drivers/target/target_core_user.c

   1 /*
   2  * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
   3  * Copyright (C) 2014 Red Hat, Inc.
   4  * Copyright (C) 2015 Arrikto, Inc.
   5  * Copyright (C) 2017 Chinamobile, Inc.
   6  *
   7  * This program is free software; you can redistribute it and/or modify it
   8  * under the terms and conditions of the GNU General Public License,
   9  * version 2, as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  14  * more details.
  15  *
  16  * You should have received a copy of the GNU General Public License along with
  17  * this program; if not, write to the Free Software Foundation, Inc.,
  18  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  19  */
  20
  21 #include <linux/spinlock.h>
  22 #include <linux/module.h>
  23 #include <linux/idr.h>
  24 #include <linux/kernel.h>
  25 #include <linux/timer.h>
  26 #include <linux/parser.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/uio_driver.h>
  29 #include <linux/radix-tree.h>
  30 #include <linux/stringify.h>
  31 #include <linux/bitops.h>
  32 #include <linux/highmem.h>
  33 #include <linux/configfs.h>
  34 #include <linux/mutex.h>
  35 #include <linux/kthread.h>
  36 #include <net/genetlink.h>
  37 #include <scsi/scsi_common.h>
  38 #include <scsi/scsi_proto.h>
  39 #include <target/target_core_base.h>
  40 #include <target/target_core_fabric.h>
  41 #include <target/target_core_backend.h>
  42
  43 #include <linux/target_core_user.h>
  44
  45 /*
  46  * Define a shared-memory interface for LIO to pass SCSI commands and
  47  * data to userspace for processing. This is to allow backends that
  48  * are too complex for in-kernel support to be possible.
  49  *
  50  * It uses the UIO framework to do a lot of the device-creation and
  51  * introspection work for us.
  52  *
  53  * See the .h file for how the ring is laid out. Note that while the
  54  * command ring is defined, the particulars of the data area are
  55  * not. Offset values in the command entry point to other locations
  56  * internal to the mmap()ed area. There is separate space outside the
  57  * command ring for data buffers. This leaves maximum flexibility for
  58  * moving buffer allocations, or even page flipping or other
  59  * allocation techniques, without altering the command ring layout.
  60  *
  61  * SECURITY:
  62  * The user process must be assumed to be malicious. There's no way to
  63  * prevent it breaking the command ring protocol if it wants, but in
  64  * order to prevent other issues we must only ever read *data* from
  65  * the shared memory area, not offsets or sizes. This applies to
  66  * command ring entries as well as the mailbox. Extra code needed for
  67  * this may have a 'UAM' comment.
  68  */
  69
  70 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
  71
  72 /* For cmd area, the size is fixed 8MB */
  73 #define CMDR_SIZE (8 * 1024 * 1024)
  74
  75 /*
  76  * For data area, the block size is PAGE_SIZE and
  77  * the total size is 256K * PAGE_SIZE.
  78  */
  79 #define DATA_BLOCK_SIZE PAGE_SIZE
  80 #define DATA_BLOCK_BITS (256 * 1024)
  81 #define DATA_SIZE (DATA_BLOCK_BITS * DATA_BLOCK_SIZE)
  82 #define DATA_BLOCK_INIT_BITS 128
  83
  84 /* The total size of the ring is 8M + 256K * PAGE_SIZE */
  85 #define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
  86
  87 /* Default maximum of the global data blocks(512K * PAGE_SIZE) */
  88 #define TCMU_GLOBAL_MAX_BLOCKS (512 * 1024)
  89
  90 static u8 tcmu_kern_cmd_reply_supported;
  91
  92 static struct device *tcmu_root_device;
  93
  94 struct tcmu_hba {
  95         u32 host_id;
  96 };
  97
  98 #define TCMU_CONFIG_LEN 256
  99
 100 struct tcmu_nl_cmd {
 101         /* wake up thread waiting for reply */
 102         struct completion complete;
 103         int cmd;
 104         int status;
 105 };
 106
 107 struct tcmu_dev {
 108         struct list_head node;
 109         struct kref kref;
 110         struct se_device se_dev;
 111
 112         char *name;
 113         struct se_hba *hba;
 114
 115 #define TCMU_DEV_BIT_OPEN 0
 116 #define TCMU_DEV_BIT_BROKEN 1
 117         unsigned long flags;
 118
 119         struct uio_info uio_info;
 120
 121         struct inode *inode;
 122
 123         struct tcmu_mailbox *mb_addr;
 124         size_t dev_size;
 125         u32 cmdr_size;
 126         u32 cmdr_last_cleaned;
 127         /* Offset of data area from start of mb */
 128         /* Must add data_off and mb_addr to get the address */
 129         size_t data_off;
 130         size_t data_size;
 131
 132         wait_queue_head_t wait_cmdr;
 133         struct mutex cmdr_lock;
 134
 135         bool waiting_global;
 136         uint32_t dbi_max;
 137         uint32_t dbi_thresh;
 138         DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
 139         struct radix_tree_root data_blocks;
 140
 141         struct idr commands;
 142         spinlock_t commands_lock;
 143
 144         struct timer_list timeout;
 145         unsigned int cmd_time_out;
 146
 147         spinlock_t nl_cmd_lock;
 148         struct tcmu_nl_cmd curr_nl_cmd;
 149         /* wake up threads waiting on curr_nl_cmd */
 150         wait_queue_head_t nl_cmd_wq;
 151
 152         char dev_config[TCMU_CONFIG_LEN];
 153 };
 154
 155 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
 156
 157 #define CMDR_OFF sizeof(struct tcmu_mailbox)
 158
 159 struct tcmu_cmd {
 160         struct se_cmd *se_cmd;
 161         struct tcmu_dev *tcmu_dev;
 162
 163         uint16_t cmd_id;
 164
 165         /* Can't use se_cmd when cleaning up expired cmds, because if
 166            cmd has been completed then accessing se_cmd is off limits */
 167         uint32_t dbi_cnt;
 168         uint32_t dbi_cur;
 169         uint32_t *dbi;
 170
 171         unsigned long deadline;
 172
 173 #define TCMU_CMD_BIT_EXPIRED 0
 174         unsigned long flags;
 175 };
 176
 177 static struct task_struct *unmap_thread;
 178 static wait_queue_head_t unmap_wait;
 179 static DEFINE_MUTEX(root_udev_mutex);
 180 static LIST_HEAD(root_udev);
 181
 182 static atomic_t global_db_count = ATOMIC_INIT(0);
 183
 184 static struct kmem_cache *tcmu_cmd_cache;
 185
 186 /* multicast group */
 187 enum tcmu_multicast_groups {
 188         TCMU_MCGRP_CONFIG,
 189 };
 190
 191 static const struct genl_multicast_group tcmu_mcgrps[] = {
 192         [TCMU_MCGRP_CONFIG] = { .name = "config", },
 193 };
 194
 195 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
 196         [TCMU_ATTR_DEVICE]      = { .type = NLA_STRING },
 197         [TCMU_ATTR_MINOR]       = { .type = NLA_U32 },
 198         [TCMU_ATTR_CMD_STATUS]  = { .type = NLA_S32 },
 199         [TCMU_ATTR_DEVICE_ID]   = { .type = NLA_U32 },
 200         [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
 201 };
 202
 203 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
 204 {
 205         struct se_device *dev;
 206         struct tcmu_dev *udev;
 207         struct tcmu_nl_cmd *nl_cmd;
 208         int dev_id, rc, ret = 0;
 209         bool is_removed = (completed_cmd == TCMU_CMD_REMOVED_DEVICE);
 210
 211         if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
 212             !info->attrs[TCMU_ATTR_DEVICE_ID]) {
 213                 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
 214                 return -EINVAL;
 215         }
 216
 217         dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
 218         rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
 219
 220         dev = target_find_device(dev_id, !is_removed);
 221         if (!dev) {
 222                 printk(KERN_ERR "tcmu nl cmd %u/%u completion could not find device with dev id %u.\n",
 223                        completed_cmd, rc, dev_id);
 224                 return -ENODEV;
 225         }
 226         udev = TCMU_DEV(dev);
 227
 228         spin_lock(&udev->nl_cmd_lock);
 229         nl_cmd = &udev->curr_nl_cmd;
 230
 231         pr_debug("genl cmd done got id %d curr %d done %d rc %d\n", dev_id,
 232                  nl_cmd->cmd, completed_cmd, rc);
 233
 234         if (nl_cmd->cmd != completed_cmd) {
 235                 printk(KERN_ERR "Mismatched commands (Expecting reply for %d. Current %d).\n",
 236                        completed_cmd, nl_cmd->cmd);
 237                 ret = -EINVAL;
 238         } else {
 239                 nl_cmd->status = rc;
 240         }
 241
 242         spin_unlock(&udev->nl_cmd_lock);
 243         if (!is_removed)
 244                  target_undepend_item(&dev->dev_group.cg_item);
 245         if (!ret)
 246                 complete(&nl_cmd->complete);
 247         return ret;
 248 }
 249
 250 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
 251 {
 252         return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
 253 }
 254
 255 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
 256 {
 257         return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
 258 }
 259
 260 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
 261                                        struct genl_info *info)
 262 {
 263         return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
 264 }
 265
 266 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
 267 {
 268         if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
 269                 tcmu_kern_cmd_reply_supported  =
 270                         nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
 271                 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
 272                        tcmu_kern_cmd_reply_supported);
 273         }
 274
 275         return 0;
 276 }
 277
 278 static const struct genl_ops tcmu_genl_ops[] = {
 279         {
 280                 .cmd    = TCMU_CMD_SET_FEATURES,
 281                 .flags  = GENL_ADMIN_PERM,
 282                 .policy = tcmu_attr_policy,
 283                 .doit   = tcmu_genl_set_features,
 284         },
 285         {
 286                 .cmd    = TCMU_CMD_ADDED_DEVICE_DONE,
 287                 .flags  = GENL_ADMIN_PERM,
 288                 .policy = tcmu_attr_policy,
 289                 .doit   = tcmu_genl_add_dev_done,
 290         },
 291         {
 292                 .cmd    = TCMU_CMD_REMOVED_DEVICE_DONE,
 293                 .flags  = GENL_ADMIN_PERM,
 294                 .policy = tcmu_attr_policy,
 295                 .doit   = tcmu_genl_rm_dev_done,
 296         },
 297         {
 298                 .cmd    = TCMU_CMD_RECONFIG_DEVICE_DONE,
 299                 .flags  = GENL_ADMIN_PERM,
 300                 .policy = tcmu_attr_policy,
 301                 .doit   = tcmu_genl_reconfig_dev_done,
 302         },
 303 };
 304
 305 /* Our generic netlink family */
 306 static struct genl_family tcmu_genl_family __ro_after_init = {
 307         .module = THIS_MODULE,
 308         .hdrsize = 0,
 309         .name = "TCM-USER",
 310         .version = 2,
 311         .maxattr = TCMU_ATTR_MAX,
 312         .mcgrps = tcmu_mcgrps,
 313         .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
 314         .netnsok = true,
 315         .ops = tcmu_genl_ops,
 316         .n_ops = ARRAY_SIZE(tcmu_genl_ops),
 317 };
 318
 319 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
 320 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
 321 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
 322 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
 323
 324 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
 325 {
 326         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 327         uint32_t i;
 328
 329         for (i = 0; i < len; i++)
 330                 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
 331 }
 332
 333 static inline bool tcmu_get_empty_block(struct tcmu_dev *udev,
 334                                         struct tcmu_cmd *tcmu_cmd)
 335 {
 336         struct page *page;
 337         int ret, dbi;
 338
 339         dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
 340         if (dbi == udev->dbi_thresh)
 341                 return false;
 342
 343         page = radix_tree_lookup(&udev->data_blocks, dbi);
 344         if (!page) {
 345                 if (atomic_add_return(1, &global_db_count) >
 346                                         TCMU_GLOBAL_MAX_BLOCKS) {
 347                         atomic_dec(&global_db_count);
 348                         return false;
 349                 }
 350
 351                 /* try to get new page from the mm */
 352                 page = alloc_page(GFP_KERNEL);
 353                 if (!page)
 354                         goto err_alloc;
 355
 356                 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
 357                 if (ret)
 358                         goto err_insert;
 359         }
 360
 361         if (dbi > udev->dbi_max)
 362                 udev->dbi_max = dbi;
 363
 364         set_bit(dbi, udev->data_bitmap);
 365         tcmu_cmd_set_dbi(tcmu_cmd, dbi);
 366
 367         return true;
 368 err_insert:
 369         __free_page(page);
 370 err_alloc:
 371         atomic_dec(&global_db_count);
 372         return false;
 373 }
 374
 375 static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
 376                                   struct tcmu_cmd *tcmu_cmd)
 377 {
 378         int i;
 379
 380         udev->waiting_global = false;
 381
 382         for (i = tcmu_cmd->dbi_cur; i < tcmu_cmd->dbi_cnt; i++) {
 383                 if (!tcmu_get_empty_block(udev, tcmu_cmd))
 384                         goto err;
 385         }
 386         return true;
 387
 388 err:
 389         udev->waiting_global = true;
 390         /* Try to wake up the unmap thread */
 391         wake_up(&unmap_wait);
 392         return false;
 393 }
 394
 395 static inline struct page *
 396 tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
 397 {
 398         return radix_tree_lookup(&udev->data_blocks, dbi);
 399 }
 400
 401 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
 402 {
 403         kfree(tcmu_cmd->dbi);
 404         kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 405 }
 406
 407 static inline size_t tcmu_cmd_get_data_length(struct tcmu_cmd *tcmu_cmd)
 408 {
 409         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 410         size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
 411
 412         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 413                 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
 414                 data_length += round_up(se_cmd->t_bidi_data_sg->length,
 415                                 DATA_BLOCK_SIZE);
 416         }
 417
 418         return data_length;
 419 }
 420
 421 static inline uint32_t tcmu_cmd_get_block_cnt(struct tcmu_cmd *tcmu_cmd)
 422 {
 423         size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
 424
 425         return data_length / DATA_BLOCK_SIZE;
 426 }
 427
 428 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
 429 {
 430         struct se_device *se_dev = se_cmd->se_dev;
 431         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 432         struct tcmu_cmd *tcmu_cmd;
 433         int cmd_id;
 434
 435         tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_KERNEL);
 436         if (!tcmu_cmd)
 437                 return NULL;
 438
 439         tcmu_cmd->se_cmd = se_cmd;
 440         tcmu_cmd->tcmu_dev = udev;
 441         if (udev->cmd_time_out)
 442                 tcmu_cmd->deadline = jiffies +
 443                                         msecs_to_jiffies(udev->cmd_time_out);
 444
 445         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 446         tcmu_cmd->dbi_cnt = tcmu_cmd_get_block_cnt(tcmu_cmd);
 447         tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
 448                                 GFP_KERNEL);
 449         if (!tcmu_cmd->dbi) {
 450                 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 451                 return NULL;
 452         }
 453
 454         idr_preload(GFP_KERNEL);
 455         spin_lock_irq(&udev->commands_lock);
 456         cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 0,
 457                 USHRT_MAX, GFP_NOWAIT);
 458         spin_unlock_irq(&udev->commands_lock);
 459         idr_preload_end();
 460
 461         if (cmd_id < 0) {
 462                 tcmu_free_cmd(tcmu_cmd);
 463                 return NULL;
 464         }
 465         tcmu_cmd->cmd_id = cmd_id;
 466
 467         return tcmu_cmd;
 468 }
 469
 470 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
 471 {
 472         unsigned long offset = offset_in_page(vaddr);
 473
 474         size = round_up(size+offset, PAGE_SIZE);
 475         vaddr -= offset;
 476
 477         while (size) {
 478                 flush_dcache_page(virt_to_page(vaddr));
 479                 size -= PAGE_SIZE;
 480         }
 481 }
 482
 483 /*
 484  * Some ring helper functions. We don't assume size is a power of 2 so
 485  * we can't use circ_buf.h.
 486  */
 487 static inline size_t spc_used(size_t head, size_t tail, size_t size)
 488 {
 489         int diff = head - tail;
 490
 491         if (diff >= 0)
 492                 return diff;
 493         else
 494                 return size + diff;
 495 }
 496
 497 static inline size_t spc_free(size_t head, size_t tail, size_t size)
 498 {
 499         /* Keep 1 byte unused or we can't tell full from empty */
 500         return (size - spc_used(head, tail, size) - 1);
 501 }
 502
 503 static inline size_t head_to_end(size_t head, size_t size)
 504 {
 505         return size - head;
 506 }
 507
 508 static inline void new_iov(struct iovec **iov, int *iov_cnt,
 509                            struct tcmu_dev *udev)
 510 {
 511         struct iovec *iovec;
 512
 513         if (*iov_cnt != 0)
 514                 (*iov)++;
 515         (*iov_cnt)++;
 516
 517         iovec = *iov;
 518         memset(iovec, 0, sizeof(struct iovec));
 519 }
 520
 521 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
 522
 523 /* offset is relative to mb_addr */
 524 static inline size_t get_block_offset_user(struct tcmu_dev *dev,
 525                 int dbi, int remaining)
 526 {
 527         return dev->data_off + dbi * DATA_BLOCK_SIZE +
 528                 DATA_BLOCK_SIZE - remaining;
 529 }
 530
 531 static inline size_t iov_tail(struct iovec *iov)
 532 {
 533         return (size_t)iov->iov_base + iov->iov_len;
 534 }
 535
 536 static int scatter_data_area(struct tcmu_dev *udev,
 537         struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
 538         unsigned int data_nents, struct iovec **iov,
 539         int *iov_cnt, bool copy_data)
 540 {
 541         int i, dbi;
 542         int block_remaining = 0;
 543         void *from, *to = NULL;
 544         size_t copy_bytes, to_offset, offset;
 545         struct scatterlist *sg;
 546         struct page *page;
 547
 548         for_each_sg(data_sg, sg, data_nents, i) {
 549                 int sg_remaining = sg->length;
 550                 from = kmap_atomic(sg_page(sg)) + sg->offset;
 551                 while (sg_remaining > 0) {
 552                         if (block_remaining == 0) {
 553                                 if (to)
 554                                         kunmap_atomic(to);
 555
 556                                 block_remaining = DATA_BLOCK_SIZE;
 557                                 dbi = tcmu_cmd_get_dbi(tcmu_cmd);
 558                                 page = tcmu_get_block_page(udev, dbi);
 559                                 to = kmap_atomic(page);
 560                         }
 561
 562                         copy_bytes = min_t(size_t, sg_remaining,
 563                                         block_remaining);
 564                         to_offset = get_block_offset_user(udev, dbi,
 565                                         block_remaining);
 566                         offset = DATA_BLOCK_SIZE - block_remaining;
 567                         to += offset;
 568
 569                         if (*iov_cnt != 0 &&
 570                             to_offset == iov_tail(*iov)) {
 571                                 (*iov)->iov_len += copy_bytes;
 572                         } else {
 573                                 new_iov(iov, iov_cnt, udev);
 574                                 (*iov)->iov_base = (void __user *)to_offset;
 575                                 (*iov)->iov_len = copy_bytes;
 576                         }
 577                         if (copy_data) {
 578                                 memcpy(to, from + sg->length - sg_remaining,
 579                                         copy_bytes);
 580                                 tcmu_flush_dcache_range(to, copy_bytes);
 581                         }
 582                         sg_remaining -= copy_bytes;
 583                         block_remaining -= copy_bytes;
 584                 }
 585                 kunmap_atomic(from - sg->offset);
 586         }
 587         if (to)
 588                 kunmap_atomic(to);
 589
 590         return 0;
 591 }
 592
 593 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 594                              bool bidi)
 595 {
 596         struct se_cmd *se_cmd = cmd->se_cmd;
 597         int i, dbi;
 598         int block_remaining = 0;
 599         void *from = NULL, *to;
 600         size_t copy_bytes, offset;
 601         struct scatterlist *sg, *data_sg;
 602         struct page *page;
 603         unsigned int data_nents;
 604         uint32_t count = 0;
 605
 606         if (!bidi) {
 607                 data_sg = se_cmd->t_data_sg;
 608                 data_nents = se_cmd->t_data_nents;
 609         } else {
 610
 611                 /*
 612                  * For bidi case, the first count blocks are for Data-Out
 613                  * buffer blocks, and before gathering the Data-In buffer
 614                  * the Data-Out buffer blocks should be discarded.
 615                  */
 616                 count = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
 617
 618                 data_sg = se_cmd->t_bidi_data_sg;
 619                 data_nents = se_cmd->t_bidi_data_nents;
 620         }
 621
 622         tcmu_cmd_set_dbi_cur(cmd, count);
 623
 624         for_each_sg(data_sg, sg, data_nents, i) {
 625                 int sg_remaining = sg->length;
 626                 to = kmap_atomic(sg_page(sg)) + sg->offset;
 627                 while (sg_remaining > 0) {
 628                         if (block_remaining == 0) {
 629                                 if (from)
 630                                         kunmap_atomic(from);
 631
 632                                 block_remaining = DATA_BLOCK_SIZE;
 633                                 dbi = tcmu_cmd_get_dbi(cmd);
 634                                 page = tcmu_get_block_page(udev, dbi);
 635                                 from = kmap_atomic(page);
 636                         }
 637                         copy_bytes = min_t(size_t, sg_remaining,
 638                                         block_remaining);
 639                         offset = DATA_BLOCK_SIZE - block_remaining;
 640                         from += offset;
 641                         tcmu_flush_dcache_range(from, copy_bytes);
 642                         memcpy(to + sg->length - sg_remaining, from,
 643                                         copy_bytes);
 644
 645                         sg_remaining -= copy_bytes;
 646                         block_remaining -= copy_bytes;
 647                 }
 648                 kunmap_atomic(to - sg->offset);
 649         }
 650         if (from)
 651                 kunmap_atomic(from);
 652 }
 653
 654 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
 655 {
 656         return DATA_BLOCK_SIZE * (thresh - bitmap_weight(bitmap, thresh));
 657 }
 658
 659 /*
 660  * We can't queue a command until we have space available on the cmd ring *and*
 661  * space available on the data area.
 662  *
 663  * Called with ring lock held.
 664  */
 665 static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 666                 size_t cmd_size, size_t data_needed)
 667 {
 668         struct tcmu_mailbox *mb = udev->mb_addr;
 669         uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
 670                                 / DATA_BLOCK_SIZE;
 671         size_t space, cmd_needed;
 672         u32 cmd_head;
 673
 674         tcmu_flush_dcache_range(mb, sizeof(*mb));
 675
 676         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 677
 678         /*
 679          * If cmd end-of-ring space is too small then we need space for a NOP plus
 680          * original cmd - cmds are internally contiguous.
 681          */
 682         if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
 683                 cmd_needed = cmd_size;
 684         else
 685                 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
 686
 687         space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
 688         if (space < cmd_needed) {
 689                 pr_debug("no cmd space: %u %u %u\n", cmd_head,
 690                        udev->cmdr_last_cleaned, udev->cmdr_size);
 691                 return false;
 692         }
 693
 694         /* try to check and get the data blocks as needed */
 695         space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
 696         if (space < data_needed) {
 697                 unsigned long blocks_left = DATA_BLOCK_BITS - udev->dbi_thresh;
 698                 unsigned long grow;
 699
 700                 if (blocks_left < blocks_needed) {
 701                         pr_debug("no data space: only %lu available, but ask for %zu\n",
 702                                         blocks_left * DATA_BLOCK_SIZE,
 703                                         data_needed);
 704                         return false;
 705                 }
 706
 707                 /* Try to expand the thresh */
 708                 if (!udev->dbi_thresh) {
 709                         /* From idle state */
 710                         uint32_t init_thresh = DATA_BLOCK_INIT_BITS;
 711
 712                         udev->dbi_thresh = max(blocks_needed, init_thresh);
 713                 } else {
 714                         /*
 715                          * Grow the data area by max(blocks needed,
 716                          * dbi_thresh / 2), but limited to the max
 717                          * DATA_BLOCK_BITS size.
 718                          */
 719                         grow = max(blocks_needed, udev->dbi_thresh / 2);
 720                         udev->dbi_thresh += grow;
 721                         if (udev->dbi_thresh > DATA_BLOCK_BITS)
 722                                 udev->dbi_thresh = DATA_BLOCK_BITS;
 723                 }
 724         }
 725
 726         return tcmu_get_empty_blocks(udev, cmd);
 727 }
 728
 729 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
 730 {
 731         return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 732                         sizeof(struct tcmu_cmd_entry));
 733 }
 734
 735 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
 736                                            size_t base_command_size)
 737 {
 738         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 739         size_t command_size;
 740
 741         command_size = base_command_size +
 742                 round_up(scsi_command_size(se_cmd->t_task_cdb),
 743                                 TCMU_OP_ALIGN_SIZE);
 744
 745         WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
 746
 747         return command_size;
 748 }
 749
 750 static sense_reason_t
 751 tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd)
 752 {
 753         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 754         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 755         size_t base_command_size, command_size;
 756         struct tcmu_mailbox *mb;
 757         struct tcmu_cmd_entry *entry;
 758         struct iovec *iov;
 759         int iov_cnt, ret;
 760         uint32_t cmd_head;
 761         uint64_t cdb_off;
 762         bool copy_to_data_area;
 763         size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
 764
 765         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
 766                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 767
 768         /*
 769          * Must be a certain minimum size for response sense info, but
 770          * also may be larger if the iov array is large.
 771          *
 772          * We prepare as many iovs as possbile for potential uses here,
 773          * because it's expensive to tell how many regions are freed in
 774          * the bitmap & global data pool, as the size calculated here
 775          * will only be used to do the checks.
 776          *
 777          * The size will be recalculated later as actually needed to save
 778          * cmd area memories.
 779          */
 780         base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
 781         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 782
 783         mutex_lock(&udev->cmdr_lock);
 784
 785         mb = udev->mb_addr;
 786         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 787         if ((command_size > (udev->cmdr_size / 2)) ||
 788             data_length > udev->data_size) {
 789                 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
 790                         "cmd ring/data area\n", command_size, data_length,
 791                         udev->cmdr_size, udev->data_size);
 792                 mutex_unlock(&udev->cmdr_lock);
 793                 return TCM_INVALID_CDB_FIELD;
 794         }
 795
 796         while (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
 797                 int ret;
 798                 DEFINE_WAIT(__wait);
 799
 800                 prepare_to_wait(&udev->wait_cmdr, &__wait, TASK_INTERRUPTIBLE);
 801
 802                 pr_debug("sleeping for ring space\n");
 803                 mutex_unlock(&udev->cmdr_lock);
 804                 if (udev->cmd_time_out)
 805                         ret = schedule_timeout(
 806                                         msecs_to_jiffies(udev->cmd_time_out));
 807                 else
 808                         ret = schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT));
 809                 finish_wait(&udev->wait_cmdr, &__wait);
 810                 if (!ret) {
 811                         pr_warn("tcmu: command timed out\n");
 812                         return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 813                 }
 814
 815                 mutex_lock(&udev->cmdr_lock);
 816
 817                 /* We dropped cmdr_lock, cmd_head is stale */
 818                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 819         }
 820
 821         /* Insert a PAD if end-of-ring space is too small */
 822         if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
 823                 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
 824
 825                 entry = (void *) mb + CMDR_OFF + cmd_head;
 826                 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
 827                 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
 828                 entry->hdr.cmd_id = 0; /* not used for PAD */
 829                 entry->hdr.kflags = 0;
 830                 entry->hdr.uflags = 0;
 831                 tcmu_flush_dcache_range(entry, sizeof(*entry));
 832
 833                 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
 834                 tcmu_flush_dcache_range(mb, sizeof(*mb));
 835
 836                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 837                 WARN_ON(cmd_head != 0);
 838         }
 839
 840         entry = (void *) mb + CMDR_OFF + cmd_head;
 841         memset(entry, 0, command_size);
 842         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
 843         entry->hdr.cmd_id = tcmu_cmd->cmd_id;
 844
 845         /* Handle allocating space from the data area */
 846         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 847         iov = &entry->req.iov[0];
 848         iov_cnt = 0;
 849         copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
 850                 || se_cmd->se_cmd_flags & SCF_BIDI);
 851         ret = scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
 852                                 se_cmd->t_data_nents, &iov, &iov_cnt,
 853                                 copy_to_data_area);
 854         if (ret) {
 855                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
 856                 mutex_unlock(&udev->cmdr_lock);
 857
 858                 pr_err("tcmu: alloc and scatter data failed\n");
 859                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 860         }
 861         entry->req.iov_cnt = iov_cnt;
 862
 863         /* Handle BIDI commands */
 864         iov_cnt = 0;
 865         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 866                 iov++;
 867                 ret = scatter_data_area(udev, tcmu_cmd,
 868                                         se_cmd->t_bidi_data_sg,
 869                                         se_cmd->t_bidi_data_nents,
 870                                         &iov, &iov_cnt, false);
 871                 if (ret) {
 872                         tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
 873                         mutex_unlock(&udev->cmdr_lock);
 874
 875                         pr_err("tcmu: alloc and scatter bidi data failed\n");
 876                         return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 877                 }
 878         }
 879         entry->req.iov_bidi_cnt = iov_cnt;
 880
 881         /*
 882          * Recalaulate the command's base size and size according
 883          * to the actual needs
 884          */
 885         base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
 886                                                        entry->req.iov_bidi_cnt);
 887         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 888
 889         tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
 890
 891         /* All offsets relative to mb_addr, not start of entry! */
 892         cdb_off = CMDR_OFF + cmd_head + base_command_size;
 893         memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
 894         entry->req.cdb_off = cdb_off;
 895         tcmu_flush_dcache_range(entry, sizeof(*entry));
 896
 897         UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
 898         tcmu_flush_dcache_range(mb, sizeof(*mb));
 899         mutex_unlock(&udev->cmdr_lock);
 900
 901         /* TODO: only if FLUSH and FUA? */
 902         uio_event_notify(&udev->uio_info);
 903
 904         if (udev->cmd_time_out)
 905                 mod_timer(&udev->timeout, round_jiffies_up(jiffies +
 906                           msecs_to_jiffies(udev->cmd_time_out)));
 907
 908         return TCM_NO_SENSE;
 909 }
 910
 911 static sense_reason_t
 912 tcmu_queue_cmd(struct se_cmd *se_cmd)
 913 {
 914         struct se_device *se_dev = se_cmd->se_dev;
 915         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 916         struct tcmu_cmd *tcmu_cmd;
 917         sense_reason_t ret;
 918
 919         tcmu_cmd = tcmu_alloc_cmd(se_cmd);
 920         if (!tcmu_cmd)
 921                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 922
 923         ret = tcmu_queue_cmd_ring(tcmu_cmd);
 924         if (ret != TCM_NO_SENSE) {
 925                 pr_err("TCMU: Could not queue command\n");
 926                 spin_lock_irq(&udev->commands_lock);
 927                 idr_remove(&udev->commands, tcmu_cmd->cmd_id);
 928                 spin_unlock_irq(&udev->commands_lock);
 929
 930                 tcmu_free_cmd(tcmu_cmd);
 931         }
 932
 933         return ret;
 934 }
 935
 936 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
 937 {
 938         struct se_cmd *se_cmd = cmd->se_cmd;
 939         struct tcmu_dev *udev = cmd->tcmu_dev;
 940
 941         /*
 942          * cmd has been completed already from timeout, just reclaim
 943          * data area space and free cmd
 944          */
 945         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
 946                 goto out;
 947
 948         tcmu_cmd_reset_dbi_cur(cmd);
 949
 950         if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
 951                 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
 952                         cmd->se_cmd);
 953                 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
 954         } else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
 955                 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
 956         } else if (se_cmd->se_cmd_flags & SCF_BIDI) {
 957                 /* Get Data-In buffer before clean up */
 958                 gather_data_area(udev, cmd, true);
 959         } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
 960                 gather_data_area(udev, cmd, false);
 961         } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
 962                 /* TODO: */
 963         } else if (se_cmd->data_direction != DMA_NONE) {
 964                 pr_warn("TCMU: data direction was %d!\n",
 965                         se_cmd->data_direction);
 966         }
 967
 968         target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
 969
 970 out:
 971         cmd->se_cmd = NULL;
 972         tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
 973         tcmu_free_cmd(cmd);
 974 }
 975
 976 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
 977 {
 978         struct tcmu_mailbox *mb;
 979         int handled = 0;
 980
 981         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
 982                 pr_err("ring broken, not handling completions\n");
 983                 return 0;
 984         }
 985
 986         mb = udev->mb_addr;
 987         tcmu_flush_dcache_range(mb, sizeof(*mb));
 988
 989         while (udev->cmdr_last_cleaned != ACCESS_ONCE(mb->cmd_tail)) {
 990
 991                 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
 992                 struct tcmu_cmd *cmd;
 993
 994                 tcmu_flush_dcache_range(entry, sizeof(*entry));
 995
 996                 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
 997                         UPDATE_HEAD(udev->cmdr_last_cleaned,
 998                                     tcmu_hdr_get_len(entry->hdr.len_op),
 999                                     udev->cmdr_size);
1000                         continue;
1001                 }
1002                 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1003
1004                 spin_lock(&udev->commands_lock);
1005                 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1006                 spin_unlock(&udev->commands_lock);
1007
1008                 if (!cmd) {
1009                         pr_err("cmd_id not found, ring is broken\n");
1010                         set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1011                         break;
1012                 }
1013
1014                 tcmu_handle_completion(cmd, entry);
1015
1016                 UPDATE_HEAD(udev->cmdr_last_cleaned,
1017                             tcmu_hdr_get_len(entry->hdr.len_op),
1018                             udev->cmdr_size);
1019
1020                 handled++;
1021         }
1022
1023         if (mb->cmd_tail == mb->cmd_head)
1024                 del_timer(&udev->timeout); /* no more pending cmds */
1025
1026         wake_up(&udev->wait_cmdr);
1027
1028         return handled;
1029 }
1030
1031 static int tcmu_check_expired_cmd(int id, void *p, void *data)
1032 {
1033         struct tcmu_cmd *cmd = p;
1034
1035         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1036                 return 0;
1037
1038         if (!time_after(jiffies, cmd->deadline))
1039                 return 0;
1040
1041         set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1042         target_complete_cmd(cmd->se_cmd, SAM_STAT_CHECK_CONDITION);
1043         cmd->se_cmd = NULL;
1044
1045         return 0;
1046 }
1047
1048 static void tcmu_device_timedout(unsigned long data)
1049 {
1050         struct tcmu_dev *udev = (struct tcmu_dev *)data;
1051         unsigned long flags;
1052
1053         spin_lock_irqsave(&udev->commands_lock, flags);
1054         idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
1055         spin_unlock_irqrestore(&udev->commands_lock, flags);
1056
1057         /* Try to wake up the ummap thread */
1058         wake_up(&unmap_wait);
1059
1060         /*
1061          * We don't need to wakeup threads on wait_cmdr since they have their
1062          * own timeout.
1063          */
1064 }
1065
1066 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1067 {
1068         struct tcmu_hba *tcmu_hba;
1069
1070         tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1071         if (!tcmu_hba)
1072                 return -ENOMEM;
1073
1074         tcmu_hba->host_id = host_id;
1075         hba->hba_ptr = tcmu_hba;
1076
1077         return 0;
1078 }
1079
1080 static void tcmu_detach_hba(struct se_hba *hba)
1081 {
1082         kfree(hba->hba_ptr);
1083         hba->hba_ptr = NULL;
1084 }
1085
1086 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1087 {
1088         struct tcmu_dev *udev;
1089
1090         udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1091         if (!udev)
1092                 return NULL;
1093         kref_init(&udev->kref);
1094
1095         udev->name = kstrdup(name, GFP_KERNEL);
1096         if (!udev->name) {
1097                 kfree(udev);
1098                 return NULL;
1099         }
1100
1101         udev->hba = hba;
1102         udev->cmd_time_out = TCMU_TIME_OUT;
1103
1104         init_waitqueue_head(&udev->wait_cmdr);
1105         mutex_init(&udev->cmdr_lock);
1106
1107         idr_init(&udev->commands);
1108         spin_lock_init(&udev->commands_lock);
1109
1110         setup_timer(&udev->timeout, tcmu_device_timedout,
1111                 (unsigned long)udev);
1112
1113         init_waitqueue_head(&udev->nl_cmd_wq);
1114         spin_lock_init(&udev->nl_cmd_lock);
1115
1116         return &udev->se_dev;
1117 }
1118
1119 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1120 {
1121         struct tcmu_dev *tcmu_dev = container_of(info, struct tcmu_dev, uio_info);
1122
1123         mutex_lock(&tcmu_dev->cmdr_lock);
1124         tcmu_handle_completions(tcmu_dev);
1125         mutex_unlock(&tcmu_dev->cmdr_lock);
1126
1127         return 0;
1128 }
1129
1130 /*
1131  * mmap code from uio.c. Copied here because we want to hook mmap()
1132  * and this stuff must come along.
1133  */
1134 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1135 {
1136         struct tcmu_dev *udev = vma->vm_private_data;
1137         struct uio_info *info = &udev->uio_info;
1138
1139         if (vma->vm_pgoff < MAX_UIO_MAPS) {
1140                 if (info->mem[vma->vm_pgoff].size == 0)
1141                         return -1;
1142                 return (int)vma->vm_pgoff;
1143         }
1144         return -1;
1145 }
1146
1147 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1148 {
1149         struct page *page;
1150         int ret;
1151
1152         mutex_lock(&udev->cmdr_lock);
1153         page = tcmu_get_block_page(udev, dbi);
1154         if (likely(page)) {
1155                 mutex_unlock(&udev->cmdr_lock);
1156                 return page;
1157         }
1158
1159         /*
1160          * Normally it shouldn't be here:
1161          * Only when the userspace has touched the blocks which
1162          * are out of the tcmu_cmd's data iov[], and will return
1163          * one zeroed page.
1164          */
1165         pr_warn("Block(%u) out of cmd's iov[] has been touched!\n", dbi);
1166         pr_warn("Mostly it will be a bug of userspace, please have a check!\n");
1167
1168         if (dbi >= udev->dbi_thresh) {
1169                 /* Extern the udev->dbi_thresh to dbi + 1 */
1170                 udev->dbi_thresh = dbi + 1;
1171                 udev->dbi_max = dbi;
1172         }
1173
1174         page = radix_tree_lookup(&udev->data_blocks, dbi);
1175         if (!page) {
1176                 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1177                 if (!page) {
1178                         mutex_unlock(&udev->cmdr_lock);
1179                         return NULL;
1180                 }
1181
1182                 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
1183                 if (ret) {
1184                         mutex_unlock(&udev->cmdr_lock);
1185                         __free_page(page);
1186                         return NULL;
1187                 }
1188
1189                 /*
1190                  * Since this case is rare in page fault routine, here we
1191                  * will allow the global_db_count >= TCMU_GLOBAL_MAX_BLOCKS
1192                  * to reduce possible page fault call trace.
1193                  */
1194                 atomic_inc(&global_db_count);
1195         }
1196         mutex_unlock(&udev->cmdr_lock);
1197
1198         return page;
1199 }
1200
1201 static int tcmu_vma_fault(struct vm_fault *vmf)
1202 {
1203         struct tcmu_dev *udev = vmf->vma->vm_private_data;
1204         struct uio_info *info = &udev->uio_info;
1205         struct page *page;
1206         unsigned long offset;
1207         void *addr;
1208
1209         int mi = tcmu_find_mem_index(vmf->vma);
1210         if (mi < 0)
1211                 return VM_FAULT_SIGBUS;
1212
1213         /*
1214          * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1215          * to use mem[N].
1216          */
1217         offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1218
1219         if (offset < udev->data_off) {
1220                 /* For the vmalloc()ed cmd area pages */
1221                 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1222                 page = vmalloc_to_page(addr);
1223         } else {
1224                 uint32_t dbi;
1225
1226                 /* For the dynamically growing data area pages */
1227                 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1228                 page = tcmu_try_get_block_page(udev, dbi);
1229                 if (!page)
1230                         return VM_FAULT_NOPAGE;
1231         }
1232
1233         get_page(page);
1234         vmf->page = page;
1235         return 0;
1236 }
1237
1238 static const struct vm_operations_struct tcmu_vm_ops = {
1239         .fault = tcmu_vma_fault,
1240 };
1241
1242 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1243 {
1244         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1245
1246         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1247         vma->vm_ops = &tcmu_vm_ops;
1248
1249         vma->vm_private_data = udev;
1250
1251         /* Ensure the mmap is exactly the right size */
1252         if (vma_pages(vma) != (TCMU_RING_SIZE >> PAGE_SHIFT))
1253                 return -EINVAL;
1254
1255         return 0;
1256 }
1257
1258 static int tcmu_open(struct uio_info *info, struct inode *inode)
1259 {
1260         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1261
1262         /* O_EXCL not supported for char devs, so fake it? */
1263         if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1264                 return -EBUSY;
1265
1266         udev->inode = inode;
1267         kref_get(&udev->kref);
1268
1269         pr_debug("open\n");
1270
1271         return 0;
1272 }
1273
1274 static void tcmu_dev_call_rcu(struct rcu_head *p)
1275 {
1276         struct se_device *dev = container_of(p, struct se_device, rcu_head);
1277         struct tcmu_dev *udev = TCMU_DEV(dev);
1278
1279         kfree(udev->uio_info.name);
1280         kfree(udev->name);
1281         kfree(udev);
1282 }
1283
1284 static void tcmu_dev_kref_release(struct kref *kref)
1285 {
1286         struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1287         struct se_device *dev = &udev->se_dev;
1288
1289         call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1290 }
1291
1292 static int tcmu_release(struct uio_info *info, struct inode *inode)
1293 {
1294         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1295
1296         clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1297
1298         pr_debug("close\n");
1299         /* release ref from open */
1300         kref_put(&udev->kref, tcmu_dev_kref_release);
1301         return 0;
1302 }
1303
1304 static void tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1305 {
1306         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1307
1308         if (!tcmu_kern_cmd_reply_supported)
1309                 return;
1310 relock:
1311         spin_lock(&udev->nl_cmd_lock);
1312
1313         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1314                 spin_unlock(&udev->nl_cmd_lock);
1315                 pr_debug("sleeping for open nl cmd\n");
1316                 wait_event(udev->nl_cmd_wq, (nl_cmd->cmd == TCMU_CMD_UNSPEC));
1317                 goto relock;
1318         }
1319
1320         memset(nl_cmd, 0, sizeof(*nl_cmd));
1321         nl_cmd->cmd = cmd;
1322         init_completion(&nl_cmd->complete);
1323
1324         spin_unlock(&udev->nl_cmd_lock);
1325 }
1326
1327 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1328 {
1329         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1330         int ret;
1331         DEFINE_WAIT(__wait);
1332
1333         if (!tcmu_kern_cmd_reply_supported)
1334                 return 0;
1335
1336         pr_debug("sleeping for nl reply\n");
1337         wait_for_completion(&nl_cmd->complete);
1338
1339         spin_lock(&udev->nl_cmd_lock);
1340         nl_cmd->cmd = TCMU_CMD_UNSPEC;
1341         ret = nl_cmd->status;
1342         nl_cmd->status = 0;
1343         spin_unlock(&udev->nl_cmd_lock);
1344
1345         wake_up_all(&udev->nl_cmd_wq);
1346
1347         return ret;;
1348 }
1349
1350 static int tcmu_netlink_event(struct tcmu_dev *udev, enum tcmu_genl_cmd cmd,
1351                               int reconfig_attr, const void *reconfig_data)
1352 {
1353         struct sk_buff *skb;
1354         void *msg_header;
1355         int ret = -ENOMEM;
1356
1357         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1358         if (!skb)
1359                 return ret;
1360
1361         msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1362         if (!msg_header)
1363                 goto free_skb;
1364
1365         ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1366         if (ret < 0)
1367                 goto free_skb;
1368
1369         ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1370         if (ret < 0)
1371                 goto free_skb;
1372
1373         ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1374         if (ret < 0)
1375                 goto free_skb;
1376
1377         if (cmd == TCMU_CMD_RECONFIG_DEVICE) {
1378                 switch (reconfig_attr) {
1379                 case TCMU_ATTR_DEV_CFG:
1380                         ret = nla_put_string(skb, reconfig_attr, reconfig_data);
1381                         break;
1382                 case TCMU_ATTR_DEV_SIZE:
1383                         ret = nla_put_u64_64bit(skb, reconfig_attr,
1384                                                 *((u64 *)reconfig_data),
1385                                                 TCMU_ATTR_PAD);
1386                         break;
1387                 case TCMU_ATTR_WRITECACHE:
1388                         ret = nla_put_u8(skb, reconfig_attr,
1389                                           *((u8 *)reconfig_data));
1390                         break;
1391                 default:
1392                         BUG();
1393                 }
1394
1395                 if (ret < 0)
1396                         goto free_skb;
1397         }
1398
1399         genlmsg_end(skb, msg_header);
1400
1401         tcmu_init_genl_cmd_reply(udev, cmd);
1402
1403         ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1404                                 TCMU_MCGRP_CONFIG, GFP_KERNEL);
1405         /* We don't care if no one is listening */
1406         if (ret == -ESRCH)
1407                 ret = 0;
1408         if (!ret)
1409                 ret = tcmu_wait_genl_cmd_reply(udev);
1410
1411         return ret;
1412 free_skb:
1413         nlmsg_free(skb);
1414         return ret;
1415 }
1416
1417 static int tcmu_update_uio_info(struct tcmu_dev *udev)
1418 {
1419         struct tcmu_hba *hba = udev->hba->hba_ptr;
1420         struct uio_info *info;
1421         size_t size, used;
1422         char *str;
1423
1424         info = &udev->uio_info;
1425         size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
1426                         udev->dev_config);
1427         size += 1; /* for \0 */
1428         str = kmalloc(size, GFP_KERNEL);
1429         if (!str)
1430                 return -ENOMEM;
1431
1432         used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
1433         if (udev->dev_config[0])
1434                 snprintf(str + used, size - used, "/%s", udev->dev_config);
1435
1436         info->name = str;
1437
1438         return 0;
1439 }
1440
1441 static int tcmu_configure_device(struct se_device *dev)
1442 {
1443         struct tcmu_dev *udev = TCMU_DEV(dev);
1444         struct uio_info *info;
1445         struct tcmu_mailbox *mb;
1446         int ret = 0;
1447
1448         ret = tcmu_update_uio_info(udev);
1449         if (ret)
1450                 return ret;
1451
1452         info = &udev->uio_info;
1453
1454         udev->mb_addr = vzalloc(CMDR_SIZE);
1455         if (!udev->mb_addr) {
1456                 ret = -ENOMEM;
1457                 goto err_vzalloc;
1458         }
1459
1460         /* mailbox fits in first part of CMDR space */
1461         udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1462         udev->data_off = CMDR_SIZE;
1463         udev->data_size = DATA_SIZE;
1464         udev->dbi_thresh = 0; /* Default in Idle state */
1465         udev->waiting_global = false;
1466
1467         /* Initialise the mailbox of the ring buffer */
1468         mb = udev->mb_addr;
1469         mb->version = TCMU_MAILBOX_VERSION;
1470         mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
1471         mb->cmdr_off = CMDR_OFF;
1472         mb->cmdr_size = udev->cmdr_size;
1473
1474         WARN_ON(!PAGE_ALIGNED(udev->data_off));
1475         WARN_ON(udev->data_size % PAGE_SIZE);
1476         WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1477
1478         INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1479
1480         info->version = __stringify(TCMU_MAILBOX_VERSION);
1481
1482         info->mem[0].name = "tcm-user command & data buffer";
1483         info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1484         info->mem[0].size = TCMU_RING_SIZE;
1485         info->mem[0].memtype = UIO_MEM_NONE;
1486
1487         info->irqcontrol = tcmu_irqcontrol;
1488         info->irq = UIO_IRQ_CUSTOM;
1489
1490         info->mmap = tcmu_mmap;
1491         info->open = tcmu_open;
1492         info->release = tcmu_release;
1493
1494         ret = uio_register_device(tcmu_root_device, info);
1495         if (ret)
1496                 goto err_register;
1497
1498         /* User can set hw_block_size before enable the device */
1499         if (dev->dev_attrib.hw_block_size == 0)
1500                 dev->dev_attrib.hw_block_size = 512;
1501         /* Other attributes can be configured in userspace */
1502         if (!dev->dev_attrib.hw_max_sectors)
1503                 dev->dev_attrib.hw_max_sectors = 128;
1504         if (!dev->dev_attrib.emulate_write_cache)
1505                 dev->dev_attrib.emulate_write_cache = 0;
1506         dev->dev_attrib.hw_queue_depth = 128;
1507
1508         /*
1509          * Get a ref incase userspace does a close on the uio device before
1510          * LIO has initiated tcmu_free_device.
1511          */
1512         kref_get(&udev->kref);
1513
1514         ret = tcmu_netlink_event(udev, TCMU_CMD_ADDED_DEVICE, 0, NULL);
1515         if (ret)
1516                 goto err_netlink;
1517
1518         mutex_lock(&root_udev_mutex);
1519         list_add(&udev->node, &root_udev);
1520         mutex_unlock(&root_udev_mutex);
1521
1522         return 0;
1523
1524 err_netlink:
1525         kref_put(&udev->kref, tcmu_dev_kref_release);
1526         uio_unregister_device(&udev->uio_info);
1527 err_register:
1528         vfree(udev->mb_addr);
1529 err_vzalloc:
1530         kfree(info->name);
1531         info->name = NULL;
1532
1533         return ret;
1534 }
1535
1536 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1537 {
1538         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1539                 kmem_cache_free(tcmu_cmd_cache, cmd);
1540                 return 0;
1541         }
1542         return -EINVAL;
1543 }
1544
1545 static bool tcmu_dev_configured(struct tcmu_dev *udev)
1546 {
1547         return udev->uio_info.uio_dev ? true : false;
1548 }
1549
1550 static void tcmu_blocks_release(struct tcmu_dev *udev)
1551 {
1552         int i;
1553         struct page *page;
1554
1555         /* Try to release all block pages */
1556         mutex_lock(&udev->cmdr_lock);
1557         for (i = 0; i <= udev->dbi_max; i++) {
1558                 page = radix_tree_delete(&udev->data_blocks, i);
1559                 if (page) {
1560                         __free_page(page);
1561                         atomic_dec(&global_db_count);
1562                 }
1563         }
1564         mutex_unlock(&udev->cmdr_lock);
1565 }
1566
1567 static void tcmu_free_device(struct se_device *dev)
1568 {
1569         struct tcmu_dev *udev = TCMU_DEV(dev);
1570
1571         /* release ref from init */
1572         kref_put(&udev->kref, tcmu_dev_kref_release);
1573 }
1574
1575 static void tcmu_destroy_device(struct se_device *dev)
1576 {
1577         struct tcmu_dev *udev = TCMU_DEV(dev);
1578         struct tcmu_cmd *cmd;
1579         bool all_expired = true;
1580         int i;
1581
1582         del_timer_sync(&udev->timeout);
1583
1584         mutex_lock(&root_udev_mutex);
1585         list_del(&udev->node);
1586         mutex_unlock(&root_udev_mutex);
1587
1588         vfree(udev->mb_addr);
1589
1590         /* Upper layer should drain all requests before calling this */
1591         spin_lock_irq(&udev->commands_lock);
1592         idr_for_each_entry(&udev->commands, cmd, i) {
1593                 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1594                         all_expired = false;
1595         }
1596         idr_destroy(&udev->commands);
1597         spin_unlock_irq(&udev->commands_lock);
1598         WARN_ON(!all_expired);
1599
1600         tcmu_blocks_release(udev);
1601
1602         tcmu_netlink_event(udev, TCMU_CMD_REMOVED_DEVICE, 0, NULL);
1603
1604         uio_unregister_device(&udev->uio_info);
1605
1606         /* release ref from configure */
1607         kref_put(&udev->kref, tcmu_dev_kref_release);
1608 }
1609
1610 enum {
1611         Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
1612         Opt_err,
1613 };
1614
1615 static match_table_t tokens = {
1616         {Opt_dev_config, "dev_config=%s"},
1617         {Opt_dev_size, "dev_size=%u"},
1618         {Opt_hw_block_size, "hw_block_size=%u"},
1619         {Opt_hw_max_sectors, "hw_max_sectors=%u"},
1620         {Opt_err, NULL}
1621 };
1622
1623 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
1624 {
1625         unsigned long tmp_ul;
1626         char *arg_p;
1627         int ret;
1628
1629         arg_p = match_strdup(arg);
1630         if (!arg_p)
1631                 return -ENOMEM;
1632
1633         ret = kstrtoul(arg_p, 0, &tmp_ul);
1634         kfree(arg_p);
1635         if (ret < 0) {
1636                 pr_err("kstrtoul() failed for dev attrib\n");
1637                 return ret;
1638         }
1639         if (!tmp_ul) {
1640                 pr_err("dev attrib must be nonzero\n");
1641                 return -EINVAL;
1642         }
1643         *dev_attrib = tmp_ul;
1644         return 0;
1645 }
1646
1647 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1648                 const char *page, ssize_t count)
1649 {
1650         struct tcmu_dev *udev = TCMU_DEV(dev);
1651         char *orig, *ptr, *opts, *arg_p;
1652         substring_t args[MAX_OPT_ARGS];
1653         int ret = 0, token;
1654
1655         opts = kstrdup(page, GFP_KERNEL);
1656         if (!opts)
1657                 return -ENOMEM;
1658
1659         orig = opts;
1660
1661         while ((ptr = strsep(&opts, ",\n")) != NULL) {
1662                 if (!*ptr)
1663                         continue;
1664
1665                 token = match_token(ptr, tokens, args);
1666                 switch (token) {
1667                 case Opt_dev_config:
1668                         if (match_strlcpy(udev->dev_config, &args[0],
1669                                           TCMU_CONFIG_LEN) == 0) {
1670                                 ret = -EINVAL;
1671                                 break;
1672                         }
1673                         pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
1674                         break;
1675                 case Opt_dev_size:
1676                         arg_p = match_strdup(&args[0]);
1677                         if (!arg_p) {
1678                                 ret = -ENOMEM;
1679                                 break;
1680                         }
1681                         ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
1682                         kfree(arg_p);
1683                         if (ret < 0)
1684                                 pr_err("kstrtoul() failed for dev_size=\n");
1685                         break;
1686                 case Opt_hw_block_size:
1687                         ret = tcmu_set_dev_attrib(&args[0],
1688                                         &(dev->dev_attrib.hw_block_size));
1689                         break;
1690                 case Opt_hw_max_sectors:
1691                         ret = tcmu_set_dev_attrib(&args[0],
1692                                         &(dev->dev_attrib.hw_max_sectors));
1693                         break;
1694                 default:
1695                         break;
1696                 }
1697
1698                 if (ret)
1699                         break;
1700         }
1701
1702         kfree(orig);
1703         return (!ret) ? count : ret;
1704 }
1705
1706 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
1707 {
1708         struct tcmu_dev *udev = TCMU_DEV(dev);
1709         ssize_t bl = 0;
1710
1711         bl = sprintf(b + bl, "Config: %s ",
1712                      udev->dev_config[0] ? udev->dev_config : "NULL");
1713         bl += sprintf(b + bl, "Size: %zu\n", udev->dev_size);
1714
1715         return bl;
1716 }
1717
1718 static sector_t tcmu_get_blocks(struct se_device *dev)
1719 {
1720         struct tcmu_dev *udev = TCMU_DEV(dev);
1721
1722         return div_u64(udev->dev_size - dev->dev_attrib.block_size,
1723                        dev->dev_attrib.block_size);
1724 }
1725
1726 static sense_reason_t
1727 tcmu_parse_cdb(struct se_cmd *cmd)
1728 {
1729         return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
1730 }
1731
1732 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
1733 {
1734         struct se_dev_attrib *da = container_of(to_config_group(item),
1735                                         struct se_dev_attrib, da_group);
1736         struct tcmu_dev *udev = container_of(da->da_dev,
1737                                         struct tcmu_dev, se_dev);
1738
1739         return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
1740 }
1741
1742 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
1743                                        size_t count)
1744 {
1745         struct se_dev_attrib *da = container_of(to_config_group(item),
1746                                         struct se_dev_attrib, da_group);
1747         struct tcmu_dev *udev = container_of(da->da_dev,
1748                                         struct tcmu_dev, se_dev);
1749         u32 val;
1750         int ret;
1751
1752         if (da->da_dev->export_count) {
1753                 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
1754                 return -EINVAL;
1755         }
1756
1757         ret = kstrtou32(page, 0, &val);
1758         if (ret < 0)
1759                 return ret;
1760
1761         udev->cmd_time_out = val * MSEC_PER_SEC;
1762         return count;
1763 }
1764 CONFIGFS_ATTR(tcmu_, cmd_time_out);
1765
1766 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
1767 {
1768         struct se_dev_attrib *da = container_of(to_config_group(item),
1769                                                 struct se_dev_attrib, da_group);
1770         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1771
1772         return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
1773 }
1774
1775 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
1776                                      size_t count)
1777 {
1778         struct se_dev_attrib *da = container_of(to_config_group(item),
1779                                                 struct se_dev_attrib, da_group);
1780         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1781         int ret, len;
1782
1783         len = strlen(page);
1784         if (!len || len > TCMU_CONFIG_LEN - 1)
1785                 return -EINVAL;
1786
1787         /* Check if device has been configured before */
1788         if (tcmu_dev_configured(udev)) {
1789                 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1790                                          TCMU_ATTR_DEV_CFG, page);
1791                 if (ret) {
1792                         pr_err("Unable to reconfigure device\n");
1793                         return ret;
1794                 }
1795                 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1796
1797                 ret = tcmu_update_uio_info(udev);
1798                 if (ret)
1799                         return ret;
1800                 return count;
1801         }
1802         strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1803
1804         return count;
1805 }
1806 CONFIGFS_ATTR(tcmu_, dev_config);
1807
1808 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
1809 {
1810         struct se_dev_attrib *da = container_of(to_config_group(item),
1811                                                 struct se_dev_attrib, da_group);
1812         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1813
1814         return snprintf(page, PAGE_SIZE, "%zu\n", udev->dev_size);
1815 }
1816
1817 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
1818                                    size_t count)
1819 {
1820         struct se_dev_attrib *da = container_of(to_config_group(item),
1821                                                 struct se_dev_attrib, da_group);
1822         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1823         u64 val;
1824         int ret;
1825
1826         ret = kstrtou64(page, 0, &val);
1827         if (ret < 0)
1828                 return ret;
1829
1830         /* Check if device has been configured before */
1831         if (tcmu_dev_configured(udev)) {
1832                 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1833                                          TCMU_ATTR_DEV_SIZE, &val);
1834                 if (ret) {
1835                         pr_err("Unable to reconfigure device\n");
1836                         return ret;
1837                 }
1838         }
1839         udev->dev_size = val;
1840         return count;
1841 }
1842 CONFIGFS_ATTR(tcmu_, dev_size);
1843
1844 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
1845                                              char *page)
1846 {
1847         struct se_dev_attrib *da = container_of(to_config_group(item),
1848                                         struct se_dev_attrib, da_group);
1849
1850         return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
1851 }
1852
1853 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
1854                                               const char *page, size_t count)
1855 {
1856         struct se_dev_attrib *da = container_of(to_config_group(item),
1857                                         struct se_dev_attrib, da_group);
1858         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1859         u8 val;
1860         int ret;
1861
1862         ret = kstrtou8(page, 0, &val);
1863         if (ret < 0)
1864                 return ret;
1865
1866         /* Check if device has been configured before */
1867         if (tcmu_dev_configured(udev)) {
1868                 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1869                                          TCMU_ATTR_WRITECACHE, &val);
1870                 if (ret) {
1871                         pr_err("Unable to reconfigure device\n");
1872                         return ret;
1873                 }
1874         }
1875
1876         da->emulate_write_cache = val;
1877         return count;
1878 }
1879 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
1880
1881 static struct configfs_attribute *tcmu_attrib_attrs[] = {
1882         &tcmu_attr_cmd_time_out,
1883         &tcmu_attr_dev_config,
1884         &tcmu_attr_dev_size,
1885         &tcmu_attr_emulate_write_cache,
1886         NULL,
1887 };
1888
1889 static struct configfs_attribute **tcmu_attrs;
1890
1891 static struct target_backend_ops tcmu_ops = {
1892         .name                   = "user",
1893         .owner                  = THIS_MODULE,
1894         .transport_flags        = TRANSPORT_FLAG_PASSTHROUGH,
1895         .attach_hba             = tcmu_attach_hba,
1896         .detach_hba             = tcmu_detach_hba,
1897         .alloc_device           = tcmu_alloc_device,
1898         .configure_device       = tcmu_configure_device,
1899         .destroy_device         = tcmu_destroy_device,
1900         .free_device            = tcmu_free_device,
1901         .parse_cdb              = tcmu_parse_cdb,
1902         .set_configfs_dev_params = tcmu_set_configfs_dev_params,
1903         .show_configfs_dev_params = tcmu_show_configfs_dev_params,
1904         .get_device_type        = sbc_get_device_type,
1905         .get_blocks             = tcmu_get_blocks,
1906         .tb_dev_attrib_attrs    = NULL,
1907 };
1908
1909 static int unmap_thread_fn(void *data)
1910 {
1911         struct tcmu_dev *udev;
1912         loff_t off;
1913         uint32_t start, end, block;
1914         struct page *page;
1915         int i;
1916
1917         while (!kthread_should_stop()) {
1918                 DEFINE_WAIT(__wait);
1919
1920                 prepare_to_wait(&unmap_wait, &__wait, TASK_INTERRUPTIBLE);
1921                 schedule();
1922                 finish_wait(&unmap_wait, &__wait);
1923
1924                 if (kthread_should_stop())
1925                         break;
1926
1927                 mutex_lock(&root_udev_mutex);
1928                 list_for_each_entry(udev, &root_udev, node) {
1929                         mutex_lock(&udev->cmdr_lock);
1930
1931                         /* Try to complete the finished commands first */
1932                         tcmu_handle_completions(udev);
1933
1934                         /* Skip the udevs waiting the global pool or in idle */
1935                         if (udev->waiting_global || !udev->dbi_thresh) {
1936                                 mutex_unlock(&udev->cmdr_lock);
1937                                 continue;
1938                         }
1939
1940                         end = udev->dbi_max + 1;
1941                         block = find_last_bit(udev->data_bitmap, end);
1942                         if (block == udev->dbi_max) {
1943                                 /*
1944                                  * The last bit is dbi_max, so there is
1945                                  * no need to shrink any blocks.
1946                                  */
1947                                 mutex_unlock(&udev->cmdr_lock);
1948                                 continue;
1949                         } else if (block == end) {
1950                                 /* The current udev will goto idle state */
1951                                 udev->dbi_thresh = start = 0;
1952                                 udev->dbi_max = 0;
1953                         } else {
1954                                 udev->dbi_thresh = start = block + 1;
1955                                 udev->dbi_max = block;
1956                         }
1957
1958                         /* Here will truncate the data area from off */
1959                         off = udev->data_off + start * DATA_BLOCK_SIZE;
1960                         unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
1961
1962                         /* Release the block pages */
1963                         for (i = start; i < end; i++) {
1964                                 page = radix_tree_delete(&udev->data_blocks, i);
1965                                 if (page) {
1966                                         __free_page(page);
1967                                         atomic_dec(&global_db_count);
1968                                 }
1969                         }
1970                         mutex_unlock(&udev->cmdr_lock);
1971                 }
1972
1973                 /*
1974                  * Try to wake up the udevs who are waiting
1975                  * for the global data pool.
1976                  */
1977                 list_for_each_entry(udev, &root_udev, node) {
1978                         if (udev->waiting_global)
1979                                 wake_up(&udev->wait_cmdr);
1980                 }
1981                 mutex_unlock(&root_udev_mutex);
1982         }
1983
1984         return 0;
1985 }
1986
1987 static int __init tcmu_module_init(void)
1988 {
1989         int ret, i, k, len = 0;
1990
1991         BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
1992
1993         tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
1994                                 sizeof(struct tcmu_cmd),
1995                                 __alignof__(struct tcmu_cmd),
1996                                 0, NULL);
1997         if (!tcmu_cmd_cache)
1998                 return -ENOMEM;
1999
2000         tcmu_root_device = root_device_register("tcm_user");
2001         if (IS_ERR(tcmu_root_device)) {
2002                 ret = PTR_ERR(tcmu_root_device);
2003                 goto out_free_cache;
2004         }
2005
2006         ret = genl_register_family(&tcmu_genl_family);
2007         if (ret < 0) {
2008                 goto out_unreg_device;
2009         }
2010
2011         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2012                 len += sizeof(struct configfs_attribute *);
2013         }
2014         for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2015                 len += sizeof(struct configfs_attribute *);
2016         }
2017         len += sizeof(struct configfs_attribute *);
2018
2019         tcmu_attrs = kzalloc(len, GFP_KERNEL);
2020         if (!tcmu_attrs) {
2021                 ret = -ENOMEM;
2022                 goto out_unreg_genl;
2023         }
2024
2025         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2026                 tcmu_attrs[i] = passthrough_attrib_attrs[i];
2027         }
2028         for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2029                 tcmu_attrs[i] = tcmu_attrib_attrs[k];
2030                 i++;
2031         }
2032         tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2033
2034         ret = transport_backend_register(&tcmu_ops);
2035         if (ret)
2036                 goto out_attrs;
2037
2038         init_waitqueue_head(&unmap_wait);
2039         unmap_thread = kthread_run(unmap_thread_fn, NULL, "tcmu_unmap");
2040         if (IS_ERR(unmap_thread)) {
2041                 ret = PTR_ERR(unmap_thread);
2042                 goto out_unreg_transport;
2043         }
2044
2045         return 0;
2046
2047 out_unreg_transport:
2048         target_backend_unregister(&tcmu_ops);
2049 out_attrs:
2050         kfree(tcmu_attrs);
2051 out_unreg_genl:
2052         genl_unregister_family(&tcmu_genl_family);
2053 out_unreg_device:
2054         root_device_unregister(tcmu_root_device);
2055 out_free_cache:
2056         kmem_cache_destroy(tcmu_cmd_cache);
2057
2058         return ret;
2059 }
2060
2061 static void __exit tcmu_module_exit(void)
2062 {
2063         kthread_stop(unmap_thread);
2064         target_backend_unregister(&tcmu_ops);
2065         kfree(tcmu_attrs);
2066         genl_unregister_family(&tcmu_genl_family);
2067         root_device_unregister(tcmu_root_device);
2068         kmem_cache_destroy(tcmu_cmd_cache);
2069 }
2070
2071 MODULE_DESCRIPTION("TCM USER subsystem plugin");
2072 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2073 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2074 MODULE_LICENSE("GPL");
2075
2076 module_init(tcmu_module_init);
2077 module_exit(tcmu_module_exit);