1 /******************************************************************************* 2 * Filename: target_core_file.c 3 * 4 * This file contains the Storage Engine <-> FILEIO transport specific functions 5 * 6 * (c) Copyright 2005-2012 RisingTide Systems LLC. 7 * 8 * Nicholas A. Bellinger <nab@kernel.org> 9 * 10 * This program is free software; you can redistribute it and/or modify 11 * it under the terms of the GNU General Public License as published by 12 * the Free Software Foundation; either version 2 of the License, or 13 * (at your option) any later version. 14 * 15 * This program is distributed in the hope that it will be useful, 16 * but WITHOUT ANY WARRANTY; without even the implied warranty of 17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 18 * GNU General Public License for more details. 19 * 20 * You should have received a copy of the GNU General Public License 21 * along with this program; if not, write to the Free Software 22 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 23 * 24 ******************************************************************************/ 25 26 #include <linux/string.h> 27 #include <linux/parser.h> 28 #include <linux/timer.h> 29 #include <linux/blkdev.h> 30 #include <linux/slab.h> 31 #include <linux/spinlock.h> 32 #include <linux/module.h> 33 #include <scsi/scsi.h> 34 #include <scsi/scsi_host.h> 35 36 #include <target/target_core_base.h> 37 #include <target/target_core_backend.h> 38 39 #include "target_core_file.h" 40 41 static inline struct fd_dev *FD_DEV(struct se_device *dev) 42 { 43 return container_of(dev, struct fd_dev, dev); 44 } 45 46 /* fd_attach_hba(): (Part of se_subsystem_api_t template) 47 * 48 * 49 */ 50 static int fd_attach_hba(struct se_hba *hba, u32 host_id) 51 { 52 struct fd_host *fd_host; 53 54 fd_host = kzalloc(sizeof(struct fd_host), GFP_KERNEL); 55 if (!fd_host) { 56 pr_err("Unable to allocate memory for struct fd_host\n"); 57 return -ENOMEM; 58 } 59 60 fd_host->fd_host_id = host_id; 61 62 hba->hba_ptr = fd_host; 63 64 pr_debug("CORE_HBA[%d] - TCM FILEIO HBA Driver %s on Generic" 65 " Target Core Stack %s\n", hba->hba_id, FD_VERSION, 66 TARGET_CORE_MOD_VERSION); 67 pr_debug("CORE_HBA[%d] - Attached FILEIO HBA: %u to Generic" 68 " MaxSectors: %u\n", 69 hba->hba_id, fd_host->fd_host_id, FD_MAX_SECTORS); 70 71 return 0; 72 } 73 74 static void fd_detach_hba(struct se_hba *hba) 75 { 76 struct fd_host *fd_host = hba->hba_ptr; 77 78 pr_debug("CORE_HBA[%d] - Detached FILEIO HBA: %u from Generic" 79 " Target Core\n", hba->hba_id, fd_host->fd_host_id); 80 81 kfree(fd_host); 82 hba->hba_ptr = NULL; 83 } 84 85 static struct se_device *fd_alloc_device(struct se_hba *hba, const char *name) 86 { 87 struct fd_dev *fd_dev; 88 struct fd_host *fd_host = hba->hba_ptr; 89 90 fd_dev = kzalloc(sizeof(struct fd_dev), GFP_KERNEL); 91 if (!fd_dev) { 92 pr_err("Unable to allocate memory for struct fd_dev\n"); 93 return NULL; 94 } 95 96 fd_dev->fd_host = fd_host; 97 98 pr_debug("FILEIO: Allocated fd_dev for %p\n", name); 99 100 return &fd_dev->dev; 101 } 102 103 static int fd_configure_device(struct se_device *dev) 104 { 105 struct fd_dev *fd_dev = FD_DEV(dev); 106 struct fd_host *fd_host = dev->se_hba->hba_ptr; 107 struct file *file; 108 struct inode *inode = NULL; 109 int flags, ret = -EINVAL; 110 111 if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) { 112 pr_err("Missing fd_dev_name=\n"); 113 return -EINVAL; 114 } 115 116 /* 117 * Use O_DSYNC by default instead of O_SYNC to forgo syncing 118 * of pure timestamp updates. 119 */ 120 flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC; 121 122 /* 123 * Optionally allow fd_buffered_io=1 to be enabled for people 124 * who want use the fs buffer cache as an WriteCache mechanism. 125 * 126 * This means that in event of a hard failure, there is a risk 127 * of silent data-loss if the SCSI client has *not* performed a 128 * forced unit access (FUA) write, or issued SYNCHRONIZE_CACHE 129 * to write-out the entire device cache. 130 */ 131 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) { 132 pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n"); 133 flags &= ~O_DSYNC; 134 } 135 136 file = filp_open(fd_dev->fd_dev_name, flags, 0600); 137 if (IS_ERR(file)) { 138 pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name); 139 ret = PTR_ERR(file); 140 goto fail; 141 } 142 fd_dev->fd_file = file; 143 /* 144 * If using a block backend with this struct file, we extract 145 * fd_dev->fd_[block,dev]_size from struct block_device. 146 * 147 * Otherwise, we use the passed fd_size= from configfs 148 */ 149 inode = file->f_mapping->host; 150 if (S_ISBLK(inode->i_mode)) { 151 struct request_queue *q = bdev_get_queue(inode->i_bdev); 152 unsigned long long dev_size; 153 154 dev->dev_attrib.hw_block_size = 155 bdev_logical_block_size(inode->i_bdev); 156 dev->dev_attrib.hw_max_sectors = queue_max_hw_sectors(q); 157 158 /* 159 * Determine the number of bytes from i_size_read() minus 160 * one (1) logical sector from underlying struct block_device 161 */ 162 dev_size = (i_size_read(file->f_mapping->host) - 163 fd_dev->fd_block_size); 164 165 pr_debug("FILEIO: Using size: %llu bytes from struct" 166 " block_device blocks: %llu logical_block_size: %d\n", 167 dev_size, div_u64(dev_size, fd_dev->fd_block_size), 168 fd_dev->fd_block_size); 169 } else { 170 if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) { 171 pr_err("FILEIO: Missing fd_dev_size=" 172 " parameter, and no backing struct" 173 " block_device\n"); 174 goto fail; 175 } 176 177 dev->dev_attrib.hw_block_size = FD_BLOCKSIZE; 178 dev->dev_attrib.hw_max_sectors = FD_MAX_SECTORS; 179 } 180 181 fd_dev->fd_block_size = dev->dev_attrib.hw_block_size; 182 183 dev->dev_attrib.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH; 184 185 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) { 186 pr_debug("FILEIO: Forcing setting of emulate_write_cache=1" 187 " with FDBD_HAS_BUFFERED_IO_WCE\n"); 188 dev->dev_attrib.emulate_write_cache = 1; 189 } 190 191 fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++; 192 fd_dev->fd_queue_depth = dev->queue_depth; 193 /* 194 * Limit WRITE_SAME w/ UNMAP=0 emulation to 8k Number of LBAs (NoLB) 195 * based upon struct iovec limit for vfs_writev() 196 */ 197 dev->dev_attrib.max_write_same_len = 0x1000; 198 199 pr_debug("CORE_FILE[%u] - Added TCM FILEIO Device ID: %u at %s," 200 " %llu total bytes\n", fd_host->fd_host_id, fd_dev->fd_dev_id, 201 fd_dev->fd_dev_name, fd_dev->fd_dev_size); 202 203 return 0; 204 fail: 205 if (fd_dev->fd_file) { 206 filp_close(fd_dev->fd_file, NULL); 207 fd_dev->fd_file = NULL; 208 } 209 return ret; 210 } 211 212 static void fd_free_device(struct se_device *dev) 213 { 214 struct fd_dev *fd_dev = FD_DEV(dev); 215 216 if (fd_dev->fd_file) { 217 filp_close(fd_dev->fd_file, NULL); 218 fd_dev->fd_file = NULL; 219 } 220 221 kfree(fd_dev); 222 } 223 224 static int fd_do_rw(struct se_cmd *cmd, struct scatterlist *sgl, 225 u32 sgl_nents, int is_write) 226 { 227 struct se_device *se_dev = cmd->se_dev; 228 struct fd_dev *dev = FD_DEV(se_dev); 229 struct file *fd = dev->fd_file; 230 struct scatterlist *sg; 231 struct iovec *iov; 232 mm_segment_t old_fs; 233 loff_t pos = (cmd->t_task_lba * se_dev->dev_attrib.block_size); 234 int ret = 0, i; 235 236 iov = kzalloc(sizeof(struct iovec) * sgl_nents, GFP_KERNEL); 237 if (!iov) { 238 pr_err("Unable to allocate fd_do_readv iov[]\n"); 239 return -ENOMEM; 240 } 241 242 for_each_sg(sgl, sg, sgl_nents, i) { 243 iov[i].iov_len = sg->length; 244 iov[i].iov_base = kmap(sg_page(sg)) + sg->offset; 245 } 246 247 old_fs = get_fs(); 248 set_fs(get_ds()); 249 250 if (is_write) 251 ret = vfs_writev(fd, &iov[0], sgl_nents, &pos); 252 else 253 ret = vfs_readv(fd, &iov[0], sgl_nents, &pos); 254 255 set_fs(old_fs); 256 257 for_each_sg(sgl, sg, sgl_nents, i) 258 kunmap(sg_page(sg)); 259 260 kfree(iov); 261 262 if (is_write) { 263 if (ret < 0 || ret != cmd->data_length) { 264 pr_err("%s() write returned %d\n", __func__, ret); 265 return (ret < 0 ? ret : -EINVAL); 266 } 267 } else { 268 /* 269 * Return zeros and GOOD status even if the READ did not return 270 * the expected virt_size for struct file w/o a backing struct 271 * block_device. 272 */ 273 if (S_ISBLK(file_inode(fd)->i_mode)) { 274 if (ret < 0 || ret != cmd->data_length) { 275 pr_err("%s() returned %d, expecting %u for " 276 "S_ISBLK\n", __func__, ret, 277 cmd->data_length); 278 return (ret < 0 ? ret : -EINVAL); 279 } 280 } else { 281 if (ret < 0) { 282 pr_err("%s() returned %d for non S_ISBLK\n", 283 __func__, ret); 284 return ret; 285 } 286 } 287 } 288 return 1; 289 } 290 291 static sense_reason_t 292 fd_execute_sync_cache(struct se_cmd *cmd) 293 { 294 struct se_device *dev = cmd->se_dev; 295 struct fd_dev *fd_dev = FD_DEV(dev); 296 int immed = (cmd->t_task_cdb[1] & 0x2); 297 loff_t start, end; 298 int ret; 299 300 /* 301 * If the Immediate bit is set, queue up the GOOD response 302 * for this SYNCHRONIZE_CACHE op 303 */ 304 if (immed) 305 target_complete_cmd(cmd, SAM_STAT_GOOD); 306 307 /* 308 * Determine if we will be flushing the entire device. 309 */ 310 if (cmd->t_task_lba == 0 && cmd->data_length == 0) { 311 start = 0; 312 end = LLONG_MAX; 313 } else { 314 start = cmd->t_task_lba * dev->dev_attrib.block_size; 315 if (cmd->data_length) 316 end = start + cmd->data_length; 317 else 318 end = LLONG_MAX; 319 } 320 321 ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1); 322 if (ret != 0) 323 pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret); 324 325 if (immed) 326 return 0; 327 328 if (ret) 329 target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION); 330 else 331 target_complete_cmd(cmd, SAM_STAT_GOOD); 332 333 return 0; 334 } 335 336 static unsigned char * 337 fd_setup_write_same_buf(struct se_cmd *cmd, struct scatterlist *sg, 338 unsigned int len) 339 { 340 struct se_device *se_dev = cmd->se_dev; 341 unsigned int block_size = se_dev->dev_attrib.block_size; 342 unsigned int i = 0, end; 343 unsigned char *buf, *p, *kmap_buf; 344 345 buf = kzalloc(min_t(unsigned int, len, PAGE_SIZE), GFP_KERNEL); 346 if (!buf) { 347 pr_err("Unable to allocate fd_execute_write_same buf\n"); 348 return NULL; 349 } 350 351 kmap_buf = kmap(sg_page(sg)) + sg->offset; 352 if (!kmap_buf) { 353 pr_err("kmap() failed in fd_setup_write_same\n"); 354 kfree(buf); 355 return NULL; 356 } 357 /* 358 * Fill local *buf to contain multiple WRITE_SAME blocks up to 359 * min(len, PAGE_SIZE) 360 */ 361 p = buf; 362 end = min_t(unsigned int, len, PAGE_SIZE); 363 364 while (i < end) { 365 memcpy(p, kmap_buf, block_size); 366 367 i += block_size; 368 p += block_size; 369 } 370 kunmap(sg_page(sg)); 371 372 return buf; 373 } 374 375 static sense_reason_t 376 fd_execute_write_same(struct se_cmd *cmd) 377 { 378 struct se_device *se_dev = cmd->se_dev; 379 struct fd_dev *fd_dev = FD_DEV(se_dev); 380 struct file *f = fd_dev->fd_file; 381 struct scatterlist *sg; 382 struct iovec *iov; 383 mm_segment_t old_fs; 384 sector_t nolb = sbc_get_write_same_sectors(cmd); 385 loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size; 386 unsigned int len, len_tmp, iov_num; 387 int i, rc; 388 unsigned char *buf; 389 390 if (!nolb) { 391 target_complete_cmd(cmd, SAM_STAT_GOOD); 392 return 0; 393 } 394 sg = &cmd->t_data_sg[0]; 395 396 if (cmd->t_data_nents > 1 || 397 sg->length != cmd->se_dev->dev_attrib.block_size) { 398 pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u" 399 " block_size: %u\n", cmd->t_data_nents, sg->length, 400 cmd->se_dev->dev_attrib.block_size); 401 return TCM_INVALID_CDB_FIELD; 402 } 403 404 len = len_tmp = nolb * se_dev->dev_attrib.block_size; 405 iov_num = DIV_ROUND_UP(len, PAGE_SIZE); 406 407 buf = fd_setup_write_same_buf(cmd, sg, len); 408 if (!buf) 409 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE; 410 411 iov = vzalloc(sizeof(struct iovec) * iov_num); 412 if (!iov) { 413 pr_err("Unable to allocate fd_execute_write_same iovecs\n"); 414 kfree(buf); 415 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE; 416 } 417 /* 418 * Map the single fabric received scatterlist block now populated 419 * in *buf into each iovec for I/O submission. 420 */ 421 for (i = 0; i < iov_num; i++) { 422 iov[i].iov_base = buf; 423 iov[i].iov_len = min_t(unsigned int, len_tmp, PAGE_SIZE); 424 len_tmp -= iov[i].iov_len; 425 } 426 427 old_fs = get_fs(); 428 set_fs(get_ds()); 429 rc = vfs_writev(f, &iov[0], iov_num, &pos); 430 set_fs(old_fs); 431 432 vfree(iov); 433 kfree(buf); 434 435 if (rc < 0 || rc != len) { 436 pr_err("vfs_writev() returned %d for write same\n", rc); 437 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE; 438 } 439 440 target_complete_cmd(cmd, SAM_STAT_GOOD); 441 return 0; 442 } 443 444 static sense_reason_t 445 fd_execute_rw(struct se_cmd *cmd) 446 { 447 struct scatterlist *sgl = cmd->t_data_sg; 448 u32 sgl_nents = cmd->t_data_nents; 449 enum dma_data_direction data_direction = cmd->data_direction; 450 struct se_device *dev = cmd->se_dev; 451 int ret = 0; 452 453 /* 454 * Call vectorized fileio functions to map struct scatterlist 455 * physical memory addresses to struct iovec virtual memory. 456 */ 457 if (data_direction == DMA_FROM_DEVICE) { 458 ret = fd_do_rw(cmd, sgl, sgl_nents, 0); 459 } else { 460 ret = fd_do_rw(cmd, sgl, sgl_nents, 1); 461 /* 462 * Perform implict vfs_fsync_range() for fd_do_writev() ops 463 * for SCSI WRITEs with Forced Unit Access (FUA) set. 464 * Allow this to happen independent of WCE=0 setting. 465 */ 466 if (ret > 0 && 467 dev->dev_attrib.emulate_fua_write > 0 && 468 (cmd->se_cmd_flags & SCF_FUA)) { 469 struct fd_dev *fd_dev = FD_DEV(dev); 470 loff_t start = cmd->t_task_lba * 471 dev->dev_attrib.block_size; 472 loff_t end = start + cmd->data_length; 473 474 vfs_fsync_range(fd_dev->fd_file, start, end, 1); 475 } 476 } 477 478 if (ret < 0) 479 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE; 480 481 if (ret) 482 target_complete_cmd(cmd, SAM_STAT_GOOD); 483 return 0; 484 } 485 486 enum { 487 Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io, Opt_err 488 }; 489 490 static match_table_t tokens = { 491 {Opt_fd_dev_name, "fd_dev_name=%s"}, 492 {Opt_fd_dev_size, "fd_dev_size=%s"}, 493 {Opt_fd_buffered_io, "fd_buffered_io=%d"}, 494 {Opt_err, NULL} 495 }; 496 497 static ssize_t fd_set_configfs_dev_params(struct se_device *dev, 498 const char *page, ssize_t count) 499 { 500 struct fd_dev *fd_dev = FD_DEV(dev); 501 char *orig, *ptr, *arg_p, *opts; 502 substring_t args[MAX_OPT_ARGS]; 503 int ret = 0, arg, token; 504 505 opts = kstrdup(page, GFP_KERNEL); 506 if (!opts) 507 return -ENOMEM; 508 509 orig = opts; 510 511 while ((ptr = strsep(&opts, ",\n")) != NULL) { 512 if (!*ptr) 513 continue; 514 515 token = match_token(ptr, tokens, args); 516 switch (token) { 517 case Opt_fd_dev_name: 518 if (match_strlcpy(fd_dev->fd_dev_name, &args[0], 519 FD_MAX_DEV_NAME) == 0) { 520 ret = -EINVAL; 521 break; 522 } 523 pr_debug("FILEIO: Referencing Path: %s\n", 524 fd_dev->fd_dev_name); 525 fd_dev->fbd_flags |= FBDF_HAS_PATH; 526 break; 527 case Opt_fd_dev_size: 528 arg_p = match_strdup(&args[0]); 529 if (!arg_p) { 530 ret = -ENOMEM; 531 break; 532 } 533 ret = strict_strtoull(arg_p, 0, &fd_dev->fd_dev_size); 534 kfree(arg_p); 535 if (ret < 0) { 536 pr_err("strict_strtoull() failed for" 537 " fd_dev_size=\n"); 538 goto out; 539 } 540 pr_debug("FILEIO: Referencing Size: %llu" 541 " bytes\n", fd_dev->fd_dev_size); 542 fd_dev->fbd_flags |= FBDF_HAS_SIZE; 543 break; 544 case Opt_fd_buffered_io: 545 match_int(args, &arg); 546 if (arg != 1) { 547 pr_err("bogus fd_buffered_io=%d value\n", arg); 548 ret = -EINVAL; 549 goto out; 550 } 551 552 pr_debug("FILEIO: Using buffered I/O" 553 " operations for struct fd_dev\n"); 554 555 fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE; 556 break; 557 default: 558 break; 559 } 560 } 561 562 out: 563 kfree(orig); 564 return (!ret) ? count : ret; 565 } 566 567 static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b) 568 { 569 struct fd_dev *fd_dev = FD_DEV(dev); 570 ssize_t bl = 0; 571 572 bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id); 573 bl += sprintf(b + bl, " File: %s Size: %llu Mode: %s\n", 574 fd_dev->fd_dev_name, fd_dev->fd_dev_size, 575 (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ? 576 "Buffered-WCE" : "O_DSYNC"); 577 return bl; 578 } 579 580 static sector_t fd_get_blocks(struct se_device *dev) 581 { 582 struct fd_dev *fd_dev = FD_DEV(dev); 583 struct file *f = fd_dev->fd_file; 584 struct inode *i = f->f_mapping->host; 585 unsigned long long dev_size; 586 /* 587 * When using a file that references an underlying struct block_device, 588 * ensure dev_size is always based on the current inode size in order 589 * to handle underlying block_device resize operations. 590 */ 591 if (S_ISBLK(i->i_mode)) 592 dev_size = (i_size_read(i) - fd_dev->fd_block_size); 593 else 594 dev_size = fd_dev->fd_dev_size; 595 596 return div_u64(dev_size, dev->dev_attrib.block_size); 597 } 598 599 static struct sbc_ops fd_sbc_ops = { 600 .execute_rw = fd_execute_rw, 601 .execute_sync_cache = fd_execute_sync_cache, 602 .execute_write_same = fd_execute_write_same, 603 }; 604 605 static sense_reason_t 606 fd_parse_cdb(struct se_cmd *cmd) 607 { 608 return sbc_parse_cdb(cmd, &fd_sbc_ops); 609 } 610 611 static struct se_subsystem_api fileio_template = { 612 .name = "fileio", 613 .inquiry_prod = "FILEIO", 614 .inquiry_rev = FD_VERSION, 615 .owner = THIS_MODULE, 616 .transport_type = TRANSPORT_PLUGIN_VHBA_PDEV, 617 .attach_hba = fd_attach_hba, 618 .detach_hba = fd_detach_hba, 619 .alloc_device = fd_alloc_device, 620 .configure_device = fd_configure_device, 621 .free_device = fd_free_device, 622 .parse_cdb = fd_parse_cdb, 623 .set_configfs_dev_params = fd_set_configfs_dev_params, 624 .show_configfs_dev_params = fd_show_configfs_dev_params, 625 .get_device_type = sbc_get_device_type, 626 .get_blocks = fd_get_blocks, 627 }; 628 629 static int __init fileio_module_init(void) 630 { 631 return transport_subsystem_register(&fileio_template); 632 } 633 634 static void __exit fileio_module_exit(void) 635 { 636 transport_subsystem_release(&fileio_template); 637 } 638 639 MODULE_DESCRIPTION("TCM FILEIO subsystem plugin"); 640 MODULE_AUTHOR("nab@Linux-iSCSI.org"); 641 MODULE_LICENSE("GPL"); 642 643 module_init(fileio_module_init); 644 module_exit(fileio_module_exit); 645