(req->error_str, sizeof(req->error_str), 1796 "error locking file %s", be_lun->dev_path); 1797 return (error); 1798 } 1799 } 1800 1801 1802 file_data->cred = crhold(curthread->td_ucred); 1803 if (params->lun_size_bytes != 0) 1804 be_lun->size_bytes = params->lun_size_bytes; 1805 else 1806 be_lun->size_bytes = vattr.va_size; 1807 /* 1808 * We set the multi thread flag for file operations because all 1809 * filesystems (in theory) are capable of allowing multiple readers 1810 * of a file at once. So we want to get the maximum possible 1811 * concurrency. 1812 */ 1813 be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; 1814 1815 /* 1816 * For files we can use any logical block size. Prefer 512 bytes 1817 * for compatibility reasons. If file's vattr.va_blocksize 1818 * (preferred I/O block size) is bigger and multiple to chosen 1819 * logical block size -- report it as physical block size. 1820 */ 1821 if (params->blocksize_bytes != 0) 1822 be_lun->blocksize = params->blocksize_bytes; 1823 else 1824 be_lun->blocksize = 512; 1825 1826 us = ps = vattr.va_blocksize; 1827 uo = po = 0; 1828 1829 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 1830 if (value != NULL) 1831 ctl_expand_number(value, &ps); 1832 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 1833 if (value != NULL) 1834 ctl_expand_number(value, &po); 1835 pss = ps / be_lun->blocksize; 1836 pos = po / be_lun->blocksize; 1837 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 1838 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 1839 be_lun->pblockexp = fls(pss) - 1; 1840 be_lun->pblockoff = (pss - pos) % pss; 1841 } 1842 1843 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 1844 if (value != NULL) 1845 ctl_expand_number(value, &us); 1846 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 1847 if (value != NULL) 1848 ctl_expand_number(value, &uo); 1849 uss = us / be_lun->blocksize; 1850 uos = uo / be_lun->blocksize; 1851 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 1852 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 1853 be_lun->ublockexp = fls(uss) - 1; 1854 be_lun->ublockoff = (uss - uos) % uss; 1855 } 1856 1857 /* 1858 * Sanity check. The media size has to be at least one 1859 * sector long. 1860 */ 1861 if (be_lun->size_bytes < be_lun->blocksize) { 1862 error = EINVAL; 1863 snprintf(req->error_str, sizeof(req->error_str), 1864 "file %s size %ju < block size %u", be_lun->dev_path, 1865 (uintmax_t)be_lun->size_bytes, be_lun->blocksize); 1866 } 1867 1868 be_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / be_lun->blocksize; 1869 return (error); 1870 } 1871 1872 static int 1873 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 1874 { 1875 struct ctl_lun_create_params *params; 1876 struct vattr vattr; 1877 struct cdev *dev; 1878 struct cdevsw *devsw; 1879 char *value; 1880 int error, atomic, maxio, unmap; 1881 off_t ps, pss, po, pos, us, uss, uo, uos; 1882 1883 params = &be_lun->params; 1884 1885 be_lun->dev_type = CTL_BE_BLOCK_DEV; 1886 be_lun->backend.dev.cdev = be_lun->vn->v_rdev; 1887 be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, 1888 &be_lun->backend.dev.dev_ref); 1889 if (be_lun->backend.dev.csw == NULL) 1890 panic("Unable to retrieve device switch"); 1891 if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) { 1892 be_lun->dispatch = ctl_be_block_dispatch_zvol; 1893 be_lun->get_lba_status = ctl_be_block_gls_zvol; 1894 atomic = maxio = CTLBLK_MAX_IO_SIZE; 1895 } else { 1896 be_lun->dispatch = ctl_be_block_dispatch_dev; 1897 atomic = 0; 1898 maxio = be_lun->backend.dev.cdev->si_iosize_max; 1899 if (maxio <= 0) 1900 maxio = DFLTPHYS; 1901 if (maxio > CTLBLK_MAX_IO_SIZE) 1902 maxio = CTLBLK_MAX_IO_SIZE; 1903 } 1904 be_lun->lun_flush = ctl_be_block_flush_dev; 1905 be_lun->getattr = ctl_be_block_getattr_dev; 1906 1907 error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); 1908 if (error) { 1909 snprintf(req->error_str, sizeof(req->error_str), 1910 "error getting vnode attributes for device %s", 1911 be_lun->dev_path); 1912 return (error); 1913 } 1914 1915 dev = be_lun->vn->v_rdev; 1916 devsw = dev->si_devsw; 1917 if (!devsw->d_ioctl) { 1918 snprintf(req->error_str, sizeof(req->error_str), 1919 "no d_ioctl for device %s!", 1920 be_lun->dev_path); 1921 return (ENODEV); 1922 } 1923 1924 error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, 1925 (caddr_t)&be_lun->blocksize, FREAD, 1926 curthread); 1927 if (error) { 1928 snprintf(req->error_str, sizeof(req->error_str), 1929 "error %d returned for DIOCGSECTORSIZE ioctl " 1930 "on %s!", error, be_lun->dev_path); 1931 return (error); 1932 } 1933 1934 /* 1935 * If the user has asked for a blocksize that is greater than the 1936 * backing device's blocksize, we can do it only if the blocksize 1937 * the user is asking for is an even multiple of the underlying 1938 * device's blocksize. 1939 */ 1940 if ((params->blocksize_bytes != 0) 1941 && (params->blocksize_bytes > be_lun->blocksize)) { 1942 uint32_t bs_multiple, tmp_blocksize; 1943 1944 bs_multiple = params->blocksize_bytes / be_lun->blocksize; 1945 1946 tmp_blocksize = bs_multiple * be_lun->blocksize; 1947 1948 if (tmp_blocksize == params->blocksize_bytes) { 1949 be_lun->blocksize = params->blocksize_bytes; 1950 } else { 1951 snprintf(req->error_str, sizeof(req->error_str), 1952 "requested blocksize %u is not an even " 1953 "multiple of backing device blocksize %u", 1954 params->blocksize_bytes, 1955 be_lun->blocksize); 1956 return (EINVAL); 1957 1958 } 1959 } else if ((params->blocksize_bytes != 0) 1960 && (params->blocksize_bytes != be_lun->blocksize)) { 1961 snprintf(req->error_str, sizeof(req->error_str), 1962 "requested blocksize %u < backing device " 1963 "blocksize %u", params->blocksize_bytes, 1964 be_lun->blocksize); 1965 return (EINVAL); 1966 } 1967 1968 error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, 1969 (caddr_t)&be_lun->size_bytes, FREAD, 1970 curthread); 1971 if (error) { 1972 snprintf(req->error_str, sizeof(req->error_str), 1973 "error %d returned for DIOCGMEDIASIZE " 1974 " ioctl on %s!", error, 1975 be_lun->dev_path); 1976 return (error); 1977 } 1978 1979 if (params->lun_size_bytes != 0) { 1980 if (params->lun_size_bytes > be_lun->size_bytes) { 1981 snprintf(req->error_str, sizeof(req->error_str), 1982 "requested LUN size %ju > backing device " 1983 "size %ju", 1984 (uintmax_t)params->lun_size_bytes, 1985 (uintmax_t)be_lun->size_bytes); 1986 return (EINVAL); 1987 } 1988 1989 be_lun->size_bytes = params->lun_size_bytes; 1990 } 1991 1992 error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, 1993 (caddr_t)&ps, FREAD, curthread); 1994 if (error) 1995 ps = po = 0; 1996 else { 1997 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, 1998 (caddr_t)&po, FREAD, curthread); 1999 if (error) 2000 po = 0; 2001 } 2002 us = ps; 2003 uo = po; 2004 2005 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblocksize"); 2006 if (value != NULL) 2007 ctl_expand_number(value, &ps); 2008 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "pblockoffset"); 2009 if (value != NULL) 2010 ctl_expand_number(value, &po); 2011 pss = ps / be_lun->blocksize; 2012 pos = po / be_lun->blocksize; 2013 if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && 2014 ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { 2015 be_lun->pblockexp = fls(pss) - 1; 2016 be_lun->pblockoff = (pss - pos) % pss; 2017 } 2018 2019 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublocksize"); 2020 if (value != NULL) 2021 ctl_expand_number(value, &us); 2022 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "ublockoffset"); 2023 if (value != NULL) 2024 ctl_expand_number(value, &uo); 2025 uss = us / be_lun->blocksize; 2026 uos = uo / be_lun->blocksize; 2027 if ((uss > 0) && (uss * be_lun->blocksize == us) && (uss >= uos) && 2028 ((uss & (uss - 1)) == 0) && (uos * be_lun->blocksize == uo)) { 2029 be_lun->ublockexp = fls(uss) - 1; 2030 be_lun->ublockoff = (uss - uos) % uss; 2031 } 2032 2033 be_lun->atomicblock = atomic / be_lun->blocksize; 2034 be_lun->opttxferlen = maxio / be_lun->blocksize; 2035 2036 if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { 2037 unmap = 1; 2038 } else { 2039 struct diocgattr_arg arg; 2040 2041 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 2042 arg.len = sizeof(arg.value.i); 2043 error = devsw->d_ioctl(dev, DIOCGATTR, 2044 (caddr_t)&arg, FREAD, curthread); 2045 unmap = (error == 0) ? arg.value.i : 0; 2046 } 2047 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); 2048 if (value != NULL) 2049 unmap = (strcmp(value, "on") == 0); 2050 if (unmap) 2051 be_lun->unmap = ctl_be_block_unmap_dev; 2052 2053 return (0); 2054 } 2055 2056 static int 2057 ctl_be_block_close(struct ctl_be_block_lun *be_lun) 2058 { 2059 DROP_GIANT(); 2060 if (be_lun->vn) { 2061 int flags = FREAD | FWRITE; 2062 2063 switch (be_lun->dev_type) { 2064 case CTL_BE_BLOCK_DEV: 2065 if (be_lun->backend.dev.csw) { 2066 dev_relthread(be_lun->backend.dev.cdev, 2067 be_lun->backend.dev.dev_ref); 2068 be_lun->backend.dev.csw = NULL; 2069 be_lun->backend.dev.cdev = NULL; 2070 } 2071 break; 2072 case CTL_BE_BLOCK_FILE: 2073 break; 2074 case CTL_BE_BLOCK_NONE: 2075 break; 2076 default: 2077 panic("Unexpected backend type."); 2078 break; 2079 } 2080 2081 (void)vn_close(be_lun->vn, flags, NOCRED, curthread); 2082 be_lun->vn = NULL; 2083 2084 switch (be_lun->dev_type) { 2085 case CTL_BE_BLOCK_DEV: 2086 break; 2087 case CTL_BE_BLOCK_FILE: 2088 if (be_lun->backend.file.cred != NULL) { 2089 crfree(be_lun->backend.file.cred); 2090 be_lun->backend.file.cred = NULL; 2091 } 2092 break; 2093 case CTL_BE_BLOCK_NONE: 2094 break; 2095 default: 2096 panic("Unexpected backend type."); 2097 break; 2098 } 2099 be_lun->dev_type = CTL_BE_BLOCK_NONE; 2100 } 2101 PICKUP_GIANT(); 2102 2103 return (0); 2104 } 2105 2106 static int 2107 ctl_be_block_open(struct ctl_be_block_softc *softc, 2108 struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) 2109 { 2110 struct nameidata nd; 2111 int flags; 2112 int error; 2113 2114 /* 2115 * XXX KDM allow a read-only option? 2116 */ 2117 flags = FREAD | FWRITE; 2118 error = 0; 2119 2120 if (rootvnode == NULL) { 2121 snprintf(req->error_str, sizeof(req->error_str), 2122 "Root filesystem is not mounted"); 2123 return (1); 2124 } 2125 2126 pwd_ensure_dirs(); 2127 2128 again: 2129 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); 2130 error = vn_open(&nd, &flags, 0, NULL); 2131 if (error) { 2132 /* 2133 * This is the only reasonable guess we can make as far as 2134 * path if the user doesn't give us a fully qualified path. 2135 * If they want to specify a file, they need to specify the 2136 * full path. 2137 */ 2138 if (be_lun->dev_path[0] != '/') { 2139 char *dev_path = "/dev/"; 2140 char *dev_name; 2141 2142 /* Try adding device path at beginning of name */ 2143 dev_name = malloc(strlen(be_lun->dev_path) 2144 + strlen(dev_path) + 1, 2145 M_CTLBLK, M_WAITOK); 2146 if (dev_name) { 2147 sprintf(dev_name, "%s%s", dev_path, 2148 be_lun->dev_path); 2149 free(be_lun->dev_path, M_CTLBLK); 2150 be_lun->dev_path = dev_name; 2151 goto again; 2152 } 2153 } 2154 snprintf(req->error_str, sizeof(req->error_str), 2155 "error opening %s: %d", be_lun->dev_path, error); 2156 return (error); 2157 } 2158 2159 NDFREE(&nd, NDF_ONLY_PNBUF); 2160 2161 be_lun->vn = nd.ni_vp; 2162 2163 /* We only support disks and files. */ 2164 if (vn_isdisk(be_lun->vn, &error)) { 2165 error = ctl_be_block_open_dev(be_lun, req); 2166 } else if (be_lun->vn->v_type == VREG) { 2167 error = ctl_be_block_open_file(be_lun, req); 2168 } else { 2169 error = EINVAL; 2170 snprintf(req->error_str, sizeof(req->error_str), 2171 "%s is not a disk or plain file", be_lun->dev_path); 2172 } 2173 VOP_UNLOCK(be_lun->vn, 0); 2174 2175 if (error != 0) { 2176 ctl_be_block_close(be_lun); 2177 return (error); 2178 } 2179 2180 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2181 be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; 2182 2183 return (0); 2184 } 2185 2186 static int 2187 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2188 { 2189 struct ctl_be_block_lun *be_lun; 2190 struct ctl_lun_create_params *params; 2191 char num_thread_str[16]; 2192 char tmpstr[32]; 2193 char *value; 2194 int retval, num_threads; 2195 int tmp_num_threads; 2196 2197 params = &req->reqdata.create; 2198 retval = 0; 2199 req->status = CTL_LUN_OK; 2200 2201 num_threads = cbb_num_threads; 2202 2203 be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); 2204 2205 be_lun->params = req->reqdata.create; 2206 be_lun->softc = softc; 2207 STAILQ_INIT(&be_lun->input_queue); 2208 STAILQ_INIT(&be_lun->config_read_queue); 2209 STAILQ_INIT(&be_lun->config_write_queue); 2210 STAILQ_INIT(&be_lun->datamove_queue); 2211 sprintf(be_lun->lunname, "cblk%d", softc->num_luns); 2212 mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); 2213 mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); 2214 ctl_init_opts(&be_lun->ctl_be_lun.options, 2215 req->num_be_args, req->kern_be_args); 2216 2217 be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, 2218 NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); 2219 2220 if (be_lun->lun_zone == NULL) { 2221 snprintf(req->error_str, sizeof(req->error_str), 2222 "error allocating UMA zone"); 2223 goto bailout_error; 2224 } 2225 2226 if (params->flags & CTL_LUN_FLAG_DEV_TYPE) 2227 be_lun->ctl_be_lun.lun_type = params->device_type; 2228 else 2229 be_lun->ctl_be_lun.lun_type = T_DIRECT; 2230 2231 if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { 2232 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); 2233 if (value == NULL) { 2234 snprintf(req->error_str, sizeof(req->error_str), 2235 "no file argument specified"); 2236 goto bailout_error; 2237 } 2238 be_lun->dev_path = strdup(value, M_CTLBLK); 2239 be_lun->blocksize = 512; 2240 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; 2241 2242 retval = ctl_be_block_open(softc, be_lun, req); 2243 if (retval != 0) { 2244 retval = 0; 2245 req->status = CTL_LUN_WARNING; 2246 } 2247 } else { 2248 /* 2249 * For processor devices, we don't have any size. 2250 */ 2251 be_lun->blocksize = 0; 2252 be_lun->pblockexp = 0; 2253 be_lun->pblockoff = 0; 2254 be_lun->ublockexp = 0; 2255 be_lun->ublockoff = 0; 2256 be_lun->size_blocks = 0; 2257 be_lun->size_bytes = 0; 2258 be_lun->ctl_be_lun.maxlba = 0; 2259 2260 /* 2261 * Default to just 1 thread for processor devices. 2262 */ 2263 num_threads = 1; 2264 } 2265 2266 /* 2267 * XXX This searching loop might be refactored to be combined with 2268 * the loop above, 2269 */ 2270 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 2271 if (value != NULL) { 2272 tmp_num_threads = strtol(value, NULL, 0); 2273 2274 /* 2275 * We don't let the user specify less than one 2276 * thread, but hope he's clueful enough not to 2277 * specify 1000 threads. 2278 */ 2279 if (tmp_num_threads < 1) { 2280 snprintf(req->error_str, sizeof(req->error_str), 2281 "invalid number of threads %s", 2282 num_thread_str); 2283 goto bailout_error; 2284 } 2285 num_threads = tmp_num_threads; 2286 } 2287 2288 be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; 2289 be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; 2290 if (be_lun->vn == NULL) 2291 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; 2292 if (be_lun->unmap != NULL) 2293 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2294 if (be_lun->dispatch != ctl_be_block_dispatch_dev) 2295 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_SERSEQ_READ; 2296 be_lun->ctl_be_lun.be_lun = be_lun; 2297 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2298 0 : (be_lun->size_blocks - 1); 2299 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2300 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2301 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2302 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2303 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2304 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2305 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2306 /* Tell the user the blocksize we ended up using */ 2307 params->lun_size_bytes = be_lun->size_bytes; 2308 params->blocksize_bytes = be_lun->blocksize; 2309 if (params->flags & CTL_LUN_FLAG_ID_REQ) { 2310 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; 2311 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; 2312 } else 2313 be_lun->ctl_be_lun.req_lun_id = 0; 2314 2315 be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; 2316 be_lun->ctl_be_lun.lun_config_status = 2317 ctl_be_block_lun_config_status; 2318 be_lun->ctl_be_lun.be = &ctl_be_block_driver; 2319 2320 if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { 2321 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", 2322 softc->num_luns); 2323 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, 2324 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2325 sizeof(tmpstr))); 2326 2327 /* Tell the user what we used for a serial number */ 2328 strncpy((char *)params->serial_num, tmpstr, 2329 MIN(sizeof(params->serial_num), sizeof(tmpstr))); 2330 } else { 2331 strncpy((char *)be_lun->ctl_be_lun.serial_num, 2332 params->serial_num, 2333 MIN(sizeof(be_lun->ctl_be_lun.serial_num), 2334 sizeof(params->serial_num))); 2335 } 2336 if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { 2337 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); 2338 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, 2339 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2340 sizeof(tmpstr))); 2341 2342 /* Tell the user what we used for a device ID */ 2343 strncpy((char *)params->device_id, tmpstr, 2344 MIN(sizeof(params->device_id), sizeof(tmpstr))); 2345 } else { 2346 strncpy((char *)be_lun->ctl_be_lun.device_id, 2347 params->device_id, 2348 MIN(sizeof(be_lun->ctl_be_lun.device_id), 2349 sizeof(params->device_id))); 2350 } 2351 2352 TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); 2353 2354 be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, 2355 taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); 2356 2357 if (be_lun->io_taskqueue == NULL) { 2358 snprintf(req->error_str, sizeof(req->error_str), 2359 "unable to create taskqueue"); 2360 goto bailout_error; 2361 } 2362 2363 /* 2364 * Note that we start the same number of threads by default for 2365 * both the file case and the block device case. For the file 2366 * case, we need multiple threads to allow concurrency, because the 2367 * vnode interface is designed to be a blocking interface. For the 2368 * block device case, ZFS zvols at least will block the caller's 2369 * context in many instances, and so we need multiple threads to 2370 * overcome that problem. Other block devices don't need as many 2371 * threads, but they shouldn't cause too many problems. 2372 * 2373 * If the user wants to just have a single thread for a block 2374 * device, he can specify that when the LUN is created, or change 2375 * the tunable/sysctl to alter the default number of threads. 2376 */ 2377 retval = taskqueue_start_threads(&be_lun->io_taskqueue, 2378 /*num threads*/num_threads, 2379 /*priority*/PWAIT, 2380 /*thread name*/ 2381 "%s taskq", be_lun->lunname); 2382 2383 if (retval != 0) 2384 goto bailout_error; 2385 2386 be_lun->num_threads = num_threads; 2387 2388 mtx_lock(&softc->lock); 2389 softc->num_luns++; 2390 STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); 2391 2392 mtx_unlock(&softc->lock); 2393 2394 retval = ctl_add_lun(&be_lun->ctl_be_lun); 2395 if (retval != 0) { 2396 mtx_lock(&softc->lock); 2397 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2398 links); 2399 softc->num_luns--; 2400 mtx_unlock(&softc->lock); 2401 snprintf(req->error_str, sizeof(req->error_str), 2402 "ctl_add_lun() returned error %d, see dmesg for " 2403 "details", retval); 2404 retval = 0; 2405 goto bailout_error; 2406 } 2407 2408 mtx_lock(&softc->lock); 2409 2410 /* 2411 * Tell the config_status routine that we're waiting so it won't 2412 * clean up the LUN in the event of an error. 2413 */ 2414 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2415 2416 while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { 2417 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2418 if (retval == EINTR) 2419 break; 2420 } 2421 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2422 2423 if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { 2424 snprintf(req->error_str, sizeof(req->error_str), 2425 "LUN configuration error, see dmesg for details"); 2426 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, 2427 links); 2428 softc->num_luns--; 2429 mtx_unlock(&softc->lock); 2430 goto bailout_error; 2431 } else { 2432 params->req_lun_id = be_lun->ctl_be_lun.lun_id; 2433 } 2434 2435 mtx_unlock(&softc->lock); 2436 2437 be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, 2438 be_lun->blocksize, 2439 DEVSTAT_ALL_SUPPORTED, 2440 be_lun->ctl_be_lun.lun_type 2441 | DEVSTAT_TYPE_IF_OTHER, 2442 DEVSTAT_PRIORITY_OTHER); 2443 2444 return (retval); 2445 2446 bailout_error: 2447 req->status = CTL_LUN_ERROR; 2448 2449 if (be_lun->io_taskqueue != NULL) 2450 taskqueue_free(be_lun->io_taskqueue); 2451 ctl_be_block_close(be_lun); 2452 if (be_lun->dev_path != NULL) 2453 free(be_lun->dev_path, M_CTLBLK); 2454 if (be_lun->lun_zone != NULL) 2455 uma_zdestroy(be_lun->lun_zone); 2456 ctl_free_opts(&be_lun->ctl_be_lun.options); 2457 mtx_destroy(&be_lun->queue_lock); 2458 mtx_destroy(&be_lun->io_lock); 2459 free(be_lun, M_CTLBLK); 2460 2461 return (retval); 2462 } 2463 2464 static int 2465 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2466 { 2467 struct ctl_lun_rm_params *params; 2468 struct ctl_be_block_lun *be_lun; 2469 int retval; 2470 2471 params = &req->reqdata.rm; 2472 2473 mtx_lock(&softc->lock); 2474 2475 be_lun = NULL; 2476 2477 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2478 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2479 break; 2480 } 2481 mtx_unlock(&softc->lock); 2482 2483 if (be_lun == NULL) { 2484 snprintf(req->error_str, sizeof(req->error_str), 2485 "LUN %u is not managed by the block backend", 2486 params->lun_id); 2487 goto bailout_error; 2488 } 2489 2490 retval = ctl_disable_lun(&be_lun->ctl_be_lun); 2491 2492 if (retval != 0) { 2493 snprintf(req->error_str, sizeof(req->error_str), 2494 "error %d returned from ctl_disable_lun() for " 2495 "LUN %d", retval, params->lun_id); 2496 goto bailout_error; 2497 2498 } 2499 2500 retval = ctl_invalidate_lun(&be_lun->ctl_be_lun); 2501 if (retval != 0) { 2502 snprintf(req->error_str, sizeof(req->error_str), 2503 "error %d returned from ctl_invalidate_lun() for " 2504 "LUN %d", retval, params->lun_id); 2505 goto bailout_error; 2506 } 2507 2508 mtx_lock(&softc->lock); 2509 2510 be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; 2511 2512 while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2513 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); 2514 if (retval == EINTR) 2515 break; 2516 } 2517 2518 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; 2519 2520 if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { 2521 snprintf(req->error_str, sizeof(req->error_str), 2522 "interrupted waiting for LUN to be freed"); 2523 mtx_unlock(&softc->lock); 2524 goto bailout_error; 2525 } 2526 2527 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); 2528 2529 softc->num_luns--; 2530 mtx_unlock(&softc->lock); 2531 2532 taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task); 2533 2534 taskqueue_free(be_lun->io_taskqueue); 2535 2536 ctl_be_block_close(be_lun); 2537 2538 if (be_lun->disk_stats != NULL) 2539 devstat_remove_entry(be_lun->disk_stats); 2540 2541 uma_zdestroy(be_lun->lun_zone); 2542 2543 ctl_free_opts(&be_lun->ctl_be_lun.options); 2544 free(be_lun->dev_path, M_CTLBLK); 2545 mtx_destroy(&be_lun->queue_lock); 2546 mtx_destroy(&be_lun->io_lock); 2547 free(be_lun, M_CTLBLK); 2548 2549 req->status = CTL_LUN_OK; 2550 2551 return (0); 2552 2553 bailout_error: 2554 2555 req->status = CTL_LUN_ERROR; 2556 2557 return (0); 2558 } 2559 2560 static int 2561 ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun, 2562 struct ctl_lun_req *req) 2563 { 2564 struct vattr vattr; 2565 int error; 2566 struct ctl_lun_create_params *params = &be_lun->params; 2567 2568 if (params->lun_size_bytes != 0) { 2569 be_lun->size_bytes = params->lun_size_bytes; 2570 } else { 2571 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); 2572 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); 2573 VOP_UNLOCK(be_lun->vn, 0); 2574 if (error != 0) { 2575 snprintf(req->error_str, sizeof(req->error_str), 2576 "error calling VOP_GETATTR() for file %s", 2577 be_lun->dev_path); 2578 return (error); 2579 } 2580 2581 be_lun->size_bytes = vattr.va_size; 2582 } 2583 2584 return (0); 2585 } 2586 2587 static int 2588 ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 2589 struct ctl_lun_req *req) 2590 { 2591 struct ctl_be_block_devdata *dev_data; 2592 int error; 2593 struct ctl_lun_create_params *params = &be_lun->params; 2594 uint64_t size_bytes; 2595 2596 dev_data = &be_lun->backend.dev; 2597 if (!dev_data->csw->d_ioctl) { 2598 snprintf(req->error_str, sizeof(req->error_str), 2599 "no d_ioctl for device %s!", be_lun->dev_path); 2600 return (ENODEV); 2601 } 2602 2603 error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, 2604 (caddr_t)&size_bytes, FREAD, 2605 curthread); 2606 if (error) { 2607 snprintf(req->error_str, sizeof(req->error_str), 2608 "error %d returned for DIOCGMEDIASIZE ioctl " 2609 "on %s!", error, be_lun->dev_path); 2610 return (error); 2611 } 2612 2613 if (params->lun_size_bytes != 0) { 2614 if (params->lun_size_bytes > size_bytes) { 2615 snprintf(req->error_str, sizeof(req->error_str), 2616 "requested LUN size %ju > backing device " 2617 "size %ju", 2618 (uintmax_t)params->lun_size_bytes, 2619 (uintmax_t)size_bytes); 2620 return (EINVAL); 2621 } 2622 2623 be_lun->size_bytes = params->lun_size_bytes; 2624 } else { 2625 be_lun->size_bytes = size_bytes; 2626 } 2627 2628 return (0); 2629 } 2630 2631 static int 2632 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) 2633 { 2634 struct ctl_lun_modify_params *params; 2635 struct ctl_be_block_lun *be_lun; 2636 uint64_t oldsize; 2637 int error; 2638 2639 params = &req->reqdata.modify; 2640 2641 mtx_lock(&softc->lock); 2642 be_lun = NULL; 2643 STAILQ_FOREACH(be_lun, &softc->lun_list, links) { 2644 if (be_lun->ctl_be_lun.lun_id == params->lun_id) 2645 break; 2646 } 2647 mtx_unlock(&softc->lock); 2648 2649 if (be_lun == NULL) { 2650 snprintf(req->error_str, sizeof(req->error_str), 2651 "LUN %u is not managed by the block backend", 2652 params->lun_id); 2653 goto bailout_error; 2654 } 2655 2656 be_lun->params.lun_size_bytes = params->lun_size_bytes; 2657 2658 oldsize = be_lun->size_bytes; 2659 if (be_lun->vn == NULL) 2660 error = ctl_be_block_open(softc, be_lun, req); 2661 else if (vn_isdisk(be_lun->vn, &error)) 2662 error = ctl_be_block_modify_dev(be_lun, req); 2663 else if (be_lun->vn->v_type == VREG) 2664 error = ctl_be_block_modify_file(be_lun, req); 2665 else 2666 error = EINVAL; 2667 2668 if (error == 0 && be_lun->size_bytes != oldsize) { 2669 be_lun->size_blocks = be_lun->size_bytes >> 2670 be_lun->blocksize_shift; 2671 2672 /* 2673 * The maximum LBA is the size - 1. 2674 * 2675 * XXX: Note that this field is being updated without locking, 2676 * which might cause problems on 32-bit architectures. 2677 */ 2678 if (be_lun->unmap != NULL) 2679 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; 2680 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 2681 0 : (be_lun->size_blocks - 1); 2682 be_lun->ctl_be_lun.blocksize = be_lun->blocksize; 2683 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; 2684 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; 2685 be_lun->ctl_be_lun.ublockexp = be_lun->ublockexp; 2686 be_lun->ctl_be_lun.ublockoff = be_lun->ublockoff; 2687 be_lun->ctl_be_lun.atomicblock = be_lun->atomicblock; 2688 be_lun->ctl_be_lun.opttxferlen = be_lun->opttxferlen; 2689 ctl_lun_capacity_changed(&be_lun->ctl_be_lun); 2690 if (oldsize == 0 && be_lun->size_blocks != 0) 2691 ctl_lun_online(&be_lun->ctl_be_lun); 2692 } 2693 2694 /* Tell the user the exact size we ended up using */ 2695 params->lun_size_bytes = be_lun->size_bytes; 2696 2697 req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; 2698 2699 return (0); 2700 2701 bailout_error: 2702 req->status = CTL_LUN_ERROR; 2703 2704 return (0); 2705 } 2706 2707 static void 2708 ctl_be_block_lun_shutdown(void *be_lun) 2709 { 2710 struct ctl_be_block_lun *lun; 2711 struct ctl_be_block_softc *softc; 2712 2713 lun = (struct ctl_be_block_lun *)be_lun; 2714 2715 softc = lun->softc; 2716 2717 mtx_lock(&softc->lock); 2718 lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; 2719 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2720 wakeup(lun); 2721 mtx_unlock(&softc->lock); 2722 2723 } 2724 2725 static void 2726 ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) 2727 { 2728 struct ctl_be_block_lun *lun; 2729 struct ctl_be_block_softc *softc; 2730 2731 lun = (struct ctl_be_block_lun *)be_lun; 2732 softc = lun->softc; 2733 2734 if (status == CTL_LUN_CONFIG_OK) { 2735 mtx_lock(&softc->lock); 2736 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2737 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) 2738 wakeup(lun); 2739 mtx_unlock(&softc->lock); 2740 2741 /* 2742 * We successfully added the LUN, attempt to enable it. 2743 */ 2744 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) { 2745 printf("%s: ctl_enable_lun() failed!\n", __func__); 2746 if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) { 2747 printf("%s: ctl_invalidate_lun() failed!\n", 2748 __func__); 2749 } 2750 } 2751 2752 return; 2753 } 2754 2755 2756 mtx_lock(&softc->lock); 2757 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; 2758 lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; 2759 wakeup(lun); 2760 mtx_unlock(&softc->lock); 2761 } 2762 2763 2764 static int 2765 ctl_be_block_config_write(union ctl_io *io) 2766 { 2767 struct ctl_be_block_lun *be_lun; 2768 struct ctl_be_lun *ctl_be_lun; 2769 int retval; 2770 2771 retval = 0; 2772 2773 DPRINTF("entered\n"); 2774 2775 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2776 CTL_PRIV_BACKEND_LUN].ptr; 2777 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2778 2779 switch (io->scsiio.cdb[0]) { 2780 case SYNCHRONIZE_CACHE: 2781 case SYNCHRONIZE_CACHE_16: 2782 case WRITE_SAME_10: 2783 case WRITE_SAME_16: 2784 case UNMAP: 2785 /* 2786 * The upper level CTL code will filter out any CDBs with 2787 * the immediate bit set and return the proper error. 2788 * 2789 * We don't really need to worry about what LBA range the 2790 * user asked to be synced out. When they issue a sync 2791 * cache command, we'll sync out the whole thing. 2792 */ 2793 mtx_lock(&be_lun->queue_lock); 2794 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, 2795 links); 2796 mtx_unlock(&be_lun->queue_lock); 2797 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); 2798 break; 2799 case START_STOP_UNIT: { 2800 struct scsi_start_stop_unit *cdb; 2801 2802 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; 2803 2804 if (cdb->how & SSS_START) 2805 retval = ctl_start_lun(ctl_be_lun); 2806 else { 2807 retval = ctl_stop_lun(ctl_be_lun); 2808 /* 2809 * XXX KDM Copan-specific offline behavior. 2810 * Figure out a reasonable way to port this? 2811 */ 2812 #ifdef NEEDTOPORT 2813 if ((retval == 0) 2814 && (cdb->byte2 & SSS_ONOFFLINE)) 2815 retval = ctl_lun_offline(ctl_be_lun); 2816 #endif 2817 } 2818 2819 /* 2820 * In general, the above routines should not fail. They 2821 * just set state for the LUN. So we've got something 2822 * pretty wrong here if we can't start or stop the LUN. 2823 */ 2824 if (retval != 0) { 2825 ctl_set_internal_failure(&io->scsiio, 2826 /*sks_valid*/ 1, 2827 /*retry_count*/ 0xf051); 2828 retval = CTL_RETVAL_COMPLETE; 2829 } else { 2830 ctl_set_success(&io->scsiio); 2831 } 2832 ctl_config_write_done(io); 2833 break; 2834 } 2835 default: 2836 ctl_set_invalid_opcode(&io->scsiio); 2837 ctl_config_write_done(io); 2838 retval = CTL_RETVAL_COMPLETE; 2839 break; 2840 } 2841 2842 return (retval); 2843 } 2844 2845 static int 2846 ctl_be_block_config_read(union ctl_io *io) 2847 { 2848 struct ctl_be_block_lun *be_lun; 2849 struct ctl_be_lun *ctl_be_lun; 2850 int retval = 0; 2851 2852 DPRINTF("entered\n"); 2853 2854 ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ 2855 CTL_PRIV_BACKEND_LUN].ptr; 2856 be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 2857 2858 switch (io->scsiio.cdb[0]) { 2859 case SERVICE_ACTION_IN: 2860 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { 2861 mtx_lock(&be_lun->queue_lock); 2862 STAILQ_INSERT_TAIL(&be_lun->config_read_queue, 2863 &io->io_hdr, links); 2864 mtx_unlock(&be_lun->queue_lock); 2865 taskqueue_enqueue(be_lun->io_taskqueue, 2866 &be_lun->io_task); 2867 retval = CTL_RETVAL_QUEUED; 2868 break; 2869 } 2870 ctl_set_invalid_field(&io->scsiio, 2871 /*sks_valid*/ 1, 2872 /*command*/ 1, 2873 /*field*/ 1, 2874 /*bit_valid*/ 1, 2875 /*bit*/ 4); 2876 ctl_config_read_done(io); 2877 retval = CTL_RETVAL_COMPLETE; 2878 break; 2879 default: 2880 ctl_set_invalid_opcode(&io->scsiio); 2881 ctl_config_read_done(io); 2882 retval = CTL_RETVAL_COMPLETE; 2883 break; 2884 } 2885 2886 return (retval); 2887 } 2888 2889 static int 2890 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) 2891 { 2892 struct ctl_be_block_lun *lun; 2893 int retval; 2894 2895 lun = (struct ctl_be_block_lun *)be_lun; 2896 retval = 0; 2897 2898 retval = sbuf_printf(sb, "\t<num_threads>"); 2899 2900 if (retval != 0) 2901 goto bailout; 2902 2903 retval = sbuf_printf(sb, "%d", lun->num_threads); 2904 2905 if (retval != 0) 2906 goto bailout; 2907 2908 retval = sbuf_printf(sb, "</num_threads>\n"); 2909 2910 bailout: 2911 2912 return (retval); 2913 } 2914 2915 static uint64_t 2916 ctl_be_block_lun_attr(void *be_lun, const char *attrname) 2917 { 2918 struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; 2919 2920 if (lun->getattr == NULL) 2921 return (UINT64_MAX); 2922 return (lun->getattr(lun, attrname)); 2923 } 2924 2925 int 2926 ctl_be_block_init(void) 2927 { 2928 struct ctl_be_block_softc *softc; 2929 int retval; 2930 2931 softc = &backend_block_softc; 2932 retval = 0; 2933 2934 mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); 2935 beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), 2936 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2937 STAILQ_INIT(&softc->disk_list); 2938 STAILQ_INIT(&softc->lun_list); 2939 2940 return (retval); 2941 } 2942