drivers/target/target_core_user.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
   4  * Copyright (C) 2014 Red Hat, Inc.
   5  * Copyright (C) 2015 Arrikto, Inc.
   6  * Copyright (C) 2017 Chinamobile, Inc.
   7  */
   8
   9 #include <linux/spinlock.h>
  10 #include <linux/module.h>
  11 #include <linux/idr.h>
  12 #include <linux/kernel.h>
  13 #include <linux/timer.h>
  14 #include <linux/parser.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/uio_driver.h>
  17 #include <linux/radix-tree.h>
  18 #include <linux/stringify.h>
  19 #include <linux/bitops.h>
  20 #include <linux/highmem.h>
  21 #include <linux/configfs.h>
  22 #include <linux/mutex.h>
  23 #include <linux/workqueue.h>
  24 #include <net/genetlink.h>
  25 #include <scsi/scsi_common.h>
  26 #include <scsi/scsi_proto.h>
  27 #include <target/target_core_base.h>
  28 #include <target/target_core_fabric.h>
  29 #include <target/target_core_backend.h>
  30
  31 #include <linux/target_core_user.h>
  32
  33 /**
  34  * DOC: Userspace I/O
  35  * Userspace I/O
  36  * -------------
  37  *
  38  * Define a shared-memory interface for LIO to pass SCSI commands and
  39  * data to userspace for processing. This is to allow backends that
  40  * are too complex for in-kernel support to be possible.
  41  *
  42  * It uses the UIO framework to do a lot of the device-creation and
  43  * introspection work for us.
  44  *
  45  * See the .h file for how the ring is laid out. Note that while the
  46  * command ring is defined, the particulars of the data area are
  47  * not. Offset values in the command entry point to other locations
  48  * internal to the mmap-ed area. There is separate space outside the
  49  * command ring for data buffers. This leaves maximum flexibility for
  50  * moving buffer allocations, or even page flipping or other
  51  * allocation techniques, without altering the command ring layout.
  52  *
  53  * SECURITY:
  54  * The user process must be assumed to be malicious. There's no way to
  55  * prevent it breaking the command ring protocol if it wants, but in
  56  * order to prevent other issues we must only ever read *data* from
  57  * the shared memory area, not offsets or sizes. This applies to
  58  * command ring entries as well as the mailbox. Extra code needed for
  59  * this may have a 'UAM' comment.
  60  */
  61
  62 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
  63
  64 /* For cmd area, the size is fixed 8MB */
  65 #define CMDR_SIZE (8 * 1024 * 1024)
  66
  67 /*
  68  * For data area, the block size is PAGE_SIZE and
  69  * the total size is 256K * PAGE_SIZE.
  70  */
  71 #define DATA_BLOCK_SIZE PAGE_SIZE
  72 #define DATA_BLOCK_SHIFT PAGE_SHIFT
  73 #define DATA_BLOCK_BITS_DEF (256 * 1024)
  74
  75 #define TCMU_MBS_TO_BLOCKS(_mbs) (_mbs << (20 - DATA_BLOCK_SHIFT))
  76 #define TCMU_BLOCKS_TO_MBS(_blocks) (_blocks >> (20 - DATA_BLOCK_SHIFT))
  77
  78 /*
  79  * Default number of global data blocks(512K * PAGE_SIZE)
  80  * when the unmap thread will be started.
  81  */
  82 #define TCMU_GLOBAL_MAX_BLOCKS_DEF (512 * 1024)
  83
  84 static u8 tcmu_kern_cmd_reply_supported;
  85 static u8 tcmu_netlink_blocked;
  86
  87 static struct device *tcmu_root_device;
  88
  89 struct tcmu_hba {
  90         u32 host_id;
  91 };
  92
  93 #define TCMU_CONFIG_LEN 256
  94
  95 static DEFINE_MUTEX(tcmu_nl_cmd_mutex);
  96 static LIST_HEAD(tcmu_nl_cmd_list);
  97
  98 struct tcmu_dev;
  99
 100 struct tcmu_nl_cmd {
 101         /* wake up thread waiting for reply */
 102         struct completion complete;
 103         struct list_head nl_list;
 104         struct tcmu_dev *udev;
 105         int cmd;
 106         int status;
 107 };
 108
 109 struct tcmu_dev {
 110         struct list_head node;
 111         struct kref kref;
 112
 113         struct se_device se_dev;
 114
 115         char *name;
 116         struct se_hba *hba;
 117
 118 #define TCMU_DEV_BIT_OPEN 0
 119 #define TCMU_DEV_BIT_BROKEN 1
 120 #define TCMU_DEV_BIT_BLOCKED 2
 121 #define TCMU_DEV_BIT_TMR_NOTIFY 3
 122         unsigned long flags;
 123
 124         struct uio_info uio_info;
 125
 126         struct inode *inode;
 127
 128         struct tcmu_mailbox *mb_addr;
 129         uint64_t dev_size;
 130         u32 cmdr_size;
 131         u32 cmdr_last_cleaned;
 132         /* Offset of data area from start of mb */
 133         /* Must add data_off and mb_addr to get the address */
 134         size_t data_off;
 135         size_t data_size;
 136         uint32_t max_blocks;
 137         size_t ring_size;
 138
 139         struct mutex cmdr_lock;
 140         struct list_head qfull_queue;
 141         struct list_head tmr_queue;
 142
 143         uint32_t dbi_max;
 144         uint32_t dbi_thresh;
 145         unsigned long *data_bitmap;
 146         struct radix_tree_root data_blocks;
 147
 148         struct idr commands;
 149
 150         struct timer_list cmd_timer;
 151         unsigned int cmd_time_out;
 152         struct list_head inflight_queue;
 153
 154         struct timer_list qfull_timer;
 155         int qfull_time_out;
 156
 157         struct list_head timedout_entry;
 158
 159         struct tcmu_nl_cmd curr_nl_cmd;
 160
 161         char dev_config[TCMU_CONFIG_LEN];
 162
 163         int nl_reply_supported;
 164 };
 165
 166 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
 167
 168 #define CMDR_OFF sizeof(struct tcmu_mailbox)
 169
 170 struct tcmu_cmd {
 171         struct se_cmd *se_cmd;
 172         struct tcmu_dev *tcmu_dev;
 173         struct list_head queue_entry;
 174
 175         uint16_t cmd_id;
 176
 177         /* Can't use se_cmd when cleaning up expired cmds, because if
 178            cmd has been completed then accessing se_cmd is off limits */
 179         uint32_t dbi_cnt;
 180         uint32_t dbi_bidi_cnt;
 181         uint32_t dbi_cur;
 182         uint32_t *dbi;
 183
 184         uint32_t data_len_bidi;
 185
 186         unsigned long deadline;
 187
 188 #define TCMU_CMD_BIT_EXPIRED 0
 189         unsigned long flags;
 190 };
 191
 192 struct tcmu_tmr {
 193         struct list_head queue_entry;
 194
 195         uint8_t tmr_type;
 196         uint32_t tmr_cmd_cnt;
 197         int16_t tmr_cmd_ids[];
 198 };
 199
 200 /*
 201  * To avoid dead lock the mutex lock order should always be:
 202  *
 203  * mutex_lock(&root_udev_mutex);
 204  * ...
 205  * mutex_lock(&tcmu_dev->cmdr_lock);
 206  * mutex_unlock(&tcmu_dev->cmdr_lock);
 207  * ...
 208  * mutex_unlock(&root_udev_mutex);
 209  */
 210 static DEFINE_MUTEX(root_udev_mutex);
 211 static LIST_HEAD(root_udev);
 212
 213 static DEFINE_SPINLOCK(timed_out_udevs_lock);
 214 static LIST_HEAD(timed_out_udevs);
 215
 216 static struct kmem_cache *tcmu_cmd_cache;
 217
 218 static atomic_t global_db_count = ATOMIC_INIT(0);
 219 static struct delayed_work tcmu_unmap_work;
 220 static int tcmu_global_max_blocks = TCMU_GLOBAL_MAX_BLOCKS_DEF;
 221
 222 static int tcmu_set_global_max_data_area(const char *str,
 223                                          const struct kernel_param *kp)
 224 {
 225         int ret, max_area_mb;
 226
 227         ret = kstrtoint(str, 10, &max_area_mb);
 228         if (ret)
 229                 return -EINVAL;
 230
 231         if (max_area_mb <= 0) {
 232                 pr_err("global_max_data_area must be larger than 0.\n");
 233                 return -EINVAL;
 234         }
 235
 236         tcmu_global_max_blocks = TCMU_MBS_TO_BLOCKS(max_area_mb);
 237         if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
 238                 schedule_delayed_work(&tcmu_unmap_work, 0);
 239         else
 240                 cancel_delayed_work_sync(&tcmu_unmap_work);
 241
 242         return 0;
 243 }
 244
 245 static int tcmu_get_global_max_data_area(char *buffer,
 246                                          const struct kernel_param *kp)
 247 {
 248         return sprintf(buffer, "%d\n", TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
 249 }
 250
 251 static const struct kernel_param_ops tcmu_global_max_data_area_op = {
 252         .set = tcmu_set_global_max_data_area,
 253         .get = tcmu_get_global_max_data_area,
 254 };
 255
 256 module_param_cb(global_max_data_area_mb, &tcmu_global_max_data_area_op, NULL,
 257                 S_IWUSR | S_IRUGO);
 258 MODULE_PARM_DESC(global_max_data_area_mb,
 259                  "Max MBs allowed to be allocated to all the tcmu device's "
 260                  "data areas.");
 261
 262 static int tcmu_get_block_netlink(char *buffer,
 263                                   const struct kernel_param *kp)
 264 {
 265         return sprintf(buffer, "%s\n", tcmu_netlink_blocked ?
 266                        "blocked" : "unblocked");
 267 }
 268
 269 static int tcmu_set_block_netlink(const char *str,
 270                                   const struct kernel_param *kp)
 271 {
 272         int ret;
 273         u8 val;
 274
 275         ret = kstrtou8(str, 0, &val);
 276         if (ret < 0)
 277                 return ret;
 278
 279         if (val > 1) {
 280                 pr_err("Invalid block netlink value %u\n", val);
 281                 return -EINVAL;
 282         }
 283
 284         tcmu_netlink_blocked = val;
 285         return 0;
 286 }
 287
 288 static const struct kernel_param_ops tcmu_block_netlink_op = {
 289         .set = tcmu_set_block_netlink,
 290         .get = tcmu_get_block_netlink,
 291 };
 292
 293 module_param_cb(block_netlink, &tcmu_block_netlink_op, NULL, S_IWUSR | S_IRUGO);
 294 MODULE_PARM_DESC(block_netlink, "Block new netlink commands.");
 295
 296 static int tcmu_fail_netlink_cmd(struct tcmu_nl_cmd *nl_cmd)
 297 {
 298         struct tcmu_dev *udev = nl_cmd->udev;
 299
 300         if (!tcmu_netlink_blocked) {
 301                 pr_err("Could not reset device's netlink interface. Netlink is not blocked.\n");
 302                 return -EBUSY;
 303         }
 304
 305         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
 306                 pr_debug("Aborting nl cmd %d on %s\n", nl_cmd->cmd, udev->name);
 307                 nl_cmd->status = -EINTR;
 308                 list_del(&nl_cmd->nl_list);
 309                 complete(&nl_cmd->complete);
 310         }
 311         return 0;
 312 }
 313
 314 static int tcmu_set_reset_netlink(const char *str,
 315                                   const struct kernel_param *kp)
 316 {
 317         struct tcmu_nl_cmd *nl_cmd, *tmp_cmd;
 318         int ret;
 319         u8 val;
 320
 321         ret = kstrtou8(str, 0, &val);
 322         if (ret < 0)
 323                 return ret;
 324
 325         if (val != 1) {
 326                 pr_err("Invalid reset netlink value %u\n", val);
 327                 return -EINVAL;
 328         }
 329
 330         mutex_lock(&tcmu_nl_cmd_mutex);
 331         list_for_each_entry_safe(nl_cmd, tmp_cmd, &tcmu_nl_cmd_list, nl_list) {
 332                 ret = tcmu_fail_netlink_cmd(nl_cmd);
 333                 if (ret)
 334                         break;
 335         }
 336         mutex_unlock(&tcmu_nl_cmd_mutex);
 337
 338         return ret;
 339 }
 340
 341 static const struct kernel_param_ops tcmu_reset_netlink_op = {
 342         .set = tcmu_set_reset_netlink,
 343 };
 344
 345 module_param_cb(reset_netlink, &tcmu_reset_netlink_op, NULL, S_IWUSR);
 346 MODULE_PARM_DESC(reset_netlink, "Reset netlink commands.");
 347
 348 /* multicast group */
 349 enum tcmu_multicast_groups {
 350         TCMU_MCGRP_CONFIG,
 351 };
 352
 353 static const struct genl_multicast_group tcmu_mcgrps[] = {
 354         [TCMU_MCGRP_CONFIG] = { .name = "config", },
 355 };
 356
 357 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
 358         [TCMU_ATTR_DEVICE]      = { .type = NLA_STRING },
 359         [TCMU_ATTR_MINOR]       = { .type = NLA_U32 },
 360         [TCMU_ATTR_CMD_STATUS]  = { .type = NLA_S32 },
 361         [TCMU_ATTR_DEVICE_ID]   = { .type = NLA_U32 },
 362         [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
 363 };
 364
 365 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
 366 {
 367         struct tcmu_dev *udev = NULL;
 368         struct tcmu_nl_cmd *nl_cmd;
 369         int dev_id, rc, ret = 0;
 370
 371         if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
 372             !info->attrs[TCMU_ATTR_DEVICE_ID]) {
 373                 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
 374                 return -EINVAL;
 375         }
 376
 377         dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
 378         rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
 379
 380         mutex_lock(&tcmu_nl_cmd_mutex);
 381         list_for_each_entry(nl_cmd, &tcmu_nl_cmd_list, nl_list) {
 382                 if (nl_cmd->udev->se_dev.dev_index == dev_id) {
 383                         udev = nl_cmd->udev;
 384                         break;
 385                 }
 386         }
 387
 388         if (!udev) {
 389                 pr_err("tcmu nl cmd %u/%d completion could not find device with dev id %u.\n",
 390                        completed_cmd, rc, dev_id);
 391                 ret = -ENODEV;
 392                 goto unlock;
 393         }
 394         list_del(&nl_cmd->nl_list);
 395
 396         pr_debug("%s genl cmd done got id %d curr %d done %d rc %d stat %d\n",
 397                  udev->name, dev_id, nl_cmd->cmd, completed_cmd, rc,
 398                  nl_cmd->status);
 399
 400         if (nl_cmd->cmd != completed_cmd) {
 401                 pr_err("Mismatched commands on %s (Expecting reply for %d. Current %d).\n",
 402                        udev->name, completed_cmd, nl_cmd->cmd);
 403                 ret = -EINVAL;
 404                 goto unlock;
 405         }
 406
 407         nl_cmd->status = rc;
 408         complete(&nl_cmd->complete);
 409 unlock:
 410         mutex_unlock(&tcmu_nl_cmd_mutex);
 411         return ret;
 412 }
 413
 414 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
 415 {
 416         return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
 417 }
 418
 419 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
 420 {
 421         return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
 422 }
 423
 424 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
 425                                        struct genl_info *info)
 426 {
 427         return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
 428 }
 429
 430 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
 431 {
 432         if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
 433                 tcmu_kern_cmd_reply_supported  =
 434                         nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
 435                 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
 436                        tcmu_kern_cmd_reply_supported);
 437         }
 438
 439         return 0;
 440 }
 441
 442 static const struct genl_small_ops tcmu_genl_ops[] = {
 443         {
 444                 .cmd    = TCMU_CMD_SET_FEATURES,
 445                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 446                 .flags  = GENL_ADMIN_PERM,
 447                 .doit   = tcmu_genl_set_features,
 448         },
 449         {
 450                 .cmd    = TCMU_CMD_ADDED_DEVICE_DONE,
 451                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 452                 .flags  = GENL_ADMIN_PERM,
 453                 .doit   = tcmu_genl_add_dev_done,
 454         },
 455         {
 456                 .cmd    = TCMU_CMD_REMOVED_DEVICE_DONE,
 457                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 458                 .flags  = GENL_ADMIN_PERM,
 459                 .doit   = tcmu_genl_rm_dev_done,
 460         },
 461         {
 462                 .cmd    = TCMU_CMD_RECONFIG_DEVICE_DONE,
 463                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 464                 .flags  = GENL_ADMIN_PERM,
 465                 .doit   = tcmu_genl_reconfig_dev_done,
 466         },
 467 };
 468
 469 /* Our generic netlink family */
 470 static struct genl_family tcmu_genl_family __ro_after_init = {
 471         .module = THIS_MODULE,
 472         .hdrsize = 0,
 473         .name = "TCM-USER",
 474         .version = 2,
 475         .maxattr = TCMU_ATTR_MAX,
 476         .policy = tcmu_attr_policy,
 477         .mcgrps = tcmu_mcgrps,
 478         .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
 479         .netnsok = true,
 480         .small_ops = tcmu_genl_ops,
 481         .n_small_ops = ARRAY_SIZE(tcmu_genl_ops),
 482 };
 483
 484 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
 485 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
 486 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
 487 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
 488
 489 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
 490 {
 491         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 492         uint32_t i;
 493
 494         for (i = 0; i < len; i++)
 495                 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
 496 }
 497
 498 static inline int tcmu_get_empty_block(struct tcmu_dev *udev,
 499                                        struct tcmu_cmd *tcmu_cmd,
 500                                        int prev_dbi, int *iov_cnt)
 501 {
 502         struct page *page;
 503         int ret, dbi;
 504
 505         dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
 506         if (dbi == udev->dbi_thresh)
 507                 return -1;
 508
 509         page = radix_tree_lookup(&udev->data_blocks, dbi);
 510         if (!page) {
 511                 if (atomic_add_return(1, &global_db_count) >
 512                                       tcmu_global_max_blocks)
 513                         schedule_delayed_work(&tcmu_unmap_work, 0);
 514
 515                 /* try to get new page from the mm */
 516                 page = alloc_page(GFP_NOIO);
 517                 if (!page)
 518                         goto err_alloc;
 519
 520                 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
 521                 if (ret)
 522                         goto err_insert;
 523         }
 524
 525         if (dbi > udev->dbi_max)
 526                 udev->dbi_max = dbi;
 527
 528         set_bit(dbi, udev->data_bitmap);
 529         tcmu_cmd_set_dbi(tcmu_cmd, dbi);
 530
 531         if (dbi != prev_dbi + 1)
 532                 *iov_cnt += 1;
 533
 534         return dbi;
 535 err_insert:
 536         __free_page(page);
 537 err_alloc:
 538         atomic_dec(&global_db_count);
 539         return -1;
 540 }
 541
 542 static int tcmu_get_empty_blocks(struct tcmu_dev *udev,
 543                                  struct tcmu_cmd *tcmu_cmd, int dbi_cnt)
 544 {
 545         /* start value of dbi + 1 must not be a valid dbi */
 546         int dbi = -2;
 547         int i, iov_cnt = 0;
 548
 549         for (i = 0; i < dbi_cnt; i++) {
 550                 dbi = tcmu_get_empty_block(udev, tcmu_cmd, dbi, &iov_cnt);
 551                 if (dbi < 0)
 552                         return -1;
 553         }
 554         return iov_cnt;
 555 }
 556
 557 static inline struct page *
 558 tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
 559 {
 560         return radix_tree_lookup(&udev->data_blocks, dbi);
 561 }
 562
 563 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
 564 {
 565         kfree(tcmu_cmd->dbi);
 566         kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 567 }
 568
 569 static inline void tcmu_cmd_set_block_cnts(struct tcmu_cmd *cmd)
 570 {
 571         int i, len;
 572         struct se_cmd *se_cmd = cmd->se_cmd;
 573
 574         cmd->dbi_cnt = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
 575
 576         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 577                 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
 578                 for (i = 0, len = 0; i < se_cmd->t_bidi_data_nents; i++)
 579                         len += se_cmd->t_bidi_data_sg[i].length;
 580                 cmd->dbi_bidi_cnt = DIV_ROUND_UP(len, DATA_BLOCK_SIZE);
 581                 cmd->dbi_cnt += cmd->dbi_bidi_cnt;
 582                 cmd->data_len_bidi = len;
 583         }
 584 }
 585
 586 static int new_block_to_iov(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 587                             struct iovec **iov, int prev_dbi, int len)
 588 {
 589         /* Get the next dbi */
 590         int dbi = tcmu_cmd_get_dbi(cmd);
 591
 592         /* Do not add more than DATA_BLOCK_SIZE to iov */
 593         if (len > DATA_BLOCK_SIZE)
 594                 len = DATA_BLOCK_SIZE;
 595
 596         /*
 597          * The following code will gather and map the blocks to the same iovec
 598          * when the blocks are all next to each other.
 599          */
 600         if (dbi != prev_dbi + 1) {
 601                 /* dbi is not next to previous dbi, so start new iov */
 602                 if (prev_dbi >= 0)
 603                         (*iov)++;
 604                 /* write offset relative to mb_addr */
 605                 (*iov)->iov_base = (void __user *)
 606                                 (udev->data_off + dbi * DATA_BLOCK_SIZE);
 607         }
 608         (*iov)->iov_len += len;
 609
 610         return dbi;
 611 }
 612
 613 static void tcmu_setup_iovs(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 614                             struct iovec **iov, int data_length)
 615 {
 616         /* start value of dbi + 1 must not be a valid dbi */
 617         int dbi = -2;
 618
 619         /* We prepare the IOVs for DMA_FROM_DEVICE transfer direction */
 620         for (; data_length > 0; data_length -= DATA_BLOCK_SIZE)
 621                 dbi = new_block_to_iov(udev, cmd, iov, dbi, data_length);
 622 }
 623
 624 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
 625 {
 626         struct se_device *se_dev = se_cmd->se_dev;
 627         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 628         struct tcmu_cmd *tcmu_cmd;
 629
 630         tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_NOIO);
 631         if (!tcmu_cmd)
 632                 return NULL;
 633
 634         INIT_LIST_HEAD(&tcmu_cmd->queue_entry);
 635         tcmu_cmd->se_cmd = se_cmd;
 636         tcmu_cmd->tcmu_dev = udev;
 637
 638         tcmu_cmd_set_block_cnts(tcmu_cmd);
 639         tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
 640                                 GFP_NOIO);
 641         if (!tcmu_cmd->dbi) {
 642                 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 643                 return NULL;
 644         }
 645
 646         return tcmu_cmd;
 647 }
 648
 649 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
 650 {
 651         unsigned long offset = offset_in_page(vaddr);
 652         void *start = vaddr - offset;
 653
 654         size = round_up(size+offset, PAGE_SIZE);
 655
 656         while (size) {
 657                 flush_dcache_page(vmalloc_to_page(start));
 658                 start += PAGE_SIZE;
 659                 size -= PAGE_SIZE;
 660         }
 661 }
 662
 663 /*
 664  * Some ring helper functions. We don't assume size is a power of 2 so
 665  * we can't use circ_buf.h.
 666  */
 667 static inline size_t spc_used(size_t head, size_t tail, size_t size)
 668 {
 669         int diff = head - tail;
 670
 671         if (diff >= 0)
 672                 return diff;
 673         else
 674                 return size + diff;
 675 }
 676
 677 static inline size_t spc_free(size_t head, size_t tail, size_t size)
 678 {
 679         /* Keep 1 byte unused or we can't tell full from empty */
 680         return (size - spc_used(head, tail, size) - 1);
 681 }
 682
 683 static inline size_t head_to_end(size_t head, size_t size)
 684 {
 685         return size - head;
 686 }
 687
 688 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
 689
 690 #define TCMU_SG_TO_DATA_AREA 1
 691 #define TCMU_DATA_AREA_TO_SG 2
 692
 693 static inline void tcmu_copy_data(struct tcmu_dev *udev,
 694                                   struct tcmu_cmd *tcmu_cmd, uint32_t direction,
 695                                   struct scatterlist *sg, unsigned int sg_nents,
 696                                   struct iovec **iov, size_t data_len)
 697 {
 698         /* start value of dbi + 1 must not be a valid dbi */
 699         int dbi = -2;
 700         size_t block_remaining, cp_len;
 701         struct sg_mapping_iter sg_iter;
 702         unsigned int sg_flags;
 703         struct page *page;
 704         void *data_page_start, *data_addr;
 705
 706         if (direction == TCMU_SG_TO_DATA_AREA)
 707                 sg_flags = SG_MITER_ATOMIC | SG_MITER_FROM_SG;
 708         else
 709                 sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG;
 710         sg_miter_start(&sg_iter, sg, sg_nents, sg_flags);
 711
 712         while (data_len) {
 713                 if (direction == TCMU_SG_TO_DATA_AREA)
 714                         dbi = new_block_to_iov(udev, tcmu_cmd, iov, dbi,
 715                                                data_len);
 716                 else
 717                         dbi = tcmu_cmd_get_dbi(tcmu_cmd);
 718                 page = tcmu_get_block_page(udev, dbi);
 719                 if (direction == TCMU_DATA_AREA_TO_SG)
 720                         flush_dcache_page(page);
 721                 data_page_start = kmap_atomic(page);
 722                 block_remaining = DATA_BLOCK_SIZE;
 723
 724                 while (block_remaining && data_len) {
 725                         if (!sg_miter_next(&sg_iter)) {
 726                                 /* set length to 0 to abort outer loop */
 727                                 data_len = 0;
 728                                 pr_debug("tcmu_move_data: aborting data copy due to exhausted sg_list\n");
 729                                 break;
 730                         }
 731                         cp_len = min3(sg_iter.length, block_remaining, data_len);
 732
 733                         data_addr = data_page_start +
 734                                     DATA_BLOCK_SIZE - block_remaining;
 735                         if (direction == TCMU_SG_TO_DATA_AREA)
 736                                 memcpy(data_addr, sg_iter.addr, cp_len);
 737                         else
 738                                 memcpy(sg_iter.addr, data_addr, cp_len);
 739
 740                         data_len -= cp_len;
 741                         block_remaining -= cp_len;
 742                         sg_iter.consumed = cp_len;
 743                 }
 744                 sg_miter_stop(&sg_iter);
 745
 746                 kunmap_atomic(data_page_start);
 747                 if (direction == TCMU_SG_TO_DATA_AREA)
 748                         flush_dcache_page(page);
 749         }
 750 }
 751
 752 static void scatter_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
 753                               struct iovec **iov)
 754 {
 755         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 756
 757         tcmu_copy_data(udev, tcmu_cmd, TCMU_SG_TO_DATA_AREA, se_cmd->t_data_sg,
 758                        se_cmd->t_data_nents, iov, se_cmd->data_length);
 759 }
 760
 761 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
 762                              bool bidi, uint32_t read_len)
 763 {
 764         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 765         struct scatterlist *data_sg;
 766         unsigned int data_nents;
 767
 768         if (!bidi) {
 769                 data_sg = se_cmd->t_data_sg;
 770                 data_nents = se_cmd->t_data_nents;
 771         } else {
 772                 /*
 773                  * For bidi case, the first count blocks are for Data-Out
 774                  * buffer blocks, and before gathering the Data-In buffer
 775                  * the Data-Out buffer blocks should be skipped.
 776                  */
 777                 tcmu_cmd_set_dbi_cur(tcmu_cmd,
 778                                      tcmu_cmd->dbi_cnt - tcmu_cmd->dbi_bidi_cnt);
 779
 780                 data_sg = se_cmd->t_bidi_data_sg;
 781                 data_nents = se_cmd->t_bidi_data_nents;
 782         }
 783
 784         tcmu_copy_data(udev, tcmu_cmd, TCMU_DATA_AREA_TO_SG, data_sg,
 785                        data_nents, NULL, read_len);
 786 }
 787
 788 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
 789 {
 790         return thresh - bitmap_weight(bitmap, thresh);
 791 }
 792
 793 /*
 794  * We can't queue a command until we have space available on the cmd ring.
 795  *
 796  * Called with ring lock held.
 797  */
 798 static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size)
 799 {
 800         struct tcmu_mailbox *mb = udev->mb_addr;
 801         size_t space, cmd_needed;
 802         u32 cmd_head;
 803
 804         tcmu_flush_dcache_range(mb, sizeof(*mb));
 805
 806         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 807
 808         /*
 809          * If cmd end-of-ring space is too small then we need space for a NOP plus
 810          * original cmd - cmds are internally contiguous.
 811          */
 812         if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
 813                 cmd_needed = cmd_size;
 814         else
 815                 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
 816
 817         space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
 818         if (space < cmd_needed) {
 819                 pr_debug("no cmd space: %u %u %u\n", cmd_head,
 820                        udev->cmdr_last_cleaned, udev->cmdr_size);
 821                 return false;
 822         }
 823         return true;
 824 }
 825
 826 /*
 827  * We have to allocate data buffers before we can queue a command.
 828  * Returns -1 on error (not enough space) or number of needed iovs on success
 829  *
 830  * Called with ring lock held.
 831  */
 832 static int tcmu_alloc_data_space(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 833                                   int *iov_bidi_cnt)
 834 {
 835         int space, iov_cnt = 0, ret = 0;
 836
 837         if (!cmd->dbi_cnt)
 838                 goto wr_iov_cnts;
 839
 840         /* try to check and get the data blocks as needed */
 841         space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
 842         if (space < cmd->dbi_cnt) {
 843                 unsigned long blocks_left =
 844                                 (udev->max_blocks - udev->dbi_thresh) + space;
 845
 846                 if (blocks_left < cmd->dbi_cnt) {
 847                         pr_debug("no data space: only %lu available, but ask for %lu\n",
 848                                         blocks_left * DATA_BLOCK_SIZE,
 849                                         cmd->dbi_cnt * DATA_BLOCK_SIZE);
 850                         return -1;
 851                 }
 852
 853                 udev->dbi_thresh += cmd->dbi_cnt;
 854                 if (udev->dbi_thresh > udev->max_blocks)
 855                         udev->dbi_thresh = udev->max_blocks;
 856         }
 857
 858         iov_cnt = tcmu_get_empty_blocks(udev, cmd,
 859                                         cmd->dbi_cnt - cmd->dbi_bidi_cnt);
 860         if (iov_cnt < 0)
 861                 return -1;
 862
 863         if (cmd->dbi_bidi_cnt) {
 864                 ret = tcmu_get_empty_blocks(udev, cmd, cmd->dbi_bidi_cnt);
 865                 if (ret < 0)
 866                         return -1;
 867         }
 868 wr_iov_cnts:
 869         *iov_bidi_cnt = ret;
 870         return iov_cnt + ret;
 871 }
 872
 873 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
 874 {
 875         return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 876                         sizeof(struct tcmu_cmd_entry));
 877 }
 878
 879 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
 880                                            size_t base_command_size)
 881 {
 882         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 883         size_t command_size;
 884
 885         command_size = base_command_size +
 886                 round_up(scsi_command_size(se_cmd->t_task_cdb),
 887                                 TCMU_OP_ALIGN_SIZE);
 888
 889         WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
 890
 891         return command_size;
 892 }
 893
 894 static void tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
 895                                  struct timer_list *timer)
 896 {
 897         if (!tmo)
 898                 return;
 899
 900         tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
 901         if (!timer_pending(timer))
 902                 mod_timer(timer, tcmu_cmd->deadline);
 903
 904         pr_debug("Timeout set up for cmd %p, dev = %s, tmo = %lu\n", tcmu_cmd,
 905                  tcmu_cmd->tcmu_dev->name, tmo / MSEC_PER_SEC);
 906 }
 907
 908 static int add_to_qfull_queue(struct tcmu_cmd *tcmu_cmd)
 909 {
 910         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 911         unsigned int tmo;
 912
 913         /*
 914          * For backwards compat if qfull_time_out is not set use
 915          * cmd_time_out and if that's not set use the default time out.
 916          */
 917         if (!udev->qfull_time_out)
 918                 return -ETIMEDOUT;
 919         else if (udev->qfull_time_out > 0)
 920                 tmo = udev->qfull_time_out;
 921         else if (udev->cmd_time_out)
 922                 tmo = udev->cmd_time_out;
 923         else
 924                 tmo = TCMU_TIME_OUT;
 925
 926         tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
 927
 928         list_add_tail(&tcmu_cmd->queue_entry, &udev->qfull_queue);
 929         pr_debug("adding cmd %p on dev %s to ring space wait queue\n",
 930                  tcmu_cmd, udev->name);
 931         return 0;
 932 }
 933
 934 static uint32_t ring_insert_padding(struct tcmu_dev *udev, size_t cmd_size)
 935 {
 936         struct tcmu_cmd_entry_hdr *hdr;
 937         struct tcmu_mailbox *mb = udev->mb_addr;
 938         uint32_t cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 939
 940         /* Insert a PAD if end-of-ring space is too small */
 941         if (head_to_end(cmd_head, udev->cmdr_size) < cmd_size) {
 942                 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
 943
 944                 hdr = (void *) mb + CMDR_OFF + cmd_head;
 945                 tcmu_hdr_set_op(&hdr->len_op, TCMU_OP_PAD);
 946                 tcmu_hdr_set_len(&hdr->len_op, pad_size);
 947                 hdr->cmd_id = 0; /* not used for PAD */
 948                 hdr->kflags = 0;
 949                 hdr->uflags = 0;
 950                 tcmu_flush_dcache_range(hdr, sizeof(*hdr));
 951
 952                 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
 953                 tcmu_flush_dcache_range(mb, sizeof(*mb));
 954
 955                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 956                 WARN_ON(cmd_head != 0);
 957         }
 958
 959         return cmd_head;
 960 }
 961
 962 /**
 963  * queue_cmd_ring - queue cmd to ring or internally
 964  * @tcmu_cmd: cmd to queue
 965  * @scsi_err: TCM error code if failure (-1) returned.
 966  *
 967  * Returns:
 968  * -1 we cannot queue internally or to the ring.
 969  *  0 success
 970  *  1 internally queued to wait for ring memory to free.
 971  */
 972 static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
 973 {
 974         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 975         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 976         size_t base_command_size, command_size;
 977         struct tcmu_mailbox *mb = udev->mb_addr;
 978         struct tcmu_cmd_entry *entry;
 979         struct iovec *iov;
 980         int iov_cnt, iov_bidi_cnt, cmd_id;
 981         uint32_t cmd_head;
 982         uint64_t cdb_off;
 983         /* size of data buffer needed */
 984         size_t data_length = (size_t)tcmu_cmd->dbi_cnt * DATA_BLOCK_SIZE;
 985
 986         *scsi_err = TCM_NO_SENSE;
 987
 988         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
 989                 *scsi_err = TCM_LUN_BUSY;
 990                 return -1;
 991         }
 992
 993         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
 994                 *scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 995                 return -1;
 996         }
 997
 998         if (!list_empty(&udev->qfull_queue))
 999                 goto queue;
1000
1001         if (data_length > udev->data_size) {
1002                 pr_warn("TCMU: Request of size %zu is too big for %zu data area\n",
1003                         data_length, udev->data_size);
1004                 *scsi_err = TCM_INVALID_CDB_FIELD;
1005                 return -1;
1006         }
1007
1008         iov_cnt = tcmu_alloc_data_space(udev, tcmu_cmd, &iov_bidi_cnt);
1009         if (iov_cnt < 0)
1010                 goto free_and_queue;
1011
1012         /*
1013          * Must be a certain minimum size for response sense info, but
1014          * also may be larger if the iov array is large.
1015          */
1016         base_command_size = tcmu_cmd_get_base_cmd_size(iov_cnt);
1017         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
1018
1019         if (command_size > (udev->cmdr_size / 2)) {
1020                 pr_warn("TCMU: Request of size %zu is too big for %u cmd ring\n",
1021                         command_size, udev->cmdr_size);
1022                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1023                 *scsi_err = TCM_INVALID_CDB_FIELD;
1024                 return -1;
1025         }
1026
1027         if (!is_ring_space_avail(udev, command_size))
1028                 /*
1029                  * Don't leave commands partially setup because the unmap
1030                  * thread might need the blocks to make forward progress.
1031                  */
1032                 goto free_and_queue;
1033
1034         cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 1, USHRT_MAX, GFP_NOWAIT);
1035         if (cmd_id < 0) {
1036                 pr_err("tcmu: Could not allocate cmd id.\n");
1037
1038                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
1039                 *scsi_err = TCM_OUT_OF_RESOURCES;
1040                 return -1;
1041         }
1042         tcmu_cmd->cmd_id = cmd_id;
1043
1044         pr_debug("allocated cmd id %u for cmd %p dev %s\n", tcmu_cmd->cmd_id,
1045                  tcmu_cmd, udev->name);
1046
1047         cmd_head = ring_insert_padding(udev, command_size);
1048
1049         entry = (void *) mb + CMDR_OFF + cmd_head;
1050         memset(entry, 0, command_size);
1051         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
1052
1053         /* prepare iov list and copy data to data area if necessary */
1054         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1055         iov = &entry->req.iov[0];
1056
1057         if (se_cmd->data_direction == DMA_TO_DEVICE ||
1058             se_cmd->se_cmd_flags & SCF_BIDI)
1059                 scatter_data_area(udev, tcmu_cmd, &iov);
1060         else
1061                 tcmu_setup_iovs(udev, tcmu_cmd, &iov, se_cmd->data_length);
1062
1063         entry->req.iov_cnt = iov_cnt - iov_bidi_cnt;
1064
1065         /* Handle BIDI commands */
1066         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1067                 iov++;
1068                 tcmu_setup_iovs(udev, tcmu_cmd, &iov, tcmu_cmd->data_len_bidi);
1069                 entry->req.iov_bidi_cnt = iov_bidi_cnt;
1070         }
1071
1072         tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out, &udev->cmd_timer);
1073
1074         entry->hdr.cmd_id = tcmu_cmd->cmd_id;
1075
1076         tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
1077
1078         /* All offsets relative to mb_addr, not start of entry! */
1079         cdb_off = CMDR_OFF + cmd_head + base_command_size;
1080         memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
1081         entry->req.cdb_off = cdb_off;
1082         tcmu_flush_dcache_range(entry, command_size);
1083
1084         UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1085         tcmu_flush_dcache_range(mb, sizeof(*mb));
1086
1087         list_add_tail(&tcmu_cmd->queue_entry, &udev->inflight_queue);
1088
1089         /* TODO: only if FLUSH and FUA? */
1090         uio_event_notify(&udev->uio_info);
1091
1092         return 0;
1093
1094 free_and_queue:
1095         tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1096         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1097
1098 queue:
1099         if (add_to_qfull_queue(tcmu_cmd)) {
1100                 *scsi_err = TCM_OUT_OF_RESOURCES;
1101                 return -1;
1102         }
1103
1104         return 1;
1105 }
1106
1107 /**
1108  * queue_tmr_ring - queue tmr info to ring or internally
1109  * @udev: related tcmu_dev
1110  * @tmr: tcmu_tmr containing tmr info to queue
1111  *
1112  * Returns:
1113  *  0 success
1114  *  1 internally queued to wait for ring memory to free.
1115  */
1116 static int
1117 queue_tmr_ring(struct tcmu_dev *udev, struct tcmu_tmr *tmr)
1118 {
1119         struct tcmu_tmr_entry *entry;
1120         int cmd_size;
1121         int id_list_sz;
1122         struct tcmu_mailbox *mb = udev->mb_addr;
1123         uint32_t cmd_head;
1124
1125         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
1126                 goto out_free;
1127
1128         id_list_sz = sizeof(tmr->tmr_cmd_ids[0]) * tmr->tmr_cmd_cnt;
1129         cmd_size = round_up(sizeof(*entry) + id_list_sz, TCMU_OP_ALIGN_SIZE);
1130
1131         if (!list_empty(&udev->tmr_queue) ||
1132             !is_ring_space_avail(udev, cmd_size)) {
1133                 list_add_tail(&tmr->queue_entry, &udev->tmr_queue);
1134                 pr_debug("adding tmr %p on dev %s to TMR ring space wait queue\n",
1135                          tmr, udev->name);
1136                 return 1;
1137         }
1138
1139         cmd_head = ring_insert_padding(udev, cmd_size);
1140
1141         entry = (void *)mb + CMDR_OFF + cmd_head;
1142         memset(entry, 0, cmd_size);
1143         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_TMR);
1144         tcmu_hdr_set_len(&entry->hdr.len_op, cmd_size);
1145         entry->tmr_type = tmr->tmr_type;
1146         entry->cmd_cnt = tmr->tmr_cmd_cnt;
1147         memcpy(&entry->cmd_ids[0], &tmr->tmr_cmd_ids[0], id_list_sz);
1148         tcmu_flush_dcache_range(entry, cmd_size);
1149
1150         UPDATE_HEAD(mb->cmd_head, cmd_size, udev->cmdr_size);
1151         tcmu_flush_dcache_range(mb, sizeof(*mb));
1152
1153         uio_event_notify(&udev->uio_info);
1154
1155 out_free:
1156         kfree(tmr);
1157
1158         return 0;
1159 }
1160
1161 static sense_reason_t
1162 tcmu_queue_cmd(struct se_cmd *se_cmd)
1163 {
1164         struct se_device *se_dev = se_cmd->se_dev;
1165         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1166         struct tcmu_cmd *tcmu_cmd;
1167         sense_reason_t scsi_ret = TCM_CHECK_CONDITION_ABORT_CMD;
1168         int ret = -1;
1169
1170         tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1171         if (!tcmu_cmd)
1172                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1173
1174         mutex_lock(&udev->cmdr_lock);
1175         if (!(se_cmd->transport_state & CMD_T_ABORTED))
1176                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1177         if (ret < 0)
1178                 tcmu_free_cmd(tcmu_cmd);
1179         else
1180                 se_cmd->priv = tcmu_cmd;
1181         mutex_unlock(&udev->cmdr_lock);
1182         return scsi_ret;
1183 }
1184
1185 static void tcmu_set_next_deadline(struct list_head *queue,
1186                                    struct timer_list *timer)
1187 {
1188         struct tcmu_cmd *cmd;
1189
1190         if (!list_empty(queue)) {
1191                 cmd = list_first_entry(queue, struct tcmu_cmd, queue_entry);
1192                 mod_timer(timer, cmd->deadline);
1193         } else
1194                 del_timer(timer);
1195 }
1196
1197 static int
1198 tcmu_tmr_type(enum tcm_tmreq_table tmf)
1199 {
1200         switch (tmf) {
1201         case TMR_ABORT_TASK:            return TCMU_TMR_ABORT_TASK;
1202         case TMR_ABORT_TASK_SET:        return TCMU_TMR_ABORT_TASK_SET;
1203         case TMR_CLEAR_ACA:             return TCMU_TMR_CLEAR_ACA;
1204         case TMR_CLEAR_TASK_SET:        return TCMU_TMR_CLEAR_TASK_SET;
1205         case TMR_LUN_RESET:             return TCMU_TMR_LUN_RESET;
1206         case TMR_TARGET_WARM_RESET:     return TCMU_TMR_TARGET_WARM_RESET;
1207         case TMR_TARGET_COLD_RESET:     return TCMU_TMR_TARGET_COLD_RESET;
1208         case TMR_LUN_RESET_PRO:         return TCMU_TMR_LUN_RESET_PRO;
1209         default:                        return TCMU_TMR_UNKNOWN;
1210         }
1211 }
1212
1213 static void
1214 tcmu_tmr_notify(struct se_device *se_dev, enum tcm_tmreq_table tmf,
1215                 struct list_head *cmd_list)
1216 {
1217         int i = 0, cmd_cnt = 0;
1218         bool unqueued = false;
1219         uint16_t *cmd_ids = NULL;
1220         struct tcmu_cmd *cmd;
1221         struct se_cmd *se_cmd;
1222         struct tcmu_tmr *tmr;
1223         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1224
1225         mutex_lock(&udev->cmdr_lock);
1226
1227         /* First we check for aborted commands in qfull_queue */
1228         list_for_each_entry(se_cmd, cmd_list, state_list) {
1229                 i++;
1230                 if (!se_cmd->priv)
1231                         continue;
1232                 cmd = se_cmd->priv;
1233                 /* Commands on qfull queue have no id yet */
1234                 if (cmd->cmd_id) {
1235                         cmd_cnt++;
1236                         continue;
1237                 }
1238                 pr_debug("Removing aborted command %p from queue on dev %s.\n",
1239                          cmd, udev->name);
1240
1241                 list_del_init(&cmd->queue_entry);
1242                 tcmu_free_cmd(cmd);
1243                 se_cmd->priv = NULL;
1244                 target_complete_cmd(se_cmd, SAM_STAT_TASK_ABORTED);
1245                 unqueued = true;
1246         }
1247         if (unqueued)
1248                 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1249
1250         if (!test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags))
1251                 goto unlock;
1252
1253         pr_debug("TMR event %d on dev %s, aborted cmds %d, afflicted cmd_ids %d\n",
1254                  tcmu_tmr_type(tmf), udev->name, i, cmd_cnt);
1255
1256         tmr = kmalloc(sizeof(*tmr) + cmd_cnt * sizeof(*cmd_ids), GFP_KERNEL);
1257         if (!tmr)
1258                 goto unlock;
1259
1260         tmr->tmr_type = tcmu_tmr_type(tmf);
1261         tmr->tmr_cmd_cnt = cmd_cnt;
1262
1263         if (cmd_cnt != 0) {
1264                 cmd_cnt = 0;
1265                 list_for_each_entry(se_cmd, cmd_list, state_list) {
1266                         if (!se_cmd->priv)
1267                                 continue;
1268                         cmd = se_cmd->priv;
1269                         if (cmd->cmd_id)
1270                                 tmr->tmr_cmd_ids[cmd_cnt++] = cmd->cmd_id;
1271                 }
1272         }
1273
1274         queue_tmr_ring(udev, tmr);
1275
1276 unlock:
1277         mutex_unlock(&udev->cmdr_lock);
1278 }
1279
1280 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
1281 {
1282         struct se_cmd *se_cmd = cmd->se_cmd;
1283         struct tcmu_dev *udev = cmd->tcmu_dev;
1284         bool read_len_valid = false;
1285         uint32_t read_len;
1286
1287         /*
1288          * cmd has been completed already from timeout, just reclaim
1289          * data area space and free cmd
1290          */
1291         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1292                 WARN_ON_ONCE(se_cmd);
1293                 goto out;
1294         }
1295
1296         list_del_init(&cmd->queue_entry);
1297
1298         tcmu_cmd_reset_dbi_cur(cmd);
1299
1300         if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1301                 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1302                         cmd->se_cmd);
1303                 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1304                 goto done;
1305         }
1306
1307         read_len = se_cmd->data_length;
1308         if (se_cmd->data_direction == DMA_FROM_DEVICE &&
1309             (entry->hdr.uflags & TCMU_UFLAG_READ_LEN) && entry->rsp.read_len) {
1310                 read_len_valid = true;
1311                 if (entry->rsp.read_len < read_len)
1312                         read_len = entry->rsp.read_len;
1313         }
1314
1315         if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1316                 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1317                 if (!read_len_valid )
1318                         goto done;
1319                 else
1320                         se_cmd->se_cmd_flags |= SCF_TREAT_READ_AS_NORMAL;
1321         }
1322         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1323                 /* Get Data-In buffer before clean up */
1324                 gather_data_area(udev, cmd, true, read_len);
1325         } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1326                 gather_data_area(udev, cmd, false, read_len);
1327         } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1328                 /* TODO: */
1329         } else if (se_cmd->data_direction != DMA_NONE) {
1330                 pr_warn("TCMU: data direction was %d!\n",
1331                         se_cmd->data_direction);
1332         }
1333
1334 done:
1335         se_cmd->priv = NULL;
1336         if (read_len_valid) {
1337                 pr_debug("read_len = %d\n", read_len);
1338                 target_complete_cmd_with_length(cmd->se_cmd,
1339                                         entry->rsp.scsi_status, read_len);
1340         } else
1341                 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1342
1343 out:
1344         tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1345         tcmu_free_cmd(cmd);
1346 }
1347
1348 static int tcmu_run_tmr_queue(struct tcmu_dev *udev)
1349 {
1350         struct tcmu_tmr *tmr, *tmp;
1351         LIST_HEAD(tmrs);
1352
1353         if (list_empty(&udev->tmr_queue))
1354                 return 1;
1355
1356         pr_debug("running %s's tmr queue\n", udev->name);
1357
1358         list_splice_init(&udev->tmr_queue, &tmrs);
1359
1360         list_for_each_entry_safe(tmr, tmp, &tmrs, queue_entry) {
1361                 list_del_init(&tmr->queue_entry);
1362
1363                 pr_debug("removing tmr %p on dev %s from queue\n",
1364                          tmr, udev->name);
1365
1366                 if (queue_tmr_ring(udev, tmr)) {
1367                         pr_debug("ran out of space during tmr queue run\n");
1368                         /*
1369                          * tmr was requeued, so just put all tmrs back in
1370                          * the queue
1371                          */
1372                         list_splice_tail(&tmrs, &udev->tmr_queue);
1373                         return 0;
1374                 }
1375         }
1376
1377         return 1;
1378 }
1379
1380 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
1381 {
1382         struct tcmu_mailbox *mb;
1383         struct tcmu_cmd *cmd;
1384         bool free_space = false;
1385
1386         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1387                 pr_err("ring broken, not handling completions\n");
1388                 return 0;
1389         }
1390
1391         mb = udev->mb_addr;
1392         tcmu_flush_dcache_range(mb, sizeof(*mb));
1393
1394         while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1395
1396                 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
1397
1398                 /*
1399                  * Flush max. up to end of cmd ring since current entry might
1400                  * be a padding that is shorter than sizeof(*entry)
1401                  */
1402                 size_t ring_left = head_to_end(udev->cmdr_last_cleaned,
1403                                                udev->cmdr_size);
1404                 tcmu_flush_dcache_range(entry, ring_left < sizeof(*entry) ?
1405                                         ring_left : sizeof(*entry));
1406
1407                 free_space = true;
1408
1409                 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD ||
1410                     tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_TMR) {
1411                         UPDATE_HEAD(udev->cmdr_last_cleaned,
1412                                     tcmu_hdr_get_len(entry->hdr.len_op),
1413                                     udev->cmdr_size);
1414                         continue;
1415                 }
1416                 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1417
1418                 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1419                 if (!cmd) {
1420                         pr_err("cmd_id %u not found, ring is broken\n",
1421                                entry->hdr.cmd_id);
1422                         set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1423                         break;
1424                 }
1425
1426                 tcmu_handle_completion(cmd, entry);
1427
1428                 UPDATE_HEAD(udev->cmdr_last_cleaned,
1429                             tcmu_hdr_get_len(entry->hdr.len_op),
1430                             udev->cmdr_size);
1431         }
1432         if (free_space)
1433                 free_space = tcmu_run_tmr_queue(udev);
1434
1435         if (atomic_read(&global_db_count) > tcmu_global_max_blocks &&
1436             idr_is_empty(&udev->commands) && list_empty(&udev->qfull_queue)) {
1437                 /*
1438                  * Allocated blocks exceeded global block limit, currently no
1439                  * more pending or waiting commands so try to reclaim blocks.
1440                  */
1441                 schedule_delayed_work(&tcmu_unmap_work, 0);
1442         }
1443         if (udev->cmd_time_out)
1444                 tcmu_set_next_deadline(&udev->inflight_queue, &udev->cmd_timer);
1445
1446         return free_space;
1447 }
1448
1449 static void tcmu_check_expired_ring_cmd(struct tcmu_cmd *cmd)
1450 {
1451         struct se_cmd *se_cmd;
1452
1453         if (!time_after_eq(jiffies, cmd->deadline))
1454                 return;
1455
1456         set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1457         list_del_init(&cmd->queue_entry);
1458         se_cmd = cmd->se_cmd;
1459         se_cmd->priv = NULL;
1460         cmd->se_cmd = NULL;
1461
1462         pr_debug("Timing out inflight cmd %u on dev %s.\n",
1463                  cmd->cmd_id, cmd->tcmu_dev->name);
1464
1465         target_complete_cmd(se_cmd, SAM_STAT_CHECK_CONDITION);
1466 }
1467
1468 static void tcmu_check_expired_queue_cmd(struct tcmu_cmd *cmd)
1469 {
1470         struct se_cmd *se_cmd;
1471
1472         if (!time_after_eq(jiffies, cmd->deadline))
1473                 return;
1474
1475         pr_debug("Timing out queued cmd %p on dev %s.\n",
1476                   cmd, cmd->tcmu_dev->name);
1477
1478         list_del_init(&cmd->queue_entry);
1479         se_cmd = cmd->se_cmd;
1480         tcmu_free_cmd(cmd);
1481
1482         se_cmd->priv = NULL;
1483         target_complete_cmd(se_cmd, SAM_STAT_TASK_SET_FULL);
1484 }
1485
1486 static void tcmu_device_timedout(struct tcmu_dev *udev)
1487 {
1488         spin_lock(&timed_out_udevs_lock);
1489         if (list_empty(&udev->timedout_entry))
1490                 list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1491         spin_unlock(&timed_out_udevs_lock);
1492
1493         schedule_delayed_work(&tcmu_unmap_work, 0);
1494 }
1495
1496 static void tcmu_cmd_timedout(struct timer_list *t)
1497 {
1498         struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1499
1500         pr_debug("%s cmd timeout has expired\n", udev->name);
1501         tcmu_device_timedout(udev);
1502 }
1503
1504 static void tcmu_qfull_timedout(struct timer_list *t)
1505 {
1506         struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1507
1508         pr_debug("%s qfull timeout has expired\n", udev->name);
1509         tcmu_device_timedout(udev);
1510 }
1511
1512 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1513 {
1514         struct tcmu_hba *tcmu_hba;
1515
1516         tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1517         if (!tcmu_hba)
1518                 return -ENOMEM;
1519
1520         tcmu_hba->host_id = host_id;
1521         hba->hba_ptr = tcmu_hba;
1522
1523         return 0;
1524 }
1525
1526 static void tcmu_detach_hba(struct se_hba *hba)
1527 {
1528         kfree(hba->hba_ptr);
1529         hba->hba_ptr = NULL;
1530 }
1531
1532 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1533 {
1534         struct tcmu_dev *udev;
1535
1536         udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1537         if (!udev)
1538                 return NULL;
1539         kref_init(&udev->kref);
1540
1541         udev->name = kstrdup(name, GFP_KERNEL);
1542         if (!udev->name) {
1543                 kfree(udev);
1544                 return NULL;
1545         }
1546
1547         udev->hba = hba;
1548         udev->cmd_time_out = TCMU_TIME_OUT;
1549         udev->qfull_time_out = -1;
1550
1551         udev->max_blocks = DATA_BLOCK_BITS_DEF;
1552         mutex_init(&udev->cmdr_lock);
1553
1554         INIT_LIST_HEAD(&udev->node);
1555         INIT_LIST_HEAD(&udev->timedout_entry);
1556         INIT_LIST_HEAD(&udev->qfull_queue);
1557         INIT_LIST_HEAD(&udev->tmr_queue);
1558         INIT_LIST_HEAD(&udev->inflight_queue);
1559         idr_init(&udev->commands);
1560
1561         timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1562         timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1563
1564         INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1565
1566         return &udev->se_dev;
1567 }
1568
1569 static void tcmu_dev_call_rcu(struct rcu_head *p)
1570 {
1571         struct se_device *dev = container_of(p, struct se_device, rcu_head);
1572         struct tcmu_dev *udev = TCMU_DEV(dev);
1573
1574         kfree(udev->uio_info.name);
1575         kfree(udev->name);
1576         kfree(udev);
1577 }
1578
1579 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1580 {
1581         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1582                 kmem_cache_free(tcmu_cmd_cache, cmd);
1583                 return 0;
1584         }
1585         return -EINVAL;
1586 }
1587
1588 static void tcmu_blocks_release(struct radix_tree_root *blocks,
1589                                 int start, int end)
1590 {
1591         int i;
1592         struct page *page;
1593
1594         for (i = start; i < end; i++) {
1595                 page = radix_tree_delete(blocks, i);
1596                 if (page) {
1597                         __free_page(page);
1598                         atomic_dec(&global_db_count);
1599                 }
1600         }
1601 }
1602
1603 static void tcmu_remove_all_queued_tmr(struct tcmu_dev *udev)
1604 {
1605         struct tcmu_tmr *tmr, *tmp;
1606
1607         list_for_each_entry_safe(tmr, tmp, &udev->tmr_queue, queue_entry) {
1608                 list_del_init(&tmr->queue_entry);
1609                 kfree(tmr);
1610         }
1611 }
1612
1613 static void tcmu_dev_kref_release(struct kref *kref)
1614 {
1615         struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1616         struct se_device *dev = &udev->se_dev;
1617         struct tcmu_cmd *cmd;
1618         bool all_expired = true;
1619         int i;
1620
1621         vfree(udev->mb_addr);
1622         udev->mb_addr = NULL;
1623
1624         spin_lock_bh(&timed_out_udevs_lock);
1625         if (!list_empty(&udev->timedout_entry))
1626                 list_del(&udev->timedout_entry);
1627         spin_unlock_bh(&timed_out_udevs_lock);
1628
1629         /* Upper layer should drain all requests before calling this */
1630         mutex_lock(&udev->cmdr_lock);
1631         idr_for_each_entry(&udev->commands, cmd, i) {
1632                 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1633                         all_expired = false;
1634         }
1635         /* There can be left over TMR cmds. Remove them. */
1636         tcmu_remove_all_queued_tmr(udev);
1637         if (!list_empty(&udev->qfull_queue))
1638                 all_expired = false;
1639         idr_destroy(&udev->commands);
1640         WARN_ON(!all_expired);
1641
1642         tcmu_blocks_release(&udev->data_blocks, 0, udev->dbi_max + 1);
1643         bitmap_free(udev->data_bitmap);
1644         mutex_unlock(&udev->cmdr_lock);
1645
1646         pr_debug("dev_kref_release\n");
1647
1648         call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1649 }
1650
1651 static void run_qfull_queue(struct tcmu_dev *udev, bool fail)
1652 {
1653         struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1654         LIST_HEAD(cmds);
1655         sense_reason_t scsi_ret;
1656         int ret;
1657
1658         if (list_empty(&udev->qfull_queue))
1659                 return;
1660
1661         pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1662
1663         list_splice_init(&udev->qfull_queue, &cmds);
1664
1665         list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, queue_entry) {
1666                 list_del_init(&tcmu_cmd->queue_entry);
1667
1668                 pr_debug("removing cmd %p on dev %s from queue\n",
1669                          tcmu_cmd, udev->name);
1670
1671                 if (fail) {
1672                         /*
1673                          * We were not able to even start the command, so
1674                          * fail with busy to allow a retry in case runner
1675                          * was only temporarily down. If the device is being
1676                          * removed then LIO core will do the right thing and
1677                          * fail the retry.
1678                          */
1679                         tcmu_cmd->se_cmd->priv = NULL;
1680                         target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1681                         tcmu_free_cmd(tcmu_cmd);
1682                         continue;
1683                 }
1684
1685                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1686                 if (ret < 0) {
1687                         pr_debug("cmd %p on dev %s failed with %u\n",
1688                                  tcmu_cmd, udev->name, scsi_ret);
1689                         /*
1690                          * Ignore scsi_ret for now. target_complete_cmd
1691                          * drops it.
1692                          */
1693                         tcmu_cmd->se_cmd->priv = NULL;
1694                         target_complete_cmd(tcmu_cmd->se_cmd,
1695                                             SAM_STAT_CHECK_CONDITION);
1696                         tcmu_free_cmd(tcmu_cmd);
1697                 } else if (ret > 0) {
1698                         pr_debug("ran out of space during cmdr queue run\n");
1699                         /*
1700                          * cmd was requeued, so just put all cmds back in
1701                          * the queue
1702                          */
1703                         list_splice_tail(&cmds, &udev->qfull_queue);
1704                         break;
1705                 }
1706         }
1707
1708         tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1709 }
1710
1711 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1712 {
1713         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1714
1715         mutex_lock(&udev->cmdr_lock);
1716         if (tcmu_handle_completions(udev))
1717                 run_qfull_queue(udev, false);
1718         mutex_unlock(&udev->cmdr_lock);
1719
1720         return 0;
1721 }
1722
1723 /*
1724  * mmap code from uio.c. Copied here because we want to hook mmap()
1725  * and this stuff must come along.
1726  */
1727 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1728 {
1729         struct tcmu_dev *udev = vma->vm_private_data;
1730         struct uio_info *info = &udev->uio_info;
1731
1732         if (vma->vm_pgoff < MAX_UIO_MAPS) {
1733                 if (info->mem[vma->vm_pgoff].size == 0)
1734                         return -1;
1735                 return (int)vma->vm_pgoff;
1736         }
1737         return -1;
1738 }
1739
1740 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1741 {
1742         struct page *page;
1743
1744         mutex_lock(&udev->cmdr_lock);
1745         page = tcmu_get_block_page(udev, dbi);
1746         if (likely(page)) {
1747                 mutex_unlock(&udev->cmdr_lock);
1748                 return page;
1749         }
1750
1751         /*
1752          * Userspace messed up and passed in a address not in the
1753          * data iov passed to it.
1754          */
1755         pr_err("Invalid addr to data block mapping  (dbi %u) on device %s\n",
1756                dbi, udev->name);
1757         page = NULL;
1758         mutex_unlock(&udev->cmdr_lock);
1759
1760         return page;
1761 }
1762
1763 static void tcmu_vma_open(struct vm_area_struct *vma)
1764 {
1765         struct tcmu_dev *udev = vma->vm_private_data;
1766
1767         pr_debug("vma_open\n");
1768
1769         kref_get(&udev->kref);
1770 }
1771
1772 static void tcmu_vma_close(struct vm_area_struct *vma)
1773 {
1774         struct tcmu_dev *udev = vma->vm_private_data;
1775
1776         pr_debug("vma_close\n");
1777
1778         /* release ref from tcmu_vma_open */
1779         kref_put(&udev->kref, tcmu_dev_kref_release);
1780 }
1781
1782 static vm_fault_t tcmu_vma_fault(struct vm_fault *vmf)
1783 {
1784         struct tcmu_dev *udev = vmf->vma->vm_private_data;
1785         struct uio_info *info = &udev->uio_info;
1786         struct page *page;
1787         unsigned long offset;
1788         void *addr;
1789
1790         int mi = tcmu_find_mem_index(vmf->vma);
1791         if (mi < 0)
1792                 return VM_FAULT_SIGBUS;
1793
1794         /*
1795          * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1796          * to use mem[N].
1797          */
1798         offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1799
1800         if (offset < udev->data_off) {
1801                 /* For the vmalloc()ed cmd area pages */
1802                 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1803                 page = vmalloc_to_page(addr);
1804         } else {
1805                 uint32_t dbi;
1806
1807                 /* For the dynamically growing data area pages */
1808                 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1809                 page = tcmu_try_get_block_page(udev, dbi);
1810                 if (!page)
1811                         return VM_FAULT_SIGBUS;
1812         }
1813
1814         get_page(page);
1815         vmf->page = page;
1816         return 0;
1817 }
1818
1819 static const struct vm_operations_struct tcmu_vm_ops = {
1820         .open = tcmu_vma_open,
1821         .close = tcmu_vma_close,
1822         .fault = tcmu_vma_fault,
1823 };
1824
1825 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1826 {
1827         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1828
1829         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1830         vma->vm_ops = &tcmu_vm_ops;
1831
1832         vma->vm_private_data = udev;
1833
1834         /* Ensure the mmap is exactly the right size */
1835         if (vma_pages(vma) != (udev->ring_size >> PAGE_SHIFT))
1836                 return -EINVAL;
1837
1838         tcmu_vma_open(vma);
1839
1840         return 0;
1841 }
1842
1843 static int tcmu_open(struct uio_info *info, struct inode *inode)
1844 {
1845         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1846
1847         /* O_EXCL not supported for char devs, so fake it? */
1848         if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1849                 return -EBUSY;
1850
1851         udev->inode = inode;
1852
1853         pr_debug("open\n");
1854
1855         return 0;
1856 }
1857
1858 static int tcmu_release(struct uio_info *info, struct inode *inode)
1859 {
1860         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1861
1862         clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1863
1864         pr_debug("close\n");
1865
1866         return 0;
1867 }
1868
1869 static int tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1870 {
1871         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1872
1873         if (!tcmu_kern_cmd_reply_supported)
1874                 return 0;
1875
1876         if (udev->nl_reply_supported <= 0)
1877                 return 0;
1878
1879         mutex_lock(&tcmu_nl_cmd_mutex);
1880
1881         if (tcmu_netlink_blocked) {
1882                 mutex_unlock(&tcmu_nl_cmd_mutex);
1883                 pr_warn("Failing nl cmd %d on %s. Interface is blocked.\n", cmd,
1884                         udev->name);
1885                 return -EAGAIN;
1886         }
1887
1888         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1889                 mutex_unlock(&tcmu_nl_cmd_mutex);
1890                 pr_warn("netlink cmd %d already executing on %s\n",
1891                          nl_cmd->cmd, udev->name);
1892                 return -EBUSY;
1893         }
1894
1895         memset(nl_cmd, 0, sizeof(*nl_cmd));
1896         nl_cmd->cmd = cmd;
1897         nl_cmd->udev = udev;
1898         init_completion(&nl_cmd->complete);
1899         INIT_LIST_HEAD(&nl_cmd->nl_list);
1900
1901         list_add_tail(&nl_cmd->nl_list, &tcmu_nl_cmd_list);
1902
1903         mutex_unlock(&tcmu_nl_cmd_mutex);
1904         return 0;
1905 }
1906
1907 static void tcmu_destroy_genl_cmd_reply(struct tcmu_dev *udev)
1908 {
1909         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1910
1911         if (!tcmu_kern_cmd_reply_supported)
1912                 return;
1913
1914         if (udev->nl_reply_supported <= 0)
1915                 return;
1916
1917         mutex_lock(&tcmu_nl_cmd_mutex);
1918
1919         list_del(&nl_cmd->nl_list);
1920         memset(nl_cmd, 0, sizeof(*nl_cmd));
1921
1922         mutex_unlock(&tcmu_nl_cmd_mutex);
1923 }
1924
1925 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1926 {
1927         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1928         int ret;
1929
1930         if (!tcmu_kern_cmd_reply_supported)
1931                 return 0;
1932
1933         if (udev->nl_reply_supported <= 0)
1934                 return 0;
1935
1936         pr_debug("sleeping for nl reply\n");
1937         wait_for_completion(&nl_cmd->complete);
1938
1939         mutex_lock(&tcmu_nl_cmd_mutex);
1940         nl_cmd->cmd = TCMU_CMD_UNSPEC;
1941         ret = nl_cmd->status;
1942         mutex_unlock(&tcmu_nl_cmd_mutex);
1943
1944         return ret;
1945 }
1946
1947 static int tcmu_netlink_event_init(struct tcmu_dev *udev,
1948                                    enum tcmu_genl_cmd cmd,
1949                                    struct sk_buff **buf, void **hdr)
1950 {
1951         struct sk_buff *skb;
1952         void *msg_header;
1953         int ret = -ENOMEM;
1954
1955         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1956         if (!skb)
1957                 return ret;
1958
1959         msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1960         if (!msg_header)
1961                 goto free_skb;
1962
1963         ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1964         if (ret < 0)
1965                 goto free_skb;
1966
1967         ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1968         if (ret < 0)
1969                 goto free_skb;
1970
1971         ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1972         if (ret < 0)
1973                 goto free_skb;
1974
1975         *buf = skb;
1976         *hdr = msg_header;
1977         return ret;
1978
1979 free_skb:
1980         nlmsg_free(skb);
1981         return ret;
1982 }
1983
1984 static int tcmu_netlink_event_send(struct tcmu_dev *udev,
1985                                    enum tcmu_genl_cmd cmd,
1986                                    struct sk_buff *skb, void *msg_header)
1987 {
1988         int ret;
1989
1990         genlmsg_end(skb, msg_header);
1991
1992         ret = tcmu_init_genl_cmd_reply(udev, cmd);
1993         if (ret) {
1994                 nlmsg_free(skb);
1995                 return ret;
1996         }
1997
1998         ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1999                                       TCMU_MCGRP_CONFIG, GFP_KERNEL);
2000
2001         /* Wait during an add as the listener may not be up yet */
2002         if (ret == 0 ||
2003            (ret == -ESRCH && cmd == TCMU_CMD_ADDED_DEVICE))
2004                 return tcmu_wait_genl_cmd_reply(udev);
2005         else
2006                 tcmu_destroy_genl_cmd_reply(udev);
2007
2008         return ret;
2009 }
2010
2011 static int tcmu_send_dev_add_event(struct tcmu_dev *udev)
2012 {
2013         struct sk_buff *skb = NULL;
2014         void *msg_header = NULL;
2015         int ret = 0;
2016
2017         ret = tcmu_netlink_event_init(udev, TCMU_CMD_ADDED_DEVICE, &skb,
2018                                       &msg_header);
2019         if (ret < 0)
2020                 return ret;
2021         return tcmu_netlink_event_send(udev, TCMU_CMD_ADDED_DEVICE, skb,
2022                                        msg_header);
2023 }
2024
2025 static int tcmu_send_dev_remove_event(struct tcmu_dev *udev)
2026 {
2027         struct sk_buff *skb = NULL;
2028         void *msg_header = NULL;
2029         int ret = 0;
2030
2031         ret = tcmu_netlink_event_init(udev, TCMU_CMD_REMOVED_DEVICE,
2032                                       &skb, &msg_header);
2033         if (ret < 0)
2034                 return ret;
2035         return tcmu_netlink_event_send(udev, TCMU_CMD_REMOVED_DEVICE,
2036                                        skb, msg_header);
2037 }
2038
2039 static int tcmu_update_uio_info(struct tcmu_dev *udev)
2040 {
2041         struct tcmu_hba *hba = udev->hba->hba_ptr;
2042         struct uio_info *info;
2043         char *str;
2044
2045         info = &udev->uio_info;
2046
2047         if (udev->dev_config[0])
2048                 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s/%s", hba->host_id,
2049                                 udev->name, udev->dev_config);
2050         else
2051                 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s", hba->host_id,
2052                                 udev->name);
2053         if (!str)
2054                 return -ENOMEM;
2055
2056         /* If the old string exists, free it */
2057         kfree(info->name);
2058         info->name = str;
2059
2060         return 0;
2061 }
2062
2063 static int tcmu_configure_device(struct se_device *dev)
2064 {
2065         struct tcmu_dev *udev = TCMU_DEV(dev);
2066         struct uio_info *info;
2067         struct tcmu_mailbox *mb;
2068         int ret = 0;
2069
2070         ret = tcmu_update_uio_info(udev);
2071         if (ret)
2072                 return ret;
2073
2074         info = &udev->uio_info;
2075
2076         mutex_lock(&udev->cmdr_lock);
2077         udev->data_bitmap = bitmap_zalloc(udev->max_blocks, GFP_KERNEL);
2078         mutex_unlock(&udev->cmdr_lock);
2079         if (!udev->data_bitmap) {
2080                 ret = -ENOMEM;
2081                 goto err_bitmap_alloc;
2082         }
2083
2084         udev->mb_addr = vzalloc(CMDR_SIZE);
2085         if (!udev->mb_addr) {
2086                 ret = -ENOMEM;
2087                 goto err_vzalloc;
2088         }
2089
2090         /* mailbox fits in first part of CMDR space */
2091         udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
2092         udev->data_off = CMDR_SIZE;
2093         udev->data_size = udev->max_blocks * DATA_BLOCK_SIZE;
2094         udev->dbi_thresh = 0; /* Default in Idle state */
2095
2096         /* Initialise the mailbox of the ring buffer */
2097         mb = udev->mb_addr;
2098         mb->version = TCMU_MAILBOX_VERSION;
2099         mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC |
2100                     TCMU_MAILBOX_FLAG_CAP_READ_LEN |
2101                     TCMU_MAILBOX_FLAG_CAP_TMR;
2102         mb->cmdr_off = CMDR_OFF;
2103         mb->cmdr_size = udev->cmdr_size;
2104
2105         WARN_ON(!PAGE_ALIGNED(udev->data_off));
2106         WARN_ON(udev->data_size % PAGE_SIZE);
2107         WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
2108
2109         info->version = __stringify(TCMU_MAILBOX_VERSION);
2110
2111         info->mem[0].name = "tcm-user command & data buffer";
2112         info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
2113         info->mem[0].size = udev->ring_size = udev->data_size + CMDR_SIZE;
2114         info->mem[0].memtype = UIO_MEM_NONE;
2115
2116         info->irqcontrol = tcmu_irqcontrol;
2117         info->irq = UIO_IRQ_CUSTOM;
2118
2119         info->mmap = tcmu_mmap;
2120         info->open = tcmu_open;
2121         info->release = tcmu_release;
2122
2123         ret = uio_register_device(tcmu_root_device, info);
2124         if (ret)
2125                 goto err_register;
2126
2127         /* User can set hw_block_size before enable the device */
2128         if (dev->dev_attrib.hw_block_size == 0)
2129                 dev->dev_attrib.hw_block_size = 512;
2130         /* Other attributes can be configured in userspace */
2131         if (!dev->dev_attrib.hw_max_sectors)
2132                 dev->dev_attrib.hw_max_sectors = 128;
2133         if (!dev->dev_attrib.emulate_write_cache)
2134                 dev->dev_attrib.emulate_write_cache = 0;
2135         dev->dev_attrib.hw_queue_depth = 128;
2136
2137         /* If user didn't explicitly disable netlink reply support, use
2138          * module scope setting.
2139          */
2140         if (udev->nl_reply_supported >= 0)
2141                 udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
2142
2143         /*
2144          * Get a ref incase userspace does a close on the uio device before
2145          * LIO has initiated tcmu_free_device.
2146          */
2147         kref_get(&udev->kref);
2148
2149         ret = tcmu_send_dev_add_event(udev);
2150         if (ret)
2151                 goto err_netlink;
2152
2153         mutex_lock(&root_udev_mutex);
2154         list_add(&udev->node, &root_udev);
2155         mutex_unlock(&root_udev_mutex);
2156
2157         return 0;
2158
2159 err_netlink:
2160         kref_put(&udev->kref, tcmu_dev_kref_release);
2161         uio_unregister_device(&udev->uio_info);
2162 err_register:
2163         vfree(udev->mb_addr);
2164         udev->mb_addr = NULL;
2165 err_vzalloc:
2166         bitmap_free(udev->data_bitmap);
2167         udev->data_bitmap = NULL;
2168 err_bitmap_alloc:
2169         kfree(info->name);
2170         info->name = NULL;
2171
2172         return ret;
2173 }
2174
2175 static void tcmu_free_device(struct se_device *dev)
2176 {
2177         struct tcmu_dev *udev = TCMU_DEV(dev);
2178
2179         /* release ref from init */
2180         kref_put(&udev->kref, tcmu_dev_kref_release);
2181 }
2182
2183 static void tcmu_destroy_device(struct se_device *dev)
2184 {
2185         struct tcmu_dev *udev = TCMU_DEV(dev);
2186
2187         del_timer_sync(&udev->cmd_timer);
2188         del_timer_sync(&udev->qfull_timer);
2189
2190         mutex_lock(&root_udev_mutex);
2191         list_del(&udev->node);
2192         mutex_unlock(&root_udev_mutex);
2193
2194         tcmu_send_dev_remove_event(udev);
2195
2196         uio_unregister_device(&udev->uio_info);
2197
2198         /* release ref from configure */
2199         kref_put(&udev->kref, tcmu_dev_kref_release);
2200 }
2201
2202 static void tcmu_unblock_dev(struct tcmu_dev *udev)
2203 {
2204         mutex_lock(&udev->cmdr_lock);
2205         clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
2206         mutex_unlock(&udev->cmdr_lock);
2207 }
2208
2209 static void tcmu_block_dev(struct tcmu_dev *udev)
2210 {
2211         mutex_lock(&udev->cmdr_lock);
2212
2213         if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2214                 goto unlock;
2215
2216         /* complete IO that has executed successfully */
2217         tcmu_handle_completions(udev);
2218         /* fail IO waiting to be queued */
2219         run_qfull_queue(udev, true);
2220
2221 unlock:
2222         mutex_unlock(&udev->cmdr_lock);
2223 }
2224
2225 static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
2226 {
2227         struct tcmu_mailbox *mb;
2228         struct tcmu_cmd *cmd;
2229         int i;
2230
2231         mutex_lock(&udev->cmdr_lock);
2232
2233         idr_for_each_entry(&udev->commands, cmd, i) {
2234                 pr_debug("removing cmd %u on dev %s from ring (is expired %d)\n",
2235                           cmd->cmd_id, udev->name,
2236                           test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags));
2237
2238                 idr_remove(&udev->commands, i);
2239                 if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
2240                         WARN_ON(!cmd->se_cmd);
2241                         list_del_init(&cmd->queue_entry);
2242                         cmd->se_cmd->priv = NULL;
2243                         if (err_level == 1) {
2244                                 /*
2245                                  * Userspace was not able to start the
2246                                  * command or it is retryable.
2247                                  */
2248                                 target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
2249                         } else {
2250                                 /* hard failure */
2251                                 target_complete_cmd(cmd->se_cmd,
2252                                                     SAM_STAT_CHECK_CONDITION);
2253                         }
2254                 }
2255                 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
2256                 tcmu_free_cmd(cmd);
2257         }
2258
2259         mb = udev->mb_addr;
2260         tcmu_flush_dcache_range(mb, sizeof(*mb));
2261         pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
2262                  mb->cmd_tail, mb->cmd_head);
2263
2264         udev->cmdr_last_cleaned = 0;
2265         mb->cmd_tail = 0;
2266         mb->cmd_head = 0;
2267         tcmu_flush_dcache_range(mb, sizeof(*mb));
2268         clear_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
2269
2270         del_timer(&udev->cmd_timer);
2271
2272         /*
2273          * ring is empty and qfull queue never contains aborted commands.
2274          * So TMRs in tmr queue do not contain relevant cmd_ids.
2275          * After a ring reset userspace should do a fresh start, so
2276          * even LUN RESET message is no longer relevant.
2277          * Therefore remove all TMRs from qfull queue
2278          */
2279         tcmu_remove_all_queued_tmr(udev);
2280
2281         run_qfull_queue(udev, false);
2282
2283         mutex_unlock(&udev->cmdr_lock);
2284 }
2285
2286 enum {
2287         Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
2288         Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_err,
2289 };
2290
2291 static match_table_t tokens = {
2292         {Opt_dev_config, "dev_config=%s"},
2293         {Opt_dev_size, "dev_size=%s"},
2294         {Opt_hw_block_size, "hw_block_size=%d"},
2295         {Opt_hw_max_sectors, "hw_max_sectors=%d"},
2296         {Opt_nl_reply_supported, "nl_reply_supported=%d"},
2297         {Opt_max_data_area_mb, "max_data_area_mb=%d"},
2298         {Opt_err, NULL}
2299 };
2300
2301 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
2302 {
2303         int val, ret;
2304
2305         ret = match_int(arg, &val);
2306         if (ret < 0) {
2307                 pr_err("match_int() failed for dev attrib. Error %d.\n",
2308                        ret);
2309                 return ret;
2310         }
2311
2312         if (val <= 0) {
2313                 pr_err("Invalid dev attrib value %d. Must be greater than zero.\n",
2314                        val);
2315                 return -EINVAL;
2316         }
2317         *dev_attrib = val;
2318         return 0;
2319 }
2320
2321 static int tcmu_set_max_blocks_param(struct tcmu_dev *udev, substring_t *arg)
2322 {
2323         int val, ret;
2324
2325         ret = match_int(arg, &val);
2326         if (ret < 0) {
2327                 pr_err("match_int() failed for max_data_area_mb=. Error %d.\n",
2328                        ret);
2329                 return ret;
2330         }
2331
2332         if (val <= 0) {
2333                 pr_err("Invalid max_data_area %d.\n", val);
2334                 return -EINVAL;
2335         }
2336
2337         mutex_lock(&udev->cmdr_lock);
2338         if (udev->data_bitmap) {
2339                 pr_err("Cannot set max_data_area_mb after it has been enabled.\n");
2340                 ret = -EINVAL;
2341                 goto unlock;
2342         }
2343
2344         udev->max_blocks = TCMU_MBS_TO_BLOCKS(val);
2345         if (udev->max_blocks > tcmu_global_max_blocks) {
2346                 pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2347                        val, TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
2348                 udev->max_blocks = tcmu_global_max_blocks;
2349         }
2350
2351 unlock:
2352         mutex_unlock(&udev->cmdr_lock);
2353         return ret;
2354 }
2355
2356 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
2357                 const char *page, ssize_t count)
2358 {
2359         struct tcmu_dev *udev = TCMU_DEV(dev);
2360         char *orig, *ptr, *opts;
2361         substring_t args[MAX_OPT_ARGS];
2362         int ret = 0, token;
2363
2364         opts = kstrdup(page, GFP_KERNEL);
2365         if (!opts)
2366                 return -ENOMEM;
2367
2368         orig = opts;
2369
2370         while ((ptr = strsep(&opts, ",\n")) != NULL) {
2371                 if (!*ptr)
2372                         continue;
2373
2374                 token = match_token(ptr, tokens, args);
2375                 switch (token) {
2376                 case Opt_dev_config:
2377                         if (match_strlcpy(udev->dev_config, &args[0],
2378                                           TCMU_CONFIG_LEN) == 0) {
2379                                 ret = -EINVAL;
2380                                 break;
2381                         }
2382                         pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
2383                         break;
2384                 case Opt_dev_size:
2385                         ret = match_u64(&args[0], &udev->dev_size);
2386                         if (ret < 0)
2387                                 pr_err("match_u64() failed for dev_size=. Error %d.\n",
2388                                        ret);
2389                         break;
2390                 case Opt_hw_block_size:
2391                         ret = tcmu_set_dev_attrib(&args[0],
2392                                         &(dev->dev_attrib.hw_block_size));
2393                         break;
2394                 case Opt_hw_max_sectors:
2395                         ret = tcmu_set_dev_attrib(&args[0],
2396                                         &(dev->dev_attrib.hw_max_sectors));
2397                         break;
2398                 case Opt_nl_reply_supported:
2399                         ret = match_int(&args[0], &udev->nl_reply_supported);
2400                         if (ret < 0)
2401                                 pr_err("match_int() failed for nl_reply_supported=. Error %d.\n",
2402                                        ret);
2403                         break;
2404                 case Opt_max_data_area_mb:
2405                         ret = tcmu_set_max_blocks_param(udev, &args[0]);
2406                         break;
2407                 default:
2408                         break;
2409                 }
2410
2411                 if (ret)
2412                         break;
2413         }
2414
2415         kfree(orig);
2416         return (!ret) ? count : ret;
2417 }
2418
2419 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2420 {
2421         struct tcmu_dev *udev = TCMU_DEV(dev);
2422         ssize_t bl = 0;
2423
2424         bl = sprintf(b + bl, "Config: %s ",
2425                      udev->dev_config[0] ? udev->dev_config : "NULL");
2426         bl += sprintf(b + bl, "Size: %llu ", udev->dev_size);
2427         bl += sprintf(b + bl, "MaxDataAreaMB: %u\n",
2428                       TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2429
2430         return bl;
2431 }
2432
2433 static sector_t tcmu_get_blocks(struct se_device *dev)
2434 {
2435         struct tcmu_dev *udev = TCMU_DEV(dev);
2436
2437         return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2438                        dev->dev_attrib.block_size);
2439 }
2440
2441 static sense_reason_t
2442 tcmu_parse_cdb(struct se_cmd *cmd)
2443 {
2444         return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2445 }
2446
2447 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2448 {
2449         struct se_dev_attrib *da = container_of(to_config_group(item),
2450                                         struct se_dev_attrib, da_group);
2451         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2452
2453         return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2454 }
2455
2456 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2457                                        size_t count)
2458 {
2459         struct se_dev_attrib *da = container_of(to_config_group(item),
2460                                         struct se_dev_attrib, da_group);
2461         struct tcmu_dev *udev = container_of(da->da_dev,
2462                                         struct tcmu_dev, se_dev);
2463         u32 val;
2464         int ret;
2465
2466         if (da->da_dev->export_count) {
2467                 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2468                 return -EINVAL;
2469         }
2470
2471         ret = kstrtou32(page, 0, &val);
2472         if (ret < 0)
2473                 return ret;
2474
2475         udev->cmd_time_out = val * MSEC_PER_SEC;
2476         return count;
2477 }
2478 CONFIGFS_ATTR(tcmu_, cmd_time_out);
2479
2480 static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2481 {
2482         struct se_dev_attrib *da = container_of(to_config_group(item),
2483                                                 struct se_dev_attrib, da_group);
2484         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2485
2486         return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2487                         udev->qfull_time_out :
2488                         udev->qfull_time_out / MSEC_PER_SEC);
2489 }
2490
2491 static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2492                                          const char *page, size_t count)
2493 {
2494         struct se_dev_attrib *da = container_of(to_config_group(item),
2495                                         struct se_dev_attrib, da_group);
2496         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2497         s32 val;
2498         int ret;
2499
2500         ret = kstrtos32(page, 0, &val);
2501         if (ret < 0)
2502                 return ret;
2503
2504         if (val >= 0) {
2505                 udev->qfull_time_out = val * MSEC_PER_SEC;
2506         } else if (val == -1) {
2507                 udev->qfull_time_out = val;
2508         } else {
2509                 printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2510                 return -EINVAL;
2511         }
2512         return count;
2513 }
2514 CONFIGFS_ATTR(tcmu_, qfull_time_out);
2515
2516 static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2517 {
2518         struct se_dev_attrib *da = container_of(to_config_group(item),
2519                                                 struct se_dev_attrib, da_group);
2520         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2521
2522         return snprintf(page, PAGE_SIZE, "%u\n",
2523                         TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2524 }
2525 CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2526
2527 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2528 {
2529         struct se_dev_attrib *da = container_of(to_config_group(item),
2530                                                 struct se_dev_attrib, da_group);
2531         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2532
2533         return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2534 }
2535
2536 static int tcmu_send_dev_config_event(struct tcmu_dev *udev,
2537                                       const char *reconfig_data)
2538 {
2539         struct sk_buff *skb = NULL;
2540         void *msg_header = NULL;
2541         int ret = 0;
2542
2543         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2544                                       &skb, &msg_header);
2545         if (ret < 0)
2546                 return ret;
2547         ret = nla_put_string(skb, TCMU_ATTR_DEV_CFG, reconfig_data);
2548         if (ret < 0) {
2549                 nlmsg_free(skb);
2550                 return ret;
2551         }
2552         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2553                                        skb, msg_header);
2554 }
2555
2556
2557 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2558                                      size_t count)
2559 {
2560         struct se_dev_attrib *da = container_of(to_config_group(item),
2561                                                 struct se_dev_attrib, da_group);
2562         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2563         int ret, len;
2564
2565         len = strlen(page);
2566         if (!len || len > TCMU_CONFIG_LEN - 1)
2567                 return -EINVAL;
2568
2569         /* Check if device has been configured before */
2570         if (target_dev_configured(&udev->se_dev)) {
2571                 ret = tcmu_send_dev_config_event(udev, page);
2572                 if (ret) {
2573                         pr_err("Unable to reconfigure device\n");
2574                         return ret;
2575                 }
2576                 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2577
2578                 ret = tcmu_update_uio_info(udev);
2579                 if (ret)
2580                         return ret;
2581                 return count;
2582         }
2583         strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2584
2585         return count;
2586 }
2587 CONFIGFS_ATTR(tcmu_, dev_config);
2588
2589 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2590 {
2591         struct se_dev_attrib *da = container_of(to_config_group(item),
2592                                                 struct se_dev_attrib, da_group);
2593         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2594
2595         return snprintf(page, PAGE_SIZE, "%llu\n", udev->dev_size);
2596 }
2597
2598 static int tcmu_send_dev_size_event(struct tcmu_dev *udev, u64 size)
2599 {
2600         struct sk_buff *skb = NULL;
2601         void *msg_header = NULL;
2602         int ret = 0;
2603
2604         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2605                                       &skb, &msg_header);
2606         if (ret < 0)
2607                 return ret;
2608         ret = nla_put_u64_64bit(skb, TCMU_ATTR_DEV_SIZE,
2609                                 size, TCMU_ATTR_PAD);
2610         if (ret < 0) {
2611                 nlmsg_free(skb);
2612                 return ret;
2613         }
2614         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2615                                        skb, msg_header);
2616 }
2617
2618 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2619                                    size_t count)
2620 {
2621         struct se_dev_attrib *da = container_of(to_config_group(item),
2622                                                 struct se_dev_attrib, da_group);
2623         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2624         u64 val;
2625         int ret;
2626
2627         ret = kstrtou64(page, 0, &val);
2628         if (ret < 0)
2629                 return ret;
2630
2631         /* Check if device has been configured before */
2632         if (target_dev_configured(&udev->se_dev)) {
2633                 ret = tcmu_send_dev_size_event(udev, val);
2634                 if (ret) {
2635                         pr_err("Unable to reconfigure device\n");
2636                         return ret;
2637                 }
2638         }
2639         udev->dev_size = val;
2640         return count;
2641 }
2642 CONFIGFS_ATTR(tcmu_, dev_size);
2643
2644 static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
2645                 char *page)
2646 {
2647         struct se_dev_attrib *da = container_of(to_config_group(item),
2648                                                 struct se_dev_attrib, da_group);
2649         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2650
2651         return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2652 }
2653
2654 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2655                 const char *page, size_t count)
2656 {
2657         struct se_dev_attrib *da = container_of(to_config_group(item),
2658                                                 struct se_dev_attrib, da_group);
2659         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2660         s8 val;
2661         int ret;
2662
2663         ret = kstrtos8(page, 0, &val);
2664         if (ret < 0)
2665                 return ret;
2666
2667         udev->nl_reply_supported = val;
2668         return count;
2669 }
2670 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2671
2672 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2673                                              char *page)
2674 {
2675         struct se_dev_attrib *da = container_of(to_config_group(item),
2676                                         struct se_dev_attrib, da_group);
2677
2678         return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2679 }
2680
2681 static int tcmu_send_emulate_write_cache(struct tcmu_dev *udev, u8 val)
2682 {
2683         struct sk_buff *skb = NULL;
2684         void *msg_header = NULL;
2685         int ret = 0;
2686
2687         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2688                                       &skb, &msg_header);
2689         if (ret < 0)
2690                 return ret;
2691         ret = nla_put_u8(skb, TCMU_ATTR_WRITECACHE, val);
2692         if (ret < 0) {
2693                 nlmsg_free(skb);
2694                 return ret;
2695         }
2696         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2697                                        skb, msg_header);
2698 }
2699
2700 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2701                                               const char *page, size_t count)
2702 {
2703         struct se_dev_attrib *da = container_of(to_config_group(item),
2704                                         struct se_dev_attrib, da_group);
2705         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2706         u8 val;
2707         int ret;
2708
2709         ret = kstrtou8(page, 0, &val);
2710         if (ret < 0)
2711                 return ret;
2712
2713         /* Check if device has been configured before */
2714         if (target_dev_configured(&udev->se_dev)) {
2715                 ret = tcmu_send_emulate_write_cache(udev, val);
2716                 if (ret) {
2717                         pr_err("Unable to reconfigure device\n");
2718                         return ret;
2719                 }
2720         }
2721
2722         da->emulate_write_cache = val;
2723         return count;
2724 }
2725 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2726
2727 static ssize_t tcmu_tmr_notification_show(struct config_item *item, char *page)
2728 {
2729         struct se_dev_attrib *da = container_of(to_config_group(item),
2730                                         struct se_dev_attrib, da_group);
2731         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2732
2733         return snprintf(page, PAGE_SIZE, "%i\n",
2734                         test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags));
2735 }
2736
2737 static ssize_t tcmu_tmr_notification_store(struct config_item *item,
2738                                            const char *page, size_t count)
2739 {
2740         struct se_dev_attrib *da = container_of(to_config_group(item),
2741                                         struct se_dev_attrib, da_group);
2742         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2743         u8 val;
2744         int ret;
2745
2746         ret = kstrtou8(page, 0, &val);
2747         if (ret < 0)
2748                 return ret;
2749         if (val > 1)
2750                 return -EINVAL;
2751
2752         if (val)
2753                 set_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
2754         else
2755                 clear_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
2756         return count;
2757 }
2758 CONFIGFS_ATTR(tcmu_, tmr_notification);
2759
2760 static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
2761 {
2762         struct se_device *se_dev = container_of(to_config_group(item),
2763                                                 struct se_device,
2764                                                 dev_action_group);
2765         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2766
2767         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2768                 return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
2769         else
2770                 return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
2771 }
2772
2773 static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
2774                                     size_t count)
2775 {
2776         struct se_device *se_dev = container_of(to_config_group(item),
2777                                                 struct se_device,
2778                                                 dev_action_group);
2779         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2780         u8 val;
2781         int ret;
2782
2783         if (!target_dev_configured(&udev->se_dev)) {
2784                 pr_err("Device is not configured.\n");
2785                 return -EINVAL;
2786         }
2787
2788         ret = kstrtou8(page, 0, &val);
2789         if (ret < 0)
2790                 return ret;
2791
2792         if (val > 1) {
2793                 pr_err("Invalid block value %d\n", val);
2794                 return -EINVAL;
2795         }
2796
2797         if (!val)
2798                 tcmu_unblock_dev(udev);
2799         else
2800                 tcmu_block_dev(udev);
2801         return count;
2802 }
2803 CONFIGFS_ATTR(tcmu_, block_dev);
2804
2805 static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
2806                                      size_t count)
2807 {
2808         struct se_device *se_dev = container_of(to_config_group(item),
2809                                                 struct se_device,
2810                                                 dev_action_group);
2811         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2812         u8 val;
2813         int ret;
2814
2815         if (!target_dev_configured(&udev->se_dev)) {
2816                 pr_err("Device is not configured.\n");
2817                 return -EINVAL;
2818         }
2819
2820         ret = kstrtou8(page, 0, &val);
2821         if (ret < 0)
2822                 return ret;
2823
2824         if (val != 1 && val != 2) {
2825                 pr_err("Invalid reset ring value %d\n", val);
2826                 return -EINVAL;
2827         }
2828
2829         tcmu_reset_ring(udev, val);
2830         return count;
2831 }
2832 CONFIGFS_ATTR_WO(tcmu_, reset_ring);
2833
2834 static struct configfs_attribute *tcmu_attrib_attrs[] = {
2835         &tcmu_attr_cmd_time_out,
2836         &tcmu_attr_qfull_time_out,
2837         &tcmu_attr_max_data_area_mb,
2838         &tcmu_attr_dev_config,
2839         &tcmu_attr_dev_size,
2840         &tcmu_attr_emulate_write_cache,
2841         &tcmu_attr_tmr_notification,
2842         &tcmu_attr_nl_reply_supported,
2843         NULL,
2844 };
2845
2846 static struct configfs_attribute **tcmu_attrs;
2847
2848 static struct configfs_attribute *tcmu_action_attrs[] = {
2849         &tcmu_attr_block_dev,
2850         &tcmu_attr_reset_ring,
2851         NULL,
2852 };
2853
2854 static struct target_backend_ops tcmu_ops = {
2855         .name                   = "user",
2856         .owner                  = THIS_MODULE,
2857         .transport_flags_default = TRANSPORT_FLAG_PASSTHROUGH,
2858         .transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR |
2859                                       TRANSPORT_FLAG_PASSTHROUGH_ALUA,
2860         .attach_hba             = tcmu_attach_hba,
2861         .detach_hba             = tcmu_detach_hba,
2862         .alloc_device           = tcmu_alloc_device,
2863         .configure_device       = tcmu_configure_device,
2864         .destroy_device         = tcmu_destroy_device,
2865         .free_device            = tcmu_free_device,
2866         .parse_cdb              = tcmu_parse_cdb,
2867         .tmr_notify             = tcmu_tmr_notify,
2868         .set_configfs_dev_params = tcmu_set_configfs_dev_params,
2869         .show_configfs_dev_params = tcmu_show_configfs_dev_params,
2870         .get_device_type        = sbc_get_device_type,
2871         .get_blocks             = tcmu_get_blocks,
2872         .tb_dev_action_attrs    = tcmu_action_attrs,
2873 };
2874
2875 static void find_free_blocks(void)
2876 {
2877         struct tcmu_dev *udev;
2878         loff_t off;
2879         u32 start, end, block, total_freed = 0;
2880
2881         if (atomic_read(&global_db_count) <= tcmu_global_max_blocks)
2882                 return;
2883
2884         mutex_lock(&root_udev_mutex);
2885         list_for_each_entry(udev, &root_udev, node) {
2886                 mutex_lock(&udev->cmdr_lock);
2887
2888                 if (!target_dev_configured(&udev->se_dev)) {
2889                         mutex_unlock(&udev->cmdr_lock);
2890                         continue;
2891                 }
2892
2893                 /* Try to complete the finished commands first */
2894                 if (tcmu_handle_completions(udev))
2895                         run_qfull_queue(udev, false);
2896
2897                 /* Skip the udevs in idle */
2898                 if (!udev->dbi_thresh) {
2899                         mutex_unlock(&udev->cmdr_lock);
2900                         continue;
2901                 }
2902
2903                 end = udev->dbi_max + 1;
2904                 block = find_last_bit(udev->data_bitmap, end);
2905                 if (block == udev->dbi_max) {
2906                         /*
2907                          * The last bit is dbi_max, so it is not possible
2908                          * reclaim any blocks.
2909                          */
2910                         mutex_unlock(&udev->cmdr_lock);
2911                         continue;
2912                 } else if (block == end) {
2913                         /* The current udev will goto idle state */
2914                         udev->dbi_thresh = start = 0;
2915                         udev->dbi_max = 0;
2916                 } else {
2917                         udev->dbi_thresh = start = block + 1;
2918                         udev->dbi_max = block;
2919                 }
2920
2921                 /* Here will truncate the data area from off */
2922                 off = udev->data_off + start * DATA_BLOCK_SIZE;
2923                 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
2924
2925                 /* Release the block pages */
2926                 tcmu_blocks_release(&udev->data_blocks, start, end);
2927                 mutex_unlock(&udev->cmdr_lock);
2928
2929                 total_freed += end - start;
2930                 pr_debug("Freed %u blocks (total %u) from %s.\n", end - start,
2931                          total_freed, udev->name);
2932         }
2933         mutex_unlock(&root_udev_mutex);
2934
2935         if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
2936                 schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
2937 }
2938
2939 static void check_timedout_devices(void)
2940 {
2941         struct tcmu_dev *udev, *tmp_dev;
2942         struct tcmu_cmd *cmd, *tmp_cmd;
2943         LIST_HEAD(devs);
2944
2945         spin_lock_bh(&timed_out_udevs_lock);
2946         list_splice_init(&timed_out_udevs, &devs);
2947
2948         list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
2949                 list_del_init(&udev->timedout_entry);
2950                 spin_unlock_bh(&timed_out_udevs_lock);
2951
2952                 mutex_lock(&udev->cmdr_lock);
2953
2954                 /*
2955                  * If cmd_time_out is disabled but qfull is set deadline
2956                  * will only reflect the qfull timeout. Ignore it.
2957                  */
2958                 if (udev->cmd_time_out) {
2959                         list_for_each_entry_safe(cmd, tmp_cmd,
2960                                                  &udev->inflight_queue,
2961                                                  queue_entry) {
2962                                 tcmu_check_expired_ring_cmd(cmd);
2963                         }
2964                         tcmu_set_next_deadline(&udev->inflight_queue,
2965                                                &udev->cmd_timer);
2966                 }
2967                 list_for_each_entry_safe(cmd, tmp_cmd, &udev->qfull_queue,
2968                                          queue_entry) {
2969                         tcmu_check_expired_queue_cmd(cmd);
2970                 }
2971                 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
2972
2973                 mutex_unlock(&udev->cmdr_lock);
2974
2975                 spin_lock_bh(&timed_out_udevs_lock);
2976         }
2977
2978         spin_unlock_bh(&timed_out_udevs_lock);
2979 }
2980
2981 static void tcmu_unmap_work_fn(struct work_struct *work)
2982 {
2983         check_timedout_devices();
2984         find_free_blocks();
2985 }
2986
2987 static int __init tcmu_module_init(void)
2988 {
2989         int ret, i, k, len = 0;
2990
2991         BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2992
2993         INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
2994
2995         tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2996                                 sizeof(struct tcmu_cmd),
2997                                 __alignof__(struct tcmu_cmd),
2998                                 0, NULL);
2999         if (!tcmu_cmd_cache)
3000                 return -ENOMEM;
3001
3002         tcmu_root_device = root_device_register("tcm_user");
3003         if (IS_ERR(tcmu_root_device)) {
3004                 ret = PTR_ERR(tcmu_root_device);
3005                 goto out_free_cache;
3006         }
3007
3008         ret = genl_register_family(&tcmu_genl_family);
3009         if (ret < 0) {
3010                 goto out_unreg_device;
3011         }
3012
3013         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3014                 len += sizeof(struct configfs_attribute *);
3015         for (i = 0; passthrough_pr_attrib_attrs[i] != NULL; i++)
3016                 len += sizeof(struct configfs_attribute *);
3017         for (i = 0; tcmu_attrib_attrs[i] != NULL; i++)
3018                 len += sizeof(struct configfs_attribute *);
3019         len += sizeof(struct configfs_attribute *);
3020
3021         tcmu_attrs = kzalloc(len, GFP_KERNEL);
3022         if (!tcmu_attrs) {
3023                 ret = -ENOMEM;
3024                 goto out_unreg_genl;
3025         }
3026
3027         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3028                 tcmu_attrs[i] = passthrough_attrib_attrs[i];
3029         for (k = 0; passthrough_pr_attrib_attrs[k] != NULL; k++)
3030                 tcmu_attrs[i++] = passthrough_pr_attrib_attrs[k];
3031         for (k = 0; tcmu_attrib_attrs[k] != NULL; k++)
3032                 tcmu_attrs[i++] = tcmu_attrib_attrs[k];
3033         tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
3034
3035         ret = transport_backend_register(&tcmu_ops);
3036         if (ret)
3037                 goto out_attrs;
3038
3039         return 0;
3040
3041 out_attrs:
3042         kfree(tcmu_attrs);
3043 out_unreg_genl:
3044         genl_unregister_family(&tcmu_genl_family);
3045 out_unreg_device:
3046         root_device_unregister(tcmu_root_device);
3047 out_free_cache:
3048         kmem_cache_destroy(tcmu_cmd_cache);
3049
3050         return ret;
3051 }
3052
3053 static void __exit tcmu_module_exit(void)
3054 {
3055         cancel_delayed_work_sync(&tcmu_unmap_work);
3056         target_backend_unregister(&tcmu_ops);
3057         kfree(tcmu_attrs);
3058         genl_unregister_family(&tcmu_genl_family);
3059         root_device_unregister(tcmu_root_device);
3060         kmem_cache_destroy(tcmu_cmd_cache);
3061 }
3062
3063 MODULE_DESCRIPTION("TCM USER subsystem plugin");
3064 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
3065 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
3066 MODULE_LICENSE("GPL");
3067
3068 module_init(tcmu_module_init);
3069 module_exit(tcmu_module_exit);