drivers/target/target_core_user.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
   4  * Copyright (C) 2014 Red Hat, Inc.
   5  * Copyright (C) 2015 Arrikto, Inc.
   6  * Copyright (C) 2017 Chinamobile, Inc.
   7  */
   8
   9 #include <linux/spinlock.h>
  10 #include <linux/module.h>
  11 #include <linux/kernel.h>
  12 #include <linux/timer.h>
  13 #include <linux/parser.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/uio_driver.h>
  16 #include <linux/xarray.h>
  17 #include <linux/stringify.h>
  18 #include <linux/bitops.h>
  19 #include <linux/highmem.h>
  20 #include <linux/configfs.h>
  21 #include <linux/mutex.h>
  22 #include <linux/workqueue.h>
  23 #include <net/genetlink.h>
  24 #include <scsi/scsi_common.h>
  25 #include <scsi/scsi_proto.h>
  26 #include <target/target_core_base.h>
  27 #include <target/target_core_fabric.h>
  28 #include <target/target_core_backend.h>
  29
  30 #include <linux/target_core_user.h>
  31
  32 /**
  33  * DOC: Userspace I/O
  34  * Userspace I/O
  35  * -------------
  36  *
  37  * Define a shared-memory interface for LIO to pass SCSI commands and
  38  * data to userspace for processing. This is to allow backends that
  39  * are too complex for in-kernel support to be possible.
  40  *
  41  * It uses the UIO framework to do a lot of the device-creation and
  42  * introspection work for us.
  43  *
  44  * See the .h file for how the ring is laid out. Note that while the
  45  * command ring is defined, the particulars of the data area are
  46  * not. Offset values in the command entry point to other locations
  47  * internal to the mmap-ed area. There is separate space outside the
  48  * command ring for data buffers. This leaves maximum flexibility for
  49  * moving buffer allocations, or even page flipping or other
  50  * allocation techniques, without altering the command ring layout.
  51  *
  52  * SECURITY:
  53  * The user process must be assumed to be malicious. There's no way to
  54  * prevent it breaking the command ring protocol if it wants, but in
  55  * order to prevent other issues we must only ever read *data* from
  56  * the shared memory area, not offsets or sizes. This applies to
  57  * command ring entries as well as the mailbox. Extra code needed for
  58  * this may have a 'UAM' comment.
  59  */
  60
  61 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
  62
  63 /* For mailbox plus cmd ring, the size is fixed 8MB */
  64 #define MB_CMDR_SIZE (8 * 1024 * 1024)
  65 /* Offset of cmd ring is size of mailbox */
  66 #define CMDR_OFF sizeof(struct tcmu_mailbox)
  67 #define CMDR_SIZE (MB_CMDR_SIZE - CMDR_OFF)
  68
  69 /*
  70  * For data area, the default block size is PAGE_SIZE and
  71  * the default total size is 256K * PAGE_SIZE.
  72  */
  73 #define DATA_PAGES_PER_BLK_DEF 1
  74 #define DATA_AREA_PAGES_DEF (256 * 1024)
  75
  76 #define TCMU_MBS_TO_PAGES(_mbs) ((size_t)_mbs << (20 - PAGE_SHIFT))
  77 #define TCMU_PAGES_TO_MBS(_pages) (_pages >> (20 - PAGE_SHIFT))
  78
  79 /*
  80  * Default number of global data blocks(512K * PAGE_SIZE)
  81  * when the unmap thread will be started.
  82  */
  83 #define TCMU_GLOBAL_MAX_PAGES_DEF (512 * 1024)
  84
  85 static u8 tcmu_kern_cmd_reply_supported;
  86 static u8 tcmu_netlink_blocked;
  87
  88 static struct device *tcmu_root_device;
  89
  90 struct tcmu_hba {
  91         u32 host_id;
  92 };
  93
  94 #define TCMU_CONFIG_LEN 256
  95
  96 static DEFINE_MUTEX(tcmu_nl_cmd_mutex);
  97 static LIST_HEAD(tcmu_nl_cmd_list);
  98
  99 struct tcmu_dev;
 100
 101 struct tcmu_nl_cmd {
 102         /* wake up thread waiting for reply */
 103         struct completion complete;
 104         struct list_head nl_list;
 105         struct tcmu_dev *udev;
 106         int cmd;
 107         int status;
 108 };
 109
 110 struct tcmu_dev {
 111         struct list_head node;
 112         struct kref kref;
 113
 114         struct se_device se_dev;
 115         struct se_dev_plug se_plug;
 116
 117         char *name;
 118         struct se_hba *hba;
 119
 120 #define TCMU_DEV_BIT_OPEN 0
 121 #define TCMU_DEV_BIT_BROKEN 1
 122 #define TCMU_DEV_BIT_BLOCKED 2
 123 #define TCMU_DEV_BIT_TMR_NOTIFY 3
 124 #define TCMU_DEV_BIT_PLUGGED 4
 125         unsigned long flags;
 126
 127         struct uio_info uio_info;
 128
 129         struct inode *inode;
 130
 131         uint64_t dev_size;
 132
 133         struct tcmu_mailbox *mb_addr;
 134         void *cmdr;
 135         u32 cmdr_size;
 136         u32 cmdr_last_cleaned;
 137         /* Offset of data area from start of mb */
 138         /* Must add data_off and mb_addr to get the address */
 139         size_t data_off;
 140         int data_area_mb;
 141         uint32_t max_blocks;
 142         size_t mmap_pages;
 143
 144         struct mutex cmdr_lock;
 145         struct list_head qfull_queue;
 146         struct list_head tmr_queue;
 147
 148         uint32_t dbi_max;
 149         uint32_t dbi_thresh;
 150         unsigned long *data_bitmap;
 151         struct xarray data_pages;
 152         uint32_t data_pages_per_blk;
 153         uint32_t data_blk_size;
 154
 155         struct xarray commands;
 156
 157         struct timer_list cmd_timer;
 158         unsigned int cmd_time_out;
 159         struct list_head inflight_queue;
 160
 161         struct timer_list qfull_timer;
 162         int qfull_time_out;
 163
 164         struct list_head timedout_entry;
 165
 166         struct tcmu_nl_cmd curr_nl_cmd;
 167
 168         char dev_config[TCMU_CONFIG_LEN];
 169
 170         int nl_reply_supported;
 171 };
 172
 173 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
 174
 175 struct tcmu_cmd {
 176         struct se_cmd *se_cmd;
 177         struct tcmu_dev *tcmu_dev;
 178         struct list_head queue_entry;
 179
 180         uint16_t cmd_id;
 181
 182         /* Can't use se_cmd when cleaning up expired cmds, because if
 183            cmd has been completed then accessing se_cmd is off limits */
 184         uint32_t dbi_cnt;
 185         uint32_t dbi_bidi_cnt;
 186         uint32_t dbi_cur;
 187         uint32_t *dbi;
 188
 189         uint32_t data_len_bidi;
 190
 191         unsigned long deadline;
 192
 193 #define TCMU_CMD_BIT_EXPIRED 0
 194         unsigned long flags;
 195 };
 196
 197 struct tcmu_tmr {
 198         struct list_head queue_entry;
 199
 200         uint8_t tmr_type;
 201         uint32_t tmr_cmd_cnt;
 202         int16_t tmr_cmd_ids[];
 203 };
 204
 205 /*
 206  * To avoid dead lock the mutex lock order should always be:
 207  *
 208  * mutex_lock(&root_udev_mutex);
 209  * ...
 210  * mutex_lock(&tcmu_dev->cmdr_lock);
 211  * mutex_unlock(&tcmu_dev->cmdr_lock);
 212  * ...
 213  * mutex_unlock(&root_udev_mutex);
 214  */
 215 static DEFINE_MUTEX(root_udev_mutex);
 216 static LIST_HEAD(root_udev);
 217
 218 static DEFINE_SPINLOCK(timed_out_udevs_lock);
 219 static LIST_HEAD(timed_out_udevs);
 220
 221 static struct kmem_cache *tcmu_cmd_cache;
 222
 223 static atomic_t global_page_count = ATOMIC_INIT(0);
 224 static struct delayed_work tcmu_unmap_work;
 225 static int tcmu_global_max_pages = TCMU_GLOBAL_MAX_PAGES_DEF;
 226
 227 static int tcmu_set_global_max_data_area(const char *str,
 228                                          const struct kernel_param *kp)
 229 {
 230         int ret, max_area_mb;
 231
 232         ret = kstrtoint(str, 10, &max_area_mb);
 233         if (ret)
 234                 return -EINVAL;
 235
 236         if (max_area_mb <= 0) {
 237                 pr_err("global_max_data_area must be larger than 0.\n");
 238                 return -EINVAL;
 239         }
 240
 241         tcmu_global_max_pages = TCMU_MBS_TO_PAGES(max_area_mb);
 242         if (atomic_read(&global_page_count) > tcmu_global_max_pages)
 243                 schedule_delayed_work(&tcmu_unmap_work, 0);
 244         else
 245                 cancel_delayed_work_sync(&tcmu_unmap_work);
 246
 247         return 0;
 248 }
 249
 250 static int tcmu_get_global_max_data_area(char *buffer,
 251                                          const struct kernel_param *kp)
 252 {
 253         return sprintf(buffer, "%d\n", TCMU_PAGES_TO_MBS(tcmu_global_max_pages));
 254 }
 255
 256 static const struct kernel_param_ops tcmu_global_max_data_area_op = {
 257         .set = tcmu_set_global_max_data_area,
 258         .get = tcmu_get_global_max_data_area,
 259 };
 260
 261 module_param_cb(global_max_data_area_mb, &tcmu_global_max_data_area_op, NULL,
 262                 S_IWUSR | S_IRUGO);
 263 MODULE_PARM_DESC(global_max_data_area_mb,
 264                  "Max MBs allowed to be allocated to all the tcmu device's "
 265                  "data areas.");
 266
 267 static int tcmu_get_block_netlink(char *buffer,
 268                                   const struct kernel_param *kp)
 269 {
 270         return sprintf(buffer, "%s\n", tcmu_netlink_blocked ?
 271                        "blocked" : "unblocked");
 272 }
 273
 274 static int tcmu_set_block_netlink(const char *str,
 275                                   const struct kernel_param *kp)
 276 {
 277         int ret;
 278         u8 val;
 279
 280         ret = kstrtou8(str, 0, &val);
 281         if (ret < 0)
 282                 return ret;
 283
 284         if (val > 1) {
 285                 pr_err("Invalid block netlink value %u\n", val);
 286                 return -EINVAL;
 287         }
 288
 289         tcmu_netlink_blocked = val;
 290         return 0;
 291 }
 292
 293 static const struct kernel_param_ops tcmu_block_netlink_op = {
 294         .set = tcmu_set_block_netlink,
 295         .get = tcmu_get_block_netlink,
 296 };
 297
 298 module_param_cb(block_netlink, &tcmu_block_netlink_op, NULL, S_IWUSR | S_IRUGO);
 299 MODULE_PARM_DESC(block_netlink, "Block new netlink commands.");
 300
 301 static int tcmu_fail_netlink_cmd(struct tcmu_nl_cmd *nl_cmd)
 302 {
 303         struct tcmu_dev *udev = nl_cmd->udev;
 304
 305         if (!tcmu_netlink_blocked) {
 306                 pr_err("Could not reset device's netlink interface. Netlink is not blocked.\n");
 307                 return -EBUSY;
 308         }
 309
 310         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
 311                 pr_debug("Aborting nl cmd %d on %s\n", nl_cmd->cmd, udev->name);
 312                 nl_cmd->status = -EINTR;
 313                 list_del(&nl_cmd->nl_list);
 314                 complete(&nl_cmd->complete);
 315         }
 316         return 0;
 317 }
 318
 319 static int tcmu_set_reset_netlink(const char *str,
 320                                   const struct kernel_param *kp)
 321 {
 322         struct tcmu_nl_cmd *nl_cmd, *tmp_cmd;
 323         int ret;
 324         u8 val;
 325
 326         ret = kstrtou8(str, 0, &val);
 327         if (ret < 0)
 328                 return ret;
 329
 330         if (val != 1) {
 331                 pr_err("Invalid reset netlink value %u\n", val);
 332                 return -EINVAL;
 333         }
 334
 335         mutex_lock(&tcmu_nl_cmd_mutex);
 336         list_for_each_entry_safe(nl_cmd, tmp_cmd, &tcmu_nl_cmd_list, nl_list) {
 337                 ret = tcmu_fail_netlink_cmd(nl_cmd);
 338                 if (ret)
 339                         break;
 340         }
 341         mutex_unlock(&tcmu_nl_cmd_mutex);
 342
 343         return ret;
 344 }
 345
 346 static const struct kernel_param_ops tcmu_reset_netlink_op = {
 347         .set = tcmu_set_reset_netlink,
 348 };
 349
 350 module_param_cb(reset_netlink, &tcmu_reset_netlink_op, NULL, S_IWUSR);
 351 MODULE_PARM_DESC(reset_netlink, "Reset netlink commands.");
 352
 353 /* multicast group */
 354 enum tcmu_multicast_groups {
 355         TCMU_MCGRP_CONFIG,
 356 };
 357
 358 static const struct genl_multicast_group tcmu_mcgrps[] = {
 359         [TCMU_MCGRP_CONFIG] = { .name = "config", },
 360 };
 361
 362 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
 363         [TCMU_ATTR_DEVICE]      = { .type = NLA_STRING },
 364         [TCMU_ATTR_MINOR]       = { .type = NLA_U32 },
 365         [TCMU_ATTR_CMD_STATUS]  = { .type = NLA_S32 },
 366         [TCMU_ATTR_DEVICE_ID]   = { .type = NLA_U32 },
 367         [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
 368 };
 369
 370 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
 371 {
 372         struct tcmu_dev *udev = NULL;
 373         struct tcmu_nl_cmd *nl_cmd;
 374         int dev_id, rc, ret = 0;
 375
 376         if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
 377             !info->attrs[TCMU_ATTR_DEVICE_ID]) {
 378                 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
 379                 return -EINVAL;
 380         }
 381
 382         dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
 383         rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
 384
 385         mutex_lock(&tcmu_nl_cmd_mutex);
 386         list_for_each_entry(nl_cmd, &tcmu_nl_cmd_list, nl_list) {
 387                 if (nl_cmd->udev->se_dev.dev_index == dev_id) {
 388                         udev = nl_cmd->udev;
 389                         break;
 390                 }
 391         }
 392
 393         if (!udev) {
 394                 pr_err("tcmu nl cmd %u/%d completion could not find device with dev id %u.\n",
 395                        completed_cmd, rc, dev_id);
 396                 ret = -ENODEV;
 397                 goto unlock;
 398         }
 399         list_del(&nl_cmd->nl_list);
 400
 401         pr_debug("%s genl cmd done got id %d curr %d done %d rc %d stat %d\n",
 402                  udev->name, dev_id, nl_cmd->cmd, completed_cmd, rc,
 403                  nl_cmd->status);
 404
 405         if (nl_cmd->cmd != completed_cmd) {
 406                 pr_err("Mismatched commands on %s (Expecting reply for %d. Current %d).\n",
 407                        udev->name, completed_cmd, nl_cmd->cmd);
 408                 ret = -EINVAL;
 409                 goto unlock;
 410         }
 411
 412         nl_cmd->status = rc;
 413         complete(&nl_cmd->complete);
 414 unlock:
 415         mutex_unlock(&tcmu_nl_cmd_mutex);
 416         return ret;
 417 }
 418
 419 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
 420 {
 421         return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
 422 }
 423
 424 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
 425 {
 426         return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
 427 }
 428
 429 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
 430                                        struct genl_info *info)
 431 {
 432         return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
 433 }
 434
 435 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
 436 {
 437         if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
 438                 tcmu_kern_cmd_reply_supported  =
 439                         nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
 440                 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
 441                        tcmu_kern_cmd_reply_supported);
 442         }
 443
 444         return 0;
 445 }
 446
 447 static const struct genl_small_ops tcmu_genl_ops[] = {
 448         {
 449                 .cmd    = TCMU_CMD_SET_FEATURES,
 450                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 451                 .flags  = GENL_ADMIN_PERM,
 452                 .doit   = tcmu_genl_set_features,
 453         },
 454         {
 455                 .cmd    = TCMU_CMD_ADDED_DEVICE_DONE,
 456                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 457                 .flags  = GENL_ADMIN_PERM,
 458                 .doit   = tcmu_genl_add_dev_done,
 459         },
 460         {
 461                 .cmd    = TCMU_CMD_REMOVED_DEVICE_DONE,
 462                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 463                 .flags  = GENL_ADMIN_PERM,
 464                 .doit   = tcmu_genl_rm_dev_done,
 465         },
 466         {
 467                 .cmd    = TCMU_CMD_RECONFIG_DEVICE_DONE,
 468                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 469                 .flags  = GENL_ADMIN_PERM,
 470                 .doit   = tcmu_genl_reconfig_dev_done,
 471         },
 472 };
 473
 474 /* Our generic netlink family */
 475 static struct genl_family tcmu_genl_family __ro_after_init = {
 476         .module = THIS_MODULE,
 477         .hdrsize = 0,
 478         .name = "TCM-USER",
 479         .version = 2,
 480         .maxattr = TCMU_ATTR_MAX,
 481         .policy = tcmu_attr_policy,
 482         .mcgrps = tcmu_mcgrps,
 483         .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
 484         .netnsok = true,
 485         .small_ops = tcmu_genl_ops,
 486         .n_small_ops = ARRAY_SIZE(tcmu_genl_ops),
 487 };
 488
 489 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
 490 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
 491 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
 492 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
 493
 494 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
 495 {
 496         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 497         uint32_t i;
 498
 499         for (i = 0; i < len; i++)
 500                 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
 501 }
 502
 503 static inline int tcmu_get_empty_block(struct tcmu_dev *udev,
 504                                        struct tcmu_cmd *tcmu_cmd,
 505                                        int prev_dbi, int length, int *iov_cnt)
 506 {
 507         XA_STATE(xas, &udev->data_pages, 0);
 508         struct page *page;
 509         int i, cnt, dbi, dpi;
 510         int page_cnt = DIV_ROUND_UP(length, PAGE_SIZE);
 511
 512         dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
 513         if (dbi == udev->dbi_thresh)
 514                 return -1;
 515
 516         dpi = dbi * udev->data_pages_per_blk;
 517         /* Count the number of already allocated pages */
 518         xas_set(&xas, dpi);
 519         rcu_read_lock();
 520         for (cnt = 0; xas_next(&xas) && cnt < page_cnt;)
 521                 cnt++;
 522         rcu_read_unlock();
 523
 524         for (i = cnt; i < page_cnt; i++) {
 525                 /* try to get new page from the mm */
 526                 page = alloc_page(GFP_NOIO);
 527                 if (!page)
 528                         break;
 529
 530                 if (xa_store(&udev->data_pages, dpi + i, page, GFP_NOIO)) {
 531                         __free_page(page);
 532                         break;
 533                 }
 534         }
 535         if (atomic_add_return(i - cnt, &global_page_count) >
 536                               tcmu_global_max_pages)
 537                 schedule_delayed_work(&tcmu_unmap_work, 0);
 538
 539         if (i && dbi > udev->dbi_max)
 540                 udev->dbi_max = dbi;
 541
 542         set_bit(dbi, udev->data_bitmap);
 543         tcmu_cmd_set_dbi(tcmu_cmd, dbi);
 544
 545         if (dbi != prev_dbi + 1)
 546                 *iov_cnt += 1;
 547
 548         return i == page_cnt ? dbi : -1;
 549 }
 550
 551 static int tcmu_get_empty_blocks(struct tcmu_dev *udev,
 552                                  struct tcmu_cmd *tcmu_cmd, int length)
 553 {
 554         /* start value of dbi + 1 must not be a valid dbi */
 555         int dbi = -2;
 556         int blk_data_len, iov_cnt = 0;
 557         uint32_t blk_size = udev->data_blk_size;
 558
 559         for (; length > 0; length -= blk_size) {
 560                 blk_data_len = min_t(uint32_t, length, blk_size);
 561                 dbi = tcmu_get_empty_block(udev, tcmu_cmd, dbi, blk_data_len,
 562                                            &iov_cnt);
 563                 if (dbi < 0)
 564                         return -1;
 565         }
 566         return iov_cnt;
 567 }
 568
 569 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
 570 {
 571         kfree(tcmu_cmd->dbi);
 572         kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 573 }
 574
 575 static inline void tcmu_cmd_set_block_cnts(struct tcmu_cmd *cmd)
 576 {
 577         int i, len;
 578         struct se_cmd *se_cmd = cmd->se_cmd;
 579         uint32_t blk_size = cmd->tcmu_dev->data_blk_size;
 580
 581         cmd->dbi_cnt = DIV_ROUND_UP(se_cmd->data_length, blk_size);
 582
 583         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 584                 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
 585                 for (i = 0, len = 0; i < se_cmd->t_bidi_data_nents; i++)
 586                         len += se_cmd->t_bidi_data_sg[i].length;
 587                 cmd->dbi_bidi_cnt = DIV_ROUND_UP(len, blk_size);
 588                 cmd->dbi_cnt += cmd->dbi_bidi_cnt;
 589                 cmd->data_len_bidi = len;
 590         }
 591 }
 592
 593 static int new_block_to_iov(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 594                             struct iovec **iov, int prev_dbi, int len)
 595 {
 596         /* Get the next dbi */
 597         int dbi = tcmu_cmd_get_dbi(cmd);
 598
 599         /* Do not add more than udev->data_blk_size to iov */
 600         len = min_t(int,  len, udev->data_blk_size);
 601
 602         /*
 603          * The following code will gather and map the blocks to the same iovec
 604          * when the blocks are all next to each other.
 605          */
 606         if (dbi != prev_dbi + 1) {
 607                 /* dbi is not next to previous dbi, so start new iov */
 608                 if (prev_dbi >= 0)
 609                         (*iov)++;
 610                 /* write offset relative to mb_addr */
 611                 (*iov)->iov_base = (void __user *)
 612                                    (udev->data_off + dbi * udev->data_blk_size);
 613         }
 614         (*iov)->iov_len += len;
 615
 616         return dbi;
 617 }
 618
 619 static void tcmu_setup_iovs(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 620                             struct iovec **iov, int data_length)
 621 {
 622         /* start value of dbi + 1 must not be a valid dbi */
 623         int dbi = -2;
 624
 625         /* We prepare the IOVs for DMA_FROM_DEVICE transfer direction */
 626         for (; data_length > 0; data_length -= udev->data_blk_size)
 627                 dbi = new_block_to_iov(udev, cmd, iov, dbi, data_length);
 628 }
 629
 630 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
 631 {
 632         struct se_device *se_dev = se_cmd->se_dev;
 633         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 634         struct tcmu_cmd *tcmu_cmd;
 635
 636         tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_NOIO);
 637         if (!tcmu_cmd)
 638                 return NULL;
 639
 640         INIT_LIST_HEAD(&tcmu_cmd->queue_entry);
 641         tcmu_cmd->se_cmd = se_cmd;
 642         tcmu_cmd->tcmu_dev = udev;
 643
 644         tcmu_cmd_set_block_cnts(tcmu_cmd);
 645         tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
 646                                 GFP_NOIO);
 647         if (!tcmu_cmd->dbi) {
 648                 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 649                 return NULL;
 650         }
 651
 652         return tcmu_cmd;
 653 }
 654
 655 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
 656 {
 657         unsigned long offset = offset_in_page(vaddr);
 658         void *start = vaddr - offset;
 659
 660         size = round_up(size+offset, PAGE_SIZE);
 661
 662         while (size) {
 663                 flush_dcache_page(vmalloc_to_page(start));
 664                 start += PAGE_SIZE;
 665                 size -= PAGE_SIZE;
 666         }
 667 }
 668
 669 /*
 670  * Some ring helper functions. We don't assume size is a power of 2 so
 671  * we can't use circ_buf.h.
 672  */
 673 static inline size_t spc_used(size_t head, size_t tail, size_t size)
 674 {
 675         int diff = head - tail;
 676
 677         if (diff >= 0)
 678                 return diff;
 679         else
 680                 return size + diff;
 681 }
 682
 683 static inline size_t spc_free(size_t head, size_t tail, size_t size)
 684 {
 685         /* Keep 1 byte unused or we can't tell full from empty */
 686         return (size - spc_used(head, tail, size) - 1);
 687 }
 688
 689 static inline size_t head_to_end(size_t head, size_t size)
 690 {
 691         return size - head;
 692 }
 693
 694 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
 695
 696 #define TCMU_SG_TO_DATA_AREA 1
 697 #define TCMU_DATA_AREA_TO_SG 2
 698
 699 static inline void tcmu_copy_data(struct tcmu_dev *udev,
 700                                   struct tcmu_cmd *tcmu_cmd, uint32_t direction,
 701                                   struct scatterlist *sg, unsigned int sg_nents,
 702                                   struct iovec **iov, size_t data_len)
 703 {
 704         /* start value of dbi + 1 must not be a valid dbi */
 705         int dbi = -2;
 706         size_t page_remaining, cp_len;
 707         int page_cnt, page_inx, dpi;
 708         struct sg_mapping_iter sg_iter;
 709         unsigned int sg_flags;
 710         struct page *page;
 711         void *data_page_start, *data_addr;
 712
 713         if (direction == TCMU_SG_TO_DATA_AREA)
 714                 sg_flags = SG_MITER_ATOMIC | SG_MITER_FROM_SG;
 715         else
 716                 sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG;
 717         sg_miter_start(&sg_iter, sg, sg_nents, sg_flags);
 718
 719         while (data_len) {
 720                 if (direction == TCMU_SG_TO_DATA_AREA)
 721                         dbi = new_block_to_iov(udev, tcmu_cmd, iov, dbi,
 722                                                data_len);
 723                 else
 724                         dbi = tcmu_cmd_get_dbi(tcmu_cmd);
 725
 726                 page_cnt = DIV_ROUND_UP(data_len, PAGE_SIZE);
 727                 if (page_cnt > udev->data_pages_per_blk)
 728                         page_cnt = udev->data_pages_per_blk;
 729
 730                 dpi = dbi * udev->data_pages_per_blk;
 731                 for (page_inx = 0; page_inx < page_cnt && data_len;
 732                      page_inx++, dpi++) {
 733                         page = xa_load(&udev->data_pages, dpi);
 734
 735                         if (direction == TCMU_DATA_AREA_TO_SG)
 736                                 flush_dcache_page(page);
 737                         data_page_start = kmap_atomic(page);
 738                         page_remaining = PAGE_SIZE;
 739
 740                         while (page_remaining && data_len) {
 741                                 if (!sg_miter_next(&sg_iter)) {
 742                                         /* set length to 0 to abort outer loop */
 743                                         data_len = 0;
 744                                         pr_debug("%s: aborting data copy due to exhausted sg_list\n",
 745                                                  __func__);
 746                                         break;
 747                                 }
 748                                 cp_len = min3(sg_iter.length, page_remaining,
 749                                               data_len);
 750
 751                                 data_addr = data_page_start +
 752                                             PAGE_SIZE - page_remaining;
 753                                 if (direction == TCMU_SG_TO_DATA_AREA)
 754                                         memcpy(data_addr, sg_iter.addr, cp_len);
 755                                 else
 756                                         memcpy(sg_iter.addr, data_addr, cp_len);
 757
 758                                 data_len -= cp_len;
 759                                 page_remaining -= cp_len;
 760                                 sg_iter.consumed = cp_len;
 761                         }
 762                         sg_miter_stop(&sg_iter);
 763
 764                         kunmap_atomic(data_page_start);
 765                         if (direction == TCMU_SG_TO_DATA_AREA)
 766                                 flush_dcache_page(page);
 767                 }
 768         }
 769 }
 770
 771 static void scatter_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
 772                               struct iovec **iov)
 773 {
 774         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 775
 776         tcmu_copy_data(udev, tcmu_cmd, TCMU_SG_TO_DATA_AREA, se_cmd->t_data_sg,
 777                        se_cmd->t_data_nents, iov, se_cmd->data_length);
 778 }
 779
 780 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
 781                              bool bidi, uint32_t read_len)
 782 {
 783         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 784         struct scatterlist *data_sg;
 785         unsigned int data_nents;
 786
 787         if (!bidi) {
 788                 data_sg = se_cmd->t_data_sg;
 789                 data_nents = se_cmd->t_data_nents;
 790         } else {
 791                 /*
 792                  * For bidi case, the first count blocks are for Data-Out
 793                  * buffer blocks, and before gathering the Data-In buffer
 794                  * the Data-Out buffer blocks should be skipped.
 795                  */
 796                 tcmu_cmd_set_dbi_cur(tcmu_cmd,
 797                                      tcmu_cmd->dbi_cnt - tcmu_cmd->dbi_bidi_cnt);
 798
 799                 data_sg = se_cmd->t_bidi_data_sg;
 800                 data_nents = se_cmd->t_bidi_data_nents;
 801         }
 802
 803         tcmu_copy_data(udev, tcmu_cmd, TCMU_DATA_AREA_TO_SG, data_sg,
 804                        data_nents, NULL, read_len);
 805 }
 806
 807 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
 808 {
 809         return thresh - bitmap_weight(bitmap, thresh);
 810 }
 811
 812 /*
 813  * We can't queue a command until we have space available on the cmd ring.
 814  *
 815  * Called with ring lock held.
 816  */
 817 static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size)
 818 {
 819         struct tcmu_mailbox *mb = udev->mb_addr;
 820         size_t space, cmd_needed;
 821         u32 cmd_head;
 822
 823         tcmu_flush_dcache_range(mb, sizeof(*mb));
 824
 825         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 826
 827         /*
 828          * If cmd end-of-ring space is too small then we need space for a NOP plus
 829          * original cmd - cmds are internally contiguous.
 830          */
 831         if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
 832                 cmd_needed = cmd_size;
 833         else
 834                 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
 835
 836         space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
 837         if (space < cmd_needed) {
 838                 pr_debug("no cmd space: %u %u %u\n", cmd_head,
 839                        udev->cmdr_last_cleaned, udev->cmdr_size);
 840                 return false;
 841         }
 842         return true;
 843 }
 844
 845 /*
 846  * We have to allocate data buffers before we can queue a command.
 847  * Returns -1 on error (not enough space) or number of needed iovs on success
 848  *
 849  * Called with ring lock held.
 850  */
 851 static int tcmu_alloc_data_space(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 852                                   int *iov_bidi_cnt)
 853 {
 854         int space, iov_cnt = 0, ret = 0;
 855
 856         if (!cmd->dbi_cnt)
 857                 goto wr_iov_cnts;
 858
 859         /* try to check and get the data blocks as needed */
 860         space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
 861         if (space < cmd->dbi_cnt) {
 862                 unsigned long blocks_left =
 863                                 (udev->max_blocks - udev->dbi_thresh) + space;
 864
 865                 if (blocks_left < cmd->dbi_cnt) {
 866                         pr_debug("no data space: only %lu available, but ask for %u\n",
 867                                         blocks_left * udev->data_blk_size,
 868                                         cmd->dbi_cnt * udev->data_blk_size);
 869                         return -1;
 870                 }
 871
 872                 udev->dbi_thresh += cmd->dbi_cnt;
 873                 if (udev->dbi_thresh > udev->max_blocks)
 874                         udev->dbi_thresh = udev->max_blocks;
 875         }
 876
 877         iov_cnt = tcmu_get_empty_blocks(udev, cmd, cmd->se_cmd->data_length);
 878         if (iov_cnt < 0)
 879                 return -1;
 880
 881         if (cmd->dbi_bidi_cnt) {
 882                 ret = tcmu_get_empty_blocks(udev, cmd, cmd->data_len_bidi);
 883                 if (ret < 0)
 884                         return -1;
 885         }
 886 wr_iov_cnts:
 887         *iov_bidi_cnt = ret;
 888         return iov_cnt + ret;
 889 }
 890
 891 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
 892 {
 893         return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 894                         sizeof(struct tcmu_cmd_entry));
 895 }
 896
 897 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
 898                                            size_t base_command_size)
 899 {
 900         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 901         size_t command_size;
 902
 903         command_size = base_command_size +
 904                 round_up(scsi_command_size(se_cmd->t_task_cdb),
 905                                 TCMU_OP_ALIGN_SIZE);
 906
 907         WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
 908
 909         return command_size;
 910 }
 911
 912 static void tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
 913                                  struct timer_list *timer)
 914 {
 915         if (!tmo)
 916                 return;
 917
 918         tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
 919         if (!timer_pending(timer))
 920                 mod_timer(timer, tcmu_cmd->deadline);
 921
 922         pr_debug("Timeout set up for cmd %p, dev = %s, tmo = %lu\n", tcmu_cmd,
 923                  tcmu_cmd->tcmu_dev->name, tmo / MSEC_PER_SEC);
 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
 931         /*
 932          * For backwards compat if qfull_time_out is not set use
 933          * cmd_time_out and if that's not set use the default time out.
 934          */
 935         if (!udev->qfull_time_out)
 936                 return -ETIMEDOUT;
 937         else if (udev->qfull_time_out > 0)
 938                 tmo = udev->qfull_time_out;
 939         else if (udev->cmd_time_out)
 940                 tmo = udev->cmd_time_out;
 941         else
 942                 tmo = TCMU_TIME_OUT;
 943
 944         tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
 945
 946         list_add_tail(&tcmu_cmd->queue_entry, &udev->qfull_queue);
 947         pr_debug("adding cmd %p on dev %s to ring space wait queue\n",
 948                  tcmu_cmd, udev->name);
 949         return 0;
 950 }
 951
 952 static uint32_t ring_insert_padding(struct tcmu_dev *udev, size_t cmd_size)
 953 {
 954         struct tcmu_cmd_entry_hdr *hdr;
 955         struct tcmu_mailbox *mb = udev->mb_addr;
 956         uint32_t cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 957
 958         /* Insert a PAD if end-of-ring space is too small */
 959         if (head_to_end(cmd_head, udev->cmdr_size) < cmd_size) {
 960                 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
 961
 962                 hdr = udev->cmdr + cmd_head;
 963                 tcmu_hdr_set_op(&hdr->len_op, TCMU_OP_PAD);
 964                 tcmu_hdr_set_len(&hdr->len_op, pad_size);
 965                 hdr->cmd_id = 0; /* not used for PAD */
 966                 hdr->kflags = 0;
 967                 hdr->uflags = 0;
 968                 tcmu_flush_dcache_range(hdr, sizeof(*hdr));
 969
 970                 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
 971                 tcmu_flush_dcache_range(mb, sizeof(*mb));
 972
 973                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 974                 WARN_ON(cmd_head != 0);
 975         }
 976
 977         return cmd_head;
 978 }
 979
 980 static void tcmu_unplug_device(struct se_dev_plug *se_plug)
 981 {
 982         struct se_device *se_dev = se_plug->se_dev;
 983         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 984
 985         clear_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags);
 986         uio_event_notify(&udev->uio_info);
 987 }
 988
 989 static struct se_dev_plug *tcmu_plug_device(struct se_device *se_dev)
 990 {
 991         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 992
 993         if (!test_and_set_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags))
 994                 return &udev->se_plug;
 995
 996         return NULL;
 997 }
 998
 999 /**
1000  * queue_cmd_ring - queue cmd to ring or internally
1001  * @tcmu_cmd: cmd to queue
1002  * @scsi_err: TCM error code if failure (-1) returned.
1003  *
1004  * Returns:
1005  * -1 we cannot queue internally or to the ring.
1006  *  0 success
1007  *  1 internally queued to wait for ring memory to free.
1008  */
1009 static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
1010 {
1011         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
1012         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
1013         size_t base_command_size, command_size;
1014         struct tcmu_mailbox *mb = udev->mb_addr;
1015         struct tcmu_cmd_entry *entry;
1016         struct iovec *iov;
1017         int iov_cnt, iov_bidi_cnt;
1018         uint32_t cmd_id, cmd_head;
1019         uint64_t cdb_off;
1020         uint32_t blk_size = udev->data_blk_size;
1021         /* size of data buffer needed */
1022         size_t data_length = (size_t)tcmu_cmd->dbi_cnt * blk_size;
1023
1024         *scsi_err = TCM_NO_SENSE;
1025
1026         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
1027                 *scsi_err = TCM_LUN_BUSY;
1028                 return -1;
1029         }
1030
1031         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1032                 *scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1033                 return -1;
1034         }
1035
1036         if (!list_empty(&udev->qfull_queue))
1037                 goto queue;
1038
1039         if (data_length > (size_t)udev->max_blocks * blk_size) {
1040                 pr_warn("TCMU: Request of size %zu is too big for %zu data area\n",
1041                         data_length, (size_t)udev->max_blocks * blk_size);
1042                 *scsi_err = TCM_INVALID_CDB_FIELD;
1043                 return -1;
1044         }
1045
1046         iov_cnt = tcmu_alloc_data_space(udev, tcmu_cmd, &iov_bidi_cnt);
1047         if (iov_cnt < 0)
1048                 goto free_and_queue;
1049
1050         /*
1051          * Must be a certain minimum size for response sense info, but
1052          * also may be larger if the iov array is large.
1053          */
1054         base_command_size = tcmu_cmd_get_base_cmd_size(iov_cnt);
1055         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
1056
1057         if (command_size > (udev->cmdr_size / 2)) {
1058                 pr_warn("TCMU: Request of size %zu is too big for %u cmd ring\n",
1059                         command_size, udev->cmdr_size);
1060                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1061                 *scsi_err = TCM_INVALID_CDB_FIELD;
1062                 return -1;
1063         }
1064
1065         if (!is_ring_space_avail(udev, command_size))
1066                 /*
1067                  * Don't leave commands partially setup because the unmap
1068                  * thread might need the blocks to make forward progress.
1069                  */
1070                 goto free_and_queue;
1071
1072         if (xa_alloc(&udev->commands, &cmd_id, tcmu_cmd, XA_LIMIT(1, 0xffff),
1073                      GFP_NOWAIT) < 0) {
1074                 pr_err("tcmu: Could not allocate cmd id.\n");
1075
1076                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
1077                 *scsi_err = TCM_OUT_OF_RESOURCES;
1078                 return -1;
1079         }
1080         tcmu_cmd->cmd_id = cmd_id;
1081
1082         pr_debug("allocated cmd id %u for cmd %p dev %s\n", tcmu_cmd->cmd_id,
1083                  tcmu_cmd, udev->name);
1084
1085         cmd_head = ring_insert_padding(udev, command_size);
1086
1087         entry = udev->cmdr + cmd_head;
1088         memset(entry, 0, command_size);
1089         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
1090
1091         /* prepare iov list and copy data to data area if necessary */
1092         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1093         iov = &entry->req.iov[0];
1094
1095         if (se_cmd->data_direction == DMA_TO_DEVICE ||
1096             se_cmd->se_cmd_flags & SCF_BIDI)
1097                 scatter_data_area(udev, tcmu_cmd, &iov);
1098         else
1099                 tcmu_setup_iovs(udev, tcmu_cmd, &iov, se_cmd->data_length);
1100
1101         entry->req.iov_cnt = iov_cnt - iov_bidi_cnt;
1102
1103         /* Handle BIDI commands */
1104         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1105                 iov++;
1106                 tcmu_setup_iovs(udev, tcmu_cmd, &iov, tcmu_cmd->data_len_bidi);
1107                 entry->req.iov_bidi_cnt = iov_bidi_cnt;
1108         }
1109
1110         tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out, &udev->cmd_timer);
1111
1112         entry->hdr.cmd_id = tcmu_cmd->cmd_id;
1113
1114         tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
1115
1116         /* All offsets relative to mb_addr, not start of entry! */
1117         cdb_off = CMDR_OFF + cmd_head + base_command_size;
1118         memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
1119         entry->req.cdb_off = cdb_off;
1120         tcmu_flush_dcache_range(entry, command_size);
1121
1122         UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1123         tcmu_flush_dcache_range(mb, sizeof(*mb));
1124
1125         list_add_tail(&tcmu_cmd->queue_entry, &udev->inflight_queue);
1126
1127         if (!test_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags))
1128                 uio_event_notify(&udev->uio_info);
1129
1130         return 0;
1131
1132 free_and_queue:
1133         tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1134         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1135
1136 queue:
1137         if (add_to_qfull_queue(tcmu_cmd)) {
1138                 *scsi_err = TCM_OUT_OF_RESOURCES;
1139                 return -1;
1140         }
1141
1142         return 1;
1143 }
1144
1145 /**
1146  * queue_tmr_ring - queue tmr info to ring or internally
1147  * @udev: related tcmu_dev
1148  * @tmr: tcmu_tmr containing tmr info to queue
1149  *
1150  * Returns:
1151  *  0 success
1152  *  1 internally queued to wait for ring memory to free.
1153  */
1154 static int
1155 queue_tmr_ring(struct tcmu_dev *udev, struct tcmu_tmr *tmr)
1156 {
1157         struct tcmu_tmr_entry *entry;
1158         int cmd_size;
1159         int id_list_sz;
1160         struct tcmu_mailbox *mb = udev->mb_addr;
1161         uint32_t cmd_head;
1162
1163         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
1164                 goto out_free;
1165
1166         id_list_sz = sizeof(tmr->tmr_cmd_ids[0]) * tmr->tmr_cmd_cnt;
1167         cmd_size = round_up(sizeof(*entry) + id_list_sz, TCMU_OP_ALIGN_SIZE);
1168
1169         if (!list_empty(&udev->tmr_queue) ||
1170             !is_ring_space_avail(udev, cmd_size)) {
1171                 list_add_tail(&tmr->queue_entry, &udev->tmr_queue);
1172                 pr_debug("adding tmr %p on dev %s to TMR ring space wait queue\n",
1173                          tmr, udev->name);
1174                 return 1;
1175         }
1176
1177         cmd_head = ring_insert_padding(udev, cmd_size);
1178
1179         entry = udev->cmdr + cmd_head;
1180         memset(entry, 0, cmd_size);
1181         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_TMR);
1182         tcmu_hdr_set_len(&entry->hdr.len_op, cmd_size);
1183         entry->tmr_type = tmr->tmr_type;
1184         entry->cmd_cnt = tmr->tmr_cmd_cnt;
1185         memcpy(&entry->cmd_ids[0], &tmr->tmr_cmd_ids[0], id_list_sz);
1186         tcmu_flush_dcache_range(entry, cmd_size);
1187
1188         UPDATE_HEAD(mb->cmd_head, cmd_size, udev->cmdr_size);
1189         tcmu_flush_dcache_range(mb, sizeof(*mb));
1190
1191         uio_event_notify(&udev->uio_info);
1192
1193 out_free:
1194         kfree(tmr);
1195
1196         return 0;
1197 }
1198
1199 static sense_reason_t
1200 tcmu_queue_cmd(struct se_cmd *se_cmd)
1201 {
1202         struct se_device *se_dev = se_cmd->se_dev;
1203         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1204         struct tcmu_cmd *tcmu_cmd;
1205         sense_reason_t scsi_ret = TCM_CHECK_CONDITION_ABORT_CMD;
1206         int ret = -1;
1207
1208         tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1209         if (!tcmu_cmd)
1210                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1211
1212         mutex_lock(&udev->cmdr_lock);
1213         if (!(se_cmd->transport_state & CMD_T_ABORTED))
1214                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1215         if (ret < 0)
1216                 tcmu_free_cmd(tcmu_cmd);
1217         else
1218                 se_cmd->priv = tcmu_cmd;
1219         mutex_unlock(&udev->cmdr_lock);
1220         return scsi_ret;
1221 }
1222
1223 static void tcmu_set_next_deadline(struct list_head *queue,
1224                                    struct timer_list *timer)
1225 {
1226         struct tcmu_cmd *cmd;
1227
1228         if (!list_empty(queue)) {
1229                 cmd = list_first_entry(queue, struct tcmu_cmd, queue_entry);
1230                 mod_timer(timer, cmd->deadline);
1231         } else
1232                 del_timer(timer);
1233 }
1234
1235 static int
1236 tcmu_tmr_type(enum tcm_tmreq_table tmf)
1237 {
1238         switch (tmf) {
1239         case TMR_ABORT_TASK:            return TCMU_TMR_ABORT_TASK;
1240         case TMR_ABORT_TASK_SET:        return TCMU_TMR_ABORT_TASK_SET;
1241         case TMR_CLEAR_ACA:             return TCMU_TMR_CLEAR_ACA;
1242         case TMR_CLEAR_TASK_SET:        return TCMU_TMR_CLEAR_TASK_SET;
1243         case TMR_LUN_RESET:             return TCMU_TMR_LUN_RESET;
1244         case TMR_TARGET_WARM_RESET:     return TCMU_TMR_TARGET_WARM_RESET;
1245         case TMR_TARGET_COLD_RESET:     return TCMU_TMR_TARGET_COLD_RESET;
1246         case TMR_LUN_RESET_PRO:         return TCMU_TMR_LUN_RESET_PRO;
1247         default:                        return TCMU_TMR_UNKNOWN;
1248         }
1249 }
1250
1251 static void
1252 tcmu_tmr_notify(struct se_device *se_dev, enum tcm_tmreq_table tmf,
1253                 struct list_head *cmd_list)
1254 {
1255         int i = 0, cmd_cnt = 0;
1256         bool unqueued = false;
1257         uint16_t *cmd_ids = NULL;
1258         struct tcmu_cmd *cmd;
1259         struct se_cmd *se_cmd;
1260         struct tcmu_tmr *tmr;
1261         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1262
1263         mutex_lock(&udev->cmdr_lock);
1264
1265         /* First we check for aborted commands in qfull_queue */
1266         list_for_each_entry(se_cmd, cmd_list, state_list) {
1267                 i++;
1268                 if (!se_cmd->priv)
1269                         continue;
1270                 cmd = se_cmd->priv;
1271                 /* Commands on qfull queue have no id yet */
1272                 if (cmd->cmd_id) {
1273                         cmd_cnt++;
1274                         continue;
1275                 }
1276                 pr_debug("Removing aborted command %p from queue on dev %s.\n",
1277                          cmd, udev->name);
1278
1279                 list_del_init(&cmd->queue_entry);
1280                 tcmu_free_cmd(cmd);
1281                 se_cmd->priv = NULL;
1282                 target_complete_cmd(se_cmd, SAM_STAT_TASK_ABORTED);
1283                 unqueued = true;
1284         }
1285         if (unqueued)
1286                 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1287
1288         if (!test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags))
1289                 goto unlock;
1290
1291         pr_debug("TMR event %d on dev %s, aborted cmds %d, afflicted cmd_ids %d\n",
1292                  tcmu_tmr_type(tmf), udev->name, i, cmd_cnt);
1293
1294         tmr = kmalloc(sizeof(*tmr) + cmd_cnt * sizeof(*cmd_ids), GFP_NOIO);
1295         if (!tmr)
1296                 goto unlock;
1297
1298         tmr->tmr_type = tcmu_tmr_type(tmf);
1299         tmr->tmr_cmd_cnt = cmd_cnt;
1300
1301         if (cmd_cnt != 0) {
1302                 cmd_cnt = 0;
1303                 list_for_each_entry(se_cmd, cmd_list, state_list) {
1304                         if (!se_cmd->priv)
1305                                 continue;
1306                         cmd = se_cmd->priv;
1307                         if (cmd->cmd_id)
1308                                 tmr->tmr_cmd_ids[cmd_cnt++] = cmd->cmd_id;
1309                 }
1310         }
1311
1312         queue_tmr_ring(udev, tmr);
1313
1314 unlock:
1315         mutex_unlock(&udev->cmdr_lock);
1316 }
1317
1318 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
1319 {
1320         struct se_cmd *se_cmd = cmd->se_cmd;
1321         struct tcmu_dev *udev = cmd->tcmu_dev;
1322         bool read_len_valid = false;
1323         uint32_t read_len;
1324
1325         /*
1326          * cmd has been completed already from timeout, just reclaim
1327          * data area space and free cmd
1328          */
1329         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1330                 WARN_ON_ONCE(se_cmd);
1331                 goto out;
1332         }
1333
1334         list_del_init(&cmd->queue_entry);
1335
1336         tcmu_cmd_reset_dbi_cur(cmd);
1337
1338         if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1339                 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1340                         cmd->se_cmd);
1341                 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1342                 goto done;
1343         }
1344
1345         read_len = se_cmd->data_length;
1346         if (se_cmd->data_direction == DMA_FROM_DEVICE &&
1347             (entry->hdr.uflags & TCMU_UFLAG_READ_LEN) && entry->rsp.read_len) {
1348                 read_len_valid = true;
1349                 if (entry->rsp.read_len < read_len)
1350                         read_len = entry->rsp.read_len;
1351         }
1352
1353         if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1354                 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1355                 if (!read_len_valid )
1356                         goto done;
1357                 else
1358                         se_cmd->se_cmd_flags |= SCF_TREAT_READ_AS_NORMAL;
1359         }
1360         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1361                 /* Get Data-In buffer before clean up */
1362                 gather_data_area(udev, cmd, true, read_len);
1363         } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1364                 gather_data_area(udev, cmd, false, read_len);
1365         } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1366                 /* TODO: */
1367         } else if (se_cmd->data_direction != DMA_NONE) {
1368                 pr_warn("TCMU: data direction was %d!\n",
1369                         se_cmd->data_direction);
1370         }
1371
1372 done:
1373         se_cmd->priv = NULL;
1374         if (read_len_valid) {
1375                 pr_debug("read_len = %d\n", read_len);
1376                 target_complete_cmd_with_length(cmd->se_cmd,
1377                                         entry->rsp.scsi_status, read_len);
1378         } else
1379                 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1380
1381 out:
1382         tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1383         tcmu_free_cmd(cmd);
1384 }
1385
1386 static int tcmu_run_tmr_queue(struct tcmu_dev *udev)
1387 {
1388         struct tcmu_tmr *tmr, *tmp;
1389         LIST_HEAD(tmrs);
1390
1391         if (list_empty(&udev->tmr_queue))
1392                 return 1;
1393
1394         pr_debug("running %s's tmr queue\n", udev->name);
1395
1396         list_splice_init(&udev->tmr_queue, &tmrs);
1397
1398         list_for_each_entry_safe(tmr, tmp, &tmrs, queue_entry) {
1399                 list_del_init(&tmr->queue_entry);
1400
1401                 pr_debug("removing tmr %p on dev %s from queue\n",
1402                          tmr, udev->name);
1403
1404                 if (queue_tmr_ring(udev, tmr)) {
1405                         pr_debug("ran out of space during tmr queue run\n");
1406                         /*
1407                          * tmr was requeued, so just put all tmrs back in
1408                          * the queue
1409                          */
1410                         list_splice_tail(&tmrs, &udev->tmr_queue);
1411                         return 0;
1412                 }
1413         }
1414
1415         return 1;
1416 }
1417
1418 static bool tcmu_handle_completions(struct tcmu_dev *udev)
1419 {
1420         struct tcmu_mailbox *mb;
1421         struct tcmu_cmd *cmd;
1422         bool free_space = false;
1423
1424         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1425                 pr_err("ring broken, not handling completions\n");
1426                 return false;
1427         }
1428
1429         mb = udev->mb_addr;
1430         tcmu_flush_dcache_range(mb, sizeof(*mb));
1431
1432         while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1433
1434                 struct tcmu_cmd_entry *entry = udev->cmdr + udev->cmdr_last_cleaned;
1435
1436                 /*
1437                  * Flush max. up to end of cmd ring since current entry might
1438                  * be a padding that is shorter than sizeof(*entry)
1439                  */
1440                 size_t ring_left = head_to_end(udev->cmdr_last_cleaned,
1441                                                udev->cmdr_size);
1442                 tcmu_flush_dcache_range(entry, ring_left < sizeof(*entry) ?
1443                                         ring_left : sizeof(*entry));
1444
1445                 free_space = true;
1446
1447                 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD ||
1448                     tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_TMR) {
1449                         UPDATE_HEAD(udev->cmdr_last_cleaned,
1450                                     tcmu_hdr_get_len(entry->hdr.len_op),
1451                                     udev->cmdr_size);
1452                         continue;
1453                 }
1454                 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1455
1456                 cmd = xa_erase(&udev->commands, entry->hdr.cmd_id);
1457                 if (!cmd) {
1458                         pr_err("cmd_id %u not found, ring is broken\n",
1459                                entry->hdr.cmd_id);
1460                         set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1461                         return false;
1462                 }
1463
1464                 tcmu_handle_completion(cmd, entry);
1465
1466                 UPDATE_HEAD(udev->cmdr_last_cleaned,
1467                             tcmu_hdr_get_len(entry->hdr.len_op),
1468                             udev->cmdr_size);
1469         }
1470         if (free_space)
1471                 free_space = tcmu_run_tmr_queue(udev);
1472
1473         if (atomic_read(&global_page_count) > tcmu_global_max_pages &&
1474             xa_empty(&udev->commands) && list_empty(&udev->qfull_queue)) {
1475                 /*
1476                  * Allocated blocks exceeded global block limit, currently no
1477                  * more pending or waiting commands so try to reclaim blocks.
1478                  */
1479                 schedule_delayed_work(&tcmu_unmap_work, 0);
1480         }
1481         if (udev->cmd_time_out)
1482                 tcmu_set_next_deadline(&udev->inflight_queue, &udev->cmd_timer);
1483
1484         return free_space;
1485 }
1486
1487 static void tcmu_check_expired_ring_cmd(struct tcmu_cmd *cmd)
1488 {
1489         struct se_cmd *se_cmd;
1490
1491         if (!time_after_eq(jiffies, cmd->deadline))
1492                 return;
1493
1494         set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1495         list_del_init(&cmd->queue_entry);
1496         se_cmd = cmd->se_cmd;
1497         se_cmd->priv = NULL;
1498         cmd->se_cmd = NULL;
1499
1500         pr_debug("Timing out inflight cmd %u on dev %s.\n",
1501                  cmd->cmd_id, cmd->tcmu_dev->name);
1502
1503         target_complete_cmd(se_cmd, SAM_STAT_CHECK_CONDITION);
1504 }
1505
1506 static void tcmu_check_expired_queue_cmd(struct tcmu_cmd *cmd)
1507 {
1508         struct se_cmd *se_cmd;
1509
1510         if (!time_after_eq(jiffies, cmd->deadline))
1511                 return;
1512
1513         pr_debug("Timing out queued cmd %p on dev %s.\n",
1514                   cmd, cmd->tcmu_dev->name);
1515
1516         list_del_init(&cmd->queue_entry);
1517         se_cmd = cmd->se_cmd;
1518         tcmu_free_cmd(cmd);
1519
1520         se_cmd->priv = NULL;
1521         target_complete_cmd(se_cmd, SAM_STAT_TASK_SET_FULL);
1522 }
1523
1524 static void tcmu_device_timedout(struct tcmu_dev *udev)
1525 {
1526         spin_lock(&timed_out_udevs_lock);
1527         if (list_empty(&udev->timedout_entry))
1528                 list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1529         spin_unlock(&timed_out_udevs_lock);
1530
1531         schedule_delayed_work(&tcmu_unmap_work, 0);
1532 }
1533
1534 static void tcmu_cmd_timedout(struct timer_list *t)
1535 {
1536         struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1537
1538         pr_debug("%s cmd timeout has expired\n", udev->name);
1539         tcmu_device_timedout(udev);
1540 }
1541
1542 static void tcmu_qfull_timedout(struct timer_list *t)
1543 {
1544         struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1545
1546         pr_debug("%s qfull timeout has expired\n", udev->name);
1547         tcmu_device_timedout(udev);
1548 }
1549
1550 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1551 {
1552         struct tcmu_hba *tcmu_hba;
1553
1554         tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1555         if (!tcmu_hba)
1556                 return -ENOMEM;
1557
1558         tcmu_hba->host_id = host_id;
1559         hba->hba_ptr = tcmu_hba;
1560
1561         return 0;
1562 }
1563
1564 static void tcmu_detach_hba(struct se_hba *hba)
1565 {
1566         kfree(hba->hba_ptr);
1567         hba->hba_ptr = NULL;
1568 }
1569
1570 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1571 {
1572         struct tcmu_dev *udev;
1573
1574         udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1575         if (!udev)
1576                 return NULL;
1577         kref_init(&udev->kref);
1578
1579         udev->name = kstrdup(name, GFP_KERNEL);
1580         if (!udev->name) {
1581                 kfree(udev);
1582                 return NULL;
1583         }
1584
1585         udev->hba = hba;
1586         udev->cmd_time_out = TCMU_TIME_OUT;
1587         udev->qfull_time_out = -1;
1588
1589         udev->data_pages_per_blk = DATA_PAGES_PER_BLK_DEF;
1590         udev->max_blocks = DATA_AREA_PAGES_DEF / udev->data_pages_per_blk;
1591         udev->data_area_mb = TCMU_PAGES_TO_MBS(DATA_AREA_PAGES_DEF);
1592
1593         mutex_init(&udev->cmdr_lock);
1594
1595         INIT_LIST_HEAD(&udev->node);
1596         INIT_LIST_HEAD(&udev->timedout_entry);
1597         INIT_LIST_HEAD(&udev->qfull_queue);
1598         INIT_LIST_HEAD(&udev->tmr_queue);
1599         INIT_LIST_HEAD(&udev->inflight_queue);
1600         xa_init_flags(&udev->commands, XA_FLAGS_ALLOC1);
1601
1602         timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1603         timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1604
1605         xa_init(&udev->data_pages);
1606
1607         return &udev->se_dev;
1608 }
1609
1610 static void tcmu_dev_call_rcu(struct rcu_head *p)
1611 {
1612         struct se_device *dev = container_of(p, struct se_device, rcu_head);
1613         struct tcmu_dev *udev = TCMU_DEV(dev);
1614
1615         kfree(udev->uio_info.name);
1616         kfree(udev->name);
1617         kfree(udev);
1618 }
1619
1620 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1621 {
1622         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1623                 kmem_cache_free(tcmu_cmd_cache, cmd);
1624                 return 0;
1625         }
1626         return -EINVAL;
1627 }
1628
1629 static u32 tcmu_blocks_release(struct tcmu_dev *udev, unsigned long first,
1630                                 unsigned long last)
1631 {
1632         XA_STATE(xas, &udev->data_pages, first * udev->data_pages_per_blk);
1633         struct page *page;
1634         u32 pages_freed = 0;
1635
1636         xas_lock(&xas);
1637         xas_for_each(&xas, page, (last + 1) * udev->data_pages_per_blk - 1) {
1638                 xas_store(&xas, NULL);
1639                 __free_page(page);
1640                 pages_freed++;
1641         }
1642         xas_unlock(&xas);
1643
1644         atomic_sub(pages_freed, &global_page_count);
1645
1646         return pages_freed;
1647 }
1648
1649 static void tcmu_remove_all_queued_tmr(struct tcmu_dev *udev)
1650 {
1651         struct tcmu_tmr *tmr, *tmp;
1652
1653         list_for_each_entry_safe(tmr, tmp, &udev->tmr_queue, queue_entry) {
1654                 list_del_init(&tmr->queue_entry);
1655                 kfree(tmr);
1656         }
1657 }
1658
1659 static void tcmu_dev_kref_release(struct kref *kref)
1660 {
1661         struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1662         struct se_device *dev = &udev->se_dev;
1663         struct tcmu_cmd *cmd;
1664         bool all_expired = true;
1665         unsigned long i;
1666
1667         vfree(udev->mb_addr);
1668         udev->mb_addr = NULL;
1669
1670         spin_lock_bh(&timed_out_udevs_lock);
1671         if (!list_empty(&udev->timedout_entry))
1672                 list_del(&udev->timedout_entry);
1673         spin_unlock_bh(&timed_out_udevs_lock);
1674
1675         /* Upper layer should drain all requests before calling this */
1676         mutex_lock(&udev->cmdr_lock);
1677         xa_for_each(&udev->commands, i, cmd) {
1678                 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1679                         all_expired = false;
1680         }
1681         /* There can be left over TMR cmds. Remove them. */
1682         tcmu_remove_all_queued_tmr(udev);
1683         if (!list_empty(&udev->qfull_queue))
1684                 all_expired = false;
1685         xa_destroy(&udev->commands);
1686         WARN_ON(!all_expired);
1687
1688         tcmu_blocks_release(udev, 0, udev->dbi_max);
1689         bitmap_free(udev->data_bitmap);
1690         mutex_unlock(&udev->cmdr_lock);
1691
1692         pr_debug("dev_kref_release\n");
1693
1694         call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1695 }
1696
1697 static void run_qfull_queue(struct tcmu_dev *udev, bool fail)
1698 {
1699         struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1700         LIST_HEAD(cmds);
1701         sense_reason_t scsi_ret;
1702         int ret;
1703
1704         if (list_empty(&udev->qfull_queue))
1705                 return;
1706
1707         pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1708
1709         list_splice_init(&udev->qfull_queue, &cmds);
1710
1711         list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, queue_entry) {
1712                 list_del_init(&tcmu_cmd->queue_entry);
1713
1714                 pr_debug("removing cmd %p on dev %s from queue\n",
1715                          tcmu_cmd, udev->name);
1716
1717                 if (fail) {
1718                         /*
1719                          * We were not able to even start the command, so
1720                          * fail with busy to allow a retry in case runner
1721                          * was only temporarily down. If the device is being
1722                          * removed then LIO core will do the right thing and
1723                          * fail the retry.
1724                          */
1725                         tcmu_cmd->se_cmd->priv = NULL;
1726                         target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1727                         tcmu_free_cmd(tcmu_cmd);
1728                         continue;
1729                 }
1730
1731                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1732                 if (ret < 0) {
1733                         pr_debug("cmd %p on dev %s failed with %u\n",
1734                                  tcmu_cmd, udev->name, scsi_ret);
1735                         /*
1736                          * Ignore scsi_ret for now. target_complete_cmd
1737                          * drops it.
1738                          */
1739                         tcmu_cmd->se_cmd->priv = NULL;
1740                         target_complete_cmd(tcmu_cmd->se_cmd,
1741                                             SAM_STAT_CHECK_CONDITION);
1742                         tcmu_free_cmd(tcmu_cmd);
1743                 } else if (ret > 0) {
1744                         pr_debug("ran out of space during cmdr queue run\n");
1745                         /*
1746                          * cmd was requeued, so just put all cmds back in
1747                          * the queue
1748                          */
1749                         list_splice_tail(&cmds, &udev->qfull_queue);
1750                         break;
1751                 }
1752         }
1753
1754         tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1755 }
1756
1757 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1758 {
1759         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1760
1761         mutex_lock(&udev->cmdr_lock);
1762         if (tcmu_handle_completions(udev))
1763                 run_qfull_queue(udev, false);
1764         mutex_unlock(&udev->cmdr_lock);
1765
1766         return 0;
1767 }
1768
1769 /*
1770  * mmap code from uio.c. Copied here because we want to hook mmap()
1771  * and this stuff must come along.
1772  */
1773 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1774 {
1775         struct tcmu_dev *udev = vma->vm_private_data;
1776         struct uio_info *info = &udev->uio_info;
1777
1778         if (vma->vm_pgoff < MAX_UIO_MAPS) {
1779                 if (info->mem[vma->vm_pgoff].size == 0)
1780                         return -1;
1781                 return (int)vma->vm_pgoff;
1782         }
1783         return -1;
1784 }
1785
1786 static struct page *tcmu_try_get_data_page(struct tcmu_dev *udev, uint32_t dpi)
1787 {
1788         struct page *page;
1789
1790         mutex_lock(&udev->cmdr_lock);
1791         page = xa_load(&udev->data_pages, dpi);
1792         if (likely(page)) {
1793                 mutex_unlock(&udev->cmdr_lock);
1794                 return page;
1795         }
1796
1797         /*
1798          * Userspace messed up and passed in a address not in the
1799          * data iov passed to it.
1800          */
1801         pr_err("Invalid addr to data page mapping (dpi %u) on device %s\n",
1802                dpi, udev->name);
1803         mutex_unlock(&udev->cmdr_lock);
1804
1805         return NULL;
1806 }
1807
1808 static void tcmu_vma_open(struct vm_area_struct *vma)
1809 {
1810         struct tcmu_dev *udev = vma->vm_private_data;
1811
1812         pr_debug("vma_open\n");
1813
1814         kref_get(&udev->kref);
1815 }
1816
1817 static void tcmu_vma_close(struct vm_area_struct *vma)
1818 {
1819         struct tcmu_dev *udev = vma->vm_private_data;
1820
1821         pr_debug("vma_close\n");
1822
1823         /* release ref from tcmu_vma_open */
1824         kref_put(&udev->kref, tcmu_dev_kref_release);
1825 }
1826
1827 static vm_fault_t tcmu_vma_fault(struct vm_fault *vmf)
1828 {
1829         struct tcmu_dev *udev = vmf->vma->vm_private_data;
1830         struct uio_info *info = &udev->uio_info;
1831         struct page *page;
1832         unsigned long offset;
1833         void *addr;
1834
1835         int mi = tcmu_find_mem_index(vmf->vma);
1836         if (mi < 0)
1837                 return VM_FAULT_SIGBUS;
1838
1839         /*
1840          * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1841          * to use mem[N].
1842          */
1843         offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1844
1845         if (offset < udev->data_off) {
1846                 /* For the vmalloc()ed cmd area pages */
1847                 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1848                 page = vmalloc_to_page(addr);
1849         } else {
1850                 uint32_t dpi;
1851
1852                 /* For the dynamically growing data area pages */
1853                 dpi = (offset - udev->data_off) / PAGE_SIZE;
1854                 page = tcmu_try_get_data_page(udev, dpi);
1855                 if (!page)
1856                         return VM_FAULT_SIGBUS;
1857         }
1858
1859         get_page(page);
1860         vmf->page = page;
1861         return 0;
1862 }
1863
1864 static const struct vm_operations_struct tcmu_vm_ops = {
1865         .open = tcmu_vma_open,
1866         .close = tcmu_vma_close,
1867         .fault = tcmu_vma_fault,
1868 };
1869
1870 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1871 {
1872         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1873
1874         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1875         vma->vm_ops = &tcmu_vm_ops;
1876
1877         vma->vm_private_data = udev;
1878
1879         /* Ensure the mmap is exactly the right size */
1880         if (vma_pages(vma) != udev->mmap_pages)
1881                 return -EINVAL;
1882
1883         tcmu_vma_open(vma);
1884
1885         return 0;
1886 }
1887
1888 static int tcmu_open(struct uio_info *info, struct inode *inode)
1889 {
1890         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1891
1892         /* O_EXCL not supported for char devs, so fake it? */
1893         if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1894                 return -EBUSY;
1895
1896         udev->inode = inode;
1897
1898         pr_debug("open\n");
1899
1900         return 0;
1901 }
1902
1903 static int tcmu_release(struct uio_info *info, struct inode *inode)
1904 {
1905         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1906
1907         clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1908
1909         pr_debug("close\n");
1910
1911         return 0;
1912 }
1913
1914 static int tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1915 {
1916         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1917
1918         if (!tcmu_kern_cmd_reply_supported)
1919                 return 0;
1920
1921         if (udev->nl_reply_supported <= 0)
1922                 return 0;
1923
1924         mutex_lock(&tcmu_nl_cmd_mutex);
1925
1926         if (tcmu_netlink_blocked) {
1927                 mutex_unlock(&tcmu_nl_cmd_mutex);
1928                 pr_warn("Failing nl cmd %d on %s. Interface is blocked.\n", cmd,
1929                         udev->name);
1930                 return -EAGAIN;
1931         }
1932
1933         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1934                 mutex_unlock(&tcmu_nl_cmd_mutex);
1935                 pr_warn("netlink cmd %d already executing on %s\n",
1936                          nl_cmd->cmd, udev->name);
1937                 return -EBUSY;
1938         }
1939
1940         memset(nl_cmd, 0, sizeof(*nl_cmd));
1941         nl_cmd->cmd = cmd;
1942         nl_cmd->udev = udev;
1943         init_completion(&nl_cmd->complete);
1944         INIT_LIST_HEAD(&nl_cmd->nl_list);
1945
1946         list_add_tail(&nl_cmd->nl_list, &tcmu_nl_cmd_list);
1947
1948         mutex_unlock(&tcmu_nl_cmd_mutex);
1949         return 0;
1950 }
1951
1952 static void tcmu_destroy_genl_cmd_reply(struct tcmu_dev *udev)
1953 {
1954         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1955
1956         if (!tcmu_kern_cmd_reply_supported)
1957                 return;
1958
1959         if (udev->nl_reply_supported <= 0)
1960                 return;
1961
1962         mutex_lock(&tcmu_nl_cmd_mutex);
1963
1964         list_del(&nl_cmd->nl_list);
1965         memset(nl_cmd, 0, sizeof(*nl_cmd));
1966
1967         mutex_unlock(&tcmu_nl_cmd_mutex);
1968 }
1969
1970 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1971 {
1972         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1973         int ret;
1974
1975         if (!tcmu_kern_cmd_reply_supported)
1976                 return 0;
1977
1978         if (udev->nl_reply_supported <= 0)
1979                 return 0;
1980
1981         pr_debug("sleeping for nl reply\n");
1982         wait_for_completion(&nl_cmd->complete);
1983
1984         mutex_lock(&tcmu_nl_cmd_mutex);
1985         nl_cmd->cmd = TCMU_CMD_UNSPEC;
1986         ret = nl_cmd->status;
1987         mutex_unlock(&tcmu_nl_cmd_mutex);
1988
1989         return ret;
1990 }
1991
1992 static int tcmu_netlink_event_init(struct tcmu_dev *udev,
1993                                    enum tcmu_genl_cmd cmd,
1994                                    struct sk_buff **buf, void **hdr)
1995 {
1996         struct sk_buff *skb;
1997         void *msg_header;
1998         int ret = -ENOMEM;
1999
2000         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
2001         if (!skb)
2002                 return ret;
2003
2004         msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
2005         if (!msg_header)
2006                 goto free_skb;
2007
2008         ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
2009         if (ret < 0)
2010                 goto free_skb;
2011
2012         ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
2013         if (ret < 0)
2014                 goto free_skb;
2015
2016         ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
2017         if (ret < 0)
2018                 goto free_skb;
2019
2020         *buf = skb;
2021         *hdr = msg_header;
2022         return ret;
2023
2024 free_skb:
2025         nlmsg_free(skb);
2026         return ret;
2027 }
2028
2029 static int tcmu_netlink_event_send(struct tcmu_dev *udev,
2030                                    enum tcmu_genl_cmd cmd,
2031                                    struct sk_buff *skb, void *msg_header)
2032 {
2033         int ret;
2034
2035         genlmsg_end(skb, msg_header);
2036
2037         ret = tcmu_init_genl_cmd_reply(udev, cmd);
2038         if (ret) {
2039                 nlmsg_free(skb);
2040                 return ret;
2041         }
2042
2043         ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
2044                                       TCMU_MCGRP_CONFIG, GFP_KERNEL);
2045
2046         /* Wait during an add as the listener may not be up yet */
2047         if (ret == 0 ||
2048            (ret == -ESRCH && cmd == TCMU_CMD_ADDED_DEVICE))
2049                 return tcmu_wait_genl_cmd_reply(udev);
2050         else
2051                 tcmu_destroy_genl_cmd_reply(udev);
2052
2053         return ret;
2054 }
2055
2056 static int tcmu_send_dev_add_event(struct tcmu_dev *udev)
2057 {
2058         struct sk_buff *skb = NULL;
2059         void *msg_header = NULL;
2060         int ret = 0;
2061
2062         ret = tcmu_netlink_event_init(udev, TCMU_CMD_ADDED_DEVICE, &skb,
2063                                       &msg_header);
2064         if (ret < 0)
2065                 return ret;
2066         return tcmu_netlink_event_send(udev, TCMU_CMD_ADDED_DEVICE, skb,
2067                                        msg_header);
2068 }
2069
2070 static int tcmu_send_dev_remove_event(struct tcmu_dev *udev)
2071 {
2072         struct sk_buff *skb = NULL;
2073         void *msg_header = NULL;
2074         int ret = 0;
2075
2076         ret = tcmu_netlink_event_init(udev, TCMU_CMD_REMOVED_DEVICE,
2077                                       &skb, &msg_header);
2078         if (ret < 0)
2079                 return ret;
2080         return tcmu_netlink_event_send(udev, TCMU_CMD_REMOVED_DEVICE,
2081                                        skb, msg_header);
2082 }
2083
2084 static int tcmu_update_uio_info(struct tcmu_dev *udev)
2085 {
2086         struct tcmu_hba *hba = udev->hba->hba_ptr;
2087         struct uio_info *info;
2088         char *str;
2089
2090         info = &udev->uio_info;
2091
2092         if (udev->dev_config[0])
2093                 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s/%s", hba->host_id,
2094                                 udev->name, udev->dev_config);
2095         else
2096                 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s", hba->host_id,
2097                                 udev->name);
2098         if (!str)
2099                 return -ENOMEM;
2100
2101         /* If the old string exists, free it */
2102         kfree(info->name);
2103         info->name = str;
2104
2105         return 0;
2106 }
2107
2108 static int tcmu_configure_device(struct se_device *dev)
2109 {
2110         struct tcmu_dev *udev = TCMU_DEV(dev);
2111         struct uio_info *info;
2112         struct tcmu_mailbox *mb;
2113         size_t data_size;
2114         int ret = 0;
2115
2116         ret = tcmu_update_uio_info(udev);
2117         if (ret)
2118                 return ret;
2119
2120         info = &udev->uio_info;
2121
2122         mutex_lock(&udev->cmdr_lock);
2123         udev->data_bitmap = bitmap_zalloc(udev->max_blocks, GFP_KERNEL);
2124         mutex_unlock(&udev->cmdr_lock);
2125         if (!udev->data_bitmap) {
2126                 ret = -ENOMEM;
2127                 goto err_bitmap_alloc;
2128         }
2129
2130         mb = vzalloc(MB_CMDR_SIZE);
2131         if (!mb) {
2132                 ret = -ENOMEM;
2133                 goto err_vzalloc;
2134         }
2135
2136         /* mailbox fits in first part of CMDR space */
2137         udev->mb_addr = mb;
2138         udev->cmdr = (void *)mb + CMDR_OFF;
2139         udev->cmdr_size = CMDR_SIZE;
2140         udev->data_off = MB_CMDR_SIZE;
2141         data_size = TCMU_MBS_TO_PAGES(udev->data_area_mb) << PAGE_SHIFT;
2142         udev->mmap_pages = (data_size + MB_CMDR_SIZE) >> PAGE_SHIFT;
2143         udev->data_blk_size = udev->data_pages_per_blk * PAGE_SIZE;
2144         udev->dbi_thresh = 0; /* Default in Idle state */
2145
2146         /* Initialise the mailbox of the ring buffer */
2147         mb->version = TCMU_MAILBOX_VERSION;
2148         mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC |
2149                     TCMU_MAILBOX_FLAG_CAP_READ_LEN |
2150                     TCMU_MAILBOX_FLAG_CAP_TMR;
2151         mb->cmdr_off = CMDR_OFF;
2152         mb->cmdr_size = udev->cmdr_size;
2153
2154         WARN_ON(!PAGE_ALIGNED(udev->data_off));
2155         WARN_ON(data_size % PAGE_SIZE);
2156
2157         info->version = __stringify(TCMU_MAILBOX_VERSION);
2158
2159         info->mem[0].name = "tcm-user command & data buffer";
2160         info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
2161         info->mem[0].size = data_size + MB_CMDR_SIZE;
2162         info->mem[0].memtype = UIO_MEM_NONE;
2163
2164         info->irqcontrol = tcmu_irqcontrol;
2165         info->irq = UIO_IRQ_CUSTOM;
2166
2167         info->mmap = tcmu_mmap;
2168         info->open = tcmu_open;
2169         info->release = tcmu_release;
2170
2171         ret = uio_register_device(tcmu_root_device, info);
2172         if (ret)
2173                 goto err_register;
2174
2175         /* User can set hw_block_size before enable the device */
2176         if (dev->dev_attrib.hw_block_size == 0)
2177                 dev->dev_attrib.hw_block_size = 512;
2178         /* Other attributes can be configured in userspace */
2179         if (!dev->dev_attrib.hw_max_sectors)
2180                 dev->dev_attrib.hw_max_sectors = 128;
2181         if (!dev->dev_attrib.emulate_write_cache)
2182                 dev->dev_attrib.emulate_write_cache = 0;
2183         dev->dev_attrib.hw_queue_depth = 128;
2184
2185         /* If user didn't explicitly disable netlink reply support, use
2186          * module scope setting.
2187          */
2188         if (udev->nl_reply_supported >= 0)
2189                 udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
2190
2191         /*
2192          * Get a ref incase userspace does a close on the uio device before
2193          * LIO has initiated tcmu_free_device.
2194          */
2195         kref_get(&udev->kref);
2196
2197         ret = tcmu_send_dev_add_event(udev);
2198         if (ret)
2199                 goto err_netlink;
2200
2201         mutex_lock(&root_udev_mutex);
2202         list_add(&udev->node, &root_udev);
2203         mutex_unlock(&root_udev_mutex);
2204
2205         return 0;
2206
2207 err_netlink:
2208         kref_put(&udev->kref, tcmu_dev_kref_release);
2209         uio_unregister_device(&udev->uio_info);
2210 err_register:
2211         vfree(udev->mb_addr);
2212         udev->mb_addr = NULL;
2213 err_vzalloc:
2214         bitmap_free(udev->data_bitmap);
2215         udev->data_bitmap = NULL;
2216 err_bitmap_alloc:
2217         kfree(info->name);
2218         info->name = NULL;
2219
2220         return ret;
2221 }
2222
2223 static void tcmu_free_device(struct se_device *dev)
2224 {
2225         struct tcmu_dev *udev = TCMU_DEV(dev);
2226
2227         /* release ref from init */
2228         kref_put(&udev->kref, tcmu_dev_kref_release);
2229 }
2230
2231 static void tcmu_destroy_device(struct se_device *dev)
2232 {
2233         struct tcmu_dev *udev = TCMU_DEV(dev);
2234
2235         del_timer_sync(&udev->cmd_timer);
2236         del_timer_sync(&udev->qfull_timer);
2237
2238         mutex_lock(&root_udev_mutex);
2239         list_del(&udev->node);
2240         mutex_unlock(&root_udev_mutex);
2241
2242         tcmu_send_dev_remove_event(udev);
2243
2244         uio_unregister_device(&udev->uio_info);
2245
2246         /* release ref from configure */
2247         kref_put(&udev->kref, tcmu_dev_kref_release);
2248 }
2249
2250 static void tcmu_unblock_dev(struct tcmu_dev *udev)
2251 {
2252         mutex_lock(&udev->cmdr_lock);
2253         clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
2254         mutex_unlock(&udev->cmdr_lock);
2255 }
2256
2257 static void tcmu_block_dev(struct tcmu_dev *udev)
2258 {
2259         mutex_lock(&udev->cmdr_lock);
2260
2261         if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2262                 goto unlock;
2263
2264         /* complete IO that has executed successfully */
2265         tcmu_handle_completions(udev);
2266         /* fail IO waiting to be queued */
2267         run_qfull_queue(udev, true);
2268
2269 unlock:
2270         mutex_unlock(&udev->cmdr_lock);
2271 }
2272
2273 static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
2274 {
2275         struct tcmu_mailbox *mb;
2276         struct tcmu_cmd *cmd;
2277         unsigned long i;
2278
2279         mutex_lock(&udev->cmdr_lock);
2280
2281         xa_for_each(&udev->commands, i, cmd) {
2282                 pr_debug("removing cmd %u on dev %s from ring (is expired %d)\n",
2283                           cmd->cmd_id, udev->name,
2284                           test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags));
2285
2286                 xa_erase(&udev->commands, i);
2287                 if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
2288                         WARN_ON(!cmd->se_cmd);
2289                         list_del_init(&cmd->queue_entry);
2290                         cmd->se_cmd->priv = NULL;
2291                         if (err_level == 1) {
2292                                 /*
2293                                  * Userspace was not able to start the
2294                                  * command or it is retryable.
2295                                  */
2296                                 target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
2297                         } else {
2298                                 /* hard failure */
2299                                 target_complete_cmd(cmd->se_cmd,
2300                                                     SAM_STAT_CHECK_CONDITION);
2301                         }
2302                 }
2303                 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
2304                 tcmu_free_cmd(cmd);
2305         }
2306
2307         mb = udev->mb_addr;
2308         tcmu_flush_dcache_range(mb, sizeof(*mb));
2309         pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
2310                  mb->cmd_tail, mb->cmd_head);
2311
2312         udev->cmdr_last_cleaned = 0;
2313         mb->cmd_tail = 0;
2314         mb->cmd_head = 0;
2315         tcmu_flush_dcache_range(mb, sizeof(*mb));
2316         clear_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
2317
2318         del_timer(&udev->cmd_timer);
2319
2320         /*
2321          * ring is empty and qfull queue never contains aborted commands.
2322          * So TMRs in tmr queue do not contain relevant cmd_ids.
2323          * After a ring reset userspace should do a fresh start, so
2324          * even LUN RESET message is no longer relevant.
2325          * Therefore remove all TMRs from qfull queue
2326          */
2327         tcmu_remove_all_queued_tmr(udev);
2328
2329         run_qfull_queue(udev, false);
2330
2331         mutex_unlock(&udev->cmdr_lock);
2332 }
2333
2334 enum {
2335         Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
2336         Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_data_pages_per_blk,
2337         Opt_err,
2338 };
2339
2340 static match_table_t tokens = {
2341         {Opt_dev_config, "dev_config=%s"},
2342         {Opt_dev_size, "dev_size=%s"},
2343         {Opt_hw_block_size, "hw_block_size=%d"},
2344         {Opt_hw_max_sectors, "hw_max_sectors=%d"},
2345         {Opt_nl_reply_supported, "nl_reply_supported=%d"},
2346         {Opt_max_data_area_mb, "max_data_area_mb=%d"},
2347         {Opt_data_pages_per_blk, "data_pages_per_blk=%d"},
2348         {Opt_err, NULL}
2349 };
2350
2351 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
2352 {
2353         int val, ret;
2354
2355         ret = match_int(arg, &val);
2356         if (ret < 0) {
2357                 pr_err("match_int() failed for dev attrib. Error %d.\n",
2358                        ret);
2359                 return ret;
2360         }
2361
2362         if (val <= 0) {
2363                 pr_err("Invalid dev attrib value %d. Must be greater than zero.\n",
2364                        val);
2365                 return -EINVAL;
2366         }
2367         *dev_attrib = val;
2368         return 0;
2369 }
2370
2371 static int tcmu_set_max_blocks_param(struct tcmu_dev *udev, substring_t *arg)
2372 {
2373         int val, ret;
2374         uint32_t pages_per_blk = udev->data_pages_per_blk;
2375
2376         ret = match_int(arg, &val);
2377         if (ret < 0) {
2378                 pr_err("match_int() failed for max_data_area_mb=. Error %d.\n",
2379                        ret);
2380                 return ret;
2381         }
2382         if (val <= 0) {
2383                 pr_err("Invalid max_data_area %d.\n", val);
2384                 return -EINVAL;
2385         }
2386         if (val > TCMU_PAGES_TO_MBS(tcmu_global_max_pages)) {
2387                 pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2388                        val, TCMU_PAGES_TO_MBS(tcmu_global_max_pages));
2389                 val = TCMU_PAGES_TO_MBS(tcmu_global_max_pages);
2390         }
2391         if (TCMU_MBS_TO_PAGES(val) < pages_per_blk) {
2392                 pr_err("Invalid max_data_area %d (%zu pages): smaller than data_pages_per_blk (%u pages).\n",
2393                        val, TCMU_MBS_TO_PAGES(val), pages_per_blk);
2394                 return -EINVAL;
2395         }
2396
2397         mutex_lock(&udev->cmdr_lock);
2398         if (udev->data_bitmap) {
2399                 pr_err("Cannot set max_data_area_mb after it has been enabled.\n");
2400                 ret = -EINVAL;
2401                 goto unlock;
2402         }
2403
2404         udev->data_area_mb = val;
2405         udev->max_blocks = TCMU_MBS_TO_PAGES(val) / pages_per_blk;
2406
2407 unlock:
2408         mutex_unlock(&udev->cmdr_lock);
2409         return ret;
2410 }
2411
2412 static int tcmu_set_data_pages_per_blk(struct tcmu_dev *udev, substring_t *arg)
2413 {
2414         int val, ret;
2415
2416         ret = match_int(arg, &val);
2417         if (ret < 0) {
2418                 pr_err("match_int() failed for data_pages_per_blk=. Error %d.\n",
2419                        ret);
2420                 return ret;
2421         }
2422
2423         if (val > TCMU_MBS_TO_PAGES(udev->data_area_mb)) {
2424                 pr_err("Invalid data_pages_per_blk %d: greater than max_data_area_mb %d -> %zd pages).\n",
2425                        val, udev->data_area_mb,
2426                        TCMU_MBS_TO_PAGES(udev->data_area_mb));
2427                 return -EINVAL;
2428         }
2429
2430         mutex_lock(&udev->cmdr_lock);
2431         if (udev->data_bitmap) {
2432                 pr_err("Cannot set data_pages_per_blk after it has been enabled.\n");
2433                 ret = -EINVAL;
2434                 goto unlock;
2435         }
2436
2437         udev->data_pages_per_blk = val;
2438         udev->max_blocks = TCMU_MBS_TO_PAGES(udev->data_area_mb) / val;
2439
2440 unlock:
2441         mutex_unlock(&udev->cmdr_lock);
2442         return ret;
2443 }
2444
2445 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
2446                 const char *page, ssize_t count)
2447 {
2448         struct tcmu_dev *udev = TCMU_DEV(dev);
2449         char *orig, *ptr, *opts;
2450         substring_t args[MAX_OPT_ARGS];
2451         int ret = 0, token;
2452
2453         opts = kstrdup(page, GFP_KERNEL);
2454         if (!opts)
2455                 return -ENOMEM;
2456
2457         orig = opts;
2458
2459         while ((ptr = strsep(&opts, ",\n")) != NULL) {
2460                 if (!*ptr)
2461                         continue;
2462
2463                 token = match_token(ptr, tokens, args);
2464                 switch (token) {
2465                 case Opt_dev_config:
2466                         if (match_strlcpy(udev->dev_config, &args[0],
2467                                           TCMU_CONFIG_LEN) == 0) {
2468                                 ret = -EINVAL;
2469                                 break;
2470                         }
2471                         pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
2472                         break;
2473                 case Opt_dev_size:
2474                         ret = match_u64(&args[0], &udev->dev_size);
2475                         if (ret < 0)
2476                                 pr_err("match_u64() failed for dev_size=. Error %d.\n",
2477                                        ret);
2478                         break;
2479                 case Opt_hw_block_size:
2480                         ret = tcmu_set_dev_attrib(&args[0],
2481                                         &(dev->dev_attrib.hw_block_size));
2482                         break;
2483                 case Opt_hw_max_sectors:
2484                         ret = tcmu_set_dev_attrib(&args[0],
2485                                         &(dev->dev_attrib.hw_max_sectors));
2486                         break;
2487                 case Opt_nl_reply_supported:
2488                         ret = match_int(&args[0], &udev->nl_reply_supported);
2489                         if (ret < 0)
2490                                 pr_err("match_int() failed for nl_reply_supported=. Error %d.\n",
2491                                        ret);
2492                         break;
2493                 case Opt_max_data_area_mb:
2494                         ret = tcmu_set_max_blocks_param(udev, &args[0]);
2495                         break;
2496                 case Opt_data_pages_per_blk:
2497                         ret = tcmu_set_data_pages_per_blk(udev, &args[0]);
2498                         break;
2499                 default:
2500                         break;
2501                 }
2502
2503                 if (ret)
2504                         break;
2505         }
2506
2507         kfree(orig);
2508         return (!ret) ? count : ret;
2509 }
2510
2511 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2512 {
2513         struct tcmu_dev *udev = TCMU_DEV(dev);
2514         ssize_t bl = 0;
2515
2516         bl = sprintf(b + bl, "Config: %s ",
2517                      udev->dev_config[0] ? udev->dev_config : "NULL");
2518         bl += sprintf(b + bl, "Size: %llu ", udev->dev_size);
2519         bl += sprintf(b + bl, "MaxDataAreaMB: %u ", udev->data_area_mb);
2520         bl += sprintf(b + bl, "DataPagesPerBlk: %u\n", udev->data_pages_per_blk);
2521
2522         return bl;
2523 }
2524
2525 static sector_t tcmu_get_blocks(struct se_device *dev)
2526 {
2527         struct tcmu_dev *udev = TCMU_DEV(dev);
2528
2529         return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2530                        dev->dev_attrib.block_size);
2531 }
2532
2533 static sense_reason_t
2534 tcmu_parse_cdb(struct se_cmd *cmd)
2535 {
2536         return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2537 }
2538
2539 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2540 {
2541         struct se_dev_attrib *da = container_of(to_config_group(item),
2542                                         struct se_dev_attrib, da_group);
2543         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2544
2545         return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2546 }
2547
2548 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2549                                        size_t count)
2550 {
2551         struct se_dev_attrib *da = container_of(to_config_group(item),
2552                                         struct se_dev_attrib, da_group);
2553         struct tcmu_dev *udev = container_of(da->da_dev,
2554                                         struct tcmu_dev, se_dev);
2555         u32 val;
2556         int ret;
2557
2558         if (da->da_dev->export_count) {
2559                 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2560                 return -EINVAL;
2561         }
2562
2563         ret = kstrtou32(page, 0, &val);
2564         if (ret < 0)
2565                 return ret;
2566
2567         udev->cmd_time_out = val * MSEC_PER_SEC;
2568         return count;
2569 }
2570 CONFIGFS_ATTR(tcmu_, cmd_time_out);
2571
2572 static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2573 {
2574         struct se_dev_attrib *da = container_of(to_config_group(item),
2575                                                 struct se_dev_attrib, da_group);
2576         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2577
2578         return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2579                         udev->qfull_time_out :
2580                         udev->qfull_time_out / MSEC_PER_SEC);
2581 }
2582
2583 static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2584                                          const char *page, size_t count)
2585 {
2586         struct se_dev_attrib *da = container_of(to_config_group(item),
2587                                         struct se_dev_attrib, da_group);
2588         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2589         s32 val;
2590         int ret;
2591
2592         ret = kstrtos32(page, 0, &val);
2593         if (ret < 0)
2594                 return ret;
2595
2596         if (val >= 0) {
2597                 udev->qfull_time_out = val * MSEC_PER_SEC;
2598         } else if (val == -1) {
2599                 udev->qfull_time_out = val;
2600         } else {
2601                 printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2602                 return -EINVAL;
2603         }
2604         return count;
2605 }
2606 CONFIGFS_ATTR(tcmu_, qfull_time_out);
2607
2608 static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2609 {
2610         struct se_dev_attrib *da = container_of(to_config_group(item),
2611                                                 struct se_dev_attrib, da_group);
2612         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2613
2614         return snprintf(page, PAGE_SIZE, "%u\n", udev->data_area_mb);
2615 }
2616 CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2617
2618 static ssize_t tcmu_data_pages_per_blk_show(struct config_item *item,
2619                                             char *page)
2620 {
2621         struct se_dev_attrib *da = container_of(to_config_group(item),
2622                                                 struct se_dev_attrib, da_group);
2623         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2624
2625         return snprintf(page, PAGE_SIZE, "%u\n", udev->data_pages_per_blk);
2626 }
2627 CONFIGFS_ATTR_RO(tcmu_, data_pages_per_blk);
2628
2629 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2630 {
2631         struct se_dev_attrib *da = container_of(to_config_group(item),
2632                                                 struct se_dev_attrib, da_group);
2633         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2634
2635         return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2636 }
2637
2638 static int tcmu_send_dev_config_event(struct tcmu_dev *udev,
2639                                       const char *reconfig_data)
2640 {
2641         struct sk_buff *skb = NULL;
2642         void *msg_header = NULL;
2643         int ret = 0;
2644
2645         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2646                                       &skb, &msg_header);
2647         if (ret < 0)
2648                 return ret;
2649         ret = nla_put_string(skb, TCMU_ATTR_DEV_CFG, reconfig_data);
2650         if (ret < 0) {
2651                 nlmsg_free(skb);
2652                 return ret;
2653         }
2654         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2655                                        skb, msg_header);
2656 }
2657
2658
2659 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2660                                      size_t count)
2661 {
2662         struct se_dev_attrib *da = container_of(to_config_group(item),
2663                                                 struct se_dev_attrib, da_group);
2664         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2665         int ret, len;
2666
2667         len = strlen(page);
2668         if (!len || len > TCMU_CONFIG_LEN - 1)
2669                 return -EINVAL;
2670
2671         /* Check if device has been configured before */
2672         if (target_dev_configured(&udev->se_dev)) {
2673                 ret = tcmu_send_dev_config_event(udev, page);
2674                 if (ret) {
2675                         pr_err("Unable to reconfigure device\n");
2676                         return ret;
2677                 }
2678                 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2679
2680                 ret = tcmu_update_uio_info(udev);
2681                 if (ret)
2682                         return ret;
2683                 return count;
2684         }
2685         strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2686
2687         return count;
2688 }
2689 CONFIGFS_ATTR(tcmu_, dev_config);
2690
2691 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2692 {
2693         struct se_dev_attrib *da = container_of(to_config_group(item),
2694                                                 struct se_dev_attrib, da_group);
2695         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2696
2697         return snprintf(page, PAGE_SIZE, "%llu\n", udev->dev_size);
2698 }
2699
2700 static int tcmu_send_dev_size_event(struct tcmu_dev *udev, u64 size)
2701 {
2702         struct sk_buff *skb = NULL;
2703         void *msg_header = NULL;
2704         int ret = 0;
2705
2706         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2707                                       &skb, &msg_header);
2708         if (ret < 0)
2709                 return ret;
2710         ret = nla_put_u64_64bit(skb, TCMU_ATTR_DEV_SIZE,
2711                                 size, TCMU_ATTR_PAD);
2712         if (ret < 0) {
2713                 nlmsg_free(skb);
2714                 return ret;
2715         }
2716         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2717                                        skb, msg_header);
2718 }
2719
2720 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2721                                    size_t count)
2722 {
2723         struct se_dev_attrib *da = container_of(to_config_group(item),
2724                                                 struct se_dev_attrib, da_group);
2725         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2726         u64 val;
2727         int ret;
2728
2729         ret = kstrtou64(page, 0, &val);
2730         if (ret < 0)
2731                 return ret;
2732
2733         /* Check if device has been configured before */
2734         if (target_dev_configured(&udev->se_dev)) {
2735                 ret = tcmu_send_dev_size_event(udev, val);
2736                 if (ret) {
2737                         pr_err("Unable to reconfigure device\n");
2738                         return ret;
2739                 }
2740         }
2741         udev->dev_size = val;
2742         return count;
2743 }
2744 CONFIGFS_ATTR(tcmu_, dev_size);
2745
2746 static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
2747                 char *page)
2748 {
2749         struct se_dev_attrib *da = container_of(to_config_group(item),
2750                                                 struct se_dev_attrib, da_group);
2751         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2752
2753         return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2754 }
2755
2756 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2757                 const char *page, size_t count)
2758 {
2759         struct se_dev_attrib *da = container_of(to_config_group(item),
2760                                                 struct se_dev_attrib, da_group);
2761         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2762         s8 val;
2763         int ret;
2764
2765         ret = kstrtos8(page, 0, &val);
2766         if (ret < 0)
2767                 return ret;
2768
2769         udev->nl_reply_supported = val;
2770         return count;
2771 }
2772 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2773
2774 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2775                                              char *page)
2776 {
2777         struct se_dev_attrib *da = container_of(to_config_group(item),
2778                                         struct se_dev_attrib, da_group);
2779
2780         return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2781 }
2782
2783 static int tcmu_send_emulate_write_cache(struct tcmu_dev *udev, u8 val)
2784 {
2785         struct sk_buff *skb = NULL;
2786         void *msg_header = NULL;
2787         int ret = 0;
2788
2789         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2790                                       &skb, &msg_header);
2791         if (ret < 0)
2792                 return ret;
2793         ret = nla_put_u8(skb, TCMU_ATTR_WRITECACHE, val);
2794         if (ret < 0) {
2795                 nlmsg_free(skb);
2796                 return ret;
2797         }
2798         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2799                                        skb, msg_header);
2800 }
2801
2802 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2803                                               const char *page, size_t count)
2804 {
2805         struct se_dev_attrib *da = container_of(to_config_group(item),
2806                                         struct se_dev_attrib, da_group);
2807         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2808         u8 val;
2809         int ret;
2810
2811         ret = kstrtou8(page, 0, &val);
2812         if (ret < 0)
2813                 return ret;
2814
2815         /* Check if device has been configured before */
2816         if (target_dev_configured(&udev->se_dev)) {
2817                 ret = tcmu_send_emulate_write_cache(udev, val);
2818                 if (ret) {
2819                         pr_err("Unable to reconfigure device\n");
2820                         return ret;
2821                 }
2822         }
2823
2824         da->emulate_write_cache = val;
2825         return count;
2826 }
2827 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2828
2829 static ssize_t tcmu_tmr_notification_show(struct config_item *item, char *page)
2830 {
2831         struct se_dev_attrib *da = container_of(to_config_group(item),
2832                                         struct se_dev_attrib, da_group);
2833         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2834
2835         return snprintf(page, PAGE_SIZE, "%i\n",
2836                         test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags));
2837 }
2838
2839 static ssize_t tcmu_tmr_notification_store(struct config_item *item,
2840                                            const char *page, size_t count)
2841 {
2842         struct se_dev_attrib *da = container_of(to_config_group(item),
2843                                         struct se_dev_attrib, da_group);
2844         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2845         u8 val;
2846         int ret;
2847
2848         ret = kstrtou8(page, 0, &val);
2849         if (ret < 0)
2850                 return ret;
2851         if (val > 1)
2852                 return -EINVAL;
2853
2854         if (val)
2855                 set_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
2856         else
2857                 clear_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
2858         return count;
2859 }
2860 CONFIGFS_ATTR(tcmu_, tmr_notification);
2861
2862 static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
2863 {
2864         struct se_device *se_dev = container_of(to_config_group(item),
2865                                                 struct se_device,
2866                                                 dev_action_group);
2867         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2868
2869         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2870                 return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
2871         else
2872                 return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
2873 }
2874
2875 static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
2876                                     size_t count)
2877 {
2878         struct se_device *se_dev = container_of(to_config_group(item),
2879                                                 struct se_device,
2880                                                 dev_action_group);
2881         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2882         u8 val;
2883         int ret;
2884
2885         if (!target_dev_configured(&udev->se_dev)) {
2886                 pr_err("Device is not configured.\n");
2887                 return -EINVAL;
2888         }
2889
2890         ret = kstrtou8(page, 0, &val);
2891         if (ret < 0)
2892                 return ret;
2893
2894         if (val > 1) {
2895                 pr_err("Invalid block value %d\n", val);
2896                 return -EINVAL;
2897         }
2898
2899         if (!val)
2900                 tcmu_unblock_dev(udev);
2901         else
2902                 tcmu_block_dev(udev);
2903         return count;
2904 }
2905 CONFIGFS_ATTR(tcmu_, block_dev);
2906
2907 static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
2908                                      size_t count)
2909 {
2910         struct se_device *se_dev = container_of(to_config_group(item),
2911                                                 struct se_device,
2912                                                 dev_action_group);
2913         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2914         u8 val;
2915         int ret;
2916
2917         if (!target_dev_configured(&udev->se_dev)) {
2918                 pr_err("Device is not configured.\n");
2919                 return -EINVAL;
2920         }
2921
2922         ret = kstrtou8(page, 0, &val);
2923         if (ret < 0)
2924                 return ret;
2925
2926         if (val != 1 && val != 2) {
2927                 pr_err("Invalid reset ring value %d\n", val);
2928                 return -EINVAL;
2929         }
2930
2931         tcmu_reset_ring(udev, val);
2932         return count;
2933 }
2934 CONFIGFS_ATTR_WO(tcmu_, reset_ring);
2935
2936 static struct configfs_attribute *tcmu_attrib_attrs[] = {
2937         &tcmu_attr_cmd_time_out,
2938         &tcmu_attr_qfull_time_out,
2939         &tcmu_attr_max_data_area_mb,
2940         &tcmu_attr_data_pages_per_blk,
2941         &tcmu_attr_dev_config,
2942         &tcmu_attr_dev_size,
2943         &tcmu_attr_emulate_write_cache,
2944         &tcmu_attr_tmr_notification,
2945         &tcmu_attr_nl_reply_supported,
2946         NULL,
2947 };
2948
2949 static struct configfs_attribute **tcmu_attrs;
2950
2951 static struct configfs_attribute *tcmu_action_attrs[] = {
2952         &tcmu_attr_block_dev,
2953         &tcmu_attr_reset_ring,
2954         NULL,
2955 };
2956
2957 static struct target_backend_ops tcmu_ops = {
2958         .name                   = "user",
2959         .owner                  = THIS_MODULE,
2960         .transport_flags_default = TRANSPORT_FLAG_PASSTHROUGH,
2961         .transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR |
2962                                       TRANSPORT_FLAG_PASSTHROUGH_ALUA,
2963         .attach_hba             = tcmu_attach_hba,
2964         .detach_hba             = tcmu_detach_hba,
2965         .alloc_device           = tcmu_alloc_device,
2966         .configure_device       = tcmu_configure_device,
2967         .destroy_device         = tcmu_destroy_device,
2968         .free_device            = tcmu_free_device,
2969         .unplug_device          = tcmu_unplug_device,
2970         .plug_device            = tcmu_plug_device,
2971         .parse_cdb              = tcmu_parse_cdb,
2972         .tmr_notify             = tcmu_tmr_notify,
2973         .set_configfs_dev_params = tcmu_set_configfs_dev_params,
2974         .show_configfs_dev_params = tcmu_show_configfs_dev_params,
2975         .get_device_type        = sbc_get_device_type,
2976         .get_blocks             = tcmu_get_blocks,
2977         .tb_dev_action_attrs    = tcmu_action_attrs,
2978 };
2979
2980 static void find_free_blocks(void)
2981 {
2982         struct tcmu_dev *udev;
2983         loff_t off;
2984         u32 pages_freed, total_pages_freed = 0;
2985         u32 start, end, block, total_blocks_freed = 0;
2986
2987         if (atomic_read(&global_page_count) <= tcmu_global_max_pages)
2988                 return;
2989
2990         mutex_lock(&root_udev_mutex);
2991         list_for_each_entry(udev, &root_udev, node) {
2992                 mutex_lock(&udev->cmdr_lock);
2993
2994                 if (!target_dev_configured(&udev->se_dev)) {
2995                         mutex_unlock(&udev->cmdr_lock);
2996                         continue;
2997                 }
2998
2999                 /* Try to complete the finished commands first */
3000                 if (tcmu_handle_completions(udev))
3001                         run_qfull_queue(udev, false);
3002
3003                 /* Skip the udevs in idle */
3004                 if (!udev->dbi_thresh) {
3005                         mutex_unlock(&udev->cmdr_lock);
3006                         continue;
3007                 }
3008
3009                 end = udev->dbi_max + 1;
3010                 block = find_last_bit(udev->data_bitmap, end);
3011                 if (block == udev->dbi_max) {
3012                         /*
3013                          * The last bit is dbi_max, so it is not possible
3014                          * reclaim any blocks.
3015                          */
3016                         mutex_unlock(&udev->cmdr_lock);
3017                         continue;
3018                 } else if (block == end) {
3019                         /* The current udev will goto idle state */
3020                         udev->dbi_thresh = start = 0;
3021                         udev->dbi_max = 0;
3022                 } else {
3023                         udev->dbi_thresh = start = block + 1;
3024                         udev->dbi_max = block;
3025                 }
3026
3027                 /* Here will truncate the data area from off */
3028                 off = udev->data_off + (loff_t)start * udev->data_blk_size;
3029                 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
3030
3031                 /* Release the block pages */
3032                 pages_freed = tcmu_blocks_release(udev, start, end - 1);
3033                 mutex_unlock(&udev->cmdr_lock);
3034
3035                 total_pages_freed += pages_freed;
3036                 total_blocks_freed += end - start;
3037                 pr_debug("Freed %u pages (total %u) from %u blocks (total %u) from %s.\n",
3038                          pages_freed, total_pages_freed, end - start,
3039                          total_blocks_freed, udev->name);
3040         }
3041         mutex_unlock(&root_udev_mutex);
3042
3043         if (atomic_read(&global_page_count) > tcmu_global_max_pages)
3044                 schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
3045 }
3046
3047 static void check_timedout_devices(void)
3048 {
3049         struct tcmu_dev *udev, *tmp_dev;
3050         struct tcmu_cmd *cmd, *tmp_cmd;
3051         LIST_HEAD(devs);
3052
3053         spin_lock_bh(&timed_out_udevs_lock);
3054         list_splice_init(&timed_out_udevs, &devs);
3055
3056         list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
3057                 list_del_init(&udev->timedout_entry);
3058                 spin_unlock_bh(&timed_out_udevs_lock);
3059
3060                 mutex_lock(&udev->cmdr_lock);
3061
3062                 /*
3063                  * If cmd_time_out is disabled but qfull is set deadline
3064                  * will only reflect the qfull timeout. Ignore it.
3065                  */
3066                 if (udev->cmd_time_out) {
3067                         list_for_each_entry_safe(cmd, tmp_cmd,
3068                                                  &udev->inflight_queue,
3069                                                  queue_entry) {
3070                                 tcmu_check_expired_ring_cmd(cmd);
3071                         }
3072                         tcmu_set_next_deadline(&udev->inflight_queue,
3073                                                &udev->cmd_timer);
3074                 }
3075                 list_for_each_entry_safe(cmd, tmp_cmd, &udev->qfull_queue,
3076                                          queue_entry) {
3077                         tcmu_check_expired_queue_cmd(cmd);
3078                 }
3079                 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
3080
3081                 mutex_unlock(&udev->cmdr_lock);
3082
3083                 spin_lock_bh(&timed_out_udevs_lock);
3084         }
3085
3086         spin_unlock_bh(&timed_out_udevs_lock);
3087 }
3088
3089 static void tcmu_unmap_work_fn(struct work_struct *work)
3090 {
3091         check_timedout_devices();
3092         find_free_blocks();
3093 }
3094
3095 static int __init tcmu_module_init(void)
3096 {
3097         int ret, i, k, len = 0;
3098
3099         BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
3100
3101         INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
3102
3103         tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
3104                                 sizeof(struct tcmu_cmd),
3105                                 __alignof__(struct tcmu_cmd),
3106                                 0, NULL);
3107         if (!tcmu_cmd_cache)
3108                 return -ENOMEM;
3109
3110         tcmu_root_device = root_device_register("tcm_user");
3111         if (IS_ERR(tcmu_root_device)) {
3112                 ret = PTR_ERR(tcmu_root_device);
3113                 goto out_free_cache;
3114         }
3115
3116         ret = genl_register_family(&tcmu_genl_family);
3117         if (ret < 0) {
3118                 goto out_unreg_device;
3119         }
3120
3121         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3122                 len += sizeof(struct configfs_attribute *);
3123         for (i = 0; passthrough_pr_attrib_attrs[i] != NULL; i++)
3124                 len += sizeof(struct configfs_attribute *);
3125         for (i = 0; tcmu_attrib_attrs[i] != NULL; i++)
3126                 len += sizeof(struct configfs_attribute *);
3127         len += sizeof(struct configfs_attribute *);
3128
3129         tcmu_attrs = kzalloc(len, GFP_KERNEL);
3130         if (!tcmu_attrs) {
3131                 ret = -ENOMEM;
3132                 goto out_unreg_genl;
3133         }
3134
3135         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3136                 tcmu_attrs[i] = passthrough_attrib_attrs[i];
3137         for (k = 0; passthrough_pr_attrib_attrs[k] != NULL; k++)
3138                 tcmu_attrs[i++] = passthrough_pr_attrib_attrs[k];
3139         for (k = 0; tcmu_attrib_attrs[k] != NULL; k++)
3140                 tcmu_attrs[i++] = tcmu_attrib_attrs[k];
3141         tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
3142
3143         ret = transport_backend_register(&tcmu_ops);
3144         if (ret)
3145                 goto out_attrs;
3146
3147         return 0;
3148
3149 out_attrs:
3150         kfree(tcmu_attrs);
3151 out_unreg_genl:
3152         genl_unregister_family(&tcmu_genl_family);
3153 out_unreg_device:
3154         root_device_unregister(tcmu_root_device);
3155 out_free_cache:
3156         kmem_cache_destroy(tcmu_cmd_cache);
3157
3158         return ret;
3159 }
3160
3161 static void __exit tcmu_module_exit(void)
3162 {
3163         cancel_delayed_work_sync(&tcmu_unmap_work);
3164         target_backend_unregister(&tcmu_ops);
3165         kfree(tcmu_attrs);
3166         genl_unregister_family(&tcmu_genl_family);
3167         root_device_unregister(tcmu_root_device);
3168         kmem_cache_destroy(tcmu_cmd_cache);
3169 }
3170
3171 MODULE_DESCRIPTION("TCM USER subsystem plugin");
3172 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
3173 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
3174 MODULE_LICENSE("GPL");
3175
3176 module_init(tcmu_module_init);
3177 module_exit(tcmu_module_exit);