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/idr.h>
  12 #include <linux/kernel.h>
  13 #include <linux/timer.h>
  14 #include <linux/parser.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/uio_driver.h>
  17 #include <linux/radix-tree.h>
  18 #include <linux/stringify.h>
  19 #include <linux/bitops.h>
  20 #include <linux/highmem.h>
  21 #include <linux/configfs.h>
  22 #include <linux/mutex.h>
  23 #include <linux/workqueue.h>
  24 #include <net/genetlink.h>
  25 #include <scsi/scsi_common.h>
  26 #include <scsi/scsi_proto.h>
  27 #include <target/target_core_base.h>
  28 #include <target/target_core_fabric.h>
  29 #include <target/target_core_backend.h>
  30
  31 #include <linux/target_core_user.h>
  32
  33 /**
  34  * DOC: Userspace I/O
  35  * Userspace I/O
  36  * -------------
  37  *
  38  * Define a shared-memory interface for LIO to pass SCSI commands and
  39  * data to userspace for processing. This is to allow backends that
  40  * are too complex for in-kernel support to be possible.
  41  *
  42  * It uses the UIO framework to do a lot of the device-creation and
  43  * introspection work for us.
  44  *
  45  * See the .h file for how the ring is laid out. Note that while the
  46  * command ring is defined, the particulars of the data area are
  47  * not. Offset values in the command entry point to other locations
  48  * internal to the mmap-ed area. There is separate space outside the
  49  * command ring for data buffers. This leaves maximum flexibility for
  50  * moving buffer allocations, or even page flipping or other
  51  * allocation techniques, without altering the command ring layout.
  52  *
  53  * SECURITY:
  54  * The user process must be assumed to be malicious. There's no way to
  55  * prevent it breaking the command ring protocol if it wants, but in
  56  * order to prevent other issues we must only ever read *data* from
  57  * the shared memory area, not offsets or sizes. This applies to
  58  * command ring entries as well as the mailbox. Extra code needed for
  59  * this may have a 'UAM' comment.
  60  */
  61
  62 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
  63
  64 /* For cmd area, the size is fixed 8MB */
  65 #define CMDR_SIZE (8 * 1024 * 1024)
  66
  67 /*
  68  * For data area, the block size is PAGE_SIZE and
  69  * the total size is 256K * PAGE_SIZE.
  70  */
  71 #define DATA_BLOCK_SIZE PAGE_SIZE
  72 #define DATA_BLOCK_SHIFT PAGE_SHIFT
  73 #define DATA_BLOCK_BITS_DEF (256 * 1024)
  74
  75 #define TCMU_MBS_TO_BLOCKS(_mbs) (_mbs << (20 - DATA_BLOCK_SHIFT))
  76 #define TCMU_BLOCKS_TO_MBS(_blocks) (_blocks >> (20 - DATA_BLOCK_SHIFT))
  77
  78 /*
  79  * Default number of global data blocks(512K * PAGE_SIZE)
  80  * when the unmap thread will be started.
  81  */
  82 #define TCMU_GLOBAL_MAX_BLOCKS_DEF (512 * 1024)
  83
  84 static u8 tcmu_kern_cmd_reply_supported;
  85 static u8 tcmu_netlink_blocked;
  86
  87 static struct device *tcmu_root_device;
  88
  89 struct tcmu_hba {
  90         u32 host_id;
  91 };
  92
  93 #define TCMU_CONFIG_LEN 256
  94
  95 static DEFINE_MUTEX(tcmu_nl_cmd_mutex);
  96 static LIST_HEAD(tcmu_nl_cmd_list);
  97
  98 struct tcmu_dev;
  99
 100 struct tcmu_nl_cmd {
 101         /* wake up thread waiting for reply */
 102         struct completion complete;
 103         struct list_head nl_list;
 104         struct tcmu_dev *udev;
 105         int cmd;
 106         int status;
 107 };
 108
 109 struct tcmu_dev {
 110         struct list_head node;
 111         struct kref kref;
 112
 113         struct se_device se_dev;
 114
 115         char *name;
 116         struct se_hba *hba;
 117
 118 #define TCMU_DEV_BIT_OPEN 0
 119 #define TCMU_DEV_BIT_BROKEN 1
 120 #define TCMU_DEV_BIT_BLOCKED 2
 121         unsigned long flags;
 122
 123         struct uio_info uio_info;
 124
 125         struct inode *inode;
 126
 127         struct tcmu_mailbox *mb_addr;
 128         uint64_t dev_size;
 129         u32 cmdr_size;
 130         u32 cmdr_last_cleaned;
 131         /* Offset of data area from start of mb */
 132         /* Must add data_off and mb_addr to get the address */
 133         size_t data_off;
 134         size_t data_size;
 135         uint32_t max_blocks;
 136         size_t ring_size;
 137
 138         struct mutex cmdr_lock;
 139         struct list_head qfull_queue;
 140
 141         uint32_t dbi_max;
 142         uint32_t dbi_thresh;
 143         unsigned long *data_bitmap;
 144         struct radix_tree_root data_blocks;
 145
 146         struct idr commands;
 147
 148         struct timer_list cmd_timer;
 149         unsigned int cmd_time_out;
 150         struct list_head inflight_queue;
 151
 152         struct timer_list qfull_timer;
 153         int qfull_time_out;
 154
 155         struct list_head timedout_entry;
 156
 157         struct tcmu_nl_cmd curr_nl_cmd;
 158
 159         char dev_config[TCMU_CONFIG_LEN];
 160
 161         int nl_reply_supported;
 162 };
 163
 164 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
 165
 166 #define CMDR_OFF sizeof(struct tcmu_mailbox)
 167
 168 struct tcmu_cmd {
 169         struct se_cmd *se_cmd;
 170         struct tcmu_dev *tcmu_dev;
 171         struct list_head queue_entry;
 172
 173         uint16_t cmd_id;
 174
 175         /* Can't use se_cmd when cleaning up expired cmds, because if
 176            cmd has been completed then accessing se_cmd is off limits */
 177         uint32_t dbi_cnt;
 178         uint32_t dbi_cur;
 179         uint32_t *dbi;
 180
 181         unsigned long deadline;
 182
 183 #define TCMU_CMD_BIT_EXPIRED 0
 184 #define TCMU_CMD_BIT_INFLIGHT 1
 185         unsigned long flags;
 186 };
 187 /*
 188  * To avoid dead lock the mutex lock order should always be:
 189  *
 190  * mutex_lock(&root_udev_mutex);
 191  * ...
 192  * mutex_lock(&tcmu_dev->cmdr_lock);
 193  * mutex_unlock(&tcmu_dev->cmdr_lock);
 194  * ...
 195  * mutex_unlock(&root_udev_mutex);
 196  */
 197 static DEFINE_MUTEX(root_udev_mutex);
 198 static LIST_HEAD(root_udev);
 199
 200 static DEFINE_SPINLOCK(timed_out_udevs_lock);
 201 static LIST_HEAD(timed_out_udevs);
 202
 203 static struct kmem_cache *tcmu_cmd_cache;
 204
 205 static atomic_t global_db_count = ATOMIC_INIT(0);
 206 static struct delayed_work tcmu_unmap_work;
 207 static int tcmu_global_max_blocks = TCMU_GLOBAL_MAX_BLOCKS_DEF;
 208
 209 static int tcmu_set_global_max_data_area(const char *str,
 210                                          const struct kernel_param *kp)
 211 {
 212         int ret, max_area_mb;
 213
 214         ret = kstrtoint(str, 10, &max_area_mb);
 215         if (ret)
 216                 return -EINVAL;
 217
 218         if (max_area_mb <= 0) {
 219                 pr_err("global_max_data_area must be larger than 0.\n");
 220                 return -EINVAL;
 221         }
 222
 223         tcmu_global_max_blocks = TCMU_MBS_TO_BLOCKS(max_area_mb);
 224         if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
 225                 schedule_delayed_work(&tcmu_unmap_work, 0);
 226         else
 227                 cancel_delayed_work_sync(&tcmu_unmap_work);
 228
 229         return 0;
 230 }
 231
 232 static int tcmu_get_global_max_data_area(char *buffer,
 233                                          const struct kernel_param *kp)
 234 {
 235         return sprintf(buffer, "%d", TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
 236 }
 237
 238 static const struct kernel_param_ops tcmu_global_max_data_area_op = {
 239         .set = tcmu_set_global_max_data_area,
 240         .get = tcmu_get_global_max_data_area,
 241 };
 242
 243 module_param_cb(global_max_data_area_mb, &tcmu_global_max_data_area_op, NULL,
 244                 S_IWUSR | S_IRUGO);
 245 MODULE_PARM_DESC(global_max_data_area_mb,
 246                  "Max MBs allowed to be allocated to all the tcmu device's "
 247                  "data areas.");
 248
 249 static int tcmu_get_block_netlink(char *buffer,
 250                                   const struct kernel_param *kp)
 251 {
 252         return sprintf(buffer, "%s\n", tcmu_netlink_blocked ?
 253                        "blocked" : "unblocked");
 254 }
 255
 256 static int tcmu_set_block_netlink(const char *str,
 257                                   const struct kernel_param *kp)
 258 {
 259         int ret;
 260         u8 val;
 261
 262         ret = kstrtou8(str, 0, &val);
 263         if (ret < 0)
 264                 return ret;
 265
 266         if (val > 1) {
 267                 pr_err("Invalid block netlink value %u\n", val);
 268                 return -EINVAL;
 269         }
 270
 271         tcmu_netlink_blocked = val;
 272         return 0;
 273 }
 274
 275 static const struct kernel_param_ops tcmu_block_netlink_op = {
 276         .set = tcmu_set_block_netlink,
 277         .get = tcmu_get_block_netlink,
 278 };
 279
 280 module_param_cb(block_netlink, &tcmu_block_netlink_op, NULL, S_IWUSR | S_IRUGO);
 281 MODULE_PARM_DESC(block_netlink, "Block new netlink commands.");
 282
 283 static int tcmu_fail_netlink_cmd(struct tcmu_nl_cmd *nl_cmd)
 284 {
 285         struct tcmu_dev *udev = nl_cmd->udev;
 286
 287         if (!tcmu_netlink_blocked) {
 288                 pr_err("Could not reset device's netlink interface. Netlink is not blocked.\n");
 289                 return -EBUSY;
 290         }
 291
 292         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
 293                 pr_debug("Aborting nl cmd %d on %s\n", nl_cmd->cmd, udev->name);
 294                 nl_cmd->status = -EINTR;
 295                 list_del(&nl_cmd->nl_list);
 296                 complete(&nl_cmd->complete);
 297         }
 298         return 0;
 299 }
 300
 301 static int tcmu_set_reset_netlink(const char *str,
 302                                   const struct kernel_param *kp)
 303 {
 304         struct tcmu_nl_cmd *nl_cmd, *tmp_cmd;
 305         int ret;
 306         u8 val;
 307
 308         ret = kstrtou8(str, 0, &val);
 309         if (ret < 0)
 310                 return ret;
 311
 312         if (val != 1) {
 313                 pr_err("Invalid reset netlink value %u\n", val);
 314                 return -EINVAL;
 315         }
 316
 317         mutex_lock(&tcmu_nl_cmd_mutex);
 318         list_for_each_entry_safe(nl_cmd, tmp_cmd, &tcmu_nl_cmd_list, nl_list) {
 319                 ret = tcmu_fail_netlink_cmd(nl_cmd);
 320                 if (ret)
 321                         break;
 322         }
 323         mutex_unlock(&tcmu_nl_cmd_mutex);
 324
 325         return ret;
 326 }
 327
 328 static const struct kernel_param_ops tcmu_reset_netlink_op = {
 329         .set = tcmu_set_reset_netlink,
 330 };
 331
 332 module_param_cb(reset_netlink, &tcmu_reset_netlink_op, NULL, S_IWUSR);
 333 MODULE_PARM_DESC(reset_netlink, "Reset netlink commands.");
 334
 335 /* multicast group */
 336 enum tcmu_multicast_groups {
 337         TCMU_MCGRP_CONFIG,
 338 };
 339
 340 static const struct genl_multicast_group tcmu_mcgrps[] = {
 341         [TCMU_MCGRP_CONFIG] = { .name = "config", },
 342 };
 343
 344 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
 345         [TCMU_ATTR_DEVICE]      = { .type = NLA_STRING },
 346         [TCMU_ATTR_MINOR]       = { .type = NLA_U32 },
 347         [TCMU_ATTR_CMD_STATUS]  = { .type = NLA_S32 },
 348         [TCMU_ATTR_DEVICE_ID]   = { .type = NLA_U32 },
 349         [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
 350 };
 351
 352 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
 353 {
 354         struct tcmu_dev *udev = NULL;
 355         struct tcmu_nl_cmd *nl_cmd;
 356         int dev_id, rc, ret = 0;
 357
 358         if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
 359             !info->attrs[TCMU_ATTR_DEVICE_ID]) {
 360                 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
 361                 return -EINVAL;
 362         }
 363
 364         dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
 365         rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
 366
 367         mutex_lock(&tcmu_nl_cmd_mutex);
 368         list_for_each_entry(nl_cmd, &tcmu_nl_cmd_list, nl_list) {
 369                 if (nl_cmd->udev->se_dev.dev_index == dev_id) {
 370                         udev = nl_cmd->udev;
 371                         break;
 372                 }
 373         }
 374
 375         if (!udev) {
 376                 pr_err("tcmu nl cmd %u/%d completion could not find device with dev id %u.\n",
 377                        completed_cmd, rc, dev_id);
 378                 ret = -ENODEV;
 379                 goto unlock;
 380         }
 381         list_del(&nl_cmd->nl_list);
 382
 383         pr_debug("%s genl cmd done got id %d curr %d done %d rc %d stat %d\n",
 384                  udev->name, dev_id, nl_cmd->cmd, completed_cmd, rc,
 385                  nl_cmd->status);
 386
 387         if (nl_cmd->cmd != completed_cmd) {
 388                 pr_err("Mismatched commands on %s (Expecting reply for %d. Current %d).\n",
 389                        udev->name, completed_cmd, nl_cmd->cmd);
 390                 ret = -EINVAL;
 391                 goto unlock;
 392         }
 393
 394         nl_cmd->status = rc;
 395         complete(&nl_cmd->complete);
 396 unlock:
 397         mutex_unlock(&tcmu_nl_cmd_mutex);
 398         return ret;
 399 }
 400
 401 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
 402 {
 403         return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
 404 }
 405
 406 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
 407 {
 408         return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
 409 }
 410
 411 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
 412                                        struct genl_info *info)
 413 {
 414         return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
 415 }
 416
 417 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
 418 {
 419         if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
 420                 tcmu_kern_cmd_reply_supported  =
 421                         nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
 422                 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
 423                        tcmu_kern_cmd_reply_supported);
 424         }
 425
 426         return 0;
 427 }
 428
 429 static const struct genl_ops tcmu_genl_ops[] = {
 430         {
 431                 .cmd    = TCMU_CMD_SET_FEATURES,
 432                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 433                 .flags  = GENL_ADMIN_PERM,
 434                 .doit   = tcmu_genl_set_features,
 435         },
 436         {
 437                 .cmd    = TCMU_CMD_ADDED_DEVICE_DONE,
 438                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 439                 .flags  = GENL_ADMIN_PERM,
 440                 .doit   = tcmu_genl_add_dev_done,
 441         },
 442         {
 443                 .cmd    = TCMU_CMD_REMOVED_DEVICE_DONE,
 444                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 445                 .flags  = GENL_ADMIN_PERM,
 446                 .doit   = tcmu_genl_rm_dev_done,
 447         },
 448         {
 449                 .cmd    = TCMU_CMD_RECONFIG_DEVICE_DONE,
 450                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 451                 .flags  = GENL_ADMIN_PERM,
 452                 .doit   = tcmu_genl_reconfig_dev_done,
 453         },
 454 };
 455
 456 /* Our generic netlink family */
 457 static struct genl_family tcmu_genl_family __ro_after_init = {
 458         .module = THIS_MODULE,
 459         .hdrsize = 0,
 460         .name = "TCM-USER",
 461         .version = 2,
 462         .maxattr = TCMU_ATTR_MAX,
 463         .policy = tcmu_attr_policy,
 464         .mcgrps = tcmu_mcgrps,
 465         .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
 466         .netnsok = true,
 467         .ops = tcmu_genl_ops,
 468         .n_ops = ARRAY_SIZE(tcmu_genl_ops),
 469 };
 470
 471 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
 472 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
 473 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
 474 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
 475
 476 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
 477 {
 478         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 479         uint32_t i;
 480
 481         for (i = 0; i < len; i++)
 482                 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
 483 }
 484
 485 static inline bool tcmu_get_empty_block(struct tcmu_dev *udev,
 486                                         struct tcmu_cmd *tcmu_cmd)
 487 {
 488         struct page *page;
 489         int ret, dbi;
 490
 491         dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
 492         if (dbi == udev->dbi_thresh)
 493                 return false;
 494
 495         page = radix_tree_lookup(&udev->data_blocks, dbi);
 496         if (!page) {
 497                 if (atomic_add_return(1, &global_db_count) >
 498                                       tcmu_global_max_blocks)
 499                         schedule_delayed_work(&tcmu_unmap_work, 0);
 500
 501                 /* try to get new page from the mm */
 502                 page = alloc_page(GFP_NOIO);
 503                 if (!page)
 504                         goto err_alloc;
 505
 506                 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
 507                 if (ret)
 508                         goto err_insert;
 509         }
 510
 511         if (dbi > udev->dbi_max)
 512                 udev->dbi_max = dbi;
 513
 514         set_bit(dbi, udev->data_bitmap);
 515         tcmu_cmd_set_dbi(tcmu_cmd, dbi);
 516
 517         return true;
 518 err_insert:
 519         __free_page(page);
 520 err_alloc:
 521         atomic_dec(&global_db_count);
 522         return false;
 523 }
 524
 525 static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
 526                                   struct tcmu_cmd *tcmu_cmd)
 527 {
 528         int i;
 529
 530         for (i = tcmu_cmd->dbi_cur; i < tcmu_cmd->dbi_cnt; i++) {
 531                 if (!tcmu_get_empty_block(udev, tcmu_cmd))
 532                         return false;
 533         }
 534         return true;
 535 }
 536
 537 static inline struct page *
 538 tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
 539 {
 540         return radix_tree_lookup(&udev->data_blocks, dbi);
 541 }
 542
 543 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
 544 {
 545         kfree(tcmu_cmd->dbi);
 546         kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 547 }
 548
 549 static inline size_t tcmu_cmd_get_data_length(struct tcmu_cmd *tcmu_cmd)
 550 {
 551         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 552         size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
 553
 554         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 555                 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
 556                 data_length += round_up(se_cmd->t_bidi_data_sg->length,
 557                                 DATA_BLOCK_SIZE);
 558         }
 559
 560         return data_length;
 561 }
 562
 563 static inline uint32_t tcmu_cmd_get_block_cnt(struct tcmu_cmd *tcmu_cmd)
 564 {
 565         size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
 566
 567         return data_length / DATA_BLOCK_SIZE;
 568 }
 569
 570 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
 571 {
 572         struct se_device *se_dev = se_cmd->se_dev;
 573         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 574         struct tcmu_cmd *tcmu_cmd;
 575
 576         tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_NOIO);
 577         if (!tcmu_cmd)
 578                 return NULL;
 579
 580         INIT_LIST_HEAD(&tcmu_cmd->queue_entry);
 581         tcmu_cmd->se_cmd = se_cmd;
 582         tcmu_cmd->tcmu_dev = udev;
 583
 584         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 585         tcmu_cmd->dbi_cnt = tcmu_cmd_get_block_cnt(tcmu_cmd);
 586         tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
 587                                 GFP_NOIO);
 588         if (!tcmu_cmd->dbi) {
 589                 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 590                 return NULL;
 591         }
 592
 593         return tcmu_cmd;
 594 }
 595
 596 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
 597 {
 598         unsigned long offset = offset_in_page(vaddr);
 599         void *start = vaddr - offset;
 600
 601         size = round_up(size+offset, PAGE_SIZE);
 602
 603         while (size) {
 604                 flush_dcache_page(virt_to_page(start));
 605                 start += PAGE_SIZE;
 606                 size -= PAGE_SIZE;
 607         }
 608 }
 609
 610 /*
 611  * Some ring helper functions. We don't assume size is a power of 2 so
 612  * we can't use circ_buf.h.
 613  */
 614 static inline size_t spc_used(size_t head, size_t tail, size_t size)
 615 {
 616         int diff = head - tail;
 617
 618         if (diff >= 0)
 619                 return diff;
 620         else
 621                 return size + diff;
 622 }
 623
 624 static inline size_t spc_free(size_t head, size_t tail, size_t size)
 625 {
 626         /* Keep 1 byte unused or we can't tell full from empty */
 627         return (size - spc_used(head, tail, size) - 1);
 628 }
 629
 630 static inline size_t head_to_end(size_t head, size_t size)
 631 {
 632         return size - head;
 633 }
 634
 635 static inline void new_iov(struct iovec **iov, int *iov_cnt)
 636 {
 637         struct iovec *iovec;
 638
 639         if (*iov_cnt != 0)
 640                 (*iov)++;
 641         (*iov_cnt)++;
 642
 643         iovec = *iov;
 644         memset(iovec, 0, sizeof(struct iovec));
 645 }
 646
 647 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
 648
 649 /* offset is relative to mb_addr */
 650 static inline size_t get_block_offset_user(struct tcmu_dev *dev,
 651                 int dbi, int remaining)
 652 {
 653         return dev->data_off + dbi * DATA_BLOCK_SIZE +
 654                 DATA_BLOCK_SIZE - remaining;
 655 }
 656
 657 static inline size_t iov_tail(struct iovec *iov)
 658 {
 659         return (size_t)iov->iov_base + iov->iov_len;
 660 }
 661
 662 static void scatter_data_area(struct tcmu_dev *udev,
 663         struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
 664         unsigned int data_nents, struct iovec **iov,
 665         int *iov_cnt, bool copy_data)
 666 {
 667         int i, dbi;
 668         int block_remaining = 0;
 669         void *from, *to = NULL;
 670         size_t copy_bytes, to_offset, offset;
 671         struct scatterlist *sg;
 672         struct page *page;
 673
 674         for_each_sg(data_sg, sg, data_nents, i) {
 675                 int sg_remaining = sg->length;
 676                 from = kmap_atomic(sg_page(sg)) + sg->offset;
 677                 while (sg_remaining > 0) {
 678                         if (block_remaining == 0) {
 679                                 if (to)
 680                                         kunmap_atomic(to);
 681
 682                                 block_remaining = DATA_BLOCK_SIZE;
 683                                 dbi = tcmu_cmd_get_dbi(tcmu_cmd);
 684                                 page = tcmu_get_block_page(udev, dbi);
 685                                 to = kmap_atomic(page);
 686                         }
 687
 688                         /*
 689                          * Covert to virtual offset of the ring data area.
 690                          */
 691                         to_offset = get_block_offset_user(udev, dbi,
 692                                         block_remaining);
 693
 694                         /*
 695                          * The following code will gather and map the blocks
 696                          * to the same iovec when the blocks are all next to
 697                          * each other.
 698                          */
 699                         copy_bytes = min_t(size_t, sg_remaining,
 700                                         block_remaining);
 701                         if (*iov_cnt != 0 &&
 702                             to_offset == iov_tail(*iov)) {
 703                                 /*
 704                                  * Will append to the current iovec, because
 705                                  * the current block page is next to the
 706                                  * previous one.
 707                                  */
 708                                 (*iov)->iov_len += copy_bytes;
 709                         } else {
 710                                 /*
 711                                  * Will allocate a new iovec because we are
 712                                  * first time here or the current block page
 713                                  * is not next to the previous one.
 714                                  */
 715                                 new_iov(iov, iov_cnt);
 716                                 (*iov)->iov_base = (void __user *)to_offset;
 717                                 (*iov)->iov_len = copy_bytes;
 718                         }
 719
 720                         if (copy_data) {
 721                                 offset = DATA_BLOCK_SIZE - block_remaining;
 722                                 memcpy(to + offset,
 723                                        from + sg->length - sg_remaining,
 724                                        copy_bytes);
 725                                 tcmu_flush_dcache_range(to, copy_bytes);
 726                         }
 727
 728                         sg_remaining -= copy_bytes;
 729                         block_remaining -= copy_bytes;
 730                 }
 731                 kunmap_atomic(from - sg->offset);
 732         }
 733
 734         if (to)
 735                 kunmap_atomic(to);
 736 }
 737
 738 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 739                              bool bidi, uint32_t read_len)
 740 {
 741         struct se_cmd *se_cmd = cmd->se_cmd;
 742         int i, dbi;
 743         int block_remaining = 0;
 744         void *from = NULL, *to;
 745         size_t copy_bytes, offset;
 746         struct scatterlist *sg, *data_sg;
 747         struct page *page;
 748         unsigned int data_nents;
 749         uint32_t count = 0;
 750
 751         if (!bidi) {
 752                 data_sg = se_cmd->t_data_sg;
 753                 data_nents = se_cmd->t_data_nents;
 754         } else {
 755
 756                 /*
 757                  * For bidi case, the first count blocks are for Data-Out
 758                  * buffer blocks, and before gathering the Data-In buffer
 759                  * the Data-Out buffer blocks should be discarded.
 760                  */
 761                 count = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
 762
 763                 data_sg = se_cmd->t_bidi_data_sg;
 764                 data_nents = se_cmd->t_bidi_data_nents;
 765         }
 766
 767         tcmu_cmd_set_dbi_cur(cmd, count);
 768
 769         for_each_sg(data_sg, sg, data_nents, i) {
 770                 int sg_remaining = sg->length;
 771                 to = kmap_atomic(sg_page(sg)) + sg->offset;
 772                 while (sg_remaining > 0 && read_len > 0) {
 773                         if (block_remaining == 0) {
 774                                 if (from)
 775                                         kunmap_atomic(from);
 776
 777                                 block_remaining = DATA_BLOCK_SIZE;
 778                                 dbi = tcmu_cmd_get_dbi(cmd);
 779                                 page = tcmu_get_block_page(udev, dbi);
 780                                 from = kmap_atomic(page);
 781                         }
 782                         copy_bytes = min_t(size_t, sg_remaining,
 783                                         block_remaining);
 784                         if (read_len < copy_bytes)
 785                                 copy_bytes = read_len;
 786                         offset = DATA_BLOCK_SIZE - block_remaining;
 787                         tcmu_flush_dcache_range(from, copy_bytes);
 788                         memcpy(to + sg->length - sg_remaining, from + offset,
 789                                         copy_bytes);
 790
 791                         sg_remaining -= copy_bytes;
 792                         block_remaining -= copy_bytes;
 793                         read_len -= copy_bytes;
 794                 }
 795                 kunmap_atomic(to - sg->offset);
 796                 if (read_len == 0)
 797                         break;
 798         }
 799         if (from)
 800                 kunmap_atomic(from);
 801 }
 802
 803 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
 804 {
 805         return thresh - bitmap_weight(bitmap, thresh);
 806 }
 807
 808 /*
 809  * We can't queue a command until we have space available on the cmd ring *and*
 810  * space available on the data area.
 811  *
 812  * Called with ring lock held.
 813  */
 814 static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 815                 size_t cmd_size, size_t data_needed)
 816 {
 817         struct tcmu_mailbox *mb = udev->mb_addr;
 818         uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
 819                                 / DATA_BLOCK_SIZE;
 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
 843         /* try to check and get the data blocks as needed */
 844         space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
 845         if ((space * DATA_BLOCK_SIZE) < data_needed) {
 846                 unsigned long blocks_left =
 847                                 (udev->max_blocks - udev->dbi_thresh) + space;
 848
 849                 if (blocks_left < blocks_needed) {
 850                         pr_debug("no data space: only %lu available, but ask for %zu\n",
 851                                         blocks_left * DATA_BLOCK_SIZE,
 852                                         data_needed);
 853                         return false;
 854                 }
 855
 856                 udev->dbi_thresh += blocks_needed;
 857                 if (udev->dbi_thresh > udev->max_blocks)
 858                         udev->dbi_thresh = udev->max_blocks;
 859         }
 860
 861         return tcmu_get_empty_blocks(udev, cmd);
 862 }
 863
 864 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
 865 {
 866         return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 867                         sizeof(struct tcmu_cmd_entry));
 868 }
 869
 870 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
 871                                            size_t base_command_size)
 872 {
 873         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 874         size_t command_size;
 875
 876         command_size = base_command_size +
 877                 round_up(scsi_command_size(se_cmd->t_task_cdb),
 878                                 TCMU_OP_ALIGN_SIZE);
 879
 880         WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
 881
 882         return command_size;
 883 }
 884
 885 static int tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
 886                                 struct timer_list *timer)
 887 {
 888         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 889         int cmd_id;
 890
 891         if (tcmu_cmd->cmd_id)
 892                 goto setup_timer;
 893
 894         cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 1, USHRT_MAX, GFP_NOWAIT);
 895         if (cmd_id < 0) {
 896                 pr_err("tcmu: Could not allocate cmd id.\n");
 897                 return cmd_id;
 898         }
 899         tcmu_cmd->cmd_id = cmd_id;
 900
 901         pr_debug("allocated cmd %u for dev %s tmo %lu\n", tcmu_cmd->cmd_id,
 902                  udev->name, tmo / MSEC_PER_SEC);
 903
 904 setup_timer:
 905         if (!tmo)
 906                 return 0;
 907
 908         tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
 909         if (!timer_pending(timer))
 910                 mod_timer(timer, tcmu_cmd->deadline);
 911
 912         return 0;
 913 }
 914
 915 static int add_to_qfull_queue(struct tcmu_cmd *tcmu_cmd)
 916 {
 917         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 918         unsigned int tmo;
 919         int ret;
 920
 921         /*
 922          * For backwards compat if qfull_time_out is not set use
 923          * cmd_time_out and if that's not set use the default time out.
 924          */
 925         if (!udev->qfull_time_out)
 926                 return -ETIMEDOUT;
 927         else if (udev->qfull_time_out > 0)
 928                 tmo = udev->qfull_time_out;
 929         else if (udev->cmd_time_out)
 930                 tmo = udev->cmd_time_out;
 931         else
 932                 tmo = TCMU_TIME_OUT;
 933
 934         ret = tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
 935         if (ret)
 936                 return ret;
 937
 938         list_add_tail(&tcmu_cmd->queue_entry, &udev->qfull_queue);
 939         pr_debug("adding cmd %u on dev %s to ring space wait queue\n",
 940                  tcmu_cmd->cmd_id, udev->name);
 941         return 0;
 942 }
 943
 944 /**
 945  * queue_cmd_ring - queue cmd to ring or internally
 946  * @tcmu_cmd: cmd to queue
 947  * @scsi_err: TCM error code if failure (-1) returned.
 948  *
 949  * Returns:
 950  * -1 we cannot queue internally or to the ring.
 951  *  0 success
 952  *  1 internally queued to wait for ring memory to free.
 953  */
 954 static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
 955 {
 956         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 957         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 958         size_t base_command_size, command_size;
 959         struct tcmu_mailbox *mb;
 960         struct tcmu_cmd_entry *entry;
 961         struct iovec *iov;
 962         int iov_cnt, ret;
 963         uint32_t cmd_head;
 964         uint64_t cdb_off;
 965         bool copy_to_data_area;
 966         size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
 967
 968         *scsi_err = TCM_NO_SENSE;
 969
 970         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
 971                 *scsi_err = TCM_LUN_BUSY;
 972                 return -1;
 973         }
 974
 975         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
 976                 *scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 977                 return -1;
 978         }
 979
 980         /*
 981          * Must be a certain minimum size for response sense info, but
 982          * also may be larger if the iov array is large.
 983          *
 984          * We prepare as many iovs as possbile for potential uses here,
 985          * because it's expensive to tell how many regions are freed in
 986          * the bitmap & global data pool, as the size calculated here
 987          * will only be used to do the checks.
 988          *
 989          * The size will be recalculated later as actually needed to save
 990          * cmd area memories.
 991          */
 992         base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
 993         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 994
 995         if (!list_empty(&udev->qfull_queue))
 996                 goto queue;
 997
 998         mb = udev->mb_addr;
 999         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
1000         if ((command_size > (udev->cmdr_size / 2)) ||
1001             data_length > udev->data_size) {
1002                 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
1003                         "cmd ring/data area\n", command_size, data_length,
1004                         udev->cmdr_size, udev->data_size);
1005                 *scsi_err = TCM_INVALID_CDB_FIELD;
1006                 return -1;
1007         }
1008
1009         if (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
1010                 /*
1011                  * Don't leave commands partially setup because the unmap
1012                  * thread might need the blocks to make forward progress.
1013                  */
1014                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1015                 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1016                 goto queue;
1017         }
1018
1019         /* Insert a PAD if end-of-ring space is too small */
1020         if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
1021                 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
1022
1023                 entry = (void *) mb + CMDR_OFF + cmd_head;
1024                 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
1025                 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
1026                 entry->hdr.cmd_id = 0; /* not used for PAD */
1027                 entry->hdr.kflags = 0;
1028                 entry->hdr.uflags = 0;
1029                 tcmu_flush_dcache_range(entry, sizeof(*entry));
1030
1031                 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
1032                 tcmu_flush_dcache_range(mb, sizeof(*mb));
1033
1034                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
1035                 WARN_ON(cmd_head != 0);
1036         }
1037
1038         entry = (void *) mb + CMDR_OFF + cmd_head;
1039         memset(entry, 0, command_size);
1040         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
1041
1042         /* Handle allocating space from the data area */
1043         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1044         iov = &entry->req.iov[0];
1045         iov_cnt = 0;
1046         copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
1047                 || se_cmd->se_cmd_flags & SCF_BIDI);
1048         scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
1049                           se_cmd->t_data_nents, &iov, &iov_cnt,
1050                           copy_to_data_area);
1051         entry->req.iov_cnt = iov_cnt;
1052
1053         /* Handle BIDI commands */
1054         iov_cnt = 0;
1055         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1056                 iov++;
1057                 scatter_data_area(udev, tcmu_cmd, se_cmd->t_bidi_data_sg,
1058                                   se_cmd->t_bidi_data_nents, &iov, &iov_cnt,
1059                                   false);
1060         }
1061         entry->req.iov_bidi_cnt = iov_cnt;
1062
1063         ret = tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out,
1064                                    &udev->cmd_timer);
1065         if (ret) {
1066                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
1067
1068                 *scsi_err = TCM_OUT_OF_RESOURCES;
1069                 return -1;
1070         }
1071         entry->hdr.cmd_id = tcmu_cmd->cmd_id;
1072
1073         /*
1074          * Recalaulate the command's base size and size according
1075          * to the actual needs
1076          */
1077         base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
1078                                                        entry->req.iov_bidi_cnt);
1079         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
1080
1081         tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
1082
1083         /* All offsets relative to mb_addr, not start of entry! */
1084         cdb_off = CMDR_OFF + cmd_head + base_command_size;
1085         memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
1086         entry->req.cdb_off = cdb_off;
1087         tcmu_flush_dcache_range(entry, sizeof(*entry));
1088
1089         UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1090         tcmu_flush_dcache_range(mb, sizeof(*mb));
1091
1092         list_add_tail(&tcmu_cmd->queue_entry, &udev->inflight_queue);
1093         set_bit(TCMU_CMD_BIT_INFLIGHT, &tcmu_cmd->flags);
1094
1095         /* TODO: only if FLUSH and FUA? */
1096         uio_event_notify(&udev->uio_info);
1097
1098         return 0;
1099
1100 queue:
1101         if (add_to_qfull_queue(tcmu_cmd)) {
1102                 *scsi_err = TCM_OUT_OF_RESOURCES;
1103                 return -1;
1104         }
1105
1106         return 1;
1107 }
1108
1109 static sense_reason_t
1110 tcmu_queue_cmd(struct se_cmd *se_cmd)
1111 {
1112         struct se_device *se_dev = se_cmd->se_dev;
1113         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1114         struct tcmu_cmd *tcmu_cmd;
1115         sense_reason_t scsi_ret;
1116         int ret;
1117
1118         tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1119         if (!tcmu_cmd)
1120                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1121
1122         mutex_lock(&udev->cmdr_lock);
1123         ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1124         mutex_unlock(&udev->cmdr_lock);
1125         if (ret < 0)
1126                 tcmu_free_cmd(tcmu_cmd);
1127         return scsi_ret;
1128 }
1129
1130 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
1131 {
1132         struct se_cmd *se_cmd = cmd->se_cmd;
1133         struct tcmu_dev *udev = cmd->tcmu_dev;
1134         bool read_len_valid = false;
1135         uint32_t read_len;
1136
1137         /*
1138          * cmd has been completed already from timeout, just reclaim
1139          * data area space and free cmd
1140          */
1141         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1142                 WARN_ON_ONCE(se_cmd);
1143                 goto out;
1144         }
1145
1146         list_del_init(&cmd->queue_entry);
1147
1148         tcmu_cmd_reset_dbi_cur(cmd);
1149
1150         if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1151                 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1152                         cmd->se_cmd);
1153                 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1154                 goto done;
1155         }
1156
1157         read_len = se_cmd->data_length;
1158         if (se_cmd->data_direction == DMA_FROM_DEVICE &&
1159             (entry->hdr.uflags & TCMU_UFLAG_READ_LEN) && entry->rsp.read_len) {
1160                 read_len_valid = true;
1161                 if (entry->rsp.read_len < read_len)
1162                         read_len = entry->rsp.read_len;
1163         }
1164
1165         if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1166                 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1167                 if (!read_len_valid )
1168                         goto done;
1169                 else
1170                         se_cmd->se_cmd_flags |= SCF_TREAT_READ_AS_NORMAL;
1171         }
1172         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1173                 /* Get Data-In buffer before clean up */
1174                 gather_data_area(udev, cmd, true, read_len);
1175         } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1176                 gather_data_area(udev, cmd, false, read_len);
1177         } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1178                 /* TODO: */
1179         } else if (se_cmd->data_direction != DMA_NONE) {
1180                 pr_warn("TCMU: data direction was %d!\n",
1181                         se_cmd->data_direction);
1182         }
1183
1184 done:
1185         if (read_len_valid) {
1186                 pr_debug("read_len = %d\n", read_len);
1187                 target_complete_cmd_with_length(cmd->se_cmd,
1188                                         entry->rsp.scsi_status, read_len);
1189         } else
1190                 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1191
1192 out:
1193         cmd->se_cmd = NULL;
1194         tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1195         tcmu_free_cmd(cmd);
1196 }
1197
1198 static void tcmu_set_next_deadline(struct list_head *queue,
1199                                    struct timer_list *timer)
1200 {
1201         struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1202         unsigned long deadline = 0;
1203
1204         list_for_each_entry_safe(tcmu_cmd, tmp_cmd, queue, queue_entry) {
1205                 if (!time_after(jiffies, tcmu_cmd->deadline)) {
1206                         deadline = tcmu_cmd->deadline;
1207                         break;
1208                 }
1209         }
1210
1211         if (deadline)
1212                 mod_timer(timer, deadline);
1213         else
1214                 del_timer(timer);
1215 }
1216
1217 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
1218 {
1219         struct tcmu_mailbox *mb;
1220         struct tcmu_cmd *cmd;
1221         int handled = 0;
1222
1223         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1224                 pr_err("ring broken, not handling completions\n");
1225                 return 0;
1226         }
1227
1228         mb = udev->mb_addr;
1229         tcmu_flush_dcache_range(mb, sizeof(*mb));
1230
1231         while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1232
1233                 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
1234
1235                 tcmu_flush_dcache_range(entry, sizeof(*entry));
1236
1237                 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
1238                         UPDATE_HEAD(udev->cmdr_last_cleaned,
1239                                     tcmu_hdr_get_len(entry->hdr.len_op),
1240                                     udev->cmdr_size);
1241                         continue;
1242                 }
1243                 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1244
1245                 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1246                 if (!cmd) {
1247                         pr_err("cmd_id %u not found, ring is broken\n",
1248                                entry->hdr.cmd_id);
1249                         set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1250                         break;
1251                 }
1252
1253                 tcmu_handle_completion(cmd, entry);
1254
1255                 UPDATE_HEAD(udev->cmdr_last_cleaned,
1256                             tcmu_hdr_get_len(entry->hdr.len_op),
1257                             udev->cmdr_size);
1258
1259                 handled++;
1260         }
1261
1262         if (mb->cmd_tail == mb->cmd_head) {
1263                 /* no more pending commands */
1264                 del_timer(&udev->cmd_timer);
1265
1266                 if (list_empty(&udev->qfull_queue)) {
1267                         /*
1268                          * no more pending or waiting commands so try to
1269                          * reclaim blocks if needed.
1270                          */
1271                         if (atomic_read(&global_db_count) >
1272                             tcmu_global_max_blocks)
1273                                 schedule_delayed_work(&tcmu_unmap_work, 0);
1274                 }
1275         } else if (udev->cmd_time_out) {
1276                 tcmu_set_next_deadline(&udev->inflight_queue, &udev->cmd_timer);
1277         }
1278
1279         return handled;
1280 }
1281
1282 static int tcmu_check_expired_cmd(int id, void *p, void *data)
1283 {
1284         struct tcmu_cmd *cmd = p;
1285         struct tcmu_dev *udev = cmd->tcmu_dev;
1286         u8 scsi_status;
1287         struct se_cmd *se_cmd;
1288         bool is_running;
1289
1290         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1291                 return 0;
1292
1293         if (!time_after(jiffies, cmd->deadline))
1294                 return 0;
1295
1296         is_running = test_bit(TCMU_CMD_BIT_INFLIGHT, &cmd->flags);
1297         se_cmd = cmd->se_cmd;
1298
1299         if (is_running) {
1300                 /*
1301                  * If cmd_time_out is disabled but qfull is set deadline
1302                  * will only reflect the qfull timeout. Ignore it.
1303                  */
1304                 if (!udev->cmd_time_out)
1305                         return 0;
1306
1307                 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1308                 /*
1309                  * target_complete_cmd will translate this to LUN COMM FAILURE
1310                  */
1311                 scsi_status = SAM_STAT_CHECK_CONDITION;
1312                 list_del_init(&cmd->queue_entry);
1313                 cmd->se_cmd = NULL;
1314         } else {
1315                 list_del_init(&cmd->queue_entry);
1316                 idr_remove(&udev->commands, id);
1317                 tcmu_free_cmd(cmd);
1318                 scsi_status = SAM_STAT_TASK_SET_FULL;
1319         }
1320
1321         pr_debug("Timing out cmd %u on dev %s that is %s.\n",
1322                  id, udev->name, is_running ? "inflight" : "queued");
1323
1324         target_complete_cmd(se_cmd, scsi_status);
1325         return 0;
1326 }
1327
1328 static void tcmu_device_timedout(struct tcmu_dev *udev)
1329 {
1330         spin_lock(&timed_out_udevs_lock);
1331         if (list_empty(&udev->timedout_entry))
1332                 list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1333         spin_unlock(&timed_out_udevs_lock);
1334
1335         schedule_delayed_work(&tcmu_unmap_work, 0);
1336 }
1337
1338 static void tcmu_cmd_timedout(struct timer_list *t)
1339 {
1340         struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1341
1342         pr_debug("%s cmd timeout has expired\n", udev->name);
1343         tcmu_device_timedout(udev);
1344 }
1345
1346 static void tcmu_qfull_timedout(struct timer_list *t)
1347 {
1348         struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1349
1350         pr_debug("%s qfull timeout has expired\n", udev->name);
1351         tcmu_device_timedout(udev);
1352 }
1353
1354 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1355 {
1356         struct tcmu_hba *tcmu_hba;
1357
1358         tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1359         if (!tcmu_hba)
1360                 return -ENOMEM;
1361
1362         tcmu_hba->host_id = host_id;
1363         hba->hba_ptr = tcmu_hba;
1364
1365         return 0;
1366 }
1367
1368 static void tcmu_detach_hba(struct se_hba *hba)
1369 {
1370         kfree(hba->hba_ptr);
1371         hba->hba_ptr = NULL;
1372 }
1373
1374 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1375 {
1376         struct tcmu_dev *udev;
1377
1378         udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1379         if (!udev)
1380                 return NULL;
1381         kref_init(&udev->kref);
1382
1383         udev->name = kstrdup(name, GFP_KERNEL);
1384         if (!udev->name) {
1385                 kfree(udev);
1386                 return NULL;
1387         }
1388
1389         udev->hba = hba;
1390         udev->cmd_time_out = TCMU_TIME_OUT;
1391         udev->qfull_time_out = -1;
1392
1393         udev->max_blocks = DATA_BLOCK_BITS_DEF;
1394         mutex_init(&udev->cmdr_lock);
1395
1396         INIT_LIST_HEAD(&udev->node);
1397         INIT_LIST_HEAD(&udev->timedout_entry);
1398         INIT_LIST_HEAD(&udev->qfull_queue);
1399         INIT_LIST_HEAD(&udev->inflight_queue);
1400         idr_init(&udev->commands);
1401
1402         timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1403         timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1404
1405         INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1406
1407         return &udev->se_dev;
1408 }
1409
1410 static bool run_qfull_queue(struct tcmu_dev *udev, bool fail)
1411 {
1412         struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1413         LIST_HEAD(cmds);
1414         bool drained = true;
1415         sense_reason_t scsi_ret;
1416         int ret;
1417
1418         if (list_empty(&udev->qfull_queue))
1419                 return true;
1420
1421         pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1422
1423         list_splice_init(&udev->qfull_queue, &cmds);
1424
1425         list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, queue_entry) {
1426                 list_del_init(&tcmu_cmd->queue_entry);
1427
1428                 pr_debug("removing cmd %u on dev %s from queue\n",
1429                          tcmu_cmd->cmd_id, udev->name);
1430
1431                 if (fail) {
1432                         idr_remove(&udev->commands, tcmu_cmd->cmd_id);
1433                         /*
1434                          * We were not able to even start the command, so
1435                          * fail with busy to allow a retry in case runner
1436                          * was only temporarily down. If the device is being
1437                          * removed then LIO core will do the right thing and
1438                          * fail the retry.
1439                          */
1440                         target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1441                         tcmu_free_cmd(tcmu_cmd);
1442                         continue;
1443                 }
1444
1445                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1446                 if (ret < 0) {
1447                         pr_debug("cmd %u on dev %s failed with %u\n",
1448                                  tcmu_cmd->cmd_id, udev->name, scsi_ret);
1449
1450                         idr_remove(&udev->commands, tcmu_cmd->cmd_id);
1451                         /*
1452                          * Ignore scsi_ret for now. target_complete_cmd
1453                          * drops it.
1454                          */
1455                         target_complete_cmd(tcmu_cmd->se_cmd,
1456                                             SAM_STAT_CHECK_CONDITION);
1457                         tcmu_free_cmd(tcmu_cmd);
1458                 } else if (ret > 0) {
1459                         pr_debug("ran out of space during cmdr queue run\n");
1460                         /*
1461                          * cmd was requeued, so just put all cmds back in
1462                          * the queue
1463                          */
1464                         list_splice_tail(&cmds, &udev->qfull_queue);
1465                         drained = false;
1466                         break;
1467                 }
1468         }
1469
1470         tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1471         return drained;
1472 }
1473
1474 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1475 {
1476         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1477
1478         mutex_lock(&udev->cmdr_lock);
1479         tcmu_handle_completions(udev);
1480         run_qfull_queue(udev, false);
1481         mutex_unlock(&udev->cmdr_lock);
1482
1483         return 0;
1484 }
1485
1486 /*
1487  * mmap code from uio.c. Copied here because we want to hook mmap()
1488  * and this stuff must come along.
1489  */
1490 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1491 {
1492         struct tcmu_dev *udev = vma->vm_private_data;
1493         struct uio_info *info = &udev->uio_info;
1494
1495         if (vma->vm_pgoff < MAX_UIO_MAPS) {
1496                 if (info->mem[vma->vm_pgoff].size == 0)
1497                         return -1;
1498                 return (int)vma->vm_pgoff;
1499         }
1500         return -1;
1501 }
1502
1503 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1504 {
1505         struct page *page;
1506
1507         mutex_lock(&udev->cmdr_lock);
1508         page = tcmu_get_block_page(udev, dbi);
1509         if (likely(page)) {
1510                 mutex_unlock(&udev->cmdr_lock);
1511                 return page;
1512         }
1513
1514         /*
1515          * Userspace messed up and passed in a address not in the
1516          * data iov passed to it.
1517          */
1518         pr_err("Invalid addr to data block mapping  (dbi %u) on device %s\n",
1519                dbi, udev->name);
1520         page = NULL;
1521         mutex_unlock(&udev->cmdr_lock);
1522
1523         return page;
1524 }
1525
1526 static vm_fault_t tcmu_vma_fault(struct vm_fault *vmf)
1527 {
1528         struct tcmu_dev *udev = vmf->vma->vm_private_data;
1529         struct uio_info *info = &udev->uio_info;
1530         struct page *page;
1531         unsigned long offset;
1532         void *addr;
1533
1534         int mi = tcmu_find_mem_index(vmf->vma);
1535         if (mi < 0)
1536                 return VM_FAULT_SIGBUS;
1537
1538         /*
1539          * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1540          * to use mem[N].
1541          */
1542         offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1543
1544         if (offset < udev->data_off) {
1545                 /* For the vmalloc()ed cmd area pages */
1546                 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1547                 page = vmalloc_to_page(addr);
1548         } else {
1549                 uint32_t dbi;
1550
1551                 /* For the dynamically growing data area pages */
1552                 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1553                 page = tcmu_try_get_block_page(udev, dbi);
1554                 if (!page)
1555                         return VM_FAULT_SIGBUS;
1556         }
1557
1558         get_page(page);
1559         vmf->page = page;
1560         return 0;
1561 }
1562
1563 static const struct vm_operations_struct tcmu_vm_ops = {
1564         .fault = tcmu_vma_fault,
1565 };
1566
1567 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1568 {
1569         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1570
1571         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1572         vma->vm_ops = &tcmu_vm_ops;
1573
1574         vma->vm_private_data = udev;
1575
1576         /* Ensure the mmap is exactly the right size */
1577         if (vma_pages(vma) != (udev->ring_size >> PAGE_SHIFT))
1578                 return -EINVAL;
1579
1580         return 0;
1581 }
1582
1583 static int tcmu_open(struct uio_info *info, struct inode *inode)
1584 {
1585         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1586
1587         /* O_EXCL not supported for char devs, so fake it? */
1588         if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1589                 return -EBUSY;
1590
1591         udev->inode = inode;
1592         kref_get(&udev->kref);
1593
1594         pr_debug("open\n");
1595
1596         return 0;
1597 }
1598
1599 static void tcmu_dev_call_rcu(struct rcu_head *p)
1600 {
1601         struct se_device *dev = container_of(p, struct se_device, rcu_head);
1602         struct tcmu_dev *udev = TCMU_DEV(dev);
1603
1604         kfree(udev->uio_info.name);
1605         kfree(udev->name);
1606         kfree(udev);
1607 }
1608
1609 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1610 {
1611         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1612                 kmem_cache_free(tcmu_cmd_cache, cmd);
1613                 return 0;
1614         }
1615         return -EINVAL;
1616 }
1617
1618 static void tcmu_blocks_release(struct radix_tree_root *blocks,
1619                                 int start, int end)
1620 {
1621         int i;
1622         struct page *page;
1623
1624         for (i = start; i < end; i++) {
1625                 page = radix_tree_delete(blocks, i);
1626                 if (page) {
1627                         __free_page(page);
1628                         atomic_dec(&global_db_count);
1629                 }
1630         }
1631 }
1632
1633 static void tcmu_dev_kref_release(struct kref *kref)
1634 {
1635         struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1636         struct se_device *dev = &udev->se_dev;
1637         struct tcmu_cmd *cmd;
1638         bool all_expired = true;
1639         int i;
1640
1641         vfree(udev->mb_addr);
1642         udev->mb_addr = NULL;
1643
1644         spin_lock_bh(&timed_out_udevs_lock);
1645         if (!list_empty(&udev->timedout_entry))
1646                 list_del(&udev->timedout_entry);
1647         spin_unlock_bh(&timed_out_udevs_lock);
1648
1649         /* Upper layer should drain all requests before calling this */
1650         mutex_lock(&udev->cmdr_lock);
1651         idr_for_each_entry(&udev->commands, cmd, i) {
1652                 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1653                         all_expired = false;
1654         }
1655         idr_destroy(&udev->commands);
1656         WARN_ON(!all_expired);
1657
1658         tcmu_blocks_release(&udev->data_blocks, 0, udev->dbi_max + 1);
1659         bitmap_free(udev->data_bitmap);
1660         mutex_unlock(&udev->cmdr_lock);
1661
1662         call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1663 }
1664
1665 static int tcmu_release(struct uio_info *info, struct inode *inode)
1666 {
1667         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1668
1669         clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1670
1671         pr_debug("close\n");
1672         /* release ref from open */
1673         kref_put(&udev->kref, tcmu_dev_kref_release);
1674         return 0;
1675 }
1676
1677 static int tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1678 {
1679         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1680
1681         if (!tcmu_kern_cmd_reply_supported)
1682                 return 0;
1683
1684         if (udev->nl_reply_supported <= 0)
1685                 return 0;
1686
1687         mutex_lock(&tcmu_nl_cmd_mutex);
1688
1689         if (tcmu_netlink_blocked) {
1690                 mutex_unlock(&tcmu_nl_cmd_mutex);
1691                 pr_warn("Failing nl cmd %d on %s. Interface is blocked.\n", cmd,
1692                         udev->name);
1693                 return -EAGAIN;
1694         }
1695
1696         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1697                 mutex_unlock(&tcmu_nl_cmd_mutex);
1698                 pr_warn("netlink cmd %d already executing on %s\n",
1699                          nl_cmd->cmd, udev->name);
1700                 return -EBUSY;
1701         }
1702
1703         memset(nl_cmd, 0, sizeof(*nl_cmd));
1704         nl_cmd->cmd = cmd;
1705         nl_cmd->udev = udev;
1706         init_completion(&nl_cmd->complete);
1707         INIT_LIST_HEAD(&nl_cmd->nl_list);
1708
1709         list_add_tail(&nl_cmd->nl_list, &tcmu_nl_cmd_list);
1710
1711         mutex_unlock(&tcmu_nl_cmd_mutex);
1712         return 0;
1713 }
1714
1715 static void tcmu_destroy_genl_cmd_reply(struct tcmu_dev *udev)
1716 {
1717         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1718
1719         if (!tcmu_kern_cmd_reply_supported)
1720                 return;
1721
1722         if (udev->nl_reply_supported <= 0)
1723                 return;
1724
1725         mutex_lock(&tcmu_nl_cmd_mutex);
1726
1727         list_del(&nl_cmd->nl_list);
1728         memset(nl_cmd, 0, sizeof(*nl_cmd));
1729
1730         mutex_unlock(&tcmu_nl_cmd_mutex);
1731 }
1732
1733 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1734 {
1735         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1736         int ret;
1737
1738         if (!tcmu_kern_cmd_reply_supported)
1739                 return 0;
1740
1741         if (udev->nl_reply_supported <= 0)
1742                 return 0;
1743
1744         pr_debug("sleeping for nl reply\n");
1745         wait_for_completion(&nl_cmd->complete);
1746
1747         mutex_lock(&tcmu_nl_cmd_mutex);
1748         nl_cmd->cmd = TCMU_CMD_UNSPEC;
1749         ret = nl_cmd->status;
1750         mutex_unlock(&tcmu_nl_cmd_mutex);
1751
1752         return ret;
1753 }
1754
1755 static int tcmu_netlink_event_init(struct tcmu_dev *udev,
1756                                    enum tcmu_genl_cmd cmd,
1757                                    struct sk_buff **buf, void **hdr)
1758 {
1759         struct sk_buff *skb;
1760         void *msg_header;
1761         int ret = -ENOMEM;
1762
1763         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1764         if (!skb)
1765                 return ret;
1766
1767         msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1768         if (!msg_header)
1769                 goto free_skb;
1770
1771         ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1772         if (ret < 0)
1773                 goto free_skb;
1774
1775         ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1776         if (ret < 0)
1777                 goto free_skb;
1778
1779         ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1780         if (ret < 0)
1781                 goto free_skb;
1782
1783         *buf = skb;
1784         *hdr = msg_header;
1785         return ret;
1786
1787 free_skb:
1788         nlmsg_free(skb);
1789         return ret;
1790 }
1791
1792 static int tcmu_netlink_event_send(struct tcmu_dev *udev,
1793                                    enum tcmu_genl_cmd cmd,
1794                                    struct sk_buff *skb, void *msg_header)
1795 {
1796         int ret;
1797
1798         genlmsg_end(skb, msg_header);
1799
1800         ret = tcmu_init_genl_cmd_reply(udev, cmd);
1801         if (ret) {
1802                 nlmsg_free(skb);
1803                 return ret;
1804         }
1805
1806         ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1807                                       TCMU_MCGRP_CONFIG, GFP_KERNEL);
1808
1809         /* Wait during an add as the listener may not be up yet */
1810         if (ret == 0 ||
1811            (ret == -ESRCH && cmd == TCMU_CMD_ADDED_DEVICE))
1812                 return tcmu_wait_genl_cmd_reply(udev);
1813         else
1814                 tcmu_destroy_genl_cmd_reply(udev);
1815
1816         return ret;
1817 }
1818
1819 static int tcmu_send_dev_add_event(struct tcmu_dev *udev)
1820 {
1821         struct sk_buff *skb = NULL;
1822         void *msg_header = NULL;
1823         int ret = 0;
1824
1825         ret = tcmu_netlink_event_init(udev, TCMU_CMD_ADDED_DEVICE, &skb,
1826                                       &msg_header);
1827         if (ret < 0)
1828                 return ret;
1829         return tcmu_netlink_event_send(udev, TCMU_CMD_ADDED_DEVICE, skb,
1830                                        msg_header);
1831 }
1832
1833 static int tcmu_send_dev_remove_event(struct tcmu_dev *udev)
1834 {
1835         struct sk_buff *skb = NULL;
1836         void *msg_header = NULL;
1837         int ret = 0;
1838
1839         ret = tcmu_netlink_event_init(udev, TCMU_CMD_REMOVED_DEVICE,
1840                                       &skb, &msg_header);
1841         if (ret < 0)
1842                 return ret;
1843         return tcmu_netlink_event_send(udev, TCMU_CMD_REMOVED_DEVICE,
1844                                        skb, msg_header);
1845 }
1846
1847 static int tcmu_update_uio_info(struct tcmu_dev *udev)
1848 {
1849         struct tcmu_hba *hba = udev->hba->hba_ptr;
1850         struct uio_info *info;
1851         char *str;
1852
1853         info = &udev->uio_info;
1854
1855         if (udev->dev_config[0])
1856                 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s/%s", hba->host_id,
1857                                 udev->name, udev->dev_config);
1858         else
1859                 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s", hba->host_id,
1860                                 udev->name);
1861         if (!str)
1862                 return -ENOMEM;
1863
1864         /* If the old string exists, free it */
1865         kfree(info->name);
1866         info->name = str;
1867
1868         return 0;
1869 }
1870
1871 static int tcmu_configure_device(struct se_device *dev)
1872 {
1873         struct tcmu_dev *udev = TCMU_DEV(dev);
1874         struct uio_info *info;
1875         struct tcmu_mailbox *mb;
1876         int ret = 0;
1877
1878         ret = tcmu_update_uio_info(udev);
1879         if (ret)
1880                 return ret;
1881
1882         info = &udev->uio_info;
1883
1884         mutex_lock(&udev->cmdr_lock);
1885         udev->data_bitmap = bitmap_zalloc(udev->max_blocks, GFP_KERNEL);
1886         mutex_unlock(&udev->cmdr_lock);
1887         if (!udev->data_bitmap) {
1888                 ret = -ENOMEM;
1889                 goto err_bitmap_alloc;
1890         }
1891
1892         udev->mb_addr = vzalloc(CMDR_SIZE);
1893         if (!udev->mb_addr) {
1894                 ret = -ENOMEM;
1895                 goto err_vzalloc;
1896         }
1897
1898         /* mailbox fits in first part of CMDR space */
1899         udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1900         udev->data_off = CMDR_SIZE;
1901         udev->data_size = udev->max_blocks * DATA_BLOCK_SIZE;
1902         udev->dbi_thresh = 0; /* Default in Idle state */
1903
1904         /* Initialise the mailbox of the ring buffer */
1905         mb = udev->mb_addr;
1906         mb->version = TCMU_MAILBOX_VERSION;
1907         mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC | TCMU_MAILBOX_FLAG_CAP_READ_LEN;
1908         mb->cmdr_off = CMDR_OFF;
1909         mb->cmdr_size = udev->cmdr_size;
1910
1911         WARN_ON(!PAGE_ALIGNED(udev->data_off));
1912         WARN_ON(udev->data_size % PAGE_SIZE);
1913         WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1914
1915         info->version = __stringify(TCMU_MAILBOX_VERSION);
1916
1917         info->mem[0].name = "tcm-user command & data buffer";
1918         info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1919         info->mem[0].size = udev->ring_size = udev->data_size + CMDR_SIZE;
1920         info->mem[0].memtype = UIO_MEM_NONE;
1921
1922         info->irqcontrol = tcmu_irqcontrol;
1923         info->irq = UIO_IRQ_CUSTOM;
1924
1925         info->mmap = tcmu_mmap;
1926         info->open = tcmu_open;
1927         info->release = tcmu_release;
1928
1929         ret = uio_register_device(tcmu_root_device, info);
1930         if (ret)
1931                 goto err_register;
1932
1933         /* User can set hw_block_size before enable the device */
1934         if (dev->dev_attrib.hw_block_size == 0)
1935                 dev->dev_attrib.hw_block_size = 512;
1936         /* Other attributes can be configured in userspace */
1937         if (!dev->dev_attrib.hw_max_sectors)
1938                 dev->dev_attrib.hw_max_sectors = 128;
1939         if (!dev->dev_attrib.emulate_write_cache)
1940                 dev->dev_attrib.emulate_write_cache = 0;
1941         dev->dev_attrib.hw_queue_depth = 128;
1942
1943         /* If user didn't explicitly disable netlink reply support, use
1944          * module scope setting.
1945          */
1946         if (udev->nl_reply_supported >= 0)
1947                 udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
1948
1949         /*
1950          * Get a ref incase userspace does a close on the uio device before
1951          * LIO has initiated tcmu_free_device.
1952          */
1953         kref_get(&udev->kref);
1954
1955         ret = tcmu_send_dev_add_event(udev);
1956         if (ret)
1957                 goto err_netlink;
1958
1959         mutex_lock(&root_udev_mutex);
1960         list_add(&udev->node, &root_udev);
1961         mutex_unlock(&root_udev_mutex);
1962
1963         return 0;
1964
1965 err_netlink:
1966         kref_put(&udev->kref, tcmu_dev_kref_release);
1967         uio_unregister_device(&udev->uio_info);
1968 err_register:
1969         vfree(udev->mb_addr);
1970         udev->mb_addr = NULL;
1971 err_vzalloc:
1972         bitmap_free(udev->data_bitmap);
1973         udev->data_bitmap = NULL;
1974 err_bitmap_alloc:
1975         kfree(info->name);
1976         info->name = NULL;
1977
1978         return ret;
1979 }
1980
1981 static void tcmu_free_device(struct se_device *dev)
1982 {
1983         struct tcmu_dev *udev = TCMU_DEV(dev);
1984
1985         /* release ref from init */
1986         kref_put(&udev->kref, tcmu_dev_kref_release);
1987 }
1988
1989 static void tcmu_destroy_device(struct se_device *dev)
1990 {
1991         struct tcmu_dev *udev = TCMU_DEV(dev);
1992
1993         del_timer_sync(&udev->cmd_timer);
1994         del_timer_sync(&udev->qfull_timer);
1995
1996         mutex_lock(&root_udev_mutex);
1997         list_del(&udev->node);
1998         mutex_unlock(&root_udev_mutex);
1999
2000         tcmu_send_dev_remove_event(udev);
2001
2002         uio_unregister_device(&udev->uio_info);
2003
2004         /* release ref from configure */
2005         kref_put(&udev->kref, tcmu_dev_kref_release);
2006 }
2007
2008 static void tcmu_unblock_dev(struct tcmu_dev *udev)
2009 {
2010         mutex_lock(&udev->cmdr_lock);
2011         clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
2012         mutex_unlock(&udev->cmdr_lock);
2013 }
2014
2015 static void tcmu_block_dev(struct tcmu_dev *udev)
2016 {
2017         mutex_lock(&udev->cmdr_lock);
2018
2019         if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2020                 goto unlock;
2021
2022         /* complete IO that has executed successfully */
2023         tcmu_handle_completions(udev);
2024         /* fail IO waiting to be queued */
2025         run_qfull_queue(udev, true);
2026
2027 unlock:
2028         mutex_unlock(&udev->cmdr_lock);
2029 }
2030
2031 static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
2032 {
2033         struct tcmu_mailbox *mb;
2034         struct tcmu_cmd *cmd;
2035         int i;
2036
2037         mutex_lock(&udev->cmdr_lock);
2038
2039         idr_for_each_entry(&udev->commands, cmd, i) {
2040                 if (!test_bit(TCMU_CMD_BIT_INFLIGHT, &cmd->flags))
2041                         continue;
2042
2043                 pr_debug("removing cmd %u on dev %s from ring (is expired %d)\n",
2044                           cmd->cmd_id, udev->name,
2045                           test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags));
2046
2047                 idr_remove(&udev->commands, i);
2048                 if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
2049                         WARN_ON(!cmd->se_cmd);
2050                         list_del_init(&cmd->queue_entry);
2051                         if (err_level == 1) {
2052                                 /*
2053                                  * Userspace was not able to start the
2054                                  * command or it is retryable.
2055                                  */
2056                                 target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
2057                         } else {
2058                                 /* hard failure */
2059                                 target_complete_cmd(cmd->se_cmd,
2060                                                     SAM_STAT_CHECK_CONDITION);
2061                         }
2062                 }
2063                 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
2064                 tcmu_free_cmd(cmd);
2065         }
2066
2067         mb = udev->mb_addr;
2068         tcmu_flush_dcache_range(mb, sizeof(*mb));
2069         pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
2070                  mb->cmd_tail, mb->cmd_head);
2071
2072         udev->cmdr_last_cleaned = 0;
2073         mb->cmd_tail = 0;
2074         mb->cmd_head = 0;
2075         tcmu_flush_dcache_range(mb, sizeof(*mb));
2076
2077         del_timer(&udev->cmd_timer);
2078
2079         mutex_unlock(&udev->cmdr_lock);
2080 }
2081
2082 enum {
2083         Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
2084         Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_err,
2085 };
2086
2087 static match_table_t tokens = {
2088         {Opt_dev_config, "dev_config=%s"},
2089         {Opt_dev_size, "dev_size=%s"},
2090         {Opt_hw_block_size, "hw_block_size=%d"},
2091         {Opt_hw_max_sectors, "hw_max_sectors=%d"},
2092         {Opt_nl_reply_supported, "nl_reply_supported=%d"},
2093         {Opt_max_data_area_mb, "max_data_area_mb=%d"},
2094         {Opt_err, NULL}
2095 };
2096
2097 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
2098 {
2099         int val, ret;
2100
2101         ret = match_int(arg, &val);
2102         if (ret < 0) {
2103                 pr_err("match_int() failed for dev attrib. Error %d.\n",
2104                        ret);
2105                 return ret;
2106         }
2107
2108         if (val <= 0) {
2109                 pr_err("Invalid dev attrib value %d. Must be greater than zero.\n",
2110                        val);
2111                 return -EINVAL;
2112         }
2113         *dev_attrib = val;
2114         return 0;
2115 }
2116
2117 static int tcmu_set_max_blocks_param(struct tcmu_dev *udev, substring_t *arg)
2118 {
2119         int val, ret;
2120
2121         ret = match_int(arg, &val);
2122         if (ret < 0) {
2123                 pr_err("match_int() failed for max_data_area_mb=. Error %d.\n",
2124                        ret);
2125                 return ret;
2126         }
2127
2128         if (val <= 0) {
2129                 pr_err("Invalid max_data_area %d.\n", val);
2130                 return -EINVAL;
2131         }
2132
2133         mutex_lock(&udev->cmdr_lock);
2134         if (udev->data_bitmap) {
2135                 pr_err("Cannot set max_data_area_mb after it has been enabled.\n");
2136                 ret = -EINVAL;
2137                 goto unlock;
2138         }
2139
2140         udev->max_blocks = TCMU_MBS_TO_BLOCKS(val);
2141         if (udev->max_blocks > tcmu_global_max_blocks) {
2142                 pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2143                        val, TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
2144                 udev->max_blocks = tcmu_global_max_blocks;
2145         }
2146
2147 unlock:
2148         mutex_unlock(&udev->cmdr_lock);
2149         return ret;
2150 }
2151
2152 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
2153                 const char *page, ssize_t count)
2154 {
2155         struct tcmu_dev *udev = TCMU_DEV(dev);
2156         char *orig, *ptr, *opts;
2157         substring_t args[MAX_OPT_ARGS];
2158         int ret = 0, token;
2159
2160         opts = kstrdup(page, GFP_KERNEL);
2161         if (!opts)
2162                 return -ENOMEM;
2163
2164         orig = opts;
2165
2166         while ((ptr = strsep(&opts, ",\n")) != NULL) {
2167                 if (!*ptr)
2168                         continue;
2169
2170                 token = match_token(ptr, tokens, args);
2171                 switch (token) {
2172                 case Opt_dev_config:
2173                         if (match_strlcpy(udev->dev_config, &args[0],
2174                                           TCMU_CONFIG_LEN) == 0) {
2175                                 ret = -EINVAL;
2176                                 break;
2177                         }
2178                         pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
2179                         break;
2180                 case Opt_dev_size:
2181                         ret = match_u64(&args[0], &udev->dev_size);
2182                         if (ret < 0)
2183                                 pr_err("match_u64() failed for dev_size=. Error %d.\n",
2184                                        ret);
2185                         break;
2186                 case Opt_hw_block_size:
2187                         ret = tcmu_set_dev_attrib(&args[0],
2188                                         &(dev->dev_attrib.hw_block_size));
2189                         break;
2190                 case Opt_hw_max_sectors:
2191                         ret = tcmu_set_dev_attrib(&args[0],
2192                                         &(dev->dev_attrib.hw_max_sectors));
2193                         break;
2194                 case Opt_nl_reply_supported:
2195                         ret = match_int(&args[0], &udev->nl_reply_supported);
2196                         if (ret < 0)
2197                                 pr_err("match_int() failed for nl_reply_supported=. Error %d.\n",
2198                                        ret);
2199                         break;
2200                 case Opt_max_data_area_mb:
2201                         ret = tcmu_set_max_blocks_param(udev, &args[0]);
2202                         break;
2203                 default:
2204                         break;
2205                 }
2206
2207                 if (ret)
2208                         break;
2209         }
2210
2211         kfree(orig);
2212         return (!ret) ? count : ret;
2213 }
2214
2215 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2216 {
2217         struct tcmu_dev *udev = TCMU_DEV(dev);
2218         ssize_t bl = 0;
2219
2220         bl = sprintf(b + bl, "Config: %s ",
2221                      udev->dev_config[0] ? udev->dev_config : "NULL");
2222         bl += sprintf(b + bl, "Size: %llu ", udev->dev_size);
2223         bl += sprintf(b + bl, "MaxDataAreaMB: %u\n",
2224                       TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2225
2226         return bl;
2227 }
2228
2229 static sector_t tcmu_get_blocks(struct se_device *dev)
2230 {
2231         struct tcmu_dev *udev = TCMU_DEV(dev);
2232
2233         return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2234                        dev->dev_attrib.block_size);
2235 }
2236
2237 static sense_reason_t
2238 tcmu_parse_cdb(struct se_cmd *cmd)
2239 {
2240         return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2241 }
2242
2243 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2244 {
2245         struct se_dev_attrib *da = container_of(to_config_group(item),
2246                                         struct se_dev_attrib, da_group);
2247         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2248
2249         return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2250 }
2251
2252 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2253                                        size_t count)
2254 {
2255         struct se_dev_attrib *da = container_of(to_config_group(item),
2256                                         struct se_dev_attrib, da_group);
2257         struct tcmu_dev *udev = container_of(da->da_dev,
2258                                         struct tcmu_dev, se_dev);
2259         u32 val;
2260         int ret;
2261
2262         if (da->da_dev->export_count) {
2263                 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2264                 return -EINVAL;
2265         }
2266
2267         ret = kstrtou32(page, 0, &val);
2268         if (ret < 0)
2269                 return ret;
2270
2271         udev->cmd_time_out = val * MSEC_PER_SEC;
2272         return count;
2273 }
2274 CONFIGFS_ATTR(tcmu_, cmd_time_out);
2275
2276 static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2277 {
2278         struct se_dev_attrib *da = container_of(to_config_group(item),
2279                                                 struct se_dev_attrib, da_group);
2280         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2281
2282         return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2283                         udev->qfull_time_out :
2284                         udev->qfull_time_out / MSEC_PER_SEC);
2285 }
2286
2287 static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2288                                          const char *page, size_t count)
2289 {
2290         struct se_dev_attrib *da = container_of(to_config_group(item),
2291                                         struct se_dev_attrib, da_group);
2292         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2293         s32 val;
2294         int ret;
2295
2296         ret = kstrtos32(page, 0, &val);
2297         if (ret < 0)
2298                 return ret;
2299
2300         if (val >= 0) {
2301                 udev->qfull_time_out = val * MSEC_PER_SEC;
2302         } else if (val == -1) {
2303                 udev->qfull_time_out = val;
2304         } else {
2305                 printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2306                 return -EINVAL;
2307         }
2308         return count;
2309 }
2310 CONFIGFS_ATTR(tcmu_, qfull_time_out);
2311
2312 static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2313 {
2314         struct se_dev_attrib *da = container_of(to_config_group(item),
2315                                                 struct se_dev_attrib, da_group);
2316         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2317
2318         return snprintf(page, PAGE_SIZE, "%u\n",
2319                         TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2320 }
2321 CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2322
2323 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2324 {
2325         struct se_dev_attrib *da = container_of(to_config_group(item),
2326                                                 struct se_dev_attrib, da_group);
2327         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2328
2329         return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2330 }
2331
2332 static int tcmu_send_dev_config_event(struct tcmu_dev *udev,
2333                                       const char *reconfig_data)
2334 {
2335         struct sk_buff *skb = NULL;
2336         void *msg_header = NULL;
2337         int ret = 0;
2338
2339         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2340                                       &skb, &msg_header);
2341         if (ret < 0)
2342                 return ret;
2343         ret = nla_put_string(skb, TCMU_ATTR_DEV_CFG, reconfig_data);
2344         if (ret < 0) {
2345                 nlmsg_free(skb);
2346                 return ret;
2347         }
2348         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2349                                        skb, msg_header);
2350 }
2351
2352
2353 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2354                                      size_t count)
2355 {
2356         struct se_dev_attrib *da = container_of(to_config_group(item),
2357                                                 struct se_dev_attrib, da_group);
2358         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2359         int ret, len;
2360
2361         len = strlen(page);
2362         if (!len || len > TCMU_CONFIG_LEN - 1)
2363                 return -EINVAL;
2364
2365         /* Check if device has been configured before */
2366         if (target_dev_configured(&udev->se_dev)) {
2367                 ret = tcmu_send_dev_config_event(udev, page);
2368                 if (ret) {
2369                         pr_err("Unable to reconfigure device\n");
2370                         return ret;
2371                 }
2372                 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2373
2374                 ret = tcmu_update_uio_info(udev);
2375                 if (ret)
2376                         return ret;
2377                 return count;
2378         }
2379         strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2380
2381         return count;
2382 }
2383 CONFIGFS_ATTR(tcmu_, dev_config);
2384
2385 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2386 {
2387         struct se_dev_attrib *da = container_of(to_config_group(item),
2388                                                 struct se_dev_attrib, da_group);
2389         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2390
2391         return snprintf(page, PAGE_SIZE, "%llu\n", udev->dev_size);
2392 }
2393
2394 static int tcmu_send_dev_size_event(struct tcmu_dev *udev, u64 size)
2395 {
2396         struct sk_buff *skb = NULL;
2397         void *msg_header = NULL;
2398         int ret = 0;
2399
2400         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2401                                       &skb, &msg_header);
2402         if (ret < 0)
2403                 return ret;
2404         ret = nla_put_u64_64bit(skb, TCMU_ATTR_DEV_SIZE,
2405                                 size, TCMU_ATTR_PAD);
2406         if (ret < 0) {
2407                 nlmsg_free(skb);
2408                 return ret;
2409         }
2410         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2411                                        skb, msg_header);
2412 }
2413
2414 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2415                                    size_t count)
2416 {
2417         struct se_dev_attrib *da = container_of(to_config_group(item),
2418                                                 struct se_dev_attrib, da_group);
2419         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2420         u64 val;
2421         int ret;
2422
2423         ret = kstrtou64(page, 0, &val);
2424         if (ret < 0)
2425                 return ret;
2426
2427         /* Check if device has been configured before */
2428         if (target_dev_configured(&udev->se_dev)) {
2429                 ret = tcmu_send_dev_size_event(udev, val);
2430                 if (ret) {
2431                         pr_err("Unable to reconfigure device\n");
2432                         return ret;
2433                 }
2434         }
2435         udev->dev_size = val;
2436         return count;
2437 }
2438 CONFIGFS_ATTR(tcmu_, dev_size);
2439
2440 static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
2441                 char *page)
2442 {
2443         struct se_dev_attrib *da = container_of(to_config_group(item),
2444                                                 struct se_dev_attrib, da_group);
2445         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2446
2447         return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2448 }
2449
2450 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2451                 const char *page, size_t count)
2452 {
2453         struct se_dev_attrib *da = container_of(to_config_group(item),
2454                                                 struct se_dev_attrib, da_group);
2455         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2456         s8 val;
2457         int ret;
2458
2459         ret = kstrtos8(page, 0, &val);
2460         if (ret < 0)
2461                 return ret;
2462
2463         udev->nl_reply_supported = val;
2464         return count;
2465 }
2466 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2467
2468 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2469                                              char *page)
2470 {
2471         struct se_dev_attrib *da = container_of(to_config_group(item),
2472                                         struct se_dev_attrib, da_group);
2473
2474         return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2475 }
2476
2477 static int tcmu_send_emulate_write_cache(struct tcmu_dev *udev, u8 val)
2478 {
2479         struct sk_buff *skb = NULL;
2480         void *msg_header = NULL;
2481         int ret = 0;
2482
2483         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2484                                       &skb, &msg_header);
2485         if (ret < 0)
2486                 return ret;
2487         ret = nla_put_u8(skb, TCMU_ATTR_WRITECACHE, val);
2488         if (ret < 0) {
2489                 nlmsg_free(skb);
2490                 return ret;
2491         }
2492         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2493                                        skb, msg_header);
2494 }
2495
2496 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2497                                               const char *page, size_t count)
2498 {
2499         struct se_dev_attrib *da = container_of(to_config_group(item),
2500                                         struct se_dev_attrib, da_group);
2501         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2502         u8 val;
2503         int ret;
2504
2505         ret = kstrtou8(page, 0, &val);
2506         if (ret < 0)
2507                 return ret;
2508
2509         /* Check if device has been configured before */
2510         if (target_dev_configured(&udev->se_dev)) {
2511                 ret = tcmu_send_emulate_write_cache(udev, val);
2512                 if (ret) {
2513                         pr_err("Unable to reconfigure device\n");
2514                         return ret;
2515                 }
2516         }
2517
2518         da->emulate_write_cache = val;
2519         return count;
2520 }
2521 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2522
2523 static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
2524 {
2525         struct se_device *se_dev = container_of(to_config_group(item),
2526                                                 struct se_device,
2527                                                 dev_action_group);
2528         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2529
2530         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2531                 return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
2532         else
2533                 return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
2534 }
2535
2536 static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
2537                                     size_t count)
2538 {
2539         struct se_device *se_dev = container_of(to_config_group(item),
2540                                                 struct se_device,
2541                                                 dev_action_group);
2542         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2543         u8 val;
2544         int ret;
2545
2546         if (!target_dev_configured(&udev->se_dev)) {
2547                 pr_err("Device is not configured.\n");
2548                 return -EINVAL;
2549         }
2550
2551         ret = kstrtou8(page, 0, &val);
2552         if (ret < 0)
2553                 return ret;
2554
2555         if (val > 1) {
2556                 pr_err("Invalid block value %d\n", val);
2557                 return -EINVAL;
2558         }
2559
2560         if (!val)
2561                 tcmu_unblock_dev(udev);
2562         else
2563                 tcmu_block_dev(udev);
2564         return count;
2565 }
2566 CONFIGFS_ATTR(tcmu_, block_dev);
2567
2568 static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
2569                                      size_t count)
2570 {
2571         struct se_device *se_dev = container_of(to_config_group(item),
2572                                                 struct se_device,
2573                                                 dev_action_group);
2574         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2575         u8 val;
2576         int ret;
2577
2578         if (!target_dev_configured(&udev->se_dev)) {
2579                 pr_err("Device is not configured.\n");
2580                 return -EINVAL;
2581         }
2582
2583         ret = kstrtou8(page, 0, &val);
2584         if (ret < 0)
2585                 return ret;
2586
2587         if (val != 1 && val != 2) {
2588                 pr_err("Invalid reset ring value %d\n", val);
2589                 return -EINVAL;
2590         }
2591
2592         tcmu_reset_ring(udev, val);
2593         return count;
2594 }
2595 CONFIGFS_ATTR_WO(tcmu_, reset_ring);
2596
2597 static struct configfs_attribute *tcmu_attrib_attrs[] = {
2598         &tcmu_attr_cmd_time_out,
2599         &tcmu_attr_qfull_time_out,
2600         &tcmu_attr_max_data_area_mb,
2601         &tcmu_attr_dev_config,
2602         &tcmu_attr_dev_size,
2603         &tcmu_attr_emulate_write_cache,
2604         &tcmu_attr_nl_reply_supported,
2605         NULL,
2606 };
2607
2608 static struct configfs_attribute **tcmu_attrs;
2609
2610 static struct configfs_attribute *tcmu_action_attrs[] = {
2611         &tcmu_attr_block_dev,
2612         &tcmu_attr_reset_ring,
2613         NULL,
2614 };
2615
2616 static struct target_backend_ops tcmu_ops = {
2617         .name                   = "user",
2618         .owner                  = THIS_MODULE,
2619         .transport_flags        = TRANSPORT_FLAG_PASSTHROUGH,
2620         .attach_hba             = tcmu_attach_hba,
2621         .detach_hba             = tcmu_detach_hba,
2622         .alloc_device           = tcmu_alloc_device,
2623         .configure_device       = tcmu_configure_device,
2624         .destroy_device         = tcmu_destroy_device,
2625         .free_device            = tcmu_free_device,
2626         .parse_cdb              = tcmu_parse_cdb,
2627         .set_configfs_dev_params = tcmu_set_configfs_dev_params,
2628         .show_configfs_dev_params = tcmu_show_configfs_dev_params,
2629         .get_device_type        = sbc_get_device_type,
2630         .get_blocks             = tcmu_get_blocks,
2631         .tb_dev_action_attrs    = tcmu_action_attrs,
2632 };
2633
2634 static void find_free_blocks(void)
2635 {
2636         struct tcmu_dev *udev;
2637         loff_t off;
2638         u32 start, end, block, total_freed = 0;
2639
2640         if (atomic_read(&global_db_count) <= tcmu_global_max_blocks)
2641                 return;
2642
2643         mutex_lock(&root_udev_mutex);
2644         list_for_each_entry(udev, &root_udev, node) {
2645                 mutex_lock(&udev->cmdr_lock);
2646
2647                 if (!target_dev_configured(&udev->se_dev)) {
2648                         mutex_unlock(&udev->cmdr_lock);
2649                         continue;
2650                 }
2651
2652                 /* Try to complete the finished commands first */
2653                 tcmu_handle_completions(udev);
2654
2655                 /* Skip the udevs in idle */
2656                 if (!udev->dbi_thresh) {
2657                         mutex_unlock(&udev->cmdr_lock);
2658                         continue;
2659                 }
2660
2661                 end = udev->dbi_max + 1;
2662                 block = find_last_bit(udev->data_bitmap, end);
2663                 if (block == udev->dbi_max) {
2664                         /*
2665                          * The last bit is dbi_max, so it is not possible
2666                          * reclaim any blocks.
2667                          */
2668                         mutex_unlock(&udev->cmdr_lock);
2669                         continue;
2670                 } else if (block == end) {
2671                         /* The current udev will goto idle state */
2672                         udev->dbi_thresh = start = 0;
2673                         udev->dbi_max = 0;
2674                 } else {
2675                         udev->dbi_thresh = start = block + 1;
2676                         udev->dbi_max = block;
2677                 }
2678
2679                 /* Here will truncate the data area from off */
2680                 off = udev->data_off + start * DATA_BLOCK_SIZE;
2681                 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
2682
2683                 /* Release the block pages */
2684                 tcmu_blocks_release(&udev->data_blocks, start, end);
2685                 mutex_unlock(&udev->cmdr_lock);
2686
2687                 total_freed += end - start;
2688                 pr_debug("Freed %u blocks (total %u) from %s.\n", end - start,
2689                          total_freed, udev->name);
2690         }
2691         mutex_unlock(&root_udev_mutex);
2692
2693         if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
2694                 schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
2695 }
2696
2697 static void check_timedout_devices(void)
2698 {
2699         struct tcmu_dev *udev, *tmp_dev;
2700         LIST_HEAD(devs);
2701
2702         spin_lock_bh(&timed_out_udevs_lock);
2703         list_splice_init(&timed_out_udevs, &devs);
2704
2705         list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
2706                 list_del_init(&udev->timedout_entry);
2707                 spin_unlock_bh(&timed_out_udevs_lock);
2708
2709                 mutex_lock(&udev->cmdr_lock);
2710                 idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
2711
2712                 tcmu_set_next_deadline(&udev->inflight_queue, &udev->cmd_timer);
2713                 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
2714
2715                 mutex_unlock(&udev->cmdr_lock);
2716
2717                 spin_lock_bh(&timed_out_udevs_lock);
2718         }
2719
2720         spin_unlock_bh(&timed_out_udevs_lock);
2721 }
2722
2723 static void tcmu_unmap_work_fn(struct work_struct *work)
2724 {
2725         check_timedout_devices();
2726         find_free_blocks();
2727 }
2728
2729 static int __init tcmu_module_init(void)
2730 {
2731         int ret, i, k, len = 0;
2732
2733         BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2734
2735         INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
2736
2737         tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2738                                 sizeof(struct tcmu_cmd),
2739                                 __alignof__(struct tcmu_cmd),
2740                                 0, NULL);
2741         if (!tcmu_cmd_cache)
2742                 return -ENOMEM;
2743
2744         tcmu_root_device = root_device_register("tcm_user");
2745         if (IS_ERR(tcmu_root_device)) {
2746                 ret = PTR_ERR(tcmu_root_device);
2747                 goto out_free_cache;
2748         }
2749
2750         ret = genl_register_family(&tcmu_genl_family);
2751         if (ret < 0) {
2752                 goto out_unreg_device;
2753         }
2754
2755         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2756                 len += sizeof(struct configfs_attribute *);
2757         }
2758         for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2759                 len += sizeof(struct configfs_attribute *);
2760         }
2761         len += sizeof(struct configfs_attribute *);
2762
2763         tcmu_attrs = kzalloc(len, GFP_KERNEL);
2764         if (!tcmu_attrs) {
2765                 ret = -ENOMEM;
2766                 goto out_unreg_genl;
2767         }
2768
2769         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2770                 tcmu_attrs[i] = passthrough_attrib_attrs[i];
2771         }
2772         for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2773                 tcmu_attrs[i] = tcmu_attrib_attrs[k];
2774                 i++;
2775         }
2776         tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2777
2778         ret = transport_backend_register(&tcmu_ops);
2779         if (ret)
2780                 goto out_attrs;
2781
2782         return 0;
2783
2784 out_attrs:
2785         kfree(tcmu_attrs);
2786 out_unreg_genl:
2787         genl_unregister_family(&tcmu_genl_family);
2788 out_unreg_device:
2789         root_device_unregister(tcmu_root_device);
2790 out_free_cache:
2791         kmem_cache_destroy(tcmu_cmd_cache);
2792
2793         return ret;
2794 }
2795
2796 static void __exit tcmu_module_exit(void)
2797 {
2798         cancel_delayed_work_sync(&tcmu_unmap_work);
2799         target_backend_unregister(&tcmu_ops);
2800         kfree(tcmu_attrs);
2801         genl_unregister_family(&tcmu_genl_family);
2802         root_device_unregister(tcmu_root_device);
2803         kmem_cache_destroy(tcmu_cmd_cache);
2804 }
2805
2806 MODULE_DESCRIPTION("TCM USER subsystem plugin");
2807 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2808 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2809 MODULE_LICENSE("GPL");
2810
2811 module_init(tcmu_module_init);
2812 module_exit(tcmu_module_exit);