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