drivers/target/target_core_user.c

   1 /*
   2  * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
   3  * Copyright (C) 2014 Red Hat, Inc.
   4  * Copyright (C) 2015 Arrikto, Inc.
   5  * Copyright (C) 2017 Chinamobile, Inc.
   6  *
   7  * This program is free software; you can redistribute it and/or modify it
   8  * under the terms and conditions of the GNU General Public License,
   9  * version 2, as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  14  * more details.
  15  *
  16  * You should have received a copy of the GNU General Public License along with
  17  * this program; if not, write to the Free Software Foundation, Inc.,
  18  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  19  */
  20
  21 #include <linux/spinlock.h>
  22 #include <linux/module.h>
  23 #include <linux/idr.h>
  24 #include <linux/kernel.h>
  25 #include <linux/timer.h>
  26 #include <linux/parser.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/uio_driver.h>
  29 #include <linux/radix-tree.h>
  30 #include <linux/stringify.h>
  31 #include <linux/bitops.h>
  32 #include <linux/highmem.h>
  33 #include <linux/configfs.h>
  34 #include <linux/mutex.h>
  35 #include <linux/workqueue.h>
  36 #include <net/genetlink.h>
  37 #include <scsi/scsi_common.h>
  38 #include <scsi/scsi_proto.h>
  39 #include <target/target_core_base.h>
  40 #include <target/target_core_fabric.h>
  41 #include <target/target_core_backend.h>
  42
  43 #include <linux/target_core_user.h>
  44
  45 /**
  46  * DOC: Userspace I/O
  47  * Userspace I/O
  48  * -------------
  49  *
  50  * Define a shared-memory interface for LIO to pass SCSI commands and
  51  * data to userspace for processing. This is to allow backends that
  52  * are too complex for in-kernel support to be possible.
  53  *
  54  * It uses the UIO framework to do a lot of the device-creation and
  55  * introspection work for us.
  56  *
  57  * See the .h file for how the ring is laid out. Note that while the
  58  * command ring is defined, the particulars of the data area are
  59  * not. Offset values in the command entry point to other locations
  60  * internal to the mmap-ed area. There is separate space outside the
  61  * command ring for data buffers. This leaves maximum flexibility for
  62  * moving buffer allocations, or even page flipping or other
  63  * allocation techniques, without altering the command ring layout.
  64  *
  65  * SECURITY:
  66  * The user process must be assumed to be malicious. There's no way to
  67  * prevent it breaking the command ring protocol if it wants, but in
  68  * order to prevent other issues we must only ever read *data* from
  69  * the shared memory area, not offsets or sizes. This applies to
  70  * command ring entries as well as the mailbox. Extra code needed for
  71  * this may have a 'UAM' comment.
  72  */
  73
  74 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
  75
  76 /* For cmd area, the size is fixed 8MB */
  77 #define CMDR_SIZE (8 * 1024 * 1024)
  78
  79 /*
  80  * For data area, the block size is PAGE_SIZE and
  81  * the total size is 256K * PAGE_SIZE.
  82  */
  83 #define DATA_BLOCK_SIZE PAGE_SIZE
  84 #define DATA_BLOCK_SHIFT PAGE_SHIFT
  85 #define DATA_BLOCK_BITS_DEF (256 * 1024)
  86 #define DATA_SIZE (DATA_BLOCK_BITS * DATA_BLOCK_SIZE)
  87
  88 #define TCMU_MBS_TO_BLOCKS(_mbs) (_mbs << (20 - DATA_BLOCK_SHIFT))
  89 #define TCMU_BLOCKS_TO_MBS(_blocks) (_blocks >> (20 - DATA_BLOCK_SHIFT))
  90
  91 /* The total size of the ring is 8M + 256K * PAGE_SIZE */
  92 #define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
  93
  94 /*
  95  * Default number of global data blocks(512K * PAGE_SIZE)
  96  * when the unmap thread will be started.
  97  */
  98 #define TCMU_GLOBAL_MAX_BLOCKS_DEF (512 * 1024)
  99
 100 static u8 tcmu_kern_cmd_reply_supported;
 101
 102 static struct device *tcmu_root_device;
 103
 104 struct tcmu_hba {
 105         u32 host_id;
 106 };
 107
 108 #define TCMU_CONFIG_LEN 256
 109
 110 struct tcmu_nl_cmd {
 111         /* wake up thread waiting for reply */
 112         struct completion complete;
 113         int cmd;
 114         int status;
 115 };
 116
 117 struct tcmu_dev {
 118         struct list_head node;
 119         struct kref kref;
 120
 121         struct se_device se_dev;
 122
 123         char *name;
 124         struct se_hba *hba;
 125
 126 #define TCMU_DEV_BIT_OPEN 0
 127 #define TCMU_DEV_BIT_BROKEN 1
 128 #define TCMU_DEV_BIT_BLOCKED 2
 129         unsigned long flags;
 130
 131         struct uio_info uio_info;
 132
 133         struct inode *inode;
 134
 135         struct tcmu_mailbox *mb_addr;
 136         size_t dev_size;
 137         u32 cmdr_size;
 138         u32 cmdr_last_cleaned;
 139         /* Offset of data area from start of mb */
 140         /* Must add data_off and mb_addr to get the address */
 141         size_t data_off;
 142         size_t data_size;
 143         uint32_t max_blocks;
 144         size_t ring_size;
 145
 146         struct mutex cmdr_lock;
 147         struct list_head cmdr_queue;
 148
 149         uint32_t dbi_max;
 150         uint32_t dbi_thresh;
 151         unsigned long *data_bitmap;
 152         struct radix_tree_root data_blocks;
 153
 154         struct idr commands;
 155
 156         struct timer_list cmd_timer;
 157         unsigned int cmd_time_out;
 158
 159         struct timer_list qfull_timer;
 160         int qfull_time_out;
 161
 162         struct list_head timedout_entry;
 163
 164         spinlock_t nl_cmd_lock;
 165         struct tcmu_nl_cmd curr_nl_cmd;
 166         /* wake up threads waiting on curr_nl_cmd */
 167         wait_queue_head_t nl_cmd_wq;
 168
 169         char dev_config[TCMU_CONFIG_LEN];
 170
 171         int nl_reply_supported;
 172 };
 173
 174 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
 175
 176 #define CMDR_OFF sizeof(struct tcmu_mailbox)
 177
 178 struct tcmu_cmd {
 179         struct se_cmd *se_cmd;
 180         struct tcmu_dev *tcmu_dev;
 181         struct list_head cmdr_queue_entry;
 182
 183         uint16_t cmd_id;
 184
 185         /* Can't use se_cmd when cleaning up expired cmds, because if
 186            cmd has been completed then accessing se_cmd is off limits */
 187         uint32_t dbi_cnt;
 188         uint32_t dbi_cur;
 189         uint32_t *dbi;
 190
 191         unsigned long deadline;
 192
 193 #define TCMU_CMD_BIT_EXPIRED 0
 194         unsigned long flags;
 195 };
 196 /*
 197  * To avoid dead lock the mutex lock order should always be:
 198  *
 199  * mutex_lock(&root_udev_mutex);
 200  * ...
 201  * mutex_lock(&tcmu_dev->cmdr_lock);
 202  * mutex_unlock(&tcmu_dev->cmdr_lock);
 203  * ...
 204  * mutex_unlock(&root_udev_mutex);
 205  */
 206 static DEFINE_MUTEX(root_udev_mutex);
 207 static LIST_HEAD(root_udev);
 208
 209 static DEFINE_SPINLOCK(timed_out_udevs_lock);
 210 static LIST_HEAD(timed_out_udevs);
 211
 212 static struct kmem_cache *tcmu_cmd_cache;
 213
 214 static atomic_t global_db_count = ATOMIC_INIT(0);
 215 static struct delayed_work tcmu_unmap_work;
 216 static int tcmu_global_max_blocks = TCMU_GLOBAL_MAX_BLOCKS_DEF;
 217
 218 static int tcmu_set_global_max_data_area(const char *str,
 219                                          const struct kernel_param *kp)
 220 {
 221         int ret, max_area_mb;
 222
 223         ret = kstrtoint(str, 10, &max_area_mb);
 224         if (ret)
 225                 return -EINVAL;
 226
 227         if (max_area_mb <= 0) {
 228                 pr_err("global_max_data_area must be larger than 0.\n");
 229                 return -EINVAL;
 230         }
 231
 232         tcmu_global_max_blocks = TCMU_MBS_TO_BLOCKS(max_area_mb);
 233         if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
 234                 schedule_delayed_work(&tcmu_unmap_work, 0);
 235         else
 236                 cancel_delayed_work_sync(&tcmu_unmap_work);
 237
 238         return 0;
 239 }
 240
 241 static int tcmu_get_global_max_data_area(char *buffer,
 242                                          const struct kernel_param *kp)
 243 {
 244         return sprintf(buffer, "%d", TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
 245 }
 246
 247 static const struct kernel_param_ops tcmu_global_max_data_area_op = {
 248         .set = tcmu_set_global_max_data_area,
 249         .get = tcmu_get_global_max_data_area,
 250 };
 251
 252 module_param_cb(global_max_data_area_mb, &tcmu_global_max_data_area_op, NULL,
 253                 S_IWUSR | S_IRUGO);
 254 MODULE_PARM_DESC(global_max_data_area_mb,
 255                  "Max MBs allowed to be allocated to all the tcmu device's "
 256                  "data areas.");
 257
 258 /* multicast group */
 259 enum tcmu_multicast_groups {
 260         TCMU_MCGRP_CONFIG,
 261 };
 262
 263 static const struct genl_multicast_group tcmu_mcgrps[] = {
 264         [TCMU_MCGRP_CONFIG] = { .name = "config", },
 265 };
 266
 267 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
 268         [TCMU_ATTR_DEVICE]      = { .type = NLA_STRING },
 269         [TCMU_ATTR_MINOR]       = { .type = NLA_U32 },
 270         [TCMU_ATTR_CMD_STATUS]  = { .type = NLA_S32 },
 271         [TCMU_ATTR_DEVICE_ID]   = { .type = NLA_U32 },
 272         [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
 273 };
 274
 275 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
 276 {
 277         struct se_device *dev;
 278         struct tcmu_dev *udev;
 279         struct tcmu_nl_cmd *nl_cmd;
 280         int dev_id, rc, ret = 0;
 281         bool is_removed = (completed_cmd == TCMU_CMD_REMOVED_DEVICE);
 282
 283         if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
 284             !info->attrs[TCMU_ATTR_DEVICE_ID]) {
 285                 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
 286                 return -EINVAL;
 287         }
 288
 289         dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
 290         rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
 291
 292         dev = target_find_device(dev_id, !is_removed);
 293         if (!dev) {
 294                 printk(KERN_ERR "tcmu nl cmd %u/%u completion could not find device with dev id %u.\n",
 295                        completed_cmd, rc, dev_id);
 296                 return -ENODEV;
 297         }
 298         udev = TCMU_DEV(dev);
 299
 300         spin_lock(&udev->nl_cmd_lock);
 301         nl_cmd = &udev->curr_nl_cmd;
 302
 303         pr_debug("genl cmd done got id %d curr %d done %d rc %d\n", dev_id,
 304                  nl_cmd->cmd, completed_cmd, rc);
 305
 306         if (nl_cmd->cmd != completed_cmd) {
 307                 printk(KERN_ERR "Mismatched commands (Expecting reply for %d. Current %d).\n",
 308                        completed_cmd, nl_cmd->cmd);
 309                 ret = -EINVAL;
 310         } else {
 311                 nl_cmd->status = rc;
 312         }
 313
 314         spin_unlock(&udev->nl_cmd_lock);
 315         if (!is_removed)
 316                  target_undepend_item(&dev->dev_group.cg_item);
 317         if (!ret)
 318                 complete(&nl_cmd->complete);
 319         return ret;
 320 }
 321
 322 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
 323 {
 324         return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
 325 }
 326
 327 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
 328 {
 329         return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
 330 }
 331
 332 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
 333                                        struct genl_info *info)
 334 {
 335         return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
 336 }
 337
 338 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
 339 {
 340         if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
 341                 tcmu_kern_cmd_reply_supported  =
 342                         nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
 343                 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
 344                        tcmu_kern_cmd_reply_supported);
 345         }
 346
 347         return 0;
 348 }
 349
 350 static const struct genl_ops tcmu_genl_ops[] = {
 351         {
 352                 .cmd    = TCMU_CMD_SET_FEATURES,
 353                 .flags  = GENL_ADMIN_PERM,
 354                 .policy = tcmu_attr_policy,
 355                 .doit   = tcmu_genl_set_features,
 356         },
 357         {
 358                 .cmd    = TCMU_CMD_ADDED_DEVICE_DONE,
 359                 .flags  = GENL_ADMIN_PERM,
 360                 .policy = tcmu_attr_policy,
 361                 .doit   = tcmu_genl_add_dev_done,
 362         },
 363         {
 364                 .cmd    = TCMU_CMD_REMOVED_DEVICE_DONE,
 365                 .flags  = GENL_ADMIN_PERM,
 366                 .policy = tcmu_attr_policy,
 367                 .doit   = tcmu_genl_rm_dev_done,
 368         },
 369         {
 370                 .cmd    = TCMU_CMD_RECONFIG_DEVICE_DONE,
 371                 .flags  = GENL_ADMIN_PERM,
 372                 .policy = tcmu_attr_policy,
 373                 .doit   = tcmu_genl_reconfig_dev_done,
 374         },
 375 };
 376
 377 /* Our generic netlink family */
 378 static struct genl_family tcmu_genl_family __ro_after_init = {
 379         .module = THIS_MODULE,
 380         .hdrsize = 0,
 381         .name = "TCM-USER",
 382         .version = 2,
 383         .maxattr = TCMU_ATTR_MAX,
 384         .mcgrps = tcmu_mcgrps,
 385         .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
 386         .netnsok = true,
 387         .ops = tcmu_genl_ops,
 388         .n_ops = ARRAY_SIZE(tcmu_genl_ops),
 389 };
 390
 391 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
 392 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
 393 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
 394 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
 395
 396 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
 397 {
 398         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 399         uint32_t i;
 400
 401         for (i = 0; i < len; i++)
 402                 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
 403 }
 404
 405 static inline bool tcmu_get_empty_block(struct tcmu_dev *udev,
 406                                         struct tcmu_cmd *tcmu_cmd)
 407 {
 408         struct page *page;
 409         int ret, dbi;
 410
 411         dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
 412         if (dbi == udev->dbi_thresh)
 413                 return false;
 414
 415         page = radix_tree_lookup(&udev->data_blocks, dbi);
 416         if (!page) {
 417                 if (atomic_add_return(1, &global_db_count) >
 418                                       tcmu_global_max_blocks)
 419                         schedule_delayed_work(&tcmu_unmap_work, 0);
 420
 421                 /* try to get new page from the mm */
 422                 page = alloc_page(GFP_KERNEL);
 423                 if (!page)
 424                         goto err_alloc;
 425
 426                 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
 427                 if (ret)
 428                         goto err_insert;
 429         }
 430
 431         if (dbi > udev->dbi_max)
 432                 udev->dbi_max = dbi;
 433
 434         set_bit(dbi, udev->data_bitmap);
 435         tcmu_cmd_set_dbi(tcmu_cmd, dbi);
 436
 437         return true;
 438 err_insert:
 439         __free_page(page);
 440 err_alloc:
 441         atomic_dec(&global_db_count);
 442         return false;
 443 }
 444
 445 static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
 446                                   struct tcmu_cmd *tcmu_cmd)
 447 {
 448         int i;
 449
 450         for (i = tcmu_cmd->dbi_cur; i < tcmu_cmd->dbi_cnt; i++) {
 451                 if (!tcmu_get_empty_block(udev, tcmu_cmd))
 452                         return false;
 453         }
 454         return true;
 455 }
 456
 457 static inline struct page *
 458 tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
 459 {
 460         return radix_tree_lookup(&udev->data_blocks, dbi);
 461 }
 462
 463 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
 464 {
 465         kfree(tcmu_cmd->dbi);
 466         kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 467 }
 468
 469 static inline size_t tcmu_cmd_get_data_length(struct tcmu_cmd *tcmu_cmd)
 470 {
 471         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 472         size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
 473
 474         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 475                 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
 476                 data_length += round_up(se_cmd->t_bidi_data_sg->length,
 477                                 DATA_BLOCK_SIZE);
 478         }
 479
 480         return data_length;
 481 }
 482
 483 static inline uint32_t tcmu_cmd_get_block_cnt(struct tcmu_cmd *tcmu_cmd)
 484 {
 485         size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
 486
 487         return data_length / DATA_BLOCK_SIZE;
 488 }
 489
 490 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
 491 {
 492         struct se_device *se_dev = se_cmd->se_dev;
 493         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 494         struct tcmu_cmd *tcmu_cmd;
 495
 496         tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_KERNEL);
 497         if (!tcmu_cmd)
 498                 return NULL;
 499
 500         INIT_LIST_HEAD(&tcmu_cmd->cmdr_queue_entry);
 501         tcmu_cmd->se_cmd = se_cmd;
 502         tcmu_cmd->tcmu_dev = udev;
 503
 504         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 505         tcmu_cmd->dbi_cnt = tcmu_cmd_get_block_cnt(tcmu_cmd);
 506         tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
 507                                 GFP_KERNEL);
 508         if (!tcmu_cmd->dbi) {
 509                 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 510                 return NULL;
 511         }
 512
 513         return tcmu_cmd;
 514 }
 515
 516 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
 517 {
 518         unsigned long offset = offset_in_page(vaddr);
 519         void *start = vaddr - offset;
 520
 521         size = round_up(size+offset, PAGE_SIZE);
 522
 523         while (size) {
 524                 flush_dcache_page(virt_to_page(start));
 525                 start += PAGE_SIZE;
 526                 size -= PAGE_SIZE;
 527         }
 528 }
 529
 530 /*
 531  * Some ring helper functions. We don't assume size is a power of 2 so
 532  * we can't use circ_buf.h.
 533  */
 534 static inline size_t spc_used(size_t head, size_t tail, size_t size)
 535 {
 536         int diff = head - tail;
 537
 538         if (diff >= 0)
 539                 return diff;
 540         else
 541                 return size + diff;
 542 }
 543
 544 static inline size_t spc_free(size_t head, size_t tail, size_t size)
 545 {
 546         /* Keep 1 byte unused or we can't tell full from empty */
 547         return (size - spc_used(head, tail, size) - 1);
 548 }
 549
 550 static inline size_t head_to_end(size_t head, size_t size)
 551 {
 552         return size - head;
 553 }
 554
 555 static inline void new_iov(struct iovec **iov, int *iov_cnt)
 556 {
 557         struct iovec *iovec;
 558
 559         if (*iov_cnt != 0)
 560                 (*iov)++;
 561         (*iov_cnt)++;
 562
 563         iovec = *iov;
 564         memset(iovec, 0, sizeof(struct iovec));
 565 }
 566
 567 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
 568
 569 /* offset is relative to mb_addr */
 570 static inline size_t get_block_offset_user(struct tcmu_dev *dev,
 571                 int dbi, int remaining)
 572 {
 573         return dev->data_off + dbi * DATA_BLOCK_SIZE +
 574                 DATA_BLOCK_SIZE - remaining;
 575 }
 576
 577 static inline size_t iov_tail(struct iovec *iov)
 578 {
 579         return (size_t)iov->iov_base + iov->iov_len;
 580 }
 581
 582 static void scatter_data_area(struct tcmu_dev *udev,
 583         struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
 584         unsigned int data_nents, struct iovec **iov,
 585         int *iov_cnt, bool copy_data)
 586 {
 587         int i, dbi;
 588         int block_remaining = 0;
 589         void *from, *to = NULL;
 590         size_t copy_bytes, to_offset, offset;
 591         struct scatterlist *sg;
 592         struct page *page;
 593
 594         for_each_sg(data_sg, sg, data_nents, i) {
 595                 int sg_remaining = sg->length;
 596                 from = kmap_atomic(sg_page(sg)) + sg->offset;
 597                 while (sg_remaining > 0) {
 598                         if (block_remaining == 0) {
 599                                 if (to)
 600                                         kunmap_atomic(to);
 601
 602                                 block_remaining = DATA_BLOCK_SIZE;
 603                                 dbi = tcmu_cmd_get_dbi(tcmu_cmd);
 604                                 page = tcmu_get_block_page(udev, dbi);
 605                                 to = kmap_atomic(page);
 606                         }
 607
 608                         /*
 609                          * Covert to virtual offset of the ring data area.
 610                          */
 611                         to_offset = get_block_offset_user(udev, dbi,
 612                                         block_remaining);
 613
 614                         /*
 615                          * The following code will gather and map the blocks
 616                          * to the same iovec when the blocks are all next to
 617                          * each other.
 618                          */
 619                         copy_bytes = min_t(size_t, sg_remaining,
 620                                         block_remaining);
 621                         if (*iov_cnt != 0 &&
 622                             to_offset == iov_tail(*iov)) {
 623                                 /*
 624                                  * Will append to the current iovec, because
 625                                  * the current block page is next to the
 626                                  * previous one.
 627                                  */
 628                                 (*iov)->iov_len += copy_bytes;
 629                         } else {
 630                                 /*
 631                                  * Will allocate a new iovec because we are
 632                                  * first time here or the current block page
 633                                  * is not next to the previous one.
 634                                  */
 635                                 new_iov(iov, iov_cnt);
 636                                 (*iov)->iov_base = (void __user *)to_offset;
 637                                 (*iov)->iov_len = copy_bytes;
 638                         }
 639
 640                         if (copy_data) {
 641                                 offset = DATA_BLOCK_SIZE - block_remaining;
 642                                 memcpy(to + offset,
 643                                        from + sg->length - sg_remaining,
 644                                        copy_bytes);
 645                                 tcmu_flush_dcache_range(to, copy_bytes);
 646                         }
 647
 648                         sg_remaining -= copy_bytes;
 649                         block_remaining -= copy_bytes;
 650                 }
 651                 kunmap_atomic(from - sg->offset);
 652         }
 653
 654         if (to)
 655                 kunmap_atomic(to);
 656 }
 657
 658 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 659                              bool bidi)
 660 {
 661         struct se_cmd *se_cmd = cmd->se_cmd;
 662         int i, dbi;
 663         int block_remaining = 0;
 664         void *from = NULL, *to;
 665         size_t copy_bytes, offset;
 666         struct scatterlist *sg, *data_sg;
 667         struct page *page;
 668         unsigned int data_nents;
 669         uint32_t count = 0;
 670
 671         if (!bidi) {
 672                 data_sg = se_cmd->t_data_sg;
 673                 data_nents = se_cmd->t_data_nents;
 674         } else {
 675
 676                 /*
 677                  * For bidi case, the first count blocks are for Data-Out
 678                  * buffer blocks, and before gathering the Data-In buffer
 679                  * the Data-Out buffer blocks should be discarded.
 680                  */
 681                 count = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
 682
 683                 data_sg = se_cmd->t_bidi_data_sg;
 684                 data_nents = se_cmd->t_bidi_data_nents;
 685         }
 686
 687         tcmu_cmd_set_dbi_cur(cmd, count);
 688
 689         for_each_sg(data_sg, sg, data_nents, i) {
 690                 int sg_remaining = sg->length;
 691                 to = kmap_atomic(sg_page(sg)) + sg->offset;
 692                 while (sg_remaining > 0) {
 693                         if (block_remaining == 0) {
 694                                 if (from)
 695                                         kunmap_atomic(from);
 696
 697                                 block_remaining = DATA_BLOCK_SIZE;
 698                                 dbi = tcmu_cmd_get_dbi(cmd);
 699                                 page = tcmu_get_block_page(udev, dbi);
 700                                 from = kmap_atomic(page);
 701                         }
 702                         copy_bytes = min_t(size_t, sg_remaining,
 703                                         block_remaining);
 704                         offset = DATA_BLOCK_SIZE - block_remaining;
 705                         tcmu_flush_dcache_range(from, copy_bytes);
 706                         memcpy(to + sg->length - sg_remaining, from + offset,
 707                                         copy_bytes);
 708
 709                         sg_remaining -= copy_bytes;
 710                         block_remaining -= copy_bytes;
 711                 }
 712                 kunmap_atomic(to - sg->offset);
 713         }
 714         if (from)
 715                 kunmap_atomic(from);
 716 }
 717
 718 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
 719 {
 720         return thresh - bitmap_weight(bitmap, thresh);
 721 }
 722
 723 /*
 724  * We can't queue a command until we have space available on the cmd ring *and*
 725  * space available on the data area.
 726  *
 727  * Called with ring lock held.
 728  */
 729 static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 730                 size_t cmd_size, size_t data_needed)
 731 {
 732         struct tcmu_mailbox *mb = udev->mb_addr;
 733         uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
 734                                 / DATA_BLOCK_SIZE;
 735         size_t space, cmd_needed;
 736         u32 cmd_head;
 737
 738         tcmu_flush_dcache_range(mb, sizeof(*mb));
 739
 740         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 741
 742         /*
 743          * If cmd end-of-ring space is too small then we need space for a NOP plus
 744          * original cmd - cmds are internally contiguous.
 745          */
 746         if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
 747                 cmd_needed = cmd_size;
 748         else
 749                 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
 750
 751         space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
 752         if (space < cmd_needed) {
 753                 pr_debug("no cmd space: %u %u %u\n", cmd_head,
 754                        udev->cmdr_last_cleaned, udev->cmdr_size);
 755                 return false;
 756         }
 757
 758         /* try to check and get the data blocks as needed */
 759         space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
 760         if ((space * DATA_BLOCK_SIZE) < data_needed) {
 761                 unsigned long blocks_left =
 762                                 (udev->max_blocks - udev->dbi_thresh) + space;
 763
 764                 if (blocks_left < blocks_needed) {
 765                         pr_debug("no data space: only %lu available, but ask for %zu\n",
 766                                         blocks_left * DATA_BLOCK_SIZE,
 767                                         data_needed);
 768                         return false;
 769                 }
 770
 771                 udev->dbi_thresh += blocks_needed;
 772                 if (udev->dbi_thresh > udev->max_blocks)
 773                         udev->dbi_thresh = udev->max_blocks;
 774         }
 775
 776         return tcmu_get_empty_blocks(udev, cmd);
 777 }
 778
 779 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
 780 {
 781         return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 782                         sizeof(struct tcmu_cmd_entry));
 783 }
 784
 785 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
 786                                            size_t base_command_size)
 787 {
 788         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 789         size_t command_size;
 790
 791         command_size = base_command_size +
 792                 round_up(scsi_command_size(se_cmd->t_task_cdb),
 793                                 TCMU_OP_ALIGN_SIZE);
 794
 795         WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
 796
 797         return command_size;
 798 }
 799
 800 static int tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
 801                                 struct timer_list *timer)
 802 {
 803         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 804         int cmd_id;
 805
 806         if (tcmu_cmd->cmd_id)
 807                 goto setup_timer;
 808
 809         cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 1, USHRT_MAX, GFP_NOWAIT);
 810         if (cmd_id < 0) {
 811                 pr_err("tcmu: Could not allocate cmd id.\n");
 812                 return cmd_id;
 813         }
 814         tcmu_cmd->cmd_id = cmd_id;
 815
 816         pr_debug("allocated cmd %u for dev %s tmo %lu\n", tcmu_cmd->cmd_id,
 817                  udev->name, tmo / MSEC_PER_SEC);
 818
 819 setup_timer:
 820         if (!tmo)
 821                 return 0;
 822
 823         tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
 824         mod_timer(timer, tcmu_cmd->deadline);
 825         return 0;
 826 }
 827
 828 static int add_to_cmdr_queue(struct tcmu_cmd *tcmu_cmd)
 829 {
 830         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 831         unsigned int tmo;
 832         int ret;
 833
 834         /*
 835          * For backwards compat if qfull_time_out is not set use
 836          * cmd_time_out and if that's not set use the default time out.
 837          */
 838         if (!udev->qfull_time_out)
 839                 return -ETIMEDOUT;
 840         else if (udev->qfull_time_out > 0)
 841                 tmo = udev->qfull_time_out;
 842         else if (udev->cmd_time_out)
 843                 tmo = udev->cmd_time_out;
 844         else
 845                 tmo = TCMU_TIME_OUT;
 846
 847         ret = tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
 848         if (ret)
 849                 return ret;
 850
 851         list_add_tail(&tcmu_cmd->cmdr_queue_entry, &udev->cmdr_queue);
 852         pr_debug("adding cmd %u on dev %s to ring space wait queue\n",
 853                  tcmu_cmd->cmd_id, udev->name);
 854         return 0;
 855 }
 856
 857 /**
 858  * queue_cmd_ring - queue cmd to ring or internally
 859  * @tcmu_cmd: cmd to queue
 860  * @scsi_err: TCM error code if failure (-1) returned.
 861  *
 862  * Returns:
 863  * -1 we cannot queue internally or to the ring.
 864  *  0 success
 865  *  1 internally queued to wait for ring memory to free.
 866  */
 867 static sense_reason_t queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, int *scsi_err)
 868 {
 869         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 870         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 871         size_t base_command_size, command_size;
 872         struct tcmu_mailbox *mb;
 873         struct tcmu_cmd_entry *entry;
 874         struct iovec *iov;
 875         int iov_cnt, ret;
 876         uint32_t cmd_head;
 877         uint64_t cdb_off;
 878         bool copy_to_data_area;
 879         size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
 880
 881         *scsi_err = TCM_NO_SENSE;
 882
 883         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
 884                 *scsi_err = TCM_LUN_BUSY;
 885                 return -1;
 886         }
 887
 888         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
 889                 *scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 890                 return -1;
 891         }
 892
 893         /*
 894          * Must be a certain minimum size for response sense info, but
 895          * also may be larger if the iov array is large.
 896          *
 897          * We prepare as many iovs as possbile for potential uses here,
 898          * because it's expensive to tell how many regions are freed in
 899          * the bitmap & global data pool, as the size calculated here
 900          * will only be used to do the checks.
 901          *
 902          * The size will be recalculated later as actually needed to save
 903          * cmd area memories.
 904          */
 905         base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
 906         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 907
 908         if (!list_empty(&udev->cmdr_queue))
 909                 goto queue;
 910
 911         mb = udev->mb_addr;
 912         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 913         if ((command_size > (udev->cmdr_size / 2)) ||
 914             data_length > udev->data_size) {
 915                 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
 916                         "cmd ring/data area\n", command_size, data_length,
 917                         udev->cmdr_size, udev->data_size);
 918                 *scsi_err = TCM_INVALID_CDB_FIELD;
 919                 return -1;
 920         }
 921
 922         if (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
 923                 /*
 924                  * Don't leave commands partially setup because the unmap
 925                  * thread might need the blocks to make forward progress.
 926                  */
 927                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
 928                 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 929                 goto queue;
 930         }
 931
 932         /* Insert a PAD if end-of-ring space is too small */
 933         if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
 934                 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
 935
 936                 entry = (void *) mb + CMDR_OFF + cmd_head;
 937                 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
 938                 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
 939                 entry->hdr.cmd_id = 0; /* not used for PAD */
 940                 entry->hdr.kflags = 0;
 941                 entry->hdr.uflags = 0;
 942                 tcmu_flush_dcache_range(entry, sizeof(*entry));
 943
 944                 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
 945                 tcmu_flush_dcache_range(mb, sizeof(*mb));
 946
 947                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 948                 WARN_ON(cmd_head != 0);
 949         }
 950
 951         entry = (void *) mb + CMDR_OFF + cmd_head;
 952         memset(entry, 0, command_size);
 953         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
 954
 955         /* Handle allocating space from the data area */
 956         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 957         iov = &entry->req.iov[0];
 958         iov_cnt = 0;
 959         copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
 960                 || se_cmd->se_cmd_flags & SCF_BIDI);
 961         scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
 962                           se_cmd->t_data_nents, &iov, &iov_cnt,
 963                           copy_to_data_area);
 964         entry->req.iov_cnt = iov_cnt;
 965
 966         /* Handle BIDI commands */
 967         iov_cnt = 0;
 968         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 969                 iov++;
 970                 scatter_data_area(udev, tcmu_cmd, se_cmd->t_bidi_data_sg,
 971                                   se_cmd->t_bidi_data_nents, &iov, &iov_cnt,
 972                                   false);
 973         }
 974         entry->req.iov_bidi_cnt = iov_cnt;
 975
 976         ret = tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out,
 977                                    &udev->cmd_timer);
 978         if (ret) {
 979                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
 980                 mutex_unlock(&udev->cmdr_lock);
 981
 982                 *scsi_err = TCM_OUT_OF_RESOURCES;
 983                 return -1;
 984         }
 985         entry->hdr.cmd_id = tcmu_cmd->cmd_id;
 986
 987         /*
 988          * Recalaulate the command's base size and size according
 989          * to the actual needs
 990          */
 991         base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
 992                                                        entry->req.iov_bidi_cnt);
 993         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 994
 995         tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
 996
 997         /* All offsets relative to mb_addr, not start of entry! */
 998         cdb_off = CMDR_OFF + cmd_head + base_command_size;
 999         memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
1000         entry->req.cdb_off = cdb_off;
1001         tcmu_flush_dcache_range(entry, sizeof(*entry));
1002
1003         UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1004         tcmu_flush_dcache_range(mb, sizeof(*mb));
1005
1006         /* TODO: only if FLUSH and FUA? */
1007         uio_event_notify(&udev->uio_info);
1008
1009         return 0;
1010
1011 queue:
1012         if (add_to_cmdr_queue(tcmu_cmd)) {
1013                 *scsi_err = TCM_OUT_OF_RESOURCES;
1014                 return -1;
1015         }
1016
1017         return 1;
1018 }
1019
1020 static sense_reason_t
1021 tcmu_queue_cmd(struct se_cmd *se_cmd)
1022 {
1023         struct se_device *se_dev = se_cmd->se_dev;
1024         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1025         struct tcmu_cmd *tcmu_cmd;
1026         sense_reason_t scsi_ret;
1027         int ret;
1028
1029         tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1030         if (!tcmu_cmd)
1031                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1032
1033         mutex_lock(&udev->cmdr_lock);
1034         ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1035         mutex_unlock(&udev->cmdr_lock);
1036         if (ret < 0)
1037                 tcmu_free_cmd(tcmu_cmd);
1038         return scsi_ret;
1039 }
1040
1041 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
1042 {
1043         struct se_cmd *se_cmd = cmd->se_cmd;
1044         struct tcmu_dev *udev = cmd->tcmu_dev;
1045
1046         /*
1047          * cmd has been completed already from timeout, just reclaim
1048          * data area space and free cmd
1049          */
1050         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1051                 goto out;
1052
1053         tcmu_cmd_reset_dbi_cur(cmd);
1054
1055         if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1056                 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1057                         cmd->se_cmd);
1058                 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1059         } else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1060                 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1061         } else if (se_cmd->se_cmd_flags & SCF_BIDI) {
1062                 /* Get Data-In buffer before clean up */
1063                 gather_data_area(udev, cmd, true);
1064         } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1065                 gather_data_area(udev, cmd, false);
1066         } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1067                 /* TODO: */
1068         } else if (se_cmd->data_direction != DMA_NONE) {
1069                 pr_warn("TCMU: data direction was %d!\n",
1070                         se_cmd->data_direction);
1071         }
1072
1073         target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1074
1075 out:
1076         cmd->se_cmd = NULL;
1077         tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1078         tcmu_free_cmd(cmd);
1079 }
1080
1081 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
1082 {
1083         struct tcmu_mailbox *mb;
1084         int handled = 0;
1085
1086         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1087                 pr_err("ring broken, not handling completions\n");
1088                 return 0;
1089         }
1090
1091         mb = udev->mb_addr;
1092         tcmu_flush_dcache_range(mb, sizeof(*mb));
1093
1094         while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1095
1096                 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
1097                 struct tcmu_cmd *cmd;
1098
1099                 tcmu_flush_dcache_range(entry, sizeof(*entry));
1100
1101                 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
1102                         UPDATE_HEAD(udev->cmdr_last_cleaned,
1103                                     tcmu_hdr_get_len(entry->hdr.len_op),
1104                                     udev->cmdr_size);
1105                         continue;
1106                 }
1107                 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1108
1109                 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1110                 if (!cmd) {
1111                         pr_err("cmd_id %u not found, ring is broken\n",
1112                                entry->hdr.cmd_id);
1113                         set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1114                         break;
1115                 }
1116
1117                 tcmu_handle_completion(cmd, entry);
1118
1119                 UPDATE_HEAD(udev->cmdr_last_cleaned,
1120                             tcmu_hdr_get_len(entry->hdr.len_op),
1121                             udev->cmdr_size);
1122
1123                 handled++;
1124         }
1125
1126         if (mb->cmd_tail == mb->cmd_head) {
1127                 /* no more pending commands */
1128                 del_timer(&udev->cmd_timer);
1129
1130                 if (list_empty(&udev->cmdr_queue)) {
1131                         /*
1132                          * no more pending or waiting commands so try to
1133                          * reclaim blocks if needed.
1134                          */
1135                         if (atomic_read(&global_db_count) >
1136                             tcmu_global_max_blocks)
1137                                 schedule_delayed_work(&tcmu_unmap_work, 0);
1138                 }
1139         }
1140
1141         return handled;
1142 }
1143
1144 static int tcmu_check_expired_cmd(int id, void *p, void *data)
1145 {
1146         struct tcmu_cmd *cmd = p;
1147         struct tcmu_dev *udev = cmd->tcmu_dev;
1148         u8 scsi_status;
1149         struct se_cmd *se_cmd;
1150         bool is_running;
1151
1152         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1153                 return 0;
1154
1155         if (!time_after(jiffies, cmd->deadline))
1156                 return 0;
1157
1158         is_running = list_empty(&cmd->cmdr_queue_entry);
1159         se_cmd = cmd->se_cmd;
1160
1161         if (is_running) {
1162                 /*
1163                  * If cmd_time_out is disabled but qfull is set deadline
1164                  * will only reflect the qfull timeout. Ignore it.
1165                  */
1166                 if (!udev->cmd_time_out)
1167                         return 0;
1168
1169                 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1170                 /*
1171                  * target_complete_cmd will translate this to LUN COMM FAILURE
1172                  */
1173                 scsi_status = SAM_STAT_CHECK_CONDITION;
1174         } else {
1175                 list_del_init(&cmd->cmdr_queue_entry);
1176
1177                 idr_remove(&udev->commands, id);
1178                 tcmu_free_cmd(cmd);
1179                 scsi_status = SAM_STAT_TASK_SET_FULL;
1180         }
1181
1182         pr_debug("Timing out cmd %u on dev %s that is %s.\n",
1183                  id, udev->name, is_running ? "inflight" : "queued");
1184
1185         target_complete_cmd(se_cmd, scsi_status);
1186         return 0;
1187 }
1188
1189 static void tcmu_device_timedout(struct tcmu_dev *udev)
1190 {
1191         spin_lock(&timed_out_udevs_lock);
1192         if (list_empty(&udev->timedout_entry))
1193                 list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1194         spin_unlock(&timed_out_udevs_lock);
1195
1196         schedule_delayed_work(&tcmu_unmap_work, 0);
1197 }
1198
1199 static void tcmu_cmd_timedout(struct timer_list *t)
1200 {
1201         struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1202
1203         pr_debug("%s cmd timeout has expired\n", udev->name);
1204         tcmu_device_timedout(udev);
1205 }
1206
1207 static void tcmu_qfull_timedout(struct timer_list *t)
1208 {
1209         struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1210
1211         pr_debug("%s qfull timeout has expired\n", udev->name);
1212         tcmu_device_timedout(udev);
1213 }
1214
1215 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1216 {
1217         struct tcmu_hba *tcmu_hba;
1218
1219         tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1220         if (!tcmu_hba)
1221                 return -ENOMEM;
1222
1223         tcmu_hba->host_id = host_id;
1224         hba->hba_ptr = tcmu_hba;
1225
1226         return 0;
1227 }
1228
1229 static void tcmu_detach_hba(struct se_hba *hba)
1230 {
1231         kfree(hba->hba_ptr);
1232         hba->hba_ptr = NULL;
1233 }
1234
1235 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1236 {
1237         struct tcmu_dev *udev;
1238
1239         udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1240         if (!udev)
1241                 return NULL;
1242         kref_init(&udev->kref);
1243
1244         udev->name = kstrdup(name, GFP_KERNEL);
1245         if (!udev->name) {
1246                 kfree(udev);
1247                 return NULL;
1248         }
1249
1250         udev->hba = hba;
1251         udev->cmd_time_out = TCMU_TIME_OUT;
1252         udev->qfull_time_out = -1;
1253
1254         udev->max_blocks = DATA_BLOCK_BITS_DEF;
1255         mutex_init(&udev->cmdr_lock);
1256
1257         INIT_LIST_HEAD(&udev->timedout_entry);
1258         INIT_LIST_HEAD(&udev->cmdr_queue);
1259         idr_init(&udev->commands);
1260
1261         timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1262         timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1263
1264         init_waitqueue_head(&udev->nl_cmd_wq);
1265         spin_lock_init(&udev->nl_cmd_lock);
1266
1267         INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1268
1269         return &udev->se_dev;
1270 }
1271
1272 static bool run_cmdr_queue(struct tcmu_dev *udev, bool fail)
1273 {
1274         struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1275         LIST_HEAD(cmds);
1276         bool drained = true;
1277         sense_reason_t scsi_ret;
1278         int ret;
1279
1280         if (list_empty(&udev->cmdr_queue))
1281                 return true;
1282
1283         pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1284
1285         list_splice_init(&udev->cmdr_queue, &cmds);
1286
1287         list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, cmdr_queue_entry) {
1288                 list_del_init(&tcmu_cmd->cmdr_queue_entry);
1289
1290                 pr_debug("removing cmd %u on dev %s from queue\n",
1291                          tcmu_cmd->cmd_id, udev->name);
1292
1293                 if (fail) {
1294                         idr_remove(&udev->commands, tcmu_cmd->cmd_id);
1295                         /*
1296                          * We were not able to even start the command, so
1297                          * fail with busy to allow a retry in case runner
1298                          * was only temporarily down. If the device is being
1299                          * removed then LIO core will do the right thing and
1300                          * fail the retry.
1301                          */
1302                         target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1303                         tcmu_free_cmd(tcmu_cmd);
1304                         continue;
1305                 }
1306
1307                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1308                 if (ret < 0) {
1309                         pr_debug("cmd %u on dev %s failed with %u\n",
1310                                  tcmu_cmd->cmd_id, udev->name, scsi_ret);
1311
1312                         idr_remove(&udev->commands, tcmu_cmd->cmd_id);
1313                         /*
1314                          * Ignore scsi_ret for now. target_complete_cmd
1315                          * drops it.
1316                          */
1317                         target_complete_cmd(tcmu_cmd->se_cmd,
1318                                             SAM_STAT_CHECK_CONDITION);
1319                         tcmu_free_cmd(tcmu_cmd);
1320                 } else if (ret > 0) {
1321                         pr_debug("ran out of space during cmdr queue run\n");
1322                         /*
1323                          * cmd was requeued, so just put all cmds back in
1324                          * the queue
1325                          */
1326                         list_splice_tail(&cmds, &udev->cmdr_queue);
1327                         drained = false;
1328                         goto done;
1329                 }
1330         }
1331         if (list_empty(&udev->cmdr_queue))
1332                 del_timer(&udev->qfull_timer);
1333 done:
1334         return drained;
1335 }
1336
1337 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1338 {
1339         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1340
1341         mutex_lock(&udev->cmdr_lock);
1342         tcmu_handle_completions(udev);
1343         run_cmdr_queue(udev, false);
1344         mutex_unlock(&udev->cmdr_lock);
1345
1346         return 0;
1347 }
1348
1349 /*
1350  * mmap code from uio.c. Copied here because we want to hook mmap()
1351  * and this stuff must come along.
1352  */
1353 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1354 {
1355         struct tcmu_dev *udev = vma->vm_private_data;
1356         struct uio_info *info = &udev->uio_info;
1357
1358         if (vma->vm_pgoff < MAX_UIO_MAPS) {
1359                 if (info->mem[vma->vm_pgoff].size == 0)
1360                         return -1;
1361                 return (int)vma->vm_pgoff;
1362         }
1363         return -1;
1364 }
1365
1366 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1367 {
1368         struct page *page;
1369
1370         mutex_lock(&udev->cmdr_lock);
1371         page = tcmu_get_block_page(udev, dbi);
1372         if (likely(page)) {
1373                 mutex_unlock(&udev->cmdr_lock);
1374                 return page;
1375         }
1376
1377         /*
1378          * Userspace messed up and passed in a address not in the
1379          * data iov passed to it.
1380          */
1381         pr_err("Invalid addr to data block mapping  (dbi %u) on device %s\n",
1382                dbi, udev->name);
1383         page = NULL;
1384         mutex_unlock(&udev->cmdr_lock);
1385
1386         return page;
1387 }
1388
1389 static vm_fault_t tcmu_vma_fault(struct vm_fault *vmf)
1390 {
1391         struct tcmu_dev *udev = vmf->vma->vm_private_data;
1392         struct uio_info *info = &udev->uio_info;
1393         struct page *page;
1394         unsigned long offset;
1395         void *addr;
1396
1397         int mi = tcmu_find_mem_index(vmf->vma);
1398         if (mi < 0)
1399                 return VM_FAULT_SIGBUS;
1400
1401         /*
1402          * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1403          * to use mem[N].
1404          */
1405         offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1406
1407         if (offset < udev->data_off) {
1408                 /* For the vmalloc()ed cmd area pages */
1409                 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1410                 page = vmalloc_to_page(addr);
1411         } else {
1412                 uint32_t dbi;
1413
1414                 /* For the dynamically growing data area pages */
1415                 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1416                 page = tcmu_try_get_block_page(udev, dbi);
1417                 if (!page)
1418                         return VM_FAULT_SIGBUS;
1419         }
1420
1421         get_page(page);
1422         vmf->page = page;
1423         return 0;
1424 }
1425
1426 static const struct vm_operations_struct tcmu_vm_ops = {
1427         .fault = tcmu_vma_fault,
1428 };
1429
1430 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1431 {
1432         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1433
1434         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1435         vma->vm_ops = &tcmu_vm_ops;
1436
1437         vma->vm_private_data = udev;
1438
1439         /* Ensure the mmap is exactly the right size */
1440         if (vma_pages(vma) != (udev->ring_size >> PAGE_SHIFT))
1441                 return -EINVAL;
1442
1443         return 0;
1444 }
1445
1446 static int tcmu_open(struct uio_info *info, struct inode *inode)
1447 {
1448         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1449
1450         /* O_EXCL not supported for char devs, so fake it? */
1451         if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1452                 return -EBUSY;
1453
1454         udev->inode = inode;
1455         kref_get(&udev->kref);
1456
1457         pr_debug("open\n");
1458
1459         return 0;
1460 }
1461
1462 static void tcmu_dev_call_rcu(struct rcu_head *p)
1463 {
1464         struct se_device *dev = container_of(p, struct se_device, rcu_head);
1465         struct tcmu_dev *udev = TCMU_DEV(dev);
1466
1467         kfree(udev->uio_info.name);
1468         kfree(udev->name);
1469         kfree(udev);
1470 }
1471
1472 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1473 {
1474         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1475                 kmem_cache_free(tcmu_cmd_cache, cmd);
1476                 return 0;
1477         }
1478         return -EINVAL;
1479 }
1480
1481 static void tcmu_blocks_release(struct radix_tree_root *blocks,
1482                                 int start, int end)
1483 {
1484         int i;
1485         struct page *page;
1486
1487         for (i = start; i < end; i++) {
1488                 page = radix_tree_delete(blocks, i);
1489                 if (page) {
1490                         __free_page(page);
1491                         atomic_dec(&global_db_count);
1492                 }
1493         }
1494 }
1495
1496 static void tcmu_dev_kref_release(struct kref *kref)
1497 {
1498         struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1499         struct se_device *dev = &udev->se_dev;
1500         struct tcmu_cmd *cmd;
1501         bool all_expired = true;
1502         int i;
1503
1504         vfree(udev->mb_addr);
1505         udev->mb_addr = NULL;
1506
1507         spin_lock_bh(&timed_out_udevs_lock);
1508         if (!list_empty(&udev->timedout_entry))
1509                 list_del(&udev->timedout_entry);
1510         spin_unlock_bh(&timed_out_udevs_lock);
1511
1512         /* Upper layer should drain all requests before calling this */
1513         mutex_lock(&udev->cmdr_lock);
1514         idr_for_each_entry(&udev->commands, cmd, i) {
1515                 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1516                         all_expired = false;
1517         }
1518         idr_destroy(&udev->commands);
1519         WARN_ON(!all_expired);
1520
1521         tcmu_blocks_release(&udev->data_blocks, 0, udev->dbi_max + 1);
1522         kfree(udev->data_bitmap);
1523         mutex_unlock(&udev->cmdr_lock);
1524
1525         call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1526 }
1527
1528 static int tcmu_release(struct uio_info *info, struct inode *inode)
1529 {
1530         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1531
1532         clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1533
1534         pr_debug("close\n");
1535         /* release ref from open */
1536         kref_put(&udev->kref, tcmu_dev_kref_release);
1537         return 0;
1538 }
1539
1540 static void tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1541 {
1542         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1543
1544         if (!tcmu_kern_cmd_reply_supported)
1545                 return;
1546
1547         if (udev->nl_reply_supported <= 0)
1548                 return;
1549
1550 relock:
1551         spin_lock(&udev->nl_cmd_lock);
1552
1553         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1554                 spin_unlock(&udev->nl_cmd_lock);
1555                 pr_debug("sleeping for open nl cmd\n");
1556                 wait_event(udev->nl_cmd_wq, (nl_cmd->cmd == TCMU_CMD_UNSPEC));
1557                 goto relock;
1558         }
1559
1560         memset(nl_cmd, 0, sizeof(*nl_cmd));
1561         nl_cmd->cmd = cmd;
1562         init_completion(&nl_cmd->complete);
1563
1564         spin_unlock(&udev->nl_cmd_lock);
1565 }
1566
1567 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1568 {
1569         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1570         int ret;
1571         DEFINE_WAIT(__wait);
1572
1573         if (!tcmu_kern_cmd_reply_supported)
1574                 return 0;
1575
1576         if (udev->nl_reply_supported <= 0)
1577                 return 0;
1578
1579         pr_debug("sleeping for nl reply\n");
1580         wait_for_completion(&nl_cmd->complete);
1581
1582         spin_lock(&udev->nl_cmd_lock);
1583         nl_cmd->cmd = TCMU_CMD_UNSPEC;
1584         ret = nl_cmd->status;
1585         nl_cmd->status = 0;
1586         spin_unlock(&udev->nl_cmd_lock);
1587
1588         wake_up_all(&udev->nl_cmd_wq);
1589
1590         return ret;
1591 }
1592
1593 static int tcmu_netlink_event_init(struct tcmu_dev *udev,
1594                                    enum tcmu_genl_cmd cmd,
1595                                    struct sk_buff **buf, void **hdr)
1596 {
1597         struct sk_buff *skb;
1598         void *msg_header;
1599         int ret = -ENOMEM;
1600
1601         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1602         if (!skb)
1603                 return ret;
1604
1605         msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1606         if (!msg_header)
1607                 goto free_skb;
1608
1609         ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1610         if (ret < 0)
1611                 goto free_skb;
1612
1613         ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1614         if (ret < 0)
1615                 goto free_skb;
1616
1617         ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1618         if (ret < 0)
1619                 goto free_skb;
1620
1621         *buf = skb;
1622         *hdr = msg_header;
1623         return ret;
1624
1625 free_skb:
1626         nlmsg_free(skb);
1627         return ret;
1628 }
1629
1630 static int tcmu_netlink_event_send(struct tcmu_dev *udev,
1631                                    enum tcmu_genl_cmd cmd,
1632                                    struct sk_buff **buf, void **hdr)
1633 {
1634         int ret = 0;
1635         struct sk_buff *skb = *buf;
1636         void *msg_header = *hdr;
1637
1638         genlmsg_end(skb, msg_header);
1639
1640         tcmu_init_genl_cmd_reply(udev, cmd);
1641
1642         ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1643                                       TCMU_MCGRP_CONFIG, GFP_KERNEL);
1644        /* We don't care if no one is listening */
1645         if (ret == -ESRCH)
1646                 ret = 0;
1647         if (!ret)
1648                 ret = tcmu_wait_genl_cmd_reply(udev);
1649         return ret;
1650 }
1651
1652 static int tcmu_send_dev_add_event(struct tcmu_dev *udev)
1653 {
1654         struct sk_buff *skb = NULL;
1655         void *msg_header = NULL;
1656         int ret = 0;
1657
1658         ret = tcmu_netlink_event_init(udev, TCMU_CMD_ADDED_DEVICE, &skb,
1659                                       &msg_header);
1660         if (ret < 0)
1661                 return ret;
1662         return tcmu_netlink_event_send(udev, TCMU_CMD_ADDED_DEVICE, &skb,
1663                                        &msg_header);
1664
1665 }
1666
1667 static int tcmu_send_dev_remove_event(struct tcmu_dev *udev)
1668 {
1669         struct sk_buff *skb = NULL;
1670         void *msg_header = NULL;
1671         int ret = 0;
1672
1673         ret = tcmu_netlink_event_init(udev, TCMU_CMD_REMOVED_DEVICE,
1674                                       &skb, &msg_header);
1675         if (ret < 0)
1676                 return ret;
1677         return tcmu_netlink_event_send(udev, TCMU_CMD_REMOVED_DEVICE,
1678                                        &skb, &msg_header);
1679 }
1680
1681 static int tcmu_update_uio_info(struct tcmu_dev *udev)
1682 {
1683         struct tcmu_hba *hba = udev->hba->hba_ptr;
1684         struct uio_info *info;
1685         size_t size, used;
1686         char *str;
1687
1688         info = &udev->uio_info;
1689         size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
1690                         udev->dev_config);
1691         size += 1; /* for \0 */
1692         str = kmalloc(size, GFP_KERNEL);
1693         if (!str)
1694                 return -ENOMEM;
1695
1696         used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
1697         if (udev->dev_config[0])
1698                 snprintf(str + used, size - used, "/%s", udev->dev_config);
1699
1700         /* If the old string exists, free it */
1701         kfree(info->name);
1702         info->name = str;
1703
1704         return 0;
1705 }
1706
1707 static int tcmu_configure_device(struct se_device *dev)
1708 {
1709         struct tcmu_dev *udev = TCMU_DEV(dev);
1710         struct uio_info *info;
1711         struct tcmu_mailbox *mb;
1712         int ret = 0;
1713
1714         ret = tcmu_update_uio_info(udev);
1715         if (ret)
1716                 return ret;
1717
1718         info = &udev->uio_info;
1719
1720         udev->data_bitmap = kcalloc(BITS_TO_LONGS(udev->max_blocks),
1721                                     sizeof(unsigned long),
1722                                     GFP_KERNEL);
1723         if (!udev->data_bitmap) {
1724                 ret = -ENOMEM;
1725                 goto err_bitmap_alloc;
1726         }
1727
1728         udev->mb_addr = vzalloc(CMDR_SIZE);
1729         if (!udev->mb_addr) {
1730                 ret = -ENOMEM;
1731                 goto err_vzalloc;
1732         }
1733
1734         /* mailbox fits in first part of CMDR space */
1735         udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1736         udev->data_off = CMDR_SIZE;
1737         udev->data_size = udev->max_blocks * DATA_BLOCK_SIZE;
1738         udev->dbi_thresh = 0; /* Default in Idle state */
1739
1740         /* Initialise the mailbox of the ring buffer */
1741         mb = udev->mb_addr;
1742         mb->version = TCMU_MAILBOX_VERSION;
1743         mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
1744         mb->cmdr_off = CMDR_OFF;
1745         mb->cmdr_size = udev->cmdr_size;
1746
1747         WARN_ON(!PAGE_ALIGNED(udev->data_off));
1748         WARN_ON(udev->data_size % PAGE_SIZE);
1749         WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1750
1751         info->version = __stringify(TCMU_MAILBOX_VERSION);
1752
1753         info->mem[0].name = "tcm-user command & data buffer";
1754         info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1755         info->mem[0].size = udev->ring_size = udev->data_size + CMDR_SIZE;
1756         info->mem[0].memtype = UIO_MEM_NONE;
1757
1758         info->irqcontrol = tcmu_irqcontrol;
1759         info->irq = UIO_IRQ_CUSTOM;
1760
1761         info->mmap = tcmu_mmap;
1762         info->open = tcmu_open;
1763         info->release = tcmu_release;
1764
1765         ret = uio_register_device(tcmu_root_device, info);
1766         if (ret)
1767                 goto err_register;
1768
1769         /* User can set hw_block_size before enable the device */
1770         if (dev->dev_attrib.hw_block_size == 0)
1771                 dev->dev_attrib.hw_block_size = 512;
1772         /* Other attributes can be configured in userspace */
1773         if (!dev->dev_attrib.hw_max_sectors)
1774                 dev->dev_attrib.hw_max_sectors = 128;
1775         if (!dev->dev_attrib.emulate_write_cache)
1776                 dev->dev_attrib.emulate_write_cache = 0;
1777         dev->dev_attrib.hw_queue_depth = 128;
1778
1779         /* If user didn't explicitly disable netlink reply support, use
1780          * module scope setting.
1781          */
1782         if (udev->nl_reply_supported >= 0)
1783                 udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
1784
1785         /*
1786          * Get a ref incase userspace does a close on the uio device before
1787          * LIO has initiated tcmu_free_device.
1788          */
1789         kref_get(&udev->kref);
1790
1791         ret = tcmu_send_dev_add_event(udev);
1792         if (ret)
1793                 goto err_netlink;
1794
1795         mutex_lock(&root_udev_mutex);
1796         list_add(&udev->node, &root_udev);
1797         mutex_unlock(&root_udev_mutex);
1798
1799         return 0;
1800
1801 err_netlink:
1802         kref_put(&udev->kref, tcmu_dev_kref_release);
1803         uio_unregister_device(&udev->uio_info);
1804 err_register:
1805         vfree(udev->mb_addr);
1806         udev->mb_addr = NULL;
1807 err_vzalloc:
1808         kfree(udev->data_bitmap);
1809         udev->data_bitmap = NULL;
1810 err_bitmap_alloc:
1811         kfree(info->name);
1812         info->name = NULL;
1813
1814         return ret;
1815 }
1816
1817 static bool tcmu_dev_configured(struct tcmu_dev *udev)
1818 {
1819         return udev->uio_info.uio_dev ? true : false;
1820 }
1821
1822 static void tcmu_free_device(struct se_device *dev)
1823 {
1824         struct tcmu_dev *udev = TCMU_DEV(dev);
1825
1826         /* release ref from init */
1827         kref_put(&udev->kref, tcmu_dev_kref_release);
1828 }
1829
1830 static void tcmu_destroy_device(struct se_device *dev)
1831 {
1832         struct tcmu_dev *udev = TCMU_DEV(dev);
1833
1834         del_timer_sync(&udev->cmd_timer);
1835         del_timer_sync(&udev->qfull_timer);
1836
1837         mutex_lock(&root_udev_mutex);
1838         list_del(&udev->node);
1839         mutex_unlock(&root_udev_mutex);
1840
1841         tcmu_send_dev_remove_event(udev);
1842
1843         uio_unregister_device(&udev->uio_info);
1844
1845         /* release ref from configure */
1846         kref_put(&udev->kref, tcmu_dev_kref_release);
1847 }
1848
1849 static void tcmu_unblock_dev(struct tcmu_dev *udev)
1850 {
1851         mutex_lock(&udev->cmdr_lock);
1852         clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
1853         mutex_unlock(&udev->cmdr_lock);
1854 }
1855
1856 static void tcmu_block_dev(struct tcmu_dev *udev)
1857 {
1858         mutex_lock(&udev->cmdr_lock);
1859
1860         if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
1861                 goto unlock;
1862
1863         /* complete IO that has executed successfully */
1864         tcmu_handle_completions(udev);
1865         /* fail IO waiting to be queued */
1866         run_cmdr_queue(udev, true);
1867
1868 unlock:
1869         mutex_unlock(&udev->cmdr_lock);
1870 }
1871
1872 static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
1873 {
1874         struct tcmu_mailbox *mb;
1875         struct tcmu_cmd *cmd;
1876         int i;
1877
1878         mutex_lock(&udev->cmdr_lock);
1879
1880         idr_for_each_entry(&udev->commands, cmd, i) {
1881                 if (!list_empty(&cmd->cmdr_queue_entry))
1882                         continue;
1883
1884                 pr_debug("removing cmd %u on dev %s from ring (is expired %d)\n",
1885                           cmd->cmd_id, udev->name,
1886                           test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags));
1887
1888                 idr_remove(&udev->commands, i);
1889                 if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1890                         if (err_level == 1) {
1891                                 /*
1892                                  * Userspace was not able to start the
1893                                  * command or it is retryable.
1894                                  */
1895                                 target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
1896                         } else {
1897                                 /* hard failure */
1898                                 target_complete_cmd(cmd->se_cmd,
1899                                                     SAM_STAT_CHECK_CONDITION);
1900                         }
1901                 }
1902                 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1903                 tcmu_free_cmd(cmd);
1904         }
1905
1906         mb = udev->mb_addr;
1907         tcmu_flush_dcache_range(mb, sizeof(*mb));
1908         pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
1909                  mb->cmd_tail, mb->cmd_head);
1910
1911         udev->cmdr_last_cleaned = 0;
1912         mb->cmd_tail = 0;
1913         mb->cmd_head = 0;
1914         tcmu_flush_dcache_range(mb, sizeof(*mb));
1915
1916         del_timer(&udev->cmd_timer);
1917
1918         mutex_unlock(&udev->cmdr_lock);
1919 }
1920
1921 enum {
1922         Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
1923         Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_err,
1924 };
1925
1926 static match_table_t tokens = {
1927         {Opt_dev_config, "dev_config=%s"},
1928         {Opt_dev_size, "dev_size=%u"},
1929         {Opt_hw_block_size, "hw_block_size=%u"},
1930         {Opt_hw_max_sectors, "hw_max_sectors=%u"},
1931         {Opt_nl_reply_supported, "nl_reply_supported=%d"},
1932         {Opt_max_data_area_mb, "max_data_area_mb=%u"},
1933         {Opt_err, NULL}
1934 };
1935
1936 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
1937 {
1938         unsigned long tmp_ul;
1939         char *arg_p;
1940         int ret;
1941
1942         arg_p = match_strdup(arg);
1943         if (!arg_p)
1944                 return -ENOMEM;
1945
1946         ret = kstrtoul(arg_p, 0, &tmp_ul);
1947         kfree(arg_p);
1948         if (ret < 0) {
1949                 pr_err("kstrtoul() failed for dev attrib\n");
1950                 return ret;
1951         }
1952         if (!tmp_ul) {
1953                 pr_err("dev attrib must be nonzero\n");
1954                 return -EINVAL;
1955         }
1956         *dev_attrib = tmp_ul;
1957         return 0;
1958 }
1959
1960 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1961                 const char *page, ssize_t count)
1962 {
1963         struct tcmu_dev *udev = TCMU_DEV(dev);
1964         char *orig, *ptr, *opts, *arg_p;
1965         substring_t args[MAX_OPT_ARGS];
1966         int ret = 0, token, tmpval;
1967
1968         opts = kstrdup(page, GFP_KERNEL);
1969         if (!opts)
1970                 return -ENOMEM;
1971
1972         orig = opts;
1973
1974         while ((ptr = strsep(&opts, ",\n")) != NULL) {
1975                 if (!*ptr)
1976                         continue;
1977
1978                 token = match_token(ptr, tokens, args);
1979                 switch (token) {
1980                 case Opt_dev_config:
1981                         if (match_strlcpy(udev->dev_config, &args[0],
1982                                           TCMU_CONFIG_LEN) == 0) {
1983                                 ret = -EINVAL;
1984                                 break;
1985                         }
1986                         pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
1987                         break;
1988                 case Opt_dev_size:
1989                         arg_p = match_strdup(&args[0]);
1990                         if (!arg_p) {
1991                                 ret = -ENOMEM;
1992                                 break;
1993                         }
1994                         ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
1995                         kfree(arg_p);
1996                         if (ret < 0)
1997                                 pr_err("kstrtoul() failed for dev_size=\n");
1998                         break;
1999                 case Opt_hw_block_size:
2000                         ret = tcmu_set_dev_attrib(&args[0],
2001                                         &(dev->dev_attrib.hw_block_size));
2002                         break;
2003                 case Opt_hw_max_sectors:
2004                         ret = tcmu_set_dev_attrib(&args[0],
2005                                         &(dev->dev_attrib.hw_max_sectors));
2006                         break;
2007                 case Opt_nl_reply_supported:
2008                         arg_p = match_strdup(&args[0]);
2009                         if (!arg_p) {
2010                                 ret = -ENOMEM;
2011                                 break;
2012                         }
2013                         ret = kstrtoint(arg_p, 0, &udev->nl_reply_supported);
2014                         kfree(arg_p);
2015                         if (ret < 0)
2016                                 pr_err("kstrtoint() failed for nl_reply_supported=\n");
2017                         break;
2018                 case Opt_max_data_area_mb:
2019                         if (dev->export_count) {
2020                                 pr_err("Unable to set max_data_area_mb while exports exist\n");
2021                                 ret = -EINVAL;
2022                                 break;
2023                         }
2024
2025                         arg_p = match_strdup(&args[0]);
2026                         if (!arg_p) {
2027                                 ret = -ENOMEM;
2028                                 break;
2029                         }
2030                         ret = kstrtoint(arg_p, 0, &tmpval);
2031                         kfree(arg_p);
2032                         if (ret < 0) {
2033                                 pr_err("kstrtoint() failed for max_data_area_mb=\n");
2034                                 break;
2035                         }
2036
2037                         if (tmpval <= 0) {
2038                                 pr_err("Invalid max_data_area %d\n", tmpval);
2039                                 ret = -EINVAL;
2040                                 break;
2041                         }
2042
2043                         udev->max_blocks = TCMU_MBS_TO_BLOCKS(tmpval);
2044                         if (udev->max_blocks > tcmu_global_max_blocks) {
2045                                 pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2046                                        tmpval,
2047                                        TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
2048                                 udev->max_blocks = tcmu_global_max_blocks;
2049                         }
2050                         break;
2051                 default:
2052                         break;
2053                 }
2054
2055                 if (ret)
2056                         break;
2057         }
2058
2059         kfree(orig);
2060         return (!ret) ? count : ret;
2061 }
2062
2063 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2064 {
2065         struct tcmu_dev *udev = TCMU_DEV(dev);
2066         ssize_t bl = 0;
2067
2068         bl = sprintf(b + bl, "Config: %s ",
2069                      udev->dev_config[0] ? udev->dev_config : "NULL");
2070         bl += sprintf(b + bl, "Size: %zu ", udev->dev_size);
2071         bl += sprintf(b + bl, "MaxDataAreaMB: %u\n",
2072                       TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2073
2074         return bl;
2075 }
2076
2077 static sector_t tcmu_get_blocks(struct se_device *dev)
2078 {
2079         struct tcmu_dev *udev = TCMU_DEV(dev);
2080
2081         return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2082                        dev->dev_attrib.block_size);
2083 }
2084
2085 static sense_reason_t
2086 tcmu_parse_cdb(struct se_cmd *cmd)
2087 {
2088         return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2089 }
2090
2091 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2092 {
2093         struct se_dev_attrib *da = container_of(to_config_group(item),
2094                                         struct se_dev_attrib, da_group);
2095         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2096
2097         return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2098 }
2099
2100 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2101                                        size_t count)
2102 {
2103         struct se_dev_attrib *da = container_of(to_config_group(item),
2104                                         struct se_dev_attrib, da_group);
2105         struct tcmu_dev *udev = container_of(da->da_dev,
2106                                         struct tcmu_dev, se_dev);
2107         u32 val;
2108         int ret;
2109
2110         if (da->da_dev->export_count) {
2111                 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2112                 return -EINVAL;
2113         }
2114
2115         ret = kstrtou32(page, 0, &val);
2116         if (ret < 0)
2117                 return ret;
2118
2119         udev->cmd_time_out = val * MSEC_PER_SEC;
2120         return count;
2121 }
2122 CONFIGFS_ATTR(tcmu_, cmd_time_out);
2123
2124 static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2125 {
2126         struct se_dev_attrib *da = container_of(to_config_group(item),
2127                                                 struct se_dev_attrib, da_group);
2128         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2129
2130         return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2131                         udev->qfull_time_out :
2132                         udev->qfull_time_out / MSEC_PER_SEC);
2133 }
2134
2135 static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2136                                          const char *page, size_t count)
2137 {
2138         struct se_dev_attrib *da = container_of(to_config_group(item),
2139                                         struct se_dev_attrib, da_group);
2140         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2141         s32 val;
2142         int ret;
2143
2144         ret = kstrtos32(page, 0, &val);
2145         if (ret < 0)
2146                 return ret;
2147
2148         if (val >= 0) {
2149                 udev->qfull_time_out = val * MSEC_PER_SEC;
2150         } else if (val == -1) {
2151                 udev->qfull_time_out = val;
2152         } else {
2153                 printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2154                 return -EINVAL;
2155         }
2156         return count;
2157 }
2158 CONFIGFS_ATTR(tcmu_, qfull_time_out);
2159
2160 static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2161 {
2162         struct se_dev_attrib *da = container_of(to_config_group(item),
2163                                                 struct se_dev_attrib, da_group);
2164         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2165
2166         return snprintf(page, PAGE_SIZE, "%u\n",
2167                         TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2168 }
2169 CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2170
2171 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2172 {
2173         struct se_dev_attrib *da = container_of(to_config_group(item),
2174                                                 struct se_dev_attrib, da_group);
2175         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2176
2177         return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2178 }
2179
2180 static int tcmu_send_dev_config_event(struct tcmu_dev *udev,
2181                                       const char *reconfig_data)
2182 {
2183         struct sk_buff *skb = NULL;
2184         void *msg_header = NULL;
2185         int ret = 0;
2186
2187         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2188                                       &skb, &msg_header);
2189         if (ret < 0)
2190                 return ret;
2191         ret = nla_put_string(skb, TCMU_ATTR_DEV_CFG, reconfig_data);
2192         if (ret < 0) {
2193                 nlmsg_free(skb);
2194                 return ret;
2195         }
2196         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2197                                        &skb, &msg_header);
2198 }
2199
2200
2201 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2202                                      size_t count)
2203 {
2204         struct se_dev_attrib *da = container_of(to_config_group(item),
2205                                                 struct se_dev_attrib, da_group);
2206         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2207         int ret, len;
2208
2209         len = strlen(page);
2210         if (!len || len > TCMU_CONFIG_LEN - 1)
2211                 return -EINVAL;
2212
2213         /* Check if device has been configured before */
2214         if (tcmu_dev_configured(udev)) {
2215                 ret = tcmu_send_dev_config_event(udev, page);
2216                 if (ret) {
2217                         pr_err("Unable to reconfigure device\n");
2218                         return ret;
2219                 }
2220                 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2221
2222                 ret = tcmu_update_uio_info(udev);
2223                 if (ret)
2224                         return ret;
2225                 return count;
2226         }
2227         strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2228
2229         return count;
2230 }
2231 CONFIGFS_ATTR(tcmu_, dev_config);
2232
2233 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2234 {
2235         struct se_dev_attrib *da = container_of(to_config_group(item),
2236                                                 struct se_dev_attrib, da_group);
2237         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2238
2239         return snprintf(page, PAGE_SIZE, "%zu\n", udev->dev_size);
2240 }
2241
2242 static int tcmu_send_dev_size_event(struct tcmu_dev *udev, u64 size)
2243 {
2244         struct sk_buff *skb = NULL;
2245         void *msg_header = NULL;
2246         int ret = 0;
2247
2248         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2249                                       &skb, &msg_header);
2250         if (ret < 0)
2251                 return ret;
2252         ret = nla_put_u64_64bit(skb, TCMU_ATTR_DEV_SIZE,
2253                                 size, TCMU_ATTR_PAD);
2254         if (ret < 0) {
2255                 nlmsg_free(skb);
2256                 return ret;
2257         }
2258         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2259                                        &skb, &msg_header);
2260 }
2261
2262 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2263                                    size_t count)
2264 {
2265         struct se_dev_attrib *da = container_of(to_config_group(item),
2266                                                 struct se_dev_attrib, da_group);
2267         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2268         u64 val;
2269         int ret;
2270
2271         ret = kstrtou64(page, 0, &val);
2272         if (ret < 0)
2273                 return ret;
2274
2275         /* Check if device has been configured before */
2276         if (tcmu_dev_configured(udev)) {
2277                 ret = tcmu_send_dev_size_event(udev, val);
2278                 if (ret) {
2279                         pr_err("Unable to reconfigure device\n");
2280                         return ret;
2281                 }
2282         }
2283         udev->dev_size = val;
2284         return count;
2285 }
2286 CONFIGFS_ATTR(tcmu_, dev_size);
2287
2288 static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
2289                 char *page)
2290 {
2291         struct se_dev_attrib *da = container_of(to_config_group(item),
2292                                                 struct se_dev_attrib, da_group);
2293         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2294
2295         return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2296 }
2297
2298 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2299                 const char *page, size_t count)
2300 {
2301         struct se_dev_attrib *da = container_of(to_config_group(item),
2302                                                 struct se_dev_attrib, da_group);
2303         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2304         s8 val;
2305         int ret;
2306
2307         ret = kstrtos8(page, 0, &val);
2308         if (ret < 0)
2309                 return ret;
2310
2311         udev->nl_reply_supported = val;
2312         return count;
2313 }
2314 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2315
2316 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2317                                              char *page)
2318 {
2319         struct se_dev_attrib *da = container_of(to_config_group(item),
2320                                         struct se_dev_attrib, da_group);
2321
2322         return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2323 }
2324
2325 static int tcmu_send_emulate_write_cache(struct tcmu_dev *udev, u8 val)
2326 {
2327         struct sk_buff *skb = NULL;
2328         void *msg_header = NULL;
2329         int ret = 0;
2330
2331         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2332                                       &skb, &msg_header);
2333         if (ret < 0)
2334                 return ret;
2335         ret = nla_put_u8(skb, TCMU_ATTR_WRITECACHE, val);
2336         if (ret < 0) {
2337                 nlmsg_free(skb);
2338                 return ret;
2339         }
2340         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2341                                        &skb, &msg_header);
2342 }
2343
2344 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2345                                               const char *page, size_t count)
2346 {
2347         struct se_dev_attrib *da = container_of(to_config_group(item),
2348                                         struct se_dev_attrib, da_group);
2349         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2350         u8 val;
2351         int ret;
2352
2353         ret = kstrtou8(page, 0, &val);
2354         if (ret < 0)
2355                 return ret;
2356
2357         /* Check if device has been configured before */
2358         if (tcmu_dev_configured(udev)) {
2359                 ret = tcmu_send_emulate_write_cache(udev, val);
2360                 if (ret) {
2361                         pr_err("Unable to reconfigure device\n");
2362                         return ret;
2363                 }
2364         }
2365
2366         da->emulate_write_cache = val;
2367         return count;
2368 }
2369 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2370
2371 static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
2372 {
2373         struct se_device *se_dev = container_of(to_config_group(item),
2374                                                 struct se_device,
2375                                                 dev_action_group);
2376         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2377
2378         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2379                 return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
2380         else
2381                 return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
2382 }
2383
2384 static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
2385                                     size_t count)
2386 {
2387         struct se_device *se_dev = container_of(to_config_group(item),
2388                                                 struct se_device,
2389                                                 dev_action_group);
2390         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2391         u8 val;
2392         int ret;
2393
2394         ret = kstrtou8(page, 0, &val);
2395         if (ret < 0)
2396                 return ret;
2397
2398         if (val > 1) {
2399                 pr_err("Invalid block value %d\n", val);
2400                 return -EINVAL;
2401         }
2402
2403         if (!val)
2404                 tcmu_unblock_dev(udev);
2405         else
2406                 tcmu_block_dev(udev);
2407         return count;
2408 }
2409 CONFIGFS_ATTR(tcmu_, block_dev);
2410
2411 static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
2412                                      size_t count)
2413 {
2414         struct se_device *se_dev = container_of(to_config_group(item),
2415                                                 struct se_device,
2416                                                 dev_action_group);
2417         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2418         u8 val;
2419         int ret;
2420
2421         ret = kstrtou8(page, 0, &val);
2422         if (ret < 0)
2423                 return ret;
2424
2425         if (val != 1 && val != 2) {
2426                 pr_err("Invalid reset ring value %d\n", val);
2427                 return -EINVAL;
2428         }
2429
2430         tcmu_reset_ring(udev, val);
2431         return count;
2432 }
2433 CONFIGFS_ATTR_WO(tcmu_, reset_ring);
2434
2435 static struct configfs_attribute *tcmu_attrib_attrs[] = {
2436         &tcmu_attr_cmd_time_out,
2437         &tcmu_attr_qfull_time_out,
2438         &tcmu_attr_max_data_area_mb,
2439         &tcmu_attr_dev_config,
2440         &tcmu_attr_dev_size,
2441         &tcmu_attr_emulate_write_cache,
2442         &tcmu_attr_nl_reply_supported,
2443         NULL,
2444 };
2445
2446 static struct configfs_attribute **tcmu_attrs;
2447
2448 static struct configfs_attribute *tcmu_action_attrs[] = {
2449         &tcmu_attr_block_dev,
2450         &tcmu_attr_reset_ring,
2451         NULL,
2452 };
2453
2454 static struct target_backend_ops tcmu_ops = {
2455         .name                   = "user",
2456         .owner                  = THIS_MODULE,
2457         .transport_flags        = TRANSPORT_FLAG_PASSTHROUGH,
2458         .attach_hba             = tcmu_attach_hba,
2459         .detach_hba             = tcmu_detach_hba,
2460         .alloc_device           = tcmu_alloc_device,
2461         .configure_device       = tcmu_configure_device,
2462         .destroy_device         = tcmu_destroy_device,
2463         .free_device            = tcmu_free_device,
2464         .parse_cdb              = tcmu_parse_cdb,
2465         .set_configfs_dev_params = tcmu_set_configfs_dev_params,
2466         .show_configfs_dev_params = tcmu_show_configfs_dev_params,
2467         .get_device_type        = sbc_get_device_type,
2468         .get_blocks             = tcmu_get_blocks,
2469         .tb_dev_action_attrs    = tcmu_action_attrs,
2470 };
2471
2472 static void find_free_blocks(void)
2473 {
2474         struct tcmu_dev *udev;
2475         loff_t off;
2476         u32 start, end, block, total_freed = 0;
2477
2478         if (atomic_read(&global_db_count) <= tcmu_global_max_blocks)
2479                 return;
2480
2481         mutex_lock(&root_udev_mutex);
2482         list_for_each_entry(udev, &root_udev, node) {
2483                 mutex_lock(&udev->cmdr_lock);
2484
2485                 /* Try to complete the finished commands first */
2486                 tcmu_handle_completions(udev);
2487
2488                 /* Skip the udevs in idle */
2489                 if (!udev->dbi_thresh) {
2490                         mutex_unlock(&udev->cmdr_lock);
2491                         continue;
2492                 }
2493
2494                 end = udev->dbi_max + 1;
2495                 block = find_last_bit(udev->data_bitmap, end);
2496                 if (block == udev->dbi_max) {
2497                         /*
2498                          * The last bit is dbi_max, so it is not possible
2499                          * reclaim any blocks.
2500                          */
2501                         mutex_unlock(&udev->cmdr_lock);
2502                         continue;
2503                 } else if (block == end) {
2504                         /* The current udev will goto idle state */
2505                         udev->dbi_thresh = start = 0;
2506                         udev->dbi_max = 0;
2507                 } else {
2508                         udev->dbi_thresh = start = block + 1;
2509                         udev->dbi_max = block;
2510                 }
2511
2512                 /* Here will truncate the data area from off */
2513                 off = udev->data_off + start * DATA_BLOCK_SIZE;
2514                 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
2515
2516                 /* Release the block pages */
2517                 tcmu_blocks_release(&udev->data_blocks, start, end);
2518                 mutex_unlock(&udev->cmdr_lock);
2519
2520                 total_freed += end - start;
2521                 pr_debug("Freed %u blocks (total %u) from %s.\n", end - start,
2522                          total_freed, udev->name);
2523         }
2524         mutex_unlock(&root_udev_mutex);
2525
2526         if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
2527                 schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
2528 }
2529
2530 static void check_timedout_devices(void)
2531 {
2532         struct tcmu_dev *udev, *tmp_dev;
2533         LIST_HEAD(devs);
2534
2535         spin_lock_bh(&timed_out_udevs_lock);
2536         list_splice_init(&timed_out_udevs, &devs);
2537
2538         list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
2539                 list_del_init(&udev->timedout_entry);
2540                 spin_unlock_bh(&timed_out_udevs_lock);
2541
2542                 mutex_lock(&udev->cmdr_lock);
2543                 idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
2544                 mutex_unlock(&udev->cmdr_lock);
2545
2546                 spin_lock_bh(&timed_out_udevs_lock);
2547         }
2548
2549         spin_unlock_bh(&timed_out_udevs_lock);
2550 }
2551
2552 static void tcmu_unmap_work_fn(struct work_struct *work)
2553 {
2554         check_timedout_devices();
2555         find_free_blocks();
2556 }
2557
2558 static int __init tcmu_module_init(void)
2559 {
2560         int ret, i, k, len = 0;
2561
2562         BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2563
2564         INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
2565
2566         tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2567                                 sizeof(struct tcmu_cmd),
2568                                 __alignof__(struct tcmu_cmd),
2569                                 0, NULL);
2570         if (!tcmu_cmd_cache)
2571                 return -ENOMEM;
2572
2573         tcmu_root_device = root_device_register("tcm_user");
2574         if (IS_ERR(tcmu_root_device)) {
2575                 ret = PTR_ERR(tcmu_root_device);
2576                 goto out_free_cache;
2577         }
2578
2579         ret = genl_register_family(&tcmu_genl_family);
2580         if (ret < 0) {
2581                 goto out_unreg_device;
2582         }
2583
2584         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2585                 len += sizeof(struct configfs_attribute *);
2586         }
2587         for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2588                 len += sizeof(struct configfs_attribute *);
2589         }
2590         len += sizeof(struct configfs_attribute *);
2591
2592         tcmu_attrs = kzalloc(len, GFP_KERNEL);
2593         if (!tcmu_attrs) {
2594                 ret = -ENOMEM;
2595                 goto out_unreg_genl;
2596         }
2597
2598         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2599                 tcmu_attrs[i] = passthrough_attrib_attrs[i];
2600         }
2601         for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2602                 tcmu_attrs[i] = tcmu_attrib_attrs[k];
2603                 i++;
2604         }
2605         tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2606
2607         ret = transport_backend_register(&tcmu_ops);
2608         if (ret)
2609                 goto out_attrs;
2610
2611         return 0;
2612
2613 out_attrs:
2614         kfree(tcmu_attrs);
2615 out_unreg_genl:
2616         genl_unregister_family(&tcmu_genl_family);
2617 out_unreg_device:
2618         root_device_unregister(tcmu_root_device);
2619 out_free_cache:
2620         kmem_cache_destroy(tcmu_cmd_cache);
2621
2622         return ret;
2623 }
2624
2625 static void __exit tcmu_module_exit(void)
2626 {
2627         cancel_delayed_work_sync(&tcmu_unmap_work);
2628         target_backend_unregister(&tcmu_ops);
2629         kfree(tcmu_attrs);
2630         genl_unregister_family(&tcmu_genl_family);
2631         root_device_unregister(tcmu_root_device);
2632         kmem_cache_destroy(tcmu_cmd_cache);
2633 }
2634
2635 MODULE_DESCRIPTION("TCM USER subsystem plugin");
2636 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2637 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2638 MODULE_LICENSE("GPL");
2639
2640 module_init(tcmu_module_init);
2641 module_exit(tcmu_module_exit);