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, uint32_t read_len)
 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 && read_len > 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                         if (read_len < copy_bytes)
 705                                 copy_bytes = read_len;
 706                         offset = DATA_BLOCK_SIZE - block_remaining;
 707                         tcmu_flush_dcache_range(from, copy_bytes);
 708                         memcpy(to + sg->length - sg_remaining, from + offset,
 709                                         copy_bytes);
 710
 711                         sg_remaining -= copy_bytes;
 712                         block_remaining -= copy_bytes;
 713                         read_len -= copy_bytes;
 714                 }
 715                 kunmap_atomic(to - sg->offset);
 716                 if (read_len == 0)
 717                         break;
 718         }
 719         if (from)
 720                 kunmap_atomic(from);
 721 }
 722
 723 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
 724 {
 725         return thresh - bitmap_weight(bitmap, thresh);
 726 }
 727
 728 /*
 729  * We can't queue a command until we have space available on the cmd ring *and*
 730  * space available on the data area.
 731  *
 732  * Called with ring lock held.
 733  */
 734 static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 735                 size_t cmd_size, size_t data_needed)
 736 {
 737         struct tcmu_mailbox *mb = udev->mb_addr;
 738         uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
 739                                 / DATA_BLOCK_SIZE;
 740         size_t space, cmd_needed;
 741         u32 cmd_head;
 742
 743         tcmu_flush_dcache_range(mb, sizeof(*mb));
 744
 745         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 746
 747         /*
 748          * If cmd end-of-ring space is too small then we need space for a NOP plus
 749          * original cmd - cmds are internally contiguous.
 750          */
 751         if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
 752                 cmd_needed = cmd_size;
 753         else
 754                 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
 755
 756         space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
 757         if (space < cmd_needed) {
 758                 pr_debug("no cmd space: %u %u %u\n", cmd_head,
 759                        udev->cmdr_last_cleaned, udev->cmdr_size);
 760                 return false;
 761         }
 762
 763         /* try to check and get the data blocks as needed */
 764         space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
 765         if ((space * DATA_BLOCK_SIZE) < data_needed) {
 766                 unsigned long blocks_left =
 767                                 (udev->max_blocks - udev->dbi_thresh) + space;
 768
 769                 if (blocks_left < blocks_needed) {
 770                         pr_debug("no data space: only %lu available, but ask for %zu\n",
 771                                         blocks_left * DATA_BLOCK_SIZE,
 772                                         data_needed);
 773                         return false;
 774                 }
 775
 776                 udev->dbi_thresh += blocks_needed;
 777                 if (udev->dbi_thresh > udev->max_blocks)
 778                         udev->dbi_thresh = udev->max_blocks;
 779         }
 780
 781         return tcmu_get_empty_blocks(udev, cmd);
 782 }
 783
 784 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
 785 {
 786         return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 787                         sizeof(struct tcmu_cmd_entry));
 788 }
 789
 790 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
 791                                            size_t base_command_size)
 792 {
 793         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 794         size_t command_size;
 795
 796         command_size = base_command_size +
 797                 round_up(scsi_command_size(se_cmd->t_task_cdb),
 798                                 TCMU_OP_ALIGN_SIZE);
 799
 800         WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
 801
 802         return command_size;
 803 }
 804
 805 static int tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
 806                                 struct timer_list *timer)
 807 {
 808         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 809         int cmd_id;
 810
 811         if (tcmu_cmd->cmd_id)
 812                 goto setup_timer;
 813
 814         cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 1, USHRT_MAX, GFP_NOWAIT);
 815         if (cmd_id < 0) {
 816                 pr_err("tcmu: Could not allocate cmd id.\n");
 817                 return cmd_id;
 818         }
 819         tcmu_cmd->cmd_id = cmd_id;
 820
 821         pr_debug("allocated cmd %u for dev %s tmo %lu\n", tcmu_cmd->cmd_id,
 822                  udev->name, tmo / MSEC_PER_SEC);
 823
 824 setup_timer:
 825         if (!tmo)
 826                 return 0;
 827
 828         tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
 829         mod_timer(timer, tcmu_cmd->deadline);
 830         return 0;
 831 }
 832
 833 static int add_to_cmdr_queue(struct tcmu_cmd *tcmu_cmd)
 834 {
 835         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 836         unsigned int tmo;
 837         int ret;
 838
 839         /*
 840          * For backwards compat if qfull_time_out is not set use
 841          * cmd_time_out and if that's not set use the default time out.
 842          */
 843         if (!udev->qfull_time_out)
 844                 return -ETIMEDOUT;
 845         else if (udev->qfull_time_out > 0)
 846                 tmo = udev->qfull_time_out;
 847         else if (udev->cmd_time_out)
 848                 tmo = udev->cmd_time_out;
 849         else
 850                 tmo = TCMU_TIME_OUT;
 851
 852         ret = tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
 853         if (ret)
 854                 return ret;
 855
 856         list_add_tail(&tcmu_cmd->cmdr_queue_entry, &udev->cmdr_queue);
 857         pr_debug("adding cmd %u on dev %s to ring space wait queue\n",
 858                  tcmu_cmd->cmd_id, udev->name);
 859         return 0;
 860 }
 861
 862 /**
 863  * queue_cmd_ring - queue cmd to ring or internally
 864  * @tcmu_cmd: cmd to queue
 865  * @scsi_err: TCM error code if failure (-1) returned.
 866  *
 867  * Returns:
 868  * -1 we cannot queue internally or to the ring.
 869  *  0 success
 870  *  1 internally queued to wait for ring memory to free.
 871  */
 872 static sense_reason_t queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, int *scsi_err)
 873 {
 874         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 875         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 876         size_t base_command_size, command_size;
 877         struct tcmu_mailbox *mb;
 878         struct tcmu_cmd_entry *entry;
 879         struct iovec *iov;
 880         int iov_cnt, ret;
 881         uint32_t cmd_head;
 882         uint64_t cdb_off;
 883         bool copy_to_data_area;
 884         size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
 885
 886         *scsi_err = TCM_NO_SENSE;
 887
 888         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
 889                 *scsi_err = TCM_LUN_BUSY;
 890                 return -1;
 891         }
 892
 893         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
 894                 *scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 895                 return -1;
 896         }
 897
 898         /*
 899          * Must be a certain minimum size for response sense info, but
 900          * also may be larger if the iov array is large.
 901          *
 902          * We prepare as many iovs as possbile for potential uses here,
 903          * because it's expensive to tell how many regions are freed in
 904          * the bitmap & global data pool, as the size calculated here
 905          * will only be used to do the checks.
 906          *
 907          * The size will be recalculated later as actually needed to save
 908          * cmd area memories.
 909          */
 910         base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
 911         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 912
 913         if (!list_empty(&udev->cmdr_queue))
 914                 goto queue;
 915
 916         mb = udev->mb_addr;
 917         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 918         if ((command_size > (udev->cmdr_size / 2)) ||
 919             data_length > udev->data_size) {
 920                 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
 921                         "cmd ring/data area\n", command_size, data_length,
 922                         udev->cmdr_size, udev->data_size);
 923                 *scsi_err = TCM_INVALID_CDB_FIELD;
 924                 return -1;
 925         }
 926
 927         if (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
 928                 /*
 929                  * Don't leave commands partially setup because the unmap
 930                  * thread might need the blocks to make forward progress.
 931                  */
 932                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
 933                 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 934                 goto queue;
 935         }
 936
 937         /* Insert a PAD if end-of-ring space is too small */
 938         if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
 939                 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
 940
 941                 entry = (void *) mb + CMDR_OFF + cmd_head;
 942                 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
 943                 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
 944                 entry->hdr.cmd_id = 0; /* not used for PAD */
 945                 entry->hdr.kflags = 0;
 946                 entry->hdr.uflags = 0;
 947                 tcmu_flush_dcache_range(entry, sizeof(*entry));
 948
 949                 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
 950                 tcmu_flush_dcache_range(mb, sizeof(*mb));
 951
 952                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 953                 WARN_ON(cmd_head != 0);
 954         }
 955
 956         entry = (void *) mb + CMDR_OFF + cmd_head;
 957         memset(entry, 0, command_size);
 958         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
 959
 960         /* Handle allocating space from the data area */
 961         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 962         iov = &entry->req.iov[0];
 963         iov_cnt = 0;
 964         copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
 965                 || se_cmd->se_cmd_flags & SCF_BIDI);
 966         scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
 967                           se_cmd->t_data_nents, &iov, &iov_cnt,
 968                           copy_to_data_area);
 969         entry->req.iov_cnt = iov_cnt;
 970
 971         /* Handle BIDI commands */
 972         iov_cnt = 0;
 973         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 974                 iov++;
 975                 scatter_data_area(udev, tcmu_cmd, se_cmd->t_bidi_data_sg,
 976                                   se_cmd->t_bidi_data_nents, &iov, &iov_cnt,
 977                                   false);
 978         }
 979         entry->req.iov_bidi_cnt = iov_cnt;
 980
 981         ret = tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out,
 982                                    &udev->cmd_timer);
 983         if (ret) {
 984                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
 985                 mutex_unlock(&udev->cmdr_lock);
 986
 987                 *scsi_err = TCM_OUT_OF_RESOURCES;
 988                 return -1;
 989         }
 990         entry->hdr.cmd_id = tcmu_cmd->cmd_id;
 991
 992         /*
 993          * Recalaulate the command's base size and size according
 994          * to the actual needs
 995          */
 996         base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
 997                                                        entry->req.iov_bidi_cnt);
 998         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 999
1000         tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
1001
1002         /* All offsets relative to mb_addr, not start of entry! */
1003         cdb_off = CMDR_OFF + cmd_head + base_command_size;
1004         memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
1005         entry->req.cdb_off = cdb_off;
1006         tcmu_flush_dcache_range(entry, sizeof(*entry));
1007
1008         UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1009         tcmu_flush_dcache_range(mb, sizeof(*mb));
1010
1011         /* TODO: only if FLUSH and FUA? */
1012         uio_event_notify(&udev->uio_info);
1013
1014         return 0;
1015
1016 queue:
1017         if (add_to_cmdr_queue(tcmu_cmd)) {
1018                 *scsi_err = TCM_OUT_OF_RESOURCES;
1019                 return -1;
1020         }
1021
1022         return 1;
1023 }
1024
1025 static sense_reason_t
1026 tcmu_queue_cmd(struct se_cmd *se_cmd)
1027 {
1028         struct se_device *se_dev = se_cmd->se_dev;
1029         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1030         struct tcmu_cmd *tcmu_cmd;
1031         sense_reason_t scsi_ret;
1032         int ret;
1033
1034         tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1035         if (!tcmu_cmd)
1036                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1037
1038         mutex_lock(&udev->cmdr_lock);
1039         ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1040         mutex_unlock(&udev->cmdr_lock);
1041         if (ret < 0)
1042                 tcmu_free_cmd(tcmu_cmd);
1043         return scsi_ret;
1044 }
1045
1046 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
1047 {
1048         struct se_cmd *se_cmd = cmd->se_cmd;
1049         struct tcmu_dev *udev = cmd->tcmu_dev;
1050         bool read_len_valid = false;
1051         uint32_t read_len = se_cmd->data_length;
1052
1053         /*
1054          * cmd has been completed already from timeout, just reclaim
1055          * data area space and free cmd
1056          */
1057         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1058                 goto out;
1059
1060         tcmu_cmd_reset_dbi_cur(cmd);
1061
1062         if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1063                 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1064                         cmd->se_cmd);
1065                 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1066                 goto done;
1067         }
1068
1069         if (se_cmd->data_direction == DMA_FROM_DEVICE &&
1070             (entry->hdr.uflags & TCMU_UFLAG_READ_LEN) && entry->rsp.read_len) {
1071                 read_len_valid = true;
1072                 if (entry->rsp.read_len < read_len)
1073                         read_len = entry->rsp.read_len;
1074         }
1075
1076         if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1077                 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1078                 if (!read_len_valid )
1079                         goto done;
1080                 else
1081                         se_cmd->se_cmd_flags |= SCF_TREAT_READ_AS_NORMAL;
1082         }
1083         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1084                 /* Get Data-In buffer before clean up */
1085                 gather_data_area(udev, cmd, true, read_len);
1086         } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1087                 gather_data_area(udev, cmd, false, read_len);
1088         } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1089                 /* TODO: */
1090         } else if (se_cmd->data_direction != DMA_NONE) {
1091                 pr_warn("TCMU: data direction was %d!\n",
1092                         se_cmd->data_direction);
1093         }
1094
1095 done:
1096         if (read_len_valid) {
1097                 pr_debug("read_len = %d\n", read_len);
1098                 target_complete_cmd_with_length(cmd->se_cmd,
1099                                         entry->rsp.scsi_status, read_len);
1100         } else
1101                 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1102
1103 out:
1104         cmd->se_cmd = NULL;
1105         tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1106         tcmu_free_cmd(cmd);
1107 }
1108
1109 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
1110 {
1111         struct tcmu_mailbox *mb;
1112         int handled = 0;
1113
1114         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1115                 pr_err("ring broken, not handling completions\n");
1116                 return 0;
1117         }
1118
1119         mb = udev->mb_addr;
1120         tcmu_flush_dcache_range(mb, sizeof(*mb));
1121
1122         while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1123
1124                 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
1125                 struct tcmu_cmd *cmd;
1126
1127                 tcmu_flush_dcache_range(entry, sizeof(*entry));
1128
1129                 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
1130                         UPDATE_HEAD(udev->cmdr_last_cleaned,
1131                                     tcmu_hdr_get_len(entry->hdr.len_op),
1132                                     udev->cmdr_size);
1133                         continue;
1134                 }
1135                 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1136
1137                 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1138                 if (!cmd) {
1139                         pr_err("cmd_id %u not found, ring is broken\n",
1140                                entry->hdr.cmd_id);
1141                         set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1142                         break;
1143                 }
1144
1145                 tcmu_handle_completion(cmd, entry);
1146
1147                 UPDATE_HEAD(udev->cmdr_last_cleaned,
1148                             tcmu_hdr_get_len(entry->hdr.len_op),
1149                             udev->cmdr_size);
1150
1151                 handled++;
1152         }
1153
1154         if (mb->cmd_tail == mb->cmd_head) {
1155                 /* no more pending commands */
1156                 del_timer(&udev->cmd_timer);
1157
1158                 if (list_empty(&udev->cmdr_queue)) {
1159                         /*
1160                          * no more pending or waiting commands so try to
1161                          * reclaim blocks if needed.
1162                          */
1163                         if (atomic_read(&global_db_count) >
1164                             tcmu_global_max_blocks)
1165                                 schedule_delayed_work(&tcmu_unmap_work, 0);
1166                 }
1167         }
1168
1169         return handled;
1170 }
1171
1172 static int tcmu_check_expired_cmd(int id, void *p, void *data)
1173 {
1174         struct tcmu_cmd *cmd = p;
1175         struct tcmu_dev *udev = cmd->tcmu_dev;
1176         u8 scsi_status;
1177         struct se_cmd *se_cmd;
1178         bool is_running;
1179
1180         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1181                 return 0;
1182
1183         if (!time_after(jiffies, cmd->deadline))
1184                 return 0;
1185
1186         is_running = list_empty(&cmd->cmdr_queue_entry);
1187         se_cmd = cmd->se_cmd;
1188
1189         if (is_running) {
1190                 /*
1191                  * If cmd_time_out is disabled but qfull is set deadline
1192                  * will only reflect the qfull timeout. Ignore it.
1193                  */
1194                 if (!udev->cmd_time_out)
1195                         return 0;
1196
1197                 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1198                 /*
1199                  * target_complete_cmd will translate this to LUN COMM FAILURE
1200                  */
1201                 scsi_status = SAM_STAT_CHECK_CONDITION;
1202         } else {
1203                 list_del_init(&cmd->cmdr_queue_entry);
1204
1205                 idr_remove(&udev->commands, id);
1206                 tcmu_free_cmd(cmd);
1207                 scsi_status = SAM_STAT_TASK_SET_FULL;
1208         }
1209
1210         pr_debug("Timing out cmd %u on dev %s that is %s.\n",
1211                  id, udev->name, is_running ? "inflight" : "queued");
1212
1213         target_complete_cmd(se_cmd, scsi_status);
1214         return 0;
1215 }
1216
1217 static void tcmu_device_timedout(struct tcmu_dev *udev)
1218 {
1219         spin_lock(&timed_out_udevs_lock);
1220         if (list_empty(&udev->timedout_entry))
1221                 list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1222         spin_unlock(&timed_out_udevs_lock);
1223
1224         schedule_delayed_work(&tcmu_unmap_work, 0);
1225 }
1226
1227 static void tcmu_cmd_timedout(struct timer_list *t)
1228 {
1229         struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1230
1231         pr_debug("%s cmd timeout has expired\n", udev->name);
1232         tcmu_device_timedout(udev);
1233 }
1234
1235 static void tcmu_qfull_timedout(struct timer_list *t)
1236 {
1237         struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1238
1239         pr_debug("%s qfull timeout has expired\n", udev->name);
1240         tcmu_device_timedout(udev);
1241 }
1242
1243 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1244 {
1245         struct tcmu_hba *tcmu_hba;
1246
1247         tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1248         if (!tcmu_hba)
1249                 return -ENOMEM;
1250
1251         tcmu_hba->host_id = host_id;
1252         hba->hba_ptr = tcmu_hba;
1253
1254         return 0;
1255 }
1256
1257 static void tcmu_detach_hba(struct se_hba *hba)
1258 {
1259         kfree(hba->hba_ptr);
1260         hba->hba_ptr = NULL;
1261 }
1262
1263 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1264 {
1265         struct tcmu_dev *udev;
1266
1267         udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1268         if (!udev)
1269                 return NULL;
1270         kref_init(&udev->kref);
1271
1272         udev->name = kstrdup(name, GFP_KERNEL);
1273         if (!udev->name) {
1274                 kfree(udev);
1275                 return NULL;
1276         }
1277
1278         udev->hba = hba;
1279         udev->cmd_time_out = TCMU_TIME_OUT;
1280         udev->qfull_time_out = -1;
1281
1282         udev->max_blocks = DATA_BLOCK_BITS_DEF;
1283         mutex_init(&udev->cmdr_lock);
1284
1285         INIT_LIST_HEAD(&udev->timedout_entry);
1286         INIT_LIST_HEAD(&udev->cmdr_queue);
1287         idr_init(&udev->commands);
1288
1289         timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1290         timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1291
1292         init_waitqueue_head(&udev->nl_cmd_wq);
1293         spin_lock_init(&udev->nl_cmd_lock);
1294
1295         INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1296
1297         return &udev->se_dev;
1298 }
1299
1300 static bool run_cmdr_queue(struct tcmu_dev *udev, bool fail)
1301 {
1302         struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1303         LIST_HEAD(cmds);
1304         bool drained = true;
1305         sense_reason_t scsi_ret;
1306         int ret;
1307
1308         if (list_empty(&udev->cmdr_queue))
1309                 return true;
1310
1311         pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1312
1313         list_splice_init(&udev->cmdr_queue, &cmds);
1314
1315         list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, cmdr_queue_entry) {
1316                 list_del_init(&tcmu_cmd->cmdr_queue_entry);
1317
1318                 pr_debug("removing cmd %u on dev %s from queue\n",
1319                          tcmu_cmd->cmd_id, udev->name);
1320
1321                 if (fail) {
1322                         idr_remove(&udev->commands, tcmu_cmd->cmd_id);
1323                         /*
1324                          * We were not able to even start the command, so
1325                          * fail with busy to allow a retry in case runner
1326                          * was only temporarily down. If the device is being
1327                          * removed then LIO core will do the right thing and
1328                          * fail the retry.
1329                          */
1330                         target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1331                         tcmu_free_cmd(tcmu_cmd);
1332                         continue;
1333                 }
1334
1335                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1336                 if (ret < 0) {
1337                         pr_debug("cmd %u on dev %s failed with %u\n",
1338                                  tcmu_cmd->cmd_id, udev->name, scsi_ret);
1339
1340                         idr_remove(&udev->commands, tcmu_cmd->cmd_id);
1341                         /*
1342                          * Ignore scsi_ret for now. target_complete_cmd
1343                          * drops it.
1344                          */
1345                         target_complete_cmd(tcmu_cmd->se_cmd,
1346                                             SAM_STAT_CHECK_CONDITION);
1347                         tcmu_free_cmd(tcmu_cmd);
1348                 } else if (ret > 0) {
1349                         pr_debug("ran out of space during cmdr queue run\n");
1350                         /*
1351                          * cmd was requeued, so just put all cmds back in
1352                          * the queue
1353                          */
1354                         list_splice_tail(&cmds, &udev->cmdr_queue);
1355                         drained = false;
1356                         goto done;
1357                 }
1358         }
1359         if (list_empty(&udev->cmdr_queue))
1360                 del_timer(&udev->qfull_timer);
1361 done:
1362         return drained;
1363 }
1364
1365 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1366 {
1367         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1368
1369         mutex_lock(&udev->cmdr_lock);
1370         tcmu_handle_completions(udev);
1371         run_cmdr_queue(udev, false);
1372         mutex_unlock(&udev->cmdr_lock);
1373
1374         return 0;
1375 }
1376
1377 /*
1378  * mmap code from uio.c. Copied here because we want to hook mmap()
1379  * and this stuff must come along.
1380  */
1381 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1382 {
1383         struct tcmu_dev *udev = vma->vm_private_data;
1384         struct uio_info *info = &udev->uio_info;
1385
1386         if (vma->vm_pgoff < MAX_UIO_MAPS) {
1387                 if (info->mem[vma->vm_pgoff].size == 0)
1388                         return -1;
1389                 return (int)vma->vm_pgoff;
1390         }
1391         return -1;
1392 }
1393
1394 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1395 {
1396         struct page *page;
1397
1398         mutex_lock(&udev->cmdr_lock);
1399         page = tcmu_get_block_page(udev, dbi);
1400         if (likely(page)) {
1401                 mutex_unlock(&udev->cmdr_lock);
1402                 return page;
1403         }
1404
1405         /*
1406          * Userspace messed up and passed in a address not in the
1407          * data iov passed to it.
1408          */
1409         pr_err("Invalid addr to data block mapping  (dbi %u) on device %s\n",
1410                dbi, udev->name);
1411         page = NULL;
1412         mutex_unlock(&udev->cmdr_lock);
1413
1414         return page;
1415 }
1416
1417 static vm_fault_t tcmu_vma_fault(struct vm_fault *vmf)
1418 {
1419         struct tcmu_dev *udev = vmf->vma->vm_private_data;
1420         struct uio_info *info = &udev->uio_info;
1421         struct page *page;
1422         unsigned long offset;
1423         void *addr;
1424
1425         int mi = tcmu_find_mem_index(vmf->vma);
1426         if (mi < 0)
1427                 return VM_FAULT_SIGBUS;
1428
1429         /*
1430          * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1431          * to use mem[N].
1432          */
1433         offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1434
1435         if (offset < udev->data_off) {
1436                 /* For the vmalloc()ed cmd area pages */
1437                 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1438                 page = vmalloc_to_page(addr);
1439         } else {
1440                 uint32_t dbi;
1441
1442                 /* For the dynamically growing data area pages */
1443                 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1444                 page = tcmu_try_get_block_page(udev, dbi);
1445                 if (!page)
1446                         return VM_FAULT_SIGBUS;
1447         }
1448
1449         get_page(page);
1450         vmf->page = page;
1451         return 0;
1452 }
1453
1454 static const struct vm_operations_struct tcmu_vm_ops = {
1455         .fault = tcmu_vma_fault,
1456 };
1457
1458 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1459 {
1460         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1461
1462         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1463         vma->vm_ops = &tcmu_vm_ops;
1464
1465         vma->vm_private_data = udev;
1466
1467         /* Ensure the mmap is exactly the right size */
1468         if (vma_pages(vma) != (udev->ring_size >> PAGE_SHIFT))
1469                 return -EINVAL;
1470
1471         return 0;
1472 }
1473
1474 static int tcmu_open(struct uio_info *info, struct inode *inode)
1475 {
1476         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1477
1478         /* O_EXCL not supported for char devs, so fake it? */
1479         if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1480                 return -EBUSY;
1481
1482         udev->inode = inode;
1483         kref_get(&udev->kref);
1484
1485         pr_debug("open\n");
1486
1487         return 0;
1488 }
1489
1490 static void tcmu_dev_call_rcu(struct rcu_head *p)
1491 {
1492         struct se_device *dev = container_of(p, struct se_device, rcu_head);
1493         struct tcmu_dev *udev = TCMU_DEV(dev);
1494
1495         kfree(udev->uio_info.name);
1496         kfree(udev->name);
1497         kfree(udev);
1498 }
1499
1500 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1501 {
1502         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1503                 kmem_cache_free(tcmu_cmd_cache, cmd);
1504                 return 0;
1505         }
1506         return -EINVAL;
1507 }
1508
1509 static void tcmu_blocks_release(struct radix_tree_root *blocks,
1510                                 int start, int end)
1511 {
1512         int i;
1513         struct page *page;
1514
1515         for (i = start; i < end; i++) {
1516                 page = radix_tree_delete(blocks, i);
1517                 if (page) {
1518                         __free_page(page);
1519                         atomic_dec(&global_db_count);
1520                 }
1521         }
1522 }
1523
1524 static void tcmu_dev_kref_release(struct kref *kref)
1525 {
1526         struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1527         struct se_device *dev = &udev->se_dev;
1528         struct tcmu_cmd *cmd;
1529         bool all_expired = true;
1530         int i;
1531
1532         vfree(udev->mb_addr);
1533         udev->mb_addr = NULL;
1534
1535         spin_lock_bh(&timed_out_udevs_lock);
1536         if (!list_empty(&udev->timedout_entry))
1537                 list_del(&udev->timedout_entry);
1538         spin_unlock_bh(&timed_out_udevs_lock);
1539
1540         /* Upper layer should drain all requests before calling this */
1541         mutex_lock(&udev->cmdr_lock);
1542         idr_for_each_entry(&udev->commands, cmd, i) {
1543                 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1544                         all_expired = false;
1545         }
1546         idr_destroy(&udev->commands);
1547         WARN_ON(!all_expired);
1548
1549         tcmu_blocks_release(&udev->data_blocks, 0, udev->dbi_max + 1);
1550         kfree(udev->data_bitmap);
1551         mutex_unlock(&udev->cmdr_lock);
1552
1553         call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1554 }
1555
1556 static int tcmu_release(struct uio_info *info, struct inode *inode)
1557 {
1558         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1559
1560         clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1561
1562         pr_debug("close\n");
1563         /* release ref from open */
1564         kref_put(&udev->kref, tcmu_dev_kref_release);
1565         return 0;
1566 }
1567
1568 static void tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1569 {
1570         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1571
1572         if (!tcmu_kern_cmd_reply_supported)
1573                 return;
1574
1575         if (udev->nl_reply_supported <= 0)
1576                 return;
1577
1578 relock:
1579         spin_lock(&udev->nl_cmd_lock);
1580
1581         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1582                 spin_unlock(&udev->nl_cmd_lock);
1583                 pr_debug("sleeping for open nl cmd\n");
1584                 wait_event(udev->nl_cmd_wq, (nl_cmd->cmd == TCMU_CMD_UNSPEC));
1585                 goto relock;
1586         }
1587
1588         memset(nl_cmd, 0, sizeof(*nl_cmd));
1589         nl_cmd->cmd = cmd;
1590         init_completion(&nl_cmd->complete);
1591
1592         spin_unlock(&udev->nl_cmd_lock);
1593 }
1594
1595 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1596 {
1597         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1598         int ret;
1599         DEFINE_WAIT(__wait);
1600
1601         if (!tcmu_kern_cmd_reply_supported)
1602                 return 0;
1603
1604         if (udev->nl_reply_supported <= 0)
1605                 return 0;
1606
1607         pr_debug("sleeping for nl reply\n");
1608         wait_for_completion(&nl_cmd->complete);
1609
1610         spin_lock(&udev->nl_cmd_lock);
1611         nl_cmd->cmd = TCMU_CMD_UNSPEC;
1612         ret = nl_cmd->status;
1613         nl_cmd->status = 0;
1614         spin_unlock(&udev->nl_cmd_lock);
1615
1616         wake_up_all(&udev->nl_cmd_wq);
1617
1618         return ret;
1619 }
1620
1621 static int tcmu_netlink_event_init(struct tcmu_dev *udev,
1622                                    enum tcmu_genl_cmd cmd,
1623                                    struct sk_buff **buf, void **hdr)
1624 {
1625         struct sk_buff *skb;
1626         void *msg_header;
1627         int ret = -ENOMEM;
1628
1629         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1630         if (!skb)
1631                 return ret;
1632
1633         msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1634         if (!msg_header)
1635                 goto free_skb;
1636
1637         ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1638         if (ret < 0)
1639                 goto free_skb;
1640
1641         ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1642         if (ret < 0)
1643                 goto free_skb;
1644
1645         ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1646         if (ret < 0)
1647                 goto free_skb;
1648
1649         *buf = skb;
1650         *hdr = msg_header;
1651         return ret;
1652
1653 free_skb:
1654         nlmsg_free(skb);
1655         return ret;
1656 }
1657
1658 static int tcmu_netlink_event_send(struct tcmu_dev *udev,
1659                                    enum tcmu_genl_cmd cmd,
1660                                    struct sk_buff **buf, void **hdr)
1661 {
1662         int ret = 0;
1663         struct sk_buff *skb = *buf;
1664         void *msg_header = *hdr;
1665
1666         genlmsg_end(skb, msg_header);
1667
1668         tcmu_init_genl_cmd_reply(udev, cmd);
1669
1670         ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1671                                       TCMU_MCGRP_CONFIG, GFP_KERNEL);
1672        /* We don't care if no one is listening */
1673         if (ret == -ESRCH)
1674                 ret = 0;
1675         if (!ret)
1676                 ret = tcmu_wait_genl_cmd_reply(udev);
1677         return ret;
1678 }
1679
1680 static int tcmu_send_dev_add_event(struct tcmu_dev *udev)
1681 {
1682         struct sk_buff *skb = NULL;
1683         void *msg_header = NULL;
1684         int ret = 0;
1685
1686         ret = tcmu_netlink_event_init(udev, TCMU_CMD_ADDED_DEVICE, &skb,
1687                                       &msg_header);
1688         if (ret < 0)
1689                 return ret;
1690         return tcmu_netlink_event_send(udev, TCMU_CMD_ADDED_DEVICE, &skb,
1691                                        &msg_header);
1692
1693 }
1694
1695 static int tcmu_send_dev_remove_event(struct tcmu_dev *udev)
1696 {
1697         struct sk_buff *skb = NULL;
1698         void *msg_header = NULL;
1699         int ret = 0;
1700
1701         ret = tcmu_netlink_event_init(udev, TCMU_CMD_REMOVED_DEVICE,
1702                                       &skb, &msg_header);
1703         if (ret < 0)
1704                 return ret;
1705         return tcmu_netlink_event_send(udev, TCMU_CMD_REMOVED_DEVICE,
1706                                        &skb, &msg_header);
1707 }
1708
1709 static int tcmu_update_uio_info(struct tcmu_dev *udev)
1710 {
1711         struct tcmu_hba *hba = udev->hba->hba_ptr;
1712         struct uio_info *info;
1713         size_t size, used;
1714         char *str;
1715
1716         info = &udev->uio_info;
1717         size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
1718                         udev->dev_config);
1719         size += 1; /* for \0 */
1720         str = kmalloc(size, GFP_KERNEL);
1721         if (!str)
1722                 return -ENOMEM;
1723
1724         used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
1725         if (udev->dev_config[0])
1726                 snprintf(str + used, size - used, "/%s", udev->dev_config);
1727
1728         /* If the old string exists, free it */
1729         kfree(info->name);
1730         info->name = str;
1731
1732         return 0;
1733 }
1734
1735 static int tcmu_configure_device(struct se_device *dev)
1736 {
1737         struct tcmu_dev *udev = TCMU_DEV(dev);
1738         struct uio_info *info;
1739         struct tcmu_mailbox *mb;
1740         int ret = 0;
1741
1742         ret = tcmu_update_uio_info(udev);
1743         if (ret)
1744                 return ret;
1745
1746         info = &udev->uio_info;
1747
1748         udev->data_bitmap = kcalloc(BITS_TO_LONGS(udev->max_blocks),
1749                                     sizeof(unsigned long),
1750                                     GFP_KERNEL);
1751         if (!udev->data_bitmap) {
1752                 ret = -ENOMEM;
1753                 goto err_bitmap_alloc;
1754         }
1755
1756         udev->mb_addr = vzalloc(CMDR_SIZE);
1757         if (!udev->mb_addr) {
1758                 ret = -ENOMEM;
1759                 goto err_vzalloc;
1760         }
1761
1762         /* mailbox fits in first part of CMDR space */
1763         udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1764         udev->data_off = CMDR_SIZE;
1765         udev->data_size = udev->max_blocks * DATA_BLOCK_SIZE;
1766         udev->dbi_thresh = 0; /* Default in Idle state */
1767
1768         /* Initialise the mailbox of the ring buffer */
1769         mb = udev->mb_addr;
1770         mb->version = TCMU_MAILBOX_VERSION;
1771         mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC | TCMU_MAILBOX_FLAG_CAP_READ_LEN;
1772         mb->cmdr_off = CMDR_OFF;
1773         mb->cmdr_size = udev->cmdr_size;
1774
1775         WARN_ON(!PAGE_ALIGNED(udev->data_off));
1776         WARN_ON(udev->data_size % PAGE_SIZE);
1777         WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1778
1779         info->version = __stringify(TCMU_MAILBOX_VERSION);
1780
1781         info->mem[0].name = "tcm-user command & data buffer";
1782         info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1783         info->mem[0].size = udev->ring_size = udev->data_size + CMDR_SIZE;
1784         info->mem[0].memtype = UIO_MEM_NONE;
1785
1786         info->irqcontrol = tcmu_irqcontrol;
1787         info->irq = UIO_IRQ_CUSTOM;
1788
1789         info->mmap = tcmu_mmap;
1790         info->open = tcmu_open;
1791         info->release = tcmu_release;
1792
1793         ret = uio_register_device(tcmu_root_device, info);
1794         if (ret)
1795                 goto err_register;
1796
1797         /* User can set hw_block_size before enable the device */
1798         if (dev->dev_attrib.hw_block_size == 0)
1799                 dev->dev_attrib.hw_block_size = 512;
1800         /* Other attributes can be configured in userspace */
1801         if (!dev->dev_attrib.hw_max_sectors)
1802                 dev->dev_attrib.hw_max_sectors = 128;
1803         if (!dev->dev_attrib.emulate_write_cache)
1804                 dev->dev_attrib.emulate_write_cache = 0;
1805         dev->dev_attrib.hw_queue_depth = 128;
1806
1807         /* If user didn't explicitly disable netlink reply support, use
1808          * module scope setting.
1809          */
1810         if (udev->nl_reply_supported >= 0)
1811                 udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
1812
1813         /*
1814          * Get a ref incase userspace does a close on the uio device before
1815          * LIO has initiated tcmu_free_device.
1816          */
1817         kref_get(&udev->kref);
1818
1819         ret = tcmu_send_dev_add_event(udev);
1820         if (ret)
1821                 goto err_netlink;
1822
1823         mutex_lock(&root_udev_mutex);
1824         list_add(&udev->node, &root_udev);
1825         mutex_unlock(&root_udev_mutex);
1826
1827         return 0;
1828
1829 err_netlink:
1830         kref_put(&udev->kref, tcmu_dev_kref_release);
1831         uio_unregister_device(&udev->uio_info);
1832 err_register:
1833         vfree(udev->mb_addr);
1834         udev->mb_addr = NULL;
1835 err_vzalloc:
1836         kfree(udev->data_bitmap);
1837         udev->data_bitmap = NULL;
1838 err_bitmap_alloc:
1839         kfree(info->name);
1840         info->name = NULL;
1841
1842         return ret;
1843 }
1844
1845 static bool tcmu_dev_configured(struct tcmu_dev *udev)
1846 {
1847         return udev->uio_info.uio_dev ? true : false;
1848 }
1849
1850 static void tcmu_free_device(struct se_device *dev)
1851 {
1852         struct tcmu_dev *udev = TCMU_DEV(dev);
1853
1854         /* release ref from init */
1855         kref_put(&udev->kref, tcmu_dev_kref_release);
1856 }
1857
1858 static void tcmu_destroy_device(struct se_device *dev)
1859 {
1860         struct tcmu_dev *udev = TCMU_DEV(dev);
1861
1862         del_timer_sync(&udev->cmd_timer);
1863         del_timer_sync(&udev->qfull_timer);
1864
1865         mutex_lock(&root_udev_mutex);
1866         list_del(&udev->node);
1867         mutex_unlock(&root_udev_mutex);
1868
1869         tcmu_send_dev_remove_event(udev);
1870
1871         uio_unregister_device(&udev->uio_info);
1872
1873         /* release ref from configure */
1874         kref_put(&udev->kref, tcmu_dev_kref_release);
1875 }
1876
1877 static void tcmu_unblock_dev(struct tcmu_dev *udev)
1878 {
1879         mutex_lock(&udev->cmdr_lock);
1880         clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
1881         mutex_unlock(&udev->cmdr_lock);
1882 }
1883
1884 static void tcmu_block_dev(struct tcmu_dev *udev)
1885 {
1886         mutex_lock(&udev->cmdr_lock);
1887
1888         if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
1889                 goto unlock;
1890
1891         /* complete IO that has executed successfully */
1892         tcmu_handle_completions(udev);
1893         /* fail IO waiting to be queued */
1894         run_cmdr_queue(udev, true);
1895
1896 unlock:
1897         mutex_unlock(&udev->cmdr_lock);
1898 }
1899
1900 static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
1901 {
1902         struct tcmu_mailbox *mb;
1903         struct tcmu_cmd *cmd;
1904         int i;
1905
1906         mutex_lock(&udev->cmdr_lock);
1907
1908         idr_for_each_entry(&udev->commands, cmd, i) {
1909                 if (!list_empty(&cmd->cmdr_queue_entry))
1910                         continue;
1911
1912                 pr_debug("removing cmd %u on dev %s from ring (is expired %d)\n",
1913                           cmd->cmd_id, udev->name,
1914                           test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags));
1915
1916                 idr_remove(&udev->commands, i);
1917                 if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1918                         if (err_level == 1) {
1919                                 /*
1920                                  * Userspace was not able to start the
1921                                  * command or it is retryable.
1922                                  */
1923                                 target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
1924                         } else {
1925                                 /* hard failure */
1926                                 target_complete_cmd(cmd->se_cmd,
1927                                                     SAM_STAT_CHECK_CONDITION);
1928                         }
1929                 }
1930                 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1931                 tcmu_free_cmd(cmd);
1932         }
1933
1934         mb = udev->mb_addr;
1935         tcmu_flush_dcache_range(mb, sizeof(*mb));
1936         pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
1937                  mb->cmd_tail, mb->cmd_head);
1938
1939         udev->cmdr_last_cleaned = 0;
1940         mb->cmd_tail = 0;
1941         mb->cmd_head = 0;
1942         tcmu_flush_dcache_range(mb, sizeof(*mb));
1943
1944         del_timer(&udev->cmd_timer);
1945
1946         mutex_unlock(&udev->cmdr_lock);
1947 }
1948
1949 enum {
1950         Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
1951         Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_err,
1952 };
1953
1954 static match_table_t tokens = {
1955         {Opt_dev_config, "dev_config=%s"},
1956         {Opt_dev_size, "dev_size=%u"},
1957         {Opt_hw_block_size, "hw_block_size=%u"},
1958         {Opt_hw_max_sectors, "hw_max_sectors=%u"},
1959         {Opt_nl_reply_supported, "nl_reply_supported=%d"},
1960         {Opt_max_data_area_mb, "max_data_area_mb=%u"},
1961         {Opt_err, NULL}
1962 };
1963
1964 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
1965 {
1966         unsigned long tmp_ul;
1967         char *arg_p;
1968         int ret;
1969
1970         arg_p = match_strdup(arg);
1971         if (!arg_p)
1972                 return -ENOMEM;
1973
1974         ret = kstrtoul(arg_p, 0, &tmp_ul);
1975         kfree(arg_p);
1976         if (ret < 0) {
1977                 pr_err("kstrtoul() failed for dev attrib\n");
1978                 return ret;
1979         }
1980         if (!tmp_ul) {
1981                 pr_err("dev attrib must be nonzero\n");
1982                 return -EINVAL;
1983         }
1984         *dev_attrib = tmp_ul;
1985         return 0;
1986 }
1987
1988 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1989                 const char *page, ssize_t count)
1990 {
1991         struct tcmu_dev *udev = TCMU_DEV(dev);
1992         char *orig, *ptr, *opts, *arg_p;
1993         substring_t args[MAX_OPT_ARGS];
1994         int ret = 0, token, tmpval;
1995
1996         opts = kstrdup(page, GFP_KERNEL);
1997         if (!opts)
1998                 return -ENOMEM;
1999
2000         orig = opts;
2001
2002         while ((ptr = strsep(&opts, ",\n")) != NULL) {
2003                 if (!*ptr)
2004                         continue;
2005
2006                 token = match_token(ptr, tokens, args);
2007                 switch (token) {
2008                 case Opt_dev_config:
2009                         if (match_strlcpy(udev->dev_config, &args[0],
2010                                           TCMU_CONFIG_LEN) == 0) {
2011                                 ret = -EINVAL;
2012                                 break;
2013                         }
2014                         pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
2015                         break;
2016                 case Opt_dev_size:
2017                         arg_p = match_strdup(&args[0]);
2018                         if (!arg_p) {
2019                                 ret = -ENOMEM;
2020                                 break;
2021                         }
2022                         ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
2023                         kfree(arg_p);
2024                         if (ret < 0)
2025                                 pr_err("kstrtoul() failed for dev_size=\n");
2026                         break;
2027                 case Opt_hw_block_size:
2028                         ret = tcmu_set_dev_attrib(&args[0],
2029                                         &(dev->dev_attrib.hw_block_size));
2030                         break;
2031                 case Opt_hw_max_sectors:
2032                         ret = tcmu_set_dev_attrib(&args[0],
2033                                         &(dev->dev_attrib.hw_max_sectors));
2034                         break;
2035                 case Opt_nl_reply_supported:
2036                         arg_p = match_strdup(&args[0]);
2037                         if (!arg_p) {
2038                                 ret = -ENOMEM;
2039                                 break;
2040                         }
2041                         ret = kstrtoint(arg_p, 0, &udev->nl_reply_supported);
2042                         kfree(arg_p);
2043                         if (ret < 0)
2044                                 pr_err("kstrtoint() failed for nl_reply_supported=\n");
2045                         break;
2046                 case Opt_max_data_area_mb:
2047                         if (dev->export_count) {
2048                                 pr_err("Unable to set max_data_area_mb while exports exist\n");
2049                                 ret = -EINVAL;
2050                                 break;
2051                         }
2052
2053                         arg_p = match_strdup(&args[0]);
2054                         if (!arg_p) {
2055                                 ret = -ENOMEM;
2056                                 break;
2057                         }
2058                         ret = kstrtoint(arg_p, 0, &tmpval);
2059                         kfree(arg_p);
2060                         if (ret < 0) {
2061                                 pr_err("kstrtoint() failed for max_data_area_mb=\n");
2062                                 break;
2063                         }
2064
2065                         if (tmpval <= 0) {
2066                                 pr_err("Invalid max_data_area %d\n", tmpval);
2067                                 ret = -EINVAL;
2068                                 break;
2069                         }
2070
2071                         udev->max_blocks = TCMU_MBS_TO_BLOCKS(tmpval);
2072                         if (udev->max_blocks > tcmu_global_max_blocks) {
2073                                 pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2074                                        tmpval,
2075                                        TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
2076                                 udev->max_blocks = tcmu_global_max_blocks;
2077                         }
2078                         break;
2079                 default:
2080                         break;
2081                 }
2082
2083                 if (ret)
2084                         break;
2085         }
2086
2087         kfree(orig);
2088         return (!ret) ? count : ret;
2089 }
2090
2091 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2092 {
2093         struct tcmu_dev *udev = TCMU_DEV(dev);
2094         ssize_t bl = 0;
2095
2096         bl = sprintf(b + bl, "Config: %s ",
2097                      udev->dev_config[0] ? udev->dev_config : "NULL");
2098         bl += sprintf(b + bl, "Size: %zu ", udev->dev_size);
2099         bl += sprintf(b + bl, "MaxDataAreaMB: %u\n",
2100                       TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2101
2102         return bl;
2103 }
2104
2105 static sector_t tcmu_get_blocks(struct se_device *dev)
2106 {
2107         struct tcmu_dev *udev = TCMU_DEV(dev);
2108
2109         return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2110                        dev->dev_attrib.block_size);
2111 }
2112
2113 static sense_reason_t
2114 tcmu_parse_cdb(struct se_cmd *cmd)
2115 {
2116         return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2117 }
2118
2119 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2120 {
2121         struct se_dev_attrib *da = container_of(to_config_group(item),
2122                                         struct se_dev_attrib, da_group);
2123         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2124
2125         return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2126 }
2127
2128 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2129                                        size_t count)
2130 {
2131         struct se_dev_attrib *da = container_of(to_config_group(item),
2132                                         struct se_dev_attrib, da_group);
2133         struct tcmu_dev *udev = container_of(da->da_dev,
2134                                         struct tcmu_dev, se_dev);
2135         u32 val;
2136         int ret;
2137
2138         if (da->da_dev->export_count) {
2139                 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2140                 return -EINVAL;
2141         }
2142
2143         ret = kstrtou32(page, 0, &val);
2144         if (ret < 0)
2145                 return ret;
2146
2147         udev->cmd_time_out = val * MSEC_PER_SEC;
2148         return count;
2149 }
2150 CONFIGFS_ATTR(tcmu_, cmd_time_out);
2151
2152 static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2153 {
2154         struct se_dev_attrib *da = container_of(to_config_group(item),
2155                                                 struct se_dev_attrib, da_group);
2156         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2157
2158         return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2159                         udev->qfull_time_out :
2160                         udev->qfull_time_out / MSEC_PER_SEC);
2161 }
2162
2163 static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2164                                          const char *page, size_t count)
2165 {
2166         struct se_dev_attrib *da = container_of(to_config_group(item),
2167                                         struct se_dev_attrib, da_group);
2168         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2169         s32 val;
2170         int ret;
2171
2172         ret = kstrtos32(page, 0, &val);
2173         if (ret < 0)
2174                 return ret;
2175
2176         if (val >= 0) {
2177                 udev->qfull_time_out = val * MSEC_PER_SEC;
2178         } else if (val == -1) {
2179                 udev->qfull_time_out = val;
2180         } else {
2181                 printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2182                 return -EINVAL;
2183         }
2184         return count;
2185 }
2186 CONFIGFS_ATTR(tcmu_, qfull_time_out);
2187
2188 static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2189 {
2190         struct se_dev_attrib *da = container_of(to_config_group(item),
2191                                                 struct se_dev_attrib, da_group);
2192         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2193
2194         return snprintf(page, PAGE_SIZE, "%u\n",
2195                         TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2196 }
2197 CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2198
2199 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2200 {
2201         struct se_dev_attrib *da = container_of(to_config_group(item),
2202                                                 struct se_dev_attrib, da_group);
2203         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2204
2205         return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2206 }
2207
2208 static int tcmu_send_dev_config_event(struct tcmu_dev *udev,
2209                                       const char *reconfig_data)
2210 {
2211         struct sk_buff *skb = NULL;
2212         void *msg_header = NULL;
2213         int ret = 0;
2214
2215         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2216                                       &skb, &msg_header);
2217         if (ret < 0)
2218                 return ret;
2219         ret = nla_put_string(skb, TCMU_ATTR_DEV_CFG, reconfig_data);
2220         if (ret < 0) {
2221                 nlmsg_free(skb);
2222                 return ret;
2223         }
2224         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2225                                        &skb, &msg_header);
2226 }
2227
2228
2229 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2230                                      size_t count)
2231 {
2232         struct se_dev_attrib *da = container_of(to_config_group(item),
2233                                                 struct se_dev_attrib, da_group);
2234         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2235         int ret, len;
2236
2237         len = strlen(page);
2238         if (!len || len > TCMU_CONFIG_LEN - 1)
2239                 return -EINVAL;
2240
2241         /* Check if device has been configured before */
2242         if (tcmu_dev_configured(udev)) {
2243                 ret = tcmu_send_dev_config_event(udev, page);
2244                 if (ret) {
2245                         pr_err("Unable to reconfigure device\n");
2246                         return ret;
2247                 }
2248                 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2249
2250                 ret = tcmu_update_uio_info(udev);
2251                 if (ret)
2252                         return ret;
2253                 return count;
2254         }
2255         strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2256
2257         return count;
2258 }
2259 CONFIGFS_ATTR(tcmu_, dev_config);
2260
2261 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2262 {
2263         struct se_dev_attrib *da = container_of(to_config_group(item),
2264                                                 struct se_dev_attrib, da_group);
2265         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2266
2267         return snprintf(page, PAGE_SIZE, "%zu\n", udev->dev_size);
2268 }
2269
2270 static int tcmu_send_dev_size_event(struct tcmu_dev *udev, u64 size)
2271 {
2272         struct sk_buff *skb = NULL;
2273         void *msg_header = NULL;
2274         int ret = 0;
2275
2276         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2277                                       &skb, &msg_header);
2278         if (ret < 0)
2279                 return ret;
2280         ret = nla_put_u64_64bit(skb, TCMU_ATTR_DEV_SIZE,
2281                                 size, TCMU_ATTR_PAD);
2282         if (ret < 0) {
2283                 nlmsg_free(skb);
2284                 return ret;
2285         }
2286         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2287                                        &skb, &msg_header);
2288 }
2289
2290 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2291                                    size_t count)
2292 {
2293         struct se_dev_attrib *da = container_of(to_config_group(item),
2294                                                 struct se_dev_attrib, da_group);
2295         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2296         u64 val;
2297         int ret;
2298
2299         ret = kstrtou64(page, 0, &val);
2300         if (ret < 0)
2301                 return ret;
2302
2303         /* Check if device has been configured before */
2304         if (tcmu_dev_configured(udev)) {
2305                 ret = tcmu_send_dev_size_event(udev, val);
2306                 if (ret) {
2307                         pr_err("Unable to reconfigure device\n");
2308                         return ret;
2309                 }
2310         }
2311         udev->dev_size = val;
2312         return count;
2313 }
2314 CONFIGFS_ATTR(tcmu_, dev_size);
2315
2316 static ssize_t tcmu_nl_reply_supported_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         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2322
2323         return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2324 }
2325
2326 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2327                 const char *page, size_t count)
2328 {
2329         struct se_dev_attrib *da = container_of(to_config_group(item),
2330                                                 struct se_dev_attrib, da_group);
2331         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2332         s8 val;
2333         int ret;
2334
2335         ret = kstrtos8(page, 0, &val);
2336         if (ret < 0)
2337                 return ret;
2338
2339         udev->nl_reply_supported = val;
2340         return count;
2341 }
2342 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2343
2344 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2345                                              char *page)
2346 {
2347         struct se_dev_attrib *da = container_of(to_config_group(item),
2348                                         struct se_dev_attrib, da_group);
2349
2350         return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2351 }
2352
2353 static int tcmu_send_emulate_write_cache(struct tcmu_dev *udev, u8 val)
2354 {
2355         struct sk_buff *skb = NULL;
2356         void *msg_header = NULL;
2357         int ret = 0;
2358
2359         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2360                                       &skb, &msg_header);
2361         if (ret < 0)
2362                 return ret;
2363         ret = nla_put_u8(skb, TCMU_ATTR_WRITECACHE, val);
2364         if (ret < 0) {
2365                 nlmsg_free(skb);
2366                 return ret;
2367         }
2368         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2369                                        &skb, &msg_header);
2370 }
2371
2372 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2373                                               const char *page, size_t count)
2374 {
2375         struct se_dev_attrib *da = container_of(to_config_group(item),
2376                                         struct se_dev_attrib, da_group);
2377         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2378         u8 val;
2379         int ret;
2380
2381         ret = kstrtou8(page, 0, &val);
2382         if (ret < 0)
2383                 return ret;
2384
2385         /* Check if device has been configured before */
2386         if (tcmu_dev_configured(udev)) {
2387                 ret = tcmu_send_emulate_write_cache(udev, val);
2388                 if (ret) {
2389                         pr_err("Unable to reconfigure device\n");
2390                         return ret;
2391                 }
2392         }
2393
2394         da->emulate_write_cache = val;
2395         return count;
2396 }
2397 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2398
2399 static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
2400 {
2401         struct se_device *se_dev = container_of(to_config_group(item),
2402                                                 struct se_device,
2403                                                 dev_action_group);
2404         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2405
2406         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2407                 return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
2408         else
2409                 return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
2410 }
2411
2412 static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
2413                                     size_t count)
2414 {
2415         struct se_device *se_dev = container_of(to_config_group(item),
2416                                                 struct se_device,
2417                                                 dev_action_group);
2418         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2419         u8 val;
2420         int ret;
2421
2422         ret = kstrtou8(page, 0, &val);
2423         if (ret < 0)
2424                 return ret;
2425
2426         if (val > 1) {
2427                 pr_err("Invalid block value %d\n", val);
2428                 return -EINVAL;
2429         }
2430
2431         if (!val)
2432                 tcmu_unblock_dev(udev);
2433         else
2434                 tcmu_block_dev(udev);
2435         return count;
2436 }
2437 CONFIGFS_ATTR(tcmu_, block_dev);
2438
2439 static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
2440                                      size_t count)
2441 {
2442         struct se_device *se_dev = container_of(to_config_group(item),
2443                                                 struct se_device,
2444                                                 dev_action_group);
2445         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2446         u8 val;
2447         int ret;
2448
2449         ret = kstrtou8(page, 0, &val);
2450         if (ret < 0)
2451                 return ret;
2452
2453         if (val != 1 && val != 2) {
2454                 pr_err("Invalid reset ring value %d\n", val);
2455                 return -EINVAL;
2456         }
2457
2458         tcmu_reset_ring(udev, val);
2459         return count;
2460 }
2461 CONFIGFS_ATTR_WO(tcmu_, reset_ring);
2462
2463 static struct configfs_attribute *tcmu_attrib_attrs[] = {
2464         &tcmu_attr_cmd_time_out,
2465         &tcmu_attr_qfull_time_out,
2466         &tcmu_attr_max_data_area_mb,
2467         &tcmu_attr_dev_config,
2468         &tcmu_attr_dev_size,
2469         &tcmu_attr_emulate_write_cache,
2470         &tcmu_attr_nl_reply_supported,
2471         NULL,
2472 };
2473
2474 static struct configfs_attribute **tcmu_attrs;
2475
2476 static struct configfs_attribute *tcmu_action_attrs[] = {
2477         &tcmu_attr_block_dev,
2478         &tcmu_attr_reset_ring,
2479         NULL,
2480 };
2481
2482 static struct target_backend_ops tcmu_ops = {
2483         .name                   = "user",
2484         .owner                  = THIS_MODULE,
2485         .transport_flags        = TRANSPORT_FLAG_PASSTHROUGH,
2486         .attach_hba             = tcmu_attach_hba,
2487         .detach_hba             = tcmu_detach_hba,
2488         .alloc_device           = tcmu_alloc_device,
2489         .configure_device       = tcmu_configure_device,
2490         .destroy_device         = tcmu_destroy_device,
2491         .free_device            = tcmu_free_device,
2492         .parse_cdb              = tcmu_parse_cdb,
2493         .set_configfs_dev_params = tcmu_set_configfs_dev_params,
2494         .show_configfs_dev_params = tcmu_show_configfs_dev_params,
2495         .get_device_type        = sbc_get_device_type,
2496         .get_blocks             = tcmu_get_blocks,
2497         .tb_dev_action_attrs    = tcmu_action_attrs,
2498 };
2499
2500 static void find_free_blocks(void)
2501 {
2502         struct tcmu_dev *udev;
2503         loff_t off;
2504         u32 start, end, block, total_freed = 0;
2505
2506         if (atomic_read(&global_db_count) <= tcmu_global_max_blocks)
2507                 return;
2508
2509         mutex_lock(&root_udev_mutex);
2510         list_for_each_entry(udev, &root_udev, node) {
2511                 mutex_lock(&udev->cmdr_lock);
2512
2513                 /* Try to complete the finished commands first */
2514                 tcmu_handle_completions(udev);
2515
2516                 /* Skip the udevs in idle */
2517                 if (!udev->dbi_thresh) {
2518                         mutex_unlock(&udev->cmdr_lock);
2519                         continue;
2520                 }
2521
2522                 end = udev->dbi_max + 1;
2523                 block = find_last_bit(udev->data_bitmap, end);
2524                 if (block == udev->dbi_max) {
2525                         /*
2526                          * The last bit is dbi_max, so it is not possible
2527                          * reclaim any blocks.
2528                          */
2529                         mutex_unlock(&udev->cmdr_lock);
2530                         continue;
2531                 } else if (block == end) {
2532                         /* The current udev will goto idle state */
2533                         udev->dbi_thresh = start = 0;
2534                         udev->dbi_max = 0;
2535                 } else {
2536                         udev->dbi_thresh = start = block + 1;
2537                         udev->dbi_max = block;
2538                 }
2539
2540                 /* Here will truncate the data area from off */
2541                 off = udev->data_off + start * DATA_BLOCK_SIZE;
2542                 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
2543
2544                 /* Release the block pages */
2545                 tcmu_blocks_release(&udev->data_blocks, start, end);
2546                 mutex_unlock(&udev->cmdr_lock);
2547
2548                 total_freed += end - start;
2549                 pr_debug("Freed %u blocks (total %u) from %s.\n", end - start,
2550                          total_freed, udev->name);
2551         }
2552         mutex_unlock(&root_udev_mutex);
2553
2554         if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
2555                 schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
2556 }
2557
2558 static void check_timedout_devices(void)
2559 {
2560         struct tcmu_dev *udev, *tmp_dev;
2561         LIST_HEAD(devs);
2562
2563         spin_lock_bh(&timed_out_udevs_lock);
2564         list_splice_init(&timed_out_udevs, &devs);
2565
2566         list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
2567                 list_del_init(&udev->timedout_entry);
2568                 spin_unlock_bh(&timed_out_udevs_lock);
2569
2570                 mutex_lock(&udev->cmdr_lock);
2571                 idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
2572                 mutex_unlock(&udev->cmdr_lock);
2573
2574                 spin_lock_bh(&timed_out_udevs_lock);
2575         }
2576
2577         spin_unlock_bh(&timed_out_udevs_lock);
2578 }
2579
2580 static void tcmu_unmap_work_fn(struct work_struct *work)
2581 {
2582         check_timedout_devices();
2583         find_free_blocks();
2584 }
2585
2586 static int __init tcmu_module_init(void)
2587 {
2588         int ret, i, k, len = 0;
2589
2590         BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2591
2592         INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
2593
2594         tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2595                                 sizeof(struct tcmu_cmd),
2596                                 __alignof__(struct tcmu_cmd),
2597                                 0, NULL);
2598         if (!tcmu_cmd_cache)
2599                 return -ENOMEM;
2600
2601         tcmu_root_device = root_device_register("tcm_user");
2602         if (IS_ERR(tcmu_root_device)) {
2603                 ret = PTR_ERR(tcmu_root_device);
2604                 goto out_free_cache;
2605         }
2606
2607         ret = genl_register_family(&tcmu_genl_family);
2608         if (ret < 0) {
2609                 goto out_unreg_device;
2610         }
2611
2612         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2613                 len += sizeof(struct configfs_attribute *);
2614         }
2615         for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2616                 len += sizeof(struct configfs_attribute *);
2617         }
2618         len += sizeof(struct configfs_attribute *);
2619
2620         tcmu_attrs = kzalloc(len, GFP_KERNEL);
2621         if (!tcmu_attrs) {
2622                 ret = -ENOMEM;
2623                 goto out_unreg_genl;
2624         }
2625
2626         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2627                 tcmu_attrs[i] = passthrough_attrib_attrs[i];
2628         }
2629         for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2630                 tcmu_attrs[i] = tcmu_attrib_attrs[k];
2631                 i++;
2632         }
2633         tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2634
2635         ret = transport_backend_register(&tcmu_ops);
2636         if (ret)
2637                 goto out_attrs;
2638
2639         return 0;
2640
2641 out_attrs:
2642         kfree(tcmu_attrs);
2643 out_unreg_genl:
2644         genl_unregister_family(&tcmu_genl_family);
2645 out_unreg_device:
2646         root_device_unregister(tcmu_root_device);
2647 out_free_cache:
2648         kmem_cache_destroy(tcmu_cmd_cache);
2649
2650         return ret;
2651 }
2652
2653 static void __exit tcmu_module_exit(void)
2654 {
2655         cancel_delayed_work_sync(&tcmu_unmap_work);
2656         target_backend_unregister(&tcmu_ops);
2657         kfree(tcmu_attrs);
2658         genl_unregister_family(&tcmu_genl_family);
2659         root_device_unregister(tcmu_root_device);
2660         kmem_cache_destroy(tcmu_cmd_cache);
2661 }
2662
2663 MODULE_DESCRIPTION("TCM USER subsystem plugin");
2664 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2665 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2666 MODULE_LICENSE("GPL");
2667
2668 module_init(tcmu_module_init);
2669 module_exit(tcmu_module_exit);