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