1 /* 2 * Copyright (C) 2013 Shaohua Li <shli@kernel.org> 3 * Copyright (C) 2014 Red Hat, Inc. 4 * Copyright (C) 2015 Arrikto, Inc. 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms and conditions of the GNU General Public License, 8 * version 2, as published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 * more details. 14 * 15 * You should have received a copy of the GNU General Public License along with 16 * this program; if not, write to the Free Software Foundation, Inc., 17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. 18 */ 19 20 #include <linux/spinlock.h> 21 #include <linux/module.h> 22 #include <linux/idr.h> 23 #include <linux/kernel.h> 24 #include <linux/timer.h> 25 #include <linux/parser.h> 26 #include <linux/vmalloc.h> 27 #include <linux/uio_driver.h> 28 #include <linux/stringify.h> 29 #include <linux/bitops.h> 30 #include <linux/highmem.h> 31 #include <net/genetlink.h> 32 #include <scsi/scsi_common.h> 33 #include <scsi/scsi_proto.h> 34 #include <target/target_core_base.h> 35 #include <target/target_core_fabric.h> 36 #include <target/target_core_backend.h> 37 38 #include <linux/target_core_user.h> 39 40 /* 41 * Define a shared-memory interface for LIO to pass SCSI commands and 42 * data to userspace for processing. This is to allow backends that 43 * are too complex for in-kernel support to be possible. 44 * 45 * It uses the UIO framework to do a lot of the device-creation and 46 * introspection work for us. 47 * 48 * See the .h file for how the ring is laid out. Note that while the 49 * command ring is defined, the particulars of the data area are 50 * not. Offset values in the command entry point to other locations 51 * internal to the mmap()ed area. There is separate space outside the 52 * command ring for data buffers. This leaves maximum flexibility for 53 * moving buffer allocations, or even page flipping or other 54 * allocation techniques, without altering the command ring layout. 55 * 56 * SECURITY: 57 * The user process must be assumed to be malicious. There's no way to 58 * prevent it breaking the command ring protocol if it wants, but in 59 * order to prevent other issues we must only ever read *data* from 60 * the shared memory area, not offsets or sizes. This applies to 61 * command ring entries as well as the mailbox. Extra code needed for 62 * this may have a 'UAM' comment. 63 */ 64 65 66 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC) 67 68 #define DATA_BLOCK_BITS 256 69 #define DATA_BLOCK_SIZE 4096 70 71 #define CMDR_SIZE (16 * 4096) 72 #define DATA_SIZE (DATA_BLOCK_BITS * DATA_BLOCK_SIZE) 73 74 #define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE) 75 76 static struct device *tcmu_root_device; 77 78 struct tcmu_hba { 79 u32 host_id; 80 }; 81 82 #define TCMU_CONFIG_LEN 256 83 84 struct tcmu_dev { 85 struct se_device se_dev; 86 87 char *name; 88 struct se_hba *hba; 89 90 #define TCMU_DEV_BIT_OPEN 0 91 #define TCMU_DEV_BIT_BROKEN 1 92 unsigned long flags; 93 94 struct uio_info uio_info; 95 96 struct tcmu_mailbox *mb_addr; 97 size_t dev_size; 98 u32 cmdr_size; 99 u32 cmdr_last_cleaned; 100 /* Offset of data area from start of mb */ 101 /* Must add data_off and mb_addr to get the address */ 102 size_t data_off; 103 size_t data_size; 104 105 DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS); 106 107 wait_queue_head_t wait_cmdr; 108 /* TODO should this be a mutex? */ 109 spinlock_t cmdr_lock; 110 111 struct idr commands; 112 spinlock_t commands_lock; 113 114 struct timer_list timeout; 115 116 char dev_config[TCMU_CONFIG_LEN]; 117 }; 118 119 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev) 120 121 #define CMDR_OFF sizeof(struct tcmu_mailbox) 122 123 struct tcmu_cmd { 124 struct se_cmd *se_cmd; 125 struct tcmu_dev *tcmu_dev; 126 127 uint16_t cmd_id; 128 129 /* Can't use se_cmd when cleaning up expired cmds, because if 130 cmd has been completed then accessing se_cmd is off limits */ 131 DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS); 132 133 unsigned long deadline; 134 135 #define TCMU_CMD_BIT_EXPIRED 0 136 unsigned long flags; 137 }; 138 139 static struct kmem_cache *tcmu_cmd_cache; 140 141 /* multicast group */ 142 enum tcmu_multicast_groups { 143 TCMU_MCGRP_CONFIG, 144 }; 145 146 static const struct genl_multicast_group tcmu_mcgrps[] = { 147 [TCMU_MCGRP_CONFIG] = { .name = "config", }, 148 }; 149 150 /* Our generic netlink family */ 151 static struct genl_family tcmu_genl_family __ro_after_init = { 152 .module = THIS_MODULE, 153 .hdrsize = 0, 154 .name = "TCM-USER", 155 .version = 1, 156 .maxattr = TCMU_ATTR_MAX, 157 .mcgrps = tcmu_mcgrps, 158 .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps), 159 .netnsok = true, 160 }; 161 162 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd) 163 { 164 struct se_device *se_dev = se_cmd->se_dev; 165 struct tcmu_dev *udev = TCMU_DEV(se_dev); 166 struct tcmu_cmd *tcmu_cmd; 167 int cmd_id; 168 169 tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_KERNEL); 170 if (!tcmu_cmd) 171 return NULL; 172 173 tcmu_cmd->se_cmd = se_cmd; 174 tcmu_cmd->tcmu_dev = udev; 175 tcmu_cmd->deadline = jiffies + msecs_to_jiffies(TCMU_TIME_OUT); 176 177 idr_preload(GFP_KERNEL); 178 spin_lock_irq(&udev->commands_lock); 179 cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 0, 180 USHRT_MAX, GFP_NOWAIT); 181 spin_unlock_irq(&udev->commands_lock); 182 idr_preload_end(); 183 184 if (cmd_id < 0) { 185 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd); 186 return NULL; 187 } 188 tcmu_cmd->cmd_id = cmd_id; 189 190 return tcmu_cmd; 191 } 192 193 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size) 194 { 195 unsigned long offset = offset_in_page(vaddr); 196 197 size = round_up(size+offset, PAGE_SIZE); 198 vaddr -= offset; 199 200 while (size) { 201 flush_dcache_page(virt_to_page(vaddr)); 202 size -= PAGE_SIZE; 203 } 204 } 205 206 /* 207 * Some ring helper functions. We don't assume size is a power of 2 so 208 * we can't use circ_buf.h. 209 */ 210 static inline size_t spc_used(size_t head, size_t tail, size_t size) 211 { 212 int diff = head - tail; 213 214 if (diff >= 0) 215 return diff; 216 else 217 return size + diff; 218 } 219 220 static inline size_t spc_free(size_t head, size_t tail, size_t size) 221 { 222 /* Keep 1 byte unused or we can't tell full from empty */ 223 return (size - spc_used(head, tail, size) - 1); 224 } 225 226 static inline size_t head_to_end(size_t head, size_t size) 227 { 228 return size - head; 229 } 230 231 static inline void new_iov(struct iovec **iov, int *iov_cnt, 232 struct tcmu_dev *udev) 233 { 234 struct iovec *iovec; 235 236 if (*iov_cnt != 0) 237 (*iov)++; 238 (*iov_cnt)++; 239 240 iovec = *iov; 241 memset(iovec, 0, sizeof(struct iovec)); 242 } 243 244 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size) 245 246 /* offset is relative to mb_addr */ 247 static inline size_t get_block_offset(struct tcmu_dev *dev, 248 int block, int remaining) 249 { 250 return dev->data_off + block * DATA_BLOCK_SIZE + 251 DATA_BLOCK_SIZE - remaining; 252 } 253 254 static inline size_t iov_tail(struct tcmu_dev *udev, struct iovec *iov) 255 { 256 return (size_t)iov->iov_base + iov->iov_len; 257 } 258 259 static void alloc_and_scatter_data_area(struct tcmu_dev *udev, 260 struct scatterlist *data_sg, unsigned int data_nents, 261 struct iovec **iov, int *iov_cnt, bool copy_data) 262 { 263 int i, block; 264 int block_remaining = 0; 265 void *from, *to; 266 size_t copy_bytes, to_offset; 267 struct scatterlist *sg; 268 269 for_each_sg(data_sg, sg, data_nents, i) { 270 int sg_remaining = sg->length; 271 from = kmap_atomic(sg_page(sg)) + sg->offset; 272 while (sg_remaining > 0) { 273 if (block_remaining == 0) { 274 block = find_first_zero_bit(udev->data_bitmap, 275 DATA_BLOCK_BITS); 276 block_remaining = DATA_BLOCK_SIZE; 277 set_bit(block, udev->data_bitmap); 278 } 279 copy_bytes = min_t(size_t, sg_remaining, 280 block_remaining); 281 to_offset = get_block_offset(udev, block, 282 block_remaining); 283 to = (void *)udev->mb_addr + to_offset; 284 if (*iov_cnt != 0 && 285 to_offset == iov_tail(udev, *iov)) { 286 (*iov)->iov_len += copy_bytes; 287 } else { 288 new_iov(iov, iov_cnt, udev); 289 (*iov)->iov_base = (void __user *) to_offset; 290 (*iov)->iov_len = copy_bytes; 291 } 292 if (copy_data) { 293 memcpy(to, from + sg->length - sg_remaining, 294 copy_bytes); 295 tcmu_flush_dcache_range(to, copy_bytes); 296 } 297 sg_remaining -= copy_bytes; 298 block_remaining -= copy_bytes; 299 } 300 kunmap_atomic(from - sg->offset); 301 } 302 } 303 304 static void free_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd) 305 { 306 bitmap_xor(udev->data_bitmap, udev->data_bitmap, cmd->data_bitmap, 307 DATA_BLOCK_BITS); 308 } 309 310 static void gather_data_area(struct tcmu_dev *udev, unsigned long *cmd_bitmap, 311 struct scatterlist *data_sg, unsigned int data_nents) 312 { 313 int i, block; 314 int block_remaining = 0; 315 void *from, *to; 316 size_t copy_bytes, from_offset; 317 struct scatterlist *sg; 318 319 for_each_sg(data_sg, sg, data_nents, i) { 320 int sg_remaining = sg->length; 321 to = kmap_atomic(sg_page(sg)) + sg->offset; 322 while (sg_remaining > 0) { 323 if (block_remaining == 0) { 324 block = find_first_bit(cmd_bitmap, 325 DATA_BLOCK_BITS); 326 block_remaining = DATA_BLOCK_SIZE; 327 clear_bit(block, cmd_bitmap); 328 } 329 copy_bytes = min_t(size_t, sg_remaining, 330 block_remaining); 331 from_offset = get_block_offset(udev, block, 332 block_remaining); 333 from = (void *) udev->mb_addr + from_offset; 334 tcmu_flush_dcache_range(from, copy_bytes); 335 memcpy(to + sg->length - sg_remaining, from, 336 copy_bytes); 337 338 sg_remaining -= copy_bytes; 339 block_remaining -= copy_bytes; 340 } 341 kunmap_atomic(to - sg->offset); 342 } 343 } 344 345 static inline size_t spc_bitmap_free(unsigned long *bitmap) 346 { 347 return DATA_BLOCK_SIZE * (DATA_BLOCK_BITS - 348 bitmap_weight(bitmap, DATA_BLOCK_BITS)); 349 } 350 351 /* 352 * We can't queue a command until we have space available on the cmd ring *and* 353 * space available on the data area. 354 * 355 * Called with ring lock held. 356 */ 357 static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t data_needed) 358 { 359 struct tcmu_mailbox *mb = udev->mb_addr; 360 size_t space, cmd_needed; 361 u32 cmd_head; 362 363 tcmu_flush_dcache_range(mb, sizeof(*mb)); 364 365 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */ 366 367 /* 368 * If cmd end-of-ring space is too small then we need space for a NOP plus 369 * original cmd - cmds are internally contiguous. 370 */ 371 if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size) 372 cmd_needed = cmd_size; 373 else 374 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size); 375 376 space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size); 377 if (space < cmd_needed) { 378 pr_debug("no cmd space: %u %u %u\n", cmd_head, 379 udev->cmdr_last_cleaned, udev->cmdr_size); 380 return false; 381 } 382 383 space = spc_bitmap_free(udev->data_bitmap); 384 if (space < data_needed) { 385 pr_debug("no data space: only %zu available, but ask for %zu\n", 386 space, data_needed); 387 return false; 388 } 389 390 return true; 391 } 392 393 static sense_reason_t 394 tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd) 395 { 396 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev; 397 struct se_cmd *se_cmd = tcmu_cmd->se_cmd; 398 size_t base_command_size, command_size; 399 struct tcmu_mailbox *mb; 400 struct tcmu_cmd_entry *entry; 401 struct iovec *iov; 402 int iov_cnt; 403 uint32_t cmd_head; 404 uint64_t cdb_off; 405 bool copy_to_data_area; 406 size_t data_length; 407 DECLARE_BITMAP(old_bitmap, DATA_BLOCK_BITS); 408 409 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) 410 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE; 411 412 /* 413 * Must be a certain minimum size for response sense info, but 414 * also may be larger if the iov array is large. 415 * 416 * We prepare way too many iovs for potential uses here, because it's 417 * expensive to tell how many regions are freed in the bitmap 418 */ 419 base_command_size = max(offsetof(struct tcmu_cmd_entry, 420 req.iov[se_cmd->t_bidi_data_nents + 421 se_cmd->t_data_nents]), 422 sizeof(struct tcmu_cmd_entry)); 423 command_size = base_command_size 424 + round_up(scsi_command_size(se_cmd->t_task_cdb), TCMU_OP_ALIGN_SIZE); 425 426 WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1)); 427 428 spin_lock_irq(&udev->cmdr_lock); 429 430 mb = udev->mb_addr; 431 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */ 432 data_length = se_cmd->data_length; 433 if (se_cmd->se_cmd_flags & SCF_BIDI) { 434 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents)); 435 data_length += se_cmd->t_bidi_data_sg->length; 436 } 437 if ((command_size > (udev->cmdr_size / 2)) || 438 data_length > udev->data_size) { 439 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu " 440 "cmd ring/data area\n", command_size, data_length, 441 udev->cmdr_size, udev->data_size); 442 spin_unlock_irq(&udev->cmdr_lock); 443 return TCM_INVALID_CDB_FIELD; 444 } 445 446 while (!is_ring_space_avail(udev, command_size, data_length)) { 447 int ret; 448 DEFINE_WAIT(__wait); 449 450 prepare_to_wait(&udev->wait_cmdr, &__wait, TASK_INTERRUPTIBLE); 451 452 pr_debug("sleeping for ring space\n"); 453 spin_unlock_irq(&udev->cmdr_lock); 454 ret = schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT)); 455 finish_wait(&udev->wait_cmdr, &__wait); 456 if (!ret) { 457 pr_warn("tcmu: command timed out\n"); 458 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE; 459 } 460 461 spin_lock_irq(&udev->cmdr_lock); 462 463 /* We dropped cmdr_lock, cmd_head is stale */ 464 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */ 465 } 466 467 /* Insert a PAD if end-of-ring space is too small */ 468 if (head_to_end(cmd_head, udev->cmdr_size) < command_size) { 469 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size); 470 471 entry = (void *) mb + CMDR_OFF + cmd_head; 472 tcmu_flush_dcache_range(entry, sizeof(*entry)); 473 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD); 474 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size); 475 entry->hdr.cmd_id = 0; /* not used for PAD */ 476 entry->hdr.kflags = 0; 477 entry->hdr.uflags = 0; 478 479 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size); 480 481 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */ 482 WARN_ON(cmd_head != 0); 483 } 484 485 entry = (void *) mb + CMDR_OFF + cmd_head; 486 tcmu_flush_dcache_range(entry, sizeof(*entry)); 487 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD); 488 tcmu_hdr_set_len(&entry->hdr.len_op, command_size); 489 entry->hdr.cmd_id = tcmu_cmd->cmd_id; 490 entry->hdr.kflags = 0; 491 entry->hdr.uflags = 0; 492 493 bitmap_copy(old_bitmap, udev->data_bitmap, DATA_BLOCK_BITS); 494 495 /* Handle allocating space from the data area */ 496 iov = &entry->req.iov[0]; 497 iov_cnt = 0; 498 copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE 499 || se_cmd->se_cmd_flags & SCF_BIDI); 500 alloc_and_scatter_data_area(udev, se_cmd->t_data_sg, 501 se_cmd->t_data_nents, &iov, &iov_cnt, copy_to_data_area); 502 entry->req.iov_cnt = iov_cnt; 503 entry->req.iov_dif_cnt = 0; 504 505 /* Handle BIDI commands */ 506 iov_cnt = 0; 507 alloc_and_scatter_data_area(udev, se_cmd->t_bidi_data_sg, 508 se_cmd->t_bidi_data_nents, &iov, &iov_cnt, false); 509 entry->req.iov_bidi_cnt = iov_cnt; 510 511 /* cmd's data_bitmap is what changed in process */ 512 bitmap_xor(tcmu_cmd->data_bitmap, old_bitmap, udev->data_bitmap, 513 DATA_BLOCK_BITS); 514 515 /* All offsets relative to mb_addr, not start of entry! */ 516 cdb_off = CMDR_OFF + cmd_head + base_command_size; 517 memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb)); 518 entry->req.cdb_off = cdb_off; 519 tcmu_flush_dcache_range(entry, sizeof(*entry)); 520 521 UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size); 522 tcmu_flush_dcache_range(mb, sizeof(*mb)); 523 524 spin_unlock_irq(&udev->cmdr_lock); 525 526 /* TODO: only if FLUSH and FUA? */ 527 uio_event_notify(&udev->uio_info); 528 529 mod_timer(&udev->timeout, 530 round_jiffies_up(jiffies + msecs_to_jiffies(TCMU_TIME_OUT))); 531 532 return TCM_NO_SENSE; 533 } 534 535 static sense_reason_t 536 tcmu_queue_cmd(struct se_cmd *se_cmd) 537 { 538 struct se_device *se_dev = se_cmd->se_dev; 539 struct tcmu_dev *udev = TCMU_DEV(se_dev); 540 struct tcmu_cmd *tcmu_cmd; 541 sense_reason_t ret; 542 543 tcmu_cmd = tcmu_alloc_cmd(se_cmd); 544 if (!tcmu_cmd) 545 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE; 546 547 ret = tcmu_queue_cmd_ring(tcmu_cmd); 548 if (ret != TCM_NO_SENSE) { 549 pr_err("TCMU: Could not queue command\n"); 550 spin_lock_irq(&udev->commands_lock); 551 idr_remove(&udev->commands, tcmu_cmd->cmd_id); 552 spin_unlock_irq(&udev->commands_lock); 553 554 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd); 555 } 556 557 return ret; 558 } 559 560 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry) 561 { 562 struct se_cmd *se_cmd = cmd->se_cmd; 563 struct tcmu_dev *udev = cmd->tcmu_dev; 564 565 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) { 566 /* 567 * cmd has been completed already from timeout, just reclaim 568 * data area space and free cmd 569 */ 570 free_data_area(udev, cmd); 571 572 kmem_cache_free(tcmu_cmd_cache, cmd); 573 return; 574 } 575 576 if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) { 577 free_data_area(udev, cmd); 578 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n", 579 cmd->se_cmd); 580 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION; 581 } else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) { 582 memcpy(se_cmd->sense_buffer, entry->rsp.sense_buffer, 583 se_cmd->scsi_sense_length); 584 free_data_area(udev, cmd); 585 } else if (se_cmd->se_cmd_flags & SCF_BIDI) { 586 DECLARE_BITMAP(bitmap, DATA_BLOCK_BITS); 587 588 /* Get Data-In buffer before clean up */ 589 bitmap_copy(bitmap, cmd->data_bitmap, DATA_BLOCK_BITS); 590 gather_data_area(udev, bitmap, 591 se_cmd->t_bidi_data_sg, se_cmd->t_bidi_data_nents); 592 free_data_area(udev, cmd); 593 } else if (se_cmd->data_direction == DMA_FROM_DEVICE) { 594 DECLARE_BITMAP(bitmap, DATA_BLOCK_BITS); 595 596 bitmap_copy(bitmap, cmd->data_bitmap, DATA_BLOCK_BITS); 597 gather_data_area(udev, bitmap, 598 se_cmd->t_data_sg, se_cmd->t_data_nents); 599 free_data_area(udev, cmd); 600 } else if (se_cmd->data_direction == DMA_TO_DEVICE) { 601 free_data_area(udev, cmd); 602 } else if (se_cmd->data_direction != DMA_NONE) { 603 pr_warn("TCMU: data direction was %d!\n", 604 se_cmd->data_direction); 605 } 606 607 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status); 608 cmd->se_cmd = NULL; 609 610 kmem_cache_free(tcmu_cmd_cache, cmd); 611 } 612 613 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev) 614 { 615 struct tcmu_mailbox *mb; 616 unsigned long flags; 617 int handled = 0; 618 619 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) { 620 pr_err("ring broken, not handling completions\n"); 621 return 0; 622 } 623 624 spin_lock_irqsave(&udev->cmdr_lock, flags); 625 626 mb = udev->mb_addr; 627 tcmu_flush_dcache_range(mb, sizeof(*mb)); 628 629 while (udev->cmdr_last_cleaned != ACCESS_ONCE(mb->cmd_tail)) { 630 631 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned; 632 struct tcmu_cmd *cmd; 633 634 tcmu_flush_dcache_range(entry, sizeof(*entry)); 635 636 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) { 637 UPDATE_HEAD(udev->cmdr_last_cleaned, 638 tcmu_hdr_get_len(entry->hdr.len_op), 639 udev->cmdr_size); 640 continue; 641 } 642 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD); 643 644 spin_lock(&udev->commands_lock); 645 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id); 646 spin_unlock(&udev->commands_lock); 647 648 if (!cmd) { 649 pr_err("cmd_id not found, ring is broken\n"); 650 set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags); 651 break; 652 } 653 654 tcmu_handle_completion(cmd, entry); 655 656 UPDATE_HEAD(udev->cmdr_last_cleaned, 657 tcmu_hdr_get_len(entry->hdr.len_op), 658 udev->cmdr_size); 659 660 handled++; 661 } 662 663 if (mb->cmd_tail == mb->cmd_head) 664 del_timer(&udev->timeout); /* no more pending cmds */ 665 666 spin_unlock_irqrestore(&udev->cmdr_lock, flags); 667 668 wake_up(&udev->wait_cmdr); 669 670 return handled; 671 } 672 673 static int tcmu_check_expired_cmd(int id, void *p, void *data) 674 { 675 struct tcmu_cmd *cmd = p; 676 677 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) 678 return 0; 679 680 if (!time_after(jiffies, cmd->deadline)) 681 return 0; 682 683 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags); 684 target_complete_cmd(cmd->se_cmd, SAM_STAT_CHECK_CONDITION); 685 cmd->se_cmd = NULL; 686 687 return 0; 688 } 689 690 static void tcmu_device_timedout(unsigned long data) 691 { 692 struct tcmu_dev *udev = (struct tcmu_dev *)data; 693 unsigned long flags; 694 int handled; 695 696 handled = tcmu_handle_completions(udev); 697 698 pr_warn("%d completions handled from timeout\n", handled); 699 700 spin_lock_irqsave(&udev->commands_lock, flags); 701 idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL); 702 spin_unlock_irqrestore(&udev->commands_lock, flags); 703 704 /* 705 * We don't need to wakeup threads on wait_cmdr since they have their 706 * own timeout. 707 */ 708 } 709 710 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id) 711 { 712 struct tcmu_hba *tcmu_hba; 713 714 tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL); 715 if (!tcmu_hba) 716 return -ENOMEM; 717 718 tcmu_hba->host_id = host_id; 719 hba->hba_ptr = tcmu_hba; 720 721 return 0; 722 } 723 724 static void tcmu_detach_hba(struct se_hba *hba) 725 { 726 kfree(hba->hba_ptr); 727 hba->hba_ptr = NULL; 728 } 729 730 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name) 731 { 732 struct tcmu_dev *udev; 733 734 udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL); 735 if (!udev) 736 return NULL; 737 738 udev->name = kstrdup(name, GFP_KERNEL); 739 if (!udev->name) { 740 kfree(udev); 741 return NULL; 742 } 743 744 udev->hba = hba; 745 746 init_waitqueue_head(&udev->wait_cmdr); 747 spin_lock_init(&udev->cmdr_lock); 748 749 idr_init(&udev->commands); 750 spin_lock_init(&udev->commands_lock); 751 752 setup_timer(&udev->timeout, tcmu_device_timedout, 753 (unsigned long)udev); 754 755 return &udev->se_dev; 756 } 757 758 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on) 759 { 760 struct tcmu_dev *tcmu_dev = container_of(info, struct tcmu_dev, uio_info); 761 762 tcmu_handle_completions(tcmu_dev); 763 764 return 0; 765 } 766 767 /* 768 * mmap code from uio.c. Copied here because we want to hook mmap() 769 * and this stuff must come along. 770 */ 771 static int tcmu_find_mem_index(struct vm_area_struct *vma) 772 { 773 struct tcmu_dev *udev = vma->vm_private_data; 774 struct uio_info *info = &udev->uio_info; 775 776 if (vma->vm_pgoff < MAX_UIO_MAPS) { 777 if (info->mem[vma->vm_pgoff].size == 0) 778 return -1; 779 return (int)vma->vm_pgoff; 780 } 781 return -1; 782 } 783 784 static int tcmu_vma_fault(struct vm_fault *vmf) 785 { 786 struct tcmu_dev *udev = vmf->vma->vm_private_data; 787 struct uio_info *info = &udev->uio_info; 788 struct page *page; 789 unsigned long offset; 790 void *addr; 791 792 int mi = tcmu_find_mem_index(vmf->vma); 793 if (mi < 0) 794 return VM_FAULT_SIGBUS; 795 796 /* 797 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE 798 * to use mem[N]. 799 */ 800 offset = (vmf->pgoff - mi) << PAGE_SHIFT; 801 802 addr = (void *)(unsigned long)info->mem[mi].addr + offset; 803 if (info->mem[mi].memtype == UIO_MEM_LOGICAL) 804 page = virt_to_page(addr); 805 else 806 page = vmalloc_to_page(addr); 807 get_page(page); 808 vmf->page = page; 809 return 0; 810 } 811 812 static const struct vm_operations_struct tcmu_vm_ops = { 813 .fault = tcmu_vma_fault, 814 }; 815 816 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma) 817 { 818 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info); 819 820 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; 821 vma->vm_ops = &tcmu_vm_ops; 822 823 vma->vm_private_data = udev; 824 825 /* Ensure the mmap is exactly the right size */ 826 if (vma_pages(vma) != (TCMU_RING_SIZE >> PAGE_SHIFT)) 827 return -EINVAL; 828 829 return 0; 830 } 831 832 static int tcmu_open(struct uio_info *info, struct inode *inode) 833 { 834 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info); 835 836 /* O_EXCL not supported for char devs, so fake it? */ 837 if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags)) 838 return -EBUSY; 839 840 pr_debug("open\n"); 841 842 return 0; 843 } 844 845 static int tcmu_release(struct uio_info *info, struct inode *inode) 846 { 847 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info); 848 849 clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags); 850 851 pr_debug("close\n"); 852 853 return 0; 854 } 855 856 static int tcmu_netlink_event(enum tcmu_genl_cmd cmd, const char *name, int minor) 857 { 858 struct sk_buff *skb; 859 void *msg_header; 860 int ret = -ENOMEM; 861 862 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL); 863 if (!skb) 864 return ret; 865 866 msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd); 867 if (!msg_header) 868 goto free_skb; 869 870 ret = nla_put_string(skb, TCMU_ATTR_DEVICE, name); 871 if (ret < 0) 872 goto free_skb; 873 874 ret = nla_put_u32(skb, TCMU_ATTR_MINOR, minor); 875 if (ret < 0) 876 goto free_skb; 877 878 genlmsg_end(skb, msg_header); 879 880 ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0, 881 TCMU_MCGRP_CONFIG, GFP_KERNEL); 882 883 /* We don't care if no one is listening */ 884 if (ret == -ESRCH) 885 ret = 0; 886 887 return ret; 888 free_skb: 889 nlmsg_free(skb); 890 return ret; 891 } 892 893 static int tcmu_configure_device(struct se_device *dev) 894 { 895 struct tcmu_dev *udev = TCMU_DEV(dev); 896 struct tcmu_hba *hba = udev->hba->hba_ptr; 897 struct uio_info *info; 898 struct tcmu_mailbox *mb; 899 size_t size; 900 size_t used; 901 int ret = 0; 902 char *str; 903 904 info = &udev->uio_info; 905 906 size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name, 907 udev->dev_config); 908 size += 1; /* for \0 */ 909 str = kmalloc(size, GFP_KERNEL); 910 if (!str) 911 return -ENOMEM; 912 913 used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name); 914 915 if (udev->dev_config[0]) 916 snprintf(str + used, size - used, "/%s", udev->dev_config); 917 918 info->name = str; 919 920 udev->mb_addr = vzalloc(TCMU_RING_SIZE); 921 if (!udev->mb_addr) { 922 ret = -ENOMEM; 923 goto err_vzalloc; 924 } 925 926 /* mailbox fits in first part of CMDR space */ 927 udev->cmdr_size = CMDR_SIZE - CMDR_OFF; 928 udev->data_off = CMDR_SIZE; 929 udev->data_size = TCMU_RING_SIZE - CMDR_SIZE; 930 931 mb = udev->mb_addr; 932 mb->version = TCMU_MAILBOX_VERSION; 933 mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC; 934 mb->cmdr_off = CMDR_OFF; 935 mb->cmdr_size = udev->cmdr_size; 936 937 WARN_ON(!PAGE_ALIGNED(udev->data_off)); 938 WARN_ON(udev->data_size % PAGE_SIZE); 939 WARN_ON(udev->data_size % DATA_BLOCK_SIZE); 940 941 info->version = __stringify(TCMU_MAILBOX_VERSION); 942 943 info->mem[0].name = "tcm-user command & data buffer"; 944 info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr; 945 info->mem[0].size = TCMU_RING_SIZE; 946 info->mem[0].memtype = UIO_MEM_VIRTUAL; 947 948 info->irqcontrol = tcmu_irqcontrol; 949 info->irq = UIO_IRQ_CUSTOM; 950 951 info->mmap = tcmu_mmap; 952 info->open = tcmu_open; 953 info->release = tcmu_release; 954 955 ret = uio_register_device(tcmu_root_device, info); 956 if (ret) 957 goto err_register; 958 959 /* User can set hw_block_size before enable the device */ 960 if (dev->dev_attrib.hw_block_size == 0) 961 dev->dev_attrib.hw_block_size = 512; 962 /* Other attributes can be configured in userspace */ 963 dev->dev_attrib.hw_max_sectors = 128; 964 dev->dev_attrib.hw_queue_depth = 128; 965 966 ret = tcmu_netlink_event(TCMU_CMD_ADDED_DEVICE, udev->uio_info.name, 967 udev->uio_info.uio_dev->minor); 968 if (ret) 969 goto err_netlink; 970 971 return 0; 972 973 err_netlink: 974 uio_unregister_device(&udev->uio_info); 975 err_register: 976 vfree(udev->mb_addr); 977 err_vzalloc: 978 kfree(info->name); 979 980 return ret; 981 } 982 983 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd) 984 { 985 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) { 986 kmem_cache_free(tcmu_cmd_cache, cmd); 987 return 0; 988 } 989 return -EINVAL; 990 } 991 992 static void tcmu_dev_call_rcu(struct rcu_head *p) 993 { 994 struct se_device *dev = container_of(p, struct se_device, rcu_head); 995 struct tcmu_dev *udev = TCMU_DEV(dev); 996 997 kfree(udev); 998 } 999 1000 static void tcmu_free_device(struct se_device *dev) 1001 { 1002 struct tcmu_dev *udev = TCMU_DEV(dev); 1003 struct tcmu_cmd *cmd; 1004 bool all_expired = true; 1005 int i; 1006 1007 del_timer_sync(&udev->timeout); 1008 1009 vfree(udev->mb_addr); 1010 1011 /* Upper layer should drain all requests before calling this */ 1012 spin_lock_irq(&udev->commands_lock); 1013 idr_for_each_entry(&udev->commands, cmd, i) { 1014 if (tcmu_check_and_free_pending_cmd(cmd) != 0) 1015 all_expired = false; 1016 } 1017 idr_destroy(&udev->commands); 1018 spin_unlock_irq(&udev->commands_lock); 1019 WARN_ON(!all_expired); 1020 1021 /* Device was configured */ 1022 if (udev->uio_info.uio_dev) { 1023 tcmu_netlink_event(TCMU_CMD_REMOVED_DEVICE, udev->uio_info.name, 1024 udev->uio_info.uio_dev->minor); 1025 1026 uio_unregister_device(&udev->uio_info); 1027 kfree(udev->uio_info.name); 1028 kfree(udev->name); 1029 } 1030 call_rcu(&dev->rcu_head, tcmu_dev_call_rcu); 1031 } 1032 1033 enum { 1034 Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_err, 1035 }; 1036 1037 static match_table_t tokens = { 1038 {Opt_dev_config, "dev_config=%s"}, 1039 {Opt_dev_size, "dev_size=%u"}, 1040 {Opt_hw_block_size, "hw_block_size=%u"}, 1041 {Opt_err, NULL} 1042 }; 1043 1044 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev, 1045 const char *page, ssize_t count) 1046 { 1047 struct tcmu_dev *udev = TCMU_DEV(dev); 1048 char *orig, *ptr, *opts, *arg_p; 1049 substring_t args[MAX_OPT_ARGS]; 1050 int ret = 0, token; 1051 unsigned long tmp_ul; 1052 1053 opts = kstrdup(page, GFP_KERNEL); 1054 if (!opts) 1055 return -ENOMEM; 1056 1057 orig = opts; 1058 1059 while ((ptr = strsep(&opts, ",\n")) != NULL) { 1060 if (!*ptr) 1061 continue; 1062 1063 token = match_token(ptr, tokens, args); 1064 switch (token) { 1065 case Opt_dev_config: 1066 if (match_strlcpy(udev->dev_config, &args[0], 1067 TCMU_CONFIG_LEN) == 0) { 1068 ret = -EINVAL; 1069 break; 1070 } 1071 pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config); 1072 break; 1073 case Opt_dev_size: 1074 arg_p = match_strdup(&args[0]); 1075 if (!arg_p) { 1076 ret = -ENOMEM; 1077 break; 1078 } 1079 ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size); 1080 kfree(arg_p); 1081 if (ret < 0) 1082 pr_err("kstrtoul() failed for dev_size=\n"); 1083 break; 1084 case Opt_hw_block_size: 1085 arg_p = match_strdup(&args[0]); 1086 if (!arg_p) { 1087 ret = -ENOMEM; 1088 break; 1089 } 1090 ret = kstrtoul(arg_p, 0, &tmp_ul); 1091 kfree(arg_p); 1092 if (ret < 0) { 1093 pr_err("kstrtoul() failed for hw_block_size=\n"); 1094 break; 1095 } 1096 if (!tmp_ul) { 1097 pr_err("hw_block_size must be nonzero\n"); 1098 break; 1099 } 1100 dev->dev_attrib.hw_block_size = tmp_ul; 1101 break; 1102 default: 1103 break; 1104 } 1105 } 1106 1107 kfree(orig); 1108 return (!ret) ? count : ret; 1109 } 1110 1111 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b) 1112 { 1113 struct tcmu_dev *udev = TCMU_DEV(dev); 1114 ssize_t bl = 0; 1115 1116 bl = sprintf(b + bl, "Config: %s ", 1117 udev->dev_config[0] ? udev->dev_config : "NULL"); 1118 bl += sprintf(b + bl, "Size: %zu\n", udev->dev_size); 1119 1120 return bl; 1121 } 1122 1123 static sector_t tcmu_get_blocks(struct se_device *dev) 1124 { 1125 struct tcmu_dev *udev = TCMU_DEV(dev); 1126 1127 return div_u64(udev->dev_size - dev->dev_attrib.block_size, 1128 dev->dev_attrib.block_size); 1129 } 1130 1131 static sense_reason_t 1132 tcmu_parse_cdb(struct se_cmd *cmd) 1133 { 1134 return passthrough_parse_cdb(cmd, tcmu_queue_cmd); 1135 } 1136 1137 static const struct target_backend_ops tcmu_ops = { 1138 .name = "user", 1139 .owner = THIS_MODULE, 1140 .transport_flags = TRANSPORT_FLAG_PASSTHROUGH, 1141 .attach_hba = tcmu_attach_hba, 1142 .detach_hba = tcmu_detach_hba, 1143 .alloc_device = tcmu_alloc_device, 1144 .configure_device = tcmu_configure_device, 1145 .free_device = tcmu_free_device, 1146 .parse_cdb = tcmu_parse_cdb, 1147 .set_configfs_dev_params = tcmu_set_configfs_dev_params, 1148 .show_configfs_dev_params = tcmu_show_configfs_dev_params, 1149 .get_device_type = sbc_get_device_type, 1150 .get_blocks = tcmu_get_blocks, 1151 .tb_dev_attrib_attrs = passthrough_attrib_attrs, 1152 }; 1153 1154 static int __init tcmu_module_init(void) 1155 { 1156 int ret; 1157 1158 BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0); 1159 1160 tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache", 1161 sizeof(struct tcmu_cmd), 1162 __alignof__(struct tcmu_cmd), 1163 0, NULL); 1164 if (!tcmu_cmd_cache) 1165 return -ENOMEM; 1166 1167 tcmu_root_device = root_device_register("tcm_user"); 1168 if (IS_ERR(tcmu_root_device)) { 1169 ret = PTR_ERR(tcmu_root_device); 1170 goto out_free_cache; 1171 } 1172 1173 ret = genl_register_family(&tcmu_genl_family); 1174 if (ret < 0) { 1175 goto out_unreg_device; 1176 } 1177 1178 ret = transport_backend_register(&tcmu_ops); 1179 if (ret) 1180 goto out_unreg_genl; 1181 1182 return 0; 1183 1184 out_unreg_genl: 1185 genl_unregister_family(&tcmu_genl_family); 1186 out_unreg_device: 1187 root_device_unregister(tcmu_root_device); 1188 out_free_cache: 1189 kmem_cache_destroy(tcmu_cmd_cache); 1190 1191 return ret; 1192 } 1193 1194 static void __exit tcmu_module_exit(void) 1195 { 1196 target_backend_unregister(&tcmu_ops); 1197 genl_unregister_family(&tcmu_genl_family); 1198 root_device_unregister(tcmu_root_device); 1199 kmem_cache_destroy(tcmu_cmd_cache); 1200 } 1201 1202 MODULE_DESCRIPTION("TCM USER subsystem plugin"); 1203 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>"); 1204 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>"); 1205 MODULE_LICENSE("GPL"); 1206 1207 module_init(tcmu_module_init); 1208 module_exit(tcmu_module_exit); 1209