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
 567                         if (*iov_cnt != 0 &&
 568                             to_offset == iov_tail(*iov)) {
 569                                 (*iov)->iov_len += copy_bytes;
 570                         } else {
 571                                 new_iov(iov, iov_cnt, udev);
 572                                 (*iov)->iov_base = (void __user *)to_offset;
 573                                 (*iov)->iov_len = copy_bytes;
 574                         }
 575                         if (copy_data) {
 576                                 offset = DATA_BLOCK_SIZE - block_remaining;
 577                                 memcpy(to + offset,
 578                                        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                         tcmu_flush_dcache_range(from, copy_bytes);
 641                         memcpy(to + sg->length - sg_remaining, from + offset,
 642                                         copy_bytes);
 643
 644                         sg_remaining -= copy_bytes;
 645                         block_remaining -= copy_bytes;
 646                 }
 647                 kunmap_atomic(to - sg->offset);
 648         }
 649         if (from)
 650                 kunmap_atomic(from);
 651 }
 652
 653 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
 654 {
 655         return DATA_BLOCK_SIZE * (thresh - bitmap_weight(bitmap, thresh));
 656 }
 657
 658 /*
 659  * We can't queue a command until we have space available on the cmd ring *and*
 660  * space available on the data area.
 661  *
 662  * Called with ring lock held.
 663  */
 664 static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 665                 size_t cmd_size, size_t data_needed)
 666 {
 667         struct tcmu_mailbox *mb = udev->mb_addr;
 668         uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
 669                                 / DATA_BLOCK_SIZE;
 670         size_t space, cmd_needed;
 671         u32 cmd_head;
 672
 673         tcmu_flush_dcache_range(mb, sizeof(*mb));
 674
 675         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 676
 677         /*
 678          * If cmd end-of-ring space is too small then we need space for a NOP plus
 679          * original cmd - cmds are internally contiguous.
 680          */
 681         if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
 682                 cmd_needed = cmd_size;
 683         else
 684                 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
 685
 686         space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
 687         if (space < cmd_needed) {
 688                 pr_debug("no cmd space: %u %u %u\n", cmd_head,
 689                        udev->cmdr_last_cleaned, udev->cmdr_size);
 690                 return false;
 691         }
 692
 693         /* try to check and get the data blocks as needed */
 694         space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
 695         if (space < data_needed) {
 696                 unsigned long blocks_left = DATA_BLOCK_BITS - udev->dbi_thresh;
 697                 unsigned long grow;
 698
 699                 if (blocks_left < blocks_needed) {
 700                         pr_debug("no data space: only %lu available, but ask for %zu\n",
 701                                         blocks_left * DATA_BLOCK_SIZE,
 702                                         data_needed);
 703                         return false;
 704                 }
 705
 706                 /* Try to expand the thresh */
 707                 if (!udev->dbi_thresh) {
 708                         /* From idle state */
 709                         uint32_t init_thresh = DATA_BLOCK_INIT_BITS;
 710
 711                         udev->dbi_thresh = max(blocks_needed, init_thresh);
 712                 } else {
 713                         /*
 714                          * Grow the data area by max(blocks needed,
 715                          * dbi_thresh / 2), but limited to the max
 716                          * DATA_BLOCK_BITS size.
 717                          */
 718                         grow = max(blocks_needed, udev->dbi_thresh / 2);
 719                         udev->dbi_thresh += grow;
 720                         if (udev->dbi_thresh > DATA_BLOCK_BITS)
 721                                 udev->dbi_thresh = DATA_BLOCK_BITS;
 722                 }
 723         }
 724
 725         return tcmu_get_empty_blocks(udev, cmd);
 726 }
 727
 728 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
 729 {
 730         return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 731                         sizeof(struct tcmu_cmd_entry));
 732 }
 733
 734 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
 735                                            size_t base_command_size)
 736 {
 737         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 738         size_t command_size;
 739
 740         command_size = base_command_size +
 741                 round_up(scsi_command_size(se_cmd->t_task_cdb),
 742                                 TCMU_OP_ALIGN_SIZE);
 743
 744         WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
 745
 746         return command_size;
 747 }
 748
 749 static sense_reason_t
 750 tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd)
 751 {
 752         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 753         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 754         size_t base_command_size, command_size;
 755         struct tcmu_mailbox *mb;
 756         struct tcmu_cmd_entry *entry;
 757         struct iovec *iov;
 758         int iov_cnt, ret;
 759         uint32_t cmd_head;
 760         uint64_t cdb_off;
 761         bool copy_to_data_area;
 762         size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
 763
 764         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
 765                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 766
 767         /*
 768          * Must be a certain minimum size for response sense info, but
 769          * also may be larger if the iov array is large.
 770          *
 771          * We prepare as many iovs as possbile for potential uses here,
 772          * because it's expensive to tell how many regions are freed in
 773          * the bitmap & global data pool, as the size calculated here
 774          * will only be used to do the checks.
 775          *
 776          * The size will be recalculated later as actually needed to save
 777          * cmd area memories.
 778          */
 779         base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
 780         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 781
 782         mutex_lock(&udev->cmdr_lock);
 783
 784         mb = udev->mb_addr;
 785         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 786         if ((command_size > (udev->cmdr_size / 2)) ||
 787             data_length > udev->data_size) {
 788                 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
 789                         "cmd ring/data area\n", command_size, data_length,
 790                         udev->cmdr_size, udev->data_size);
 791                 mutex_unlock(&udev->cmdr_lock);
 792                 return TCM_INVALID_CDB_FIELD;
 793         }
 794
 795         while (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
 796                 int ret;
 797                 DEFINE_WAIT(__wait);
 798
 799                 /*
 800                  * Don't leave commands partially setup because the unmap
 801                  * thread might need the blocks to make forward progress.
 802                  */
 803                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
 804                 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 805
 806                 prepare_to_wait(&udev->wait_cmdr, &__wait, TASK_INTERRUPTIBLE);
 807
 808                 pr_debug("sleeping for ring space\n");
 809                 mutex_unlock(&udev->cmdr_lock);
 810                 if (udev->cmd_time_out)
 811                         ret = schedule_timeout(
 812                                         msecs_to_jiffies(udev->cmd_time_out));
 813                 else
 814                         ret = schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT));
 815                 finish_wait(&udev->wait_cmdr, &__wait);
 816                 if (!ret) {
 817                         pr_warn("tcmu: command timed out\n");
 818                         return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 819                 }
 820
 821                 mutex_lock(&udev->cmdr_lock);
 822
 823                 /* We dropped cmdr_lock, cmd_head is stale */
 824                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 825         }
 826
 827         /* Insert a PAD if end-of-ring space is too small */
 828         if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
 829                 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
 830
 831                 entry = (void *) mb + CMDR_OFF + cmd_head;
 832                 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
 833                 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
 834                 entry->hdr.cmd_id = 0; /* not used for PAD */
 835                 entry->hdr.kflags = 0;
 836                 entry->hdr.uflags = 0;
 837                 tcmu_flush_dcache_range(entry, sizeof(*entry));
 838
 839                 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
 840                 tcmu_flush_dcache_range(mb, sizeof(*mb));
 841
 842                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 843                 WARN_ON(cmd_head != 0);
 844         }
 845
 846         entry = (void *) mb + CMDR_OFF + cmd_head;
 847         memset(entry, 0, command_size);
 848         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
 849         entry->hdr.cmd_id = tcmu_cmd->cmd_id;
 850
 851         /* Handle allocating space from the data area */
 852         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
 853         iov = &entry->req.iov[0];
 854         iov_cnt = 0;
 855         copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
 856                 || se_cmd->se_cmd_flags & SCF_BIDI);
 857         ret = scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
 858                                 se_cmd->t_data_nents, &iov, &iov_cnt,
 859                                 copy_to_data_area);
 860         if (ret) {
 861                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
 862                 mutex_unlock(&udev->cmdr_lock);
 863
 864                 pr_err("tcmu: alloc and scatter data failed\n");
 865                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 866         }
 867         entry->req.iov_cnt = iov_cnt;
 868
 869         /* Handle BIDI commands */
 870         iov_cnt = 0;
 871         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 872                 iov++;
 873                 ret = scatter_data_area(udev, tcmu_cmd,
 874                                         se_cmd->t_bidi_data_sg,
 875                                         se_cmd->t_bidi_data_nents,
 876                                         &iov, &iov_cnt, false);
 877                 if (ret) {
 878                         tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
 879                         mutex_unlock(&udev->cmdr_lock);
 880
 881                         pr_err("tcmu: alloc and scatter bidi data failed\n");
 882                         return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 883                 }
 884         }
 885         entry->req.iov_bidi_cnt = iov_cnt;
 886
 887         /*
 888          * Recalaulate the command's base size and size according
 889          * to the actual needs
 890          */
 891         base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
 892                                                        entry->req.iov_bidi_cnt);
 893         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 894
 895         tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
 896
 897         /* All offsets relative to mb_addr, not start of entry! */
 898         cdb_off = CMDR_OFF + cmd_head + base_command_size;
 899         memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
 900         entry->req.cdb_off = cdb_off;
 901         tcmu_flush_dcache_range(entry, sizeof(*entry));
 902
 903         UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
 904         tcmu_flush_dcache_range(mb, sizeof(*mb));
 905         mutex_unlock(&udev->cmdr_lock);
 906
 907         /* TODO: only if FLUSH and FUA? */
 908         uio_event_notify(&udev->uio_info);
 909
 910         if (udev->cmd_time_out)
 911                 mod_timer(&udev->timeout, round_jiffies_up(jiffies +
 912                           msecs_to_jiffies(udev->cmd_time_out)));
 913
 914         return TCM_NO_SENSE;
 915 }
 916
 917 static sense_reason_t
 918 tcmu_queue_cmd(struct se_cmd *se_cmd)
 919 {
 920         struct se_device *se_dev = se_cmd->se_dev;
 921         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 922         struct tcmu_cmd *tcmu_cmd;
 923         sense_reason_t ret;
 924
 925         tcmu_cmd = tcmu_alloc_cmd(se_cmd);
 926         if (!tcmu_cmd)
 927                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
 928
 929         ret = tcmu_queue_cmd_ring(tcmu_cmd);
 930         if (ret != TCM_NO_SENSE) {
 931                 pr_err("TCMU: Could not queue command\n");
 932                 spin_lock_irq(&udev->commands_lock);
 933                 idr_remove(&udev->commands, tcmu_cmd->cmd_id);
 934                 spin_unlock_irq(&udev->commands_lock);
 935
 936                 tcmu_free_cmd(tcmu_cmd);
 937         }
 938
 939         return ret;
 940 }
 941
 942 static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
 943 {
 944         struct se_cmd *se_cmd = cmd->se_cmd;
 945         struct tcmu_dev *udev = cmd->tcmu_dev;
 946
 947         /*
 948          * cmd has been completed already from timeout, just reclaim
 949          * data area space and free cmd
 950          */
 951         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
 952                 goto out;
 953
 954         tcmu_cmd_reset_dbi_cur(cmd);
 955
 956         if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
 957                 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
 958                         cmd->se_cmd);
 959                 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
 960         } else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
 961                 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
 962         } else if (se_cmd->se_cmd_flags & SCF_BIDI) {
 963                 /* Get Data-In buffer before clean up */
 964                 gather_data_area(udev, cmd, true);
 965         } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
 966                 gather_data_area(udev, cmd, false);
 967         } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
 968                 /* TODO: */
 969         } else if (se_cmd->data_direction != DMA_NONE) {
 970                 pr_warn("TCMU: data direction was %d!\n",
 971                         se_cmd->data_direction);
 972         }
 973
 974         target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
 975
 976 out:
 977         cmd->se_cmd = NULL;
 978         tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
 979         tcmu_free_cmd(cmd);
 980 }
 981
 982 static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
 983 {
 984         struct tcmu_mailbox *mb;
 985         int handled = 0;
 986
 987         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
 988                 pr_err("ring broken, not handling completions\n");
 989                 return 0;
 990         }
 991
 992         mb = udev->mb_addr;
 993         tcmu_flush_dcache_range(mb, sizeof(*mb));
 994
 995         while (udev->cmdr_last_cleaned != ACCESS_ONCE(mb->cmd_tail)) {
 996
 997                 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
 998                 struct tcmu_cmd *cmd;
 999
1000                 /*
1001                  * Flush max. up to end of cmd ring since current entry might
1002                  * be a padding that is shorter than sizeof(*entry)
1003                  */
1004                 size_t ring_left = head_to_end(udev->cmdr_last_cleaned,
1005                                                udev->cmdr_size);
1006                 tcmu_flush_dcache_range(entry, ring_left < sizeof(*entry) ?
1007                                         ring_left : sizeof(*entry));
1008
1009                 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
1010                         UPDATE_HEAD(udev->cmdr_last_cleaned,
1011                                     tcmu_hdr_get_len(entry->hdr.len_op),
1012                                     udev->cmdr_size);
1013                         continue;
1014                 }
1015                 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1016
1017                 spin_lock(&udev->commands_lock);
1018                 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1019                 spin_unlock(&udev->commands_lock);
1020
1021                 if (!cmd) {
1022                         pr_err("cmd_id not found, ring is broken\n");
1023                         set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1024                         break;
1025                 }
1026
1027                 tcmu_handle_completion(cmd, entry);
1028
1029                 UPDATE_HEAD(udev->cmdr_last_cleaned,
1030                             tcmu_hdr_get_len(entry->hdr.len_op),
1031                             udev->cmdr_size);
1032
1033                 handled++;
1034         }
1035
1036         if (mb->cmd_tail == mb->cmd_head)
1037                 del_timer(&udev->timeout); /* no more pending cmds */
1038
1039         wake_up(&udev->wait_cmdr);
1040
1041         return handled;
1042 }
1043
1044 static int tcmu_check_expired_cmd(int id, void *p, void *data)
1045 {
1046         struct tcmu_cmd *cmd = p;
1047
1048         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1049                 return 0;
1050
1051         if (!time_after(jiffies, cmd->deadline))
1052                 return 0;
1053
1054         set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1055         target_complete_cmd(cmd->se_cmd, SAM_STAT_CHECK_CONDITION);
1056         cmd->se_cmd = NULL;
1057
1058         return 0;
1059 }
1060
1061 static void tcmu_device_timedout(unsigned long data)
1062 {
1063         struct tcmu_dev *udev = (struct tcmu_dev *)data;
1064         unsigned long flags;
1065
1066         spin_lock_irqsave(&udev->commands_lock, flags);
1067         idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
1068         spin_unlock_irqrestore(&udev->commands_lock, flags);
1069
1070         /* Try to wake up the ummap thread */
1071         wake_up(&unmap_wait);
1072
1073         /*
1074          * We don't need to wakeup threads on wait_cmdr since they have their
1075          * own timeout.
1076          */
1077 }
1078
1079 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1080 {
1081         struct tcmu_hba *tcmu_hba;
1082
1083         tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1084         if (!tcmu_hba)
1085                 return -ENOMEM;
1086
1087         tcmu_hba->host_id = host_id;
1088         hba->hba_ptr = tcmu_hba;
1089
1090         return 0;
1091 }
1092
1093 static void tcmu_detach_hba(struct se_hba *hba)
1094 {
1095         kfree(hba->hba_ptr);
1096         hba->hba_ptr = NULL;
1097 }
1098
1099 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1100 {
1101         struct tcmu_dev *udev;
1102
1103         udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1104         if (!udev)
1105                 return NULL;
1106         kref_init(&udev->kref);
1107
1108         udev->name = kstrdup(name, GFP_KERNEL);
1109         if (!udev->name) {
1110                 kfree(udev);
1111                 return NULL;
1112         }
1113
1114         udev->hba = hba;
1115         udev->cmd_time_out = TCMU_TIME_OUT;
1116
1117         init_waitqueue_head(&udev->wait_cmdr);
1118         mutex_init(&udev->cmdr_lock);
1119
1120         idr_init(&udev->commands);
1121         spin_lock_init(&udev->commands_lock);
1122
1123         setup_timer(&udev->timeout, tcmu_device_timedout,
1124                 (unsigned long)udev);
1125
1126         init_waitqueue_head(&udev->nl_cmd_wq);
1127         spin_lock_init(&udev->nl_cmd_lock);
1128
1129         return &udev->se_dev;
1130 }
1131
1132 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1133 {
1134         struct tcmu_dev *tcmu_dev = container_of(info, struct tcmu_dev, uio_info);
1135
1136         mutex_lock(&tcmu_dev->cmdr_lock);
1137         tcmu_handle_completions(tcmu_dev);
1138         mutex_unlock(&tcmu_dev->cmdr_lock);
1139
1140         return 0;
1141 }
1142
1143 /*
1144  * mmap code from uio.c. Copied here because we want to hook mmap()
1145  * and this stuff must come along.
1146  */
1147 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1148 {
1149         struct tcmu_dev *udev = vma->vm_private_data;
1150         struct uio_info *info = &udev->uio_info;
1151
1152         if (vma->vm_pgoff < MAX_UIO_MAPS) {
1153                 if (info->mem[vma->vm_pgoff].size == 0)
1154                         return -1;
1155                 return (int)vma->vm_pgoff;
1156         }
1157         return -1;
1158 }
1159
1160 static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1161 {
1162         struct page *page;
1163         int ret;
1164
1165         mutex_lock(&udev->cmdr_lock);
1166         page = tcmu_get_block_page(udev, dbi);
1167         if (likely(page)) {
1168                 mutex_unlock(&udev->cmdr_lock);
1169                 return page;
1170         }
1171
1172         /*
1173          * Normally it shouldn't be here:
1174          * Only when the userspace has touched the blocks which
1175          * are out of the tcmu_cmd's data iov[], and will return
1176          * one zeroed page.
1177          */
1178         pr_warn("Block(%u) out of cmd's iov[] has been touched!\n", dbi);
1179         pr_warn("Mostly it will be a bug of userspace, please have a check!\n");
1180
1181         if (dbi >= udev->dbi_thresh) {
1182                 /* Extern the udev->dbi_thresh to dbi + 1 */
1183                 udev->dbi_thresh = dbi + 1;
1184                 udev->dbi_max = dbi;
1185         }
1186
1187         page = radix_tree_lookup(&udev->data_blocks, dbi);
1188         if (!page) {
1189                 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1190                 if (!page) {
1191                         mutex_unlock(&udev->cmdr_lock);
1192                         return NULL;
1193                 }
1194
1195                 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
1196                 if (ret) {
1197                         mutex_unlock(&udev->cmdr_lock);
1198                         __free_page(page);
1199                         return NULL;
1200                 }
1201
1202                 /*
1203                  * Since this case is rare in page fault routine, here we
1204                  * will allow the global_db_count >= TCMU_GLOBAL_MAX_BLOCKS
1205                  * to reduce possible page fault call trace.
1206                  */
1207                 atomic_inc(&global_db_count);
1208         }
1209         mutex_unlock(&udev->cmdr_lock);
1210
1211         return page;
1212 }
1213
1214 static int tcmu_vma_fault(struct vm_fault *vmf)
1215 {
1216         struct tcmu_dev *udev = vmf->vma->vm_private_data;
1217         struct uio_info *info = &udev->uio_info;
1218         struct page *page;
1219         unsigned long offset;
1220         void *addr;
1221
1222         int mi = tcmu_find_mem_index(vmf->vma);
1223         if (mi < 0)
1224                 return VM_FAULT_SIGBUS;
1225
1226         /*
1227          * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1228          * to use mem[N].
1229          */
1230         offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1231
1232         if (offset < udev->data_off) {
1233                 /* For the vmalloc()ed cmd area pages */
1234                 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1235                 page = vmalloc_to_page(addr);
1236         } else {
1237                 uint32_t dbi;
1238
1239                 /* For the dynamically growing data area pages */
1240                 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1241                 page = tcmu_try_get_block_page(udev, dbi);
1242                 if (!page)
1243                         return VM_FAULT_NOPAGE;
1244         }
1245
1246         get_page(page);
1247         vmf->page = page;
1248         return 0;
1249 }
1250
1251 static const struct vm_operations_struct tcmu_vm_ops = {
1252         .fault = tcmu_vma_fault,
1253 };
1254
1255 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1256 {
1257         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1258
1259         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1260         vma->vm_ops = &tcmu_vm_ops;
1261
1262         vma->vm_private_data = udev;
1263
1264         /* Ensure the mmap is exactly the right size */
1265         if (vma_pages(vma) != (TCMU_RING_SIZE >> PAGE_SHIFT))
1266                 return -EINVAL;
1267
1268         return 0;
1269 }
1270
1271 static int tcmu_open(struct uio_info *info, struct inode *inode)
1272 {
1273         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1274
1275         /* O_EXCL not supported for char devs, so fake it? */
1276         if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1277                 return -EBUSY;
1278
1279         udev->inode = inode;
1280         kref_get(&udev->kref);
1281
1282         pr_debug("open\n");
1283
1284         return 0;
1285 }
1286
1287 static void tcmu_dev_call_rcu(struct rcu_head *p)
1288 {
1289         struct se_device *dev = container_of(p, struct se_device, rcu_head);
1290         struct tcmu_dev *udev = TCMU_DEV(dev);
1291
1292         kfree(udev->uio_info.name);
1293         kfree(udev->name);
1294         kfree(udev);
1295 }
1296
1297 static void tcmu_dev_kref_release(struct kref *kref)
1298 {
1299         struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1300         struct se_device *dev = &udev->se_dev;
1301
1302         call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1303 }
1304
1305 static int tcmu_release(struct uio_info *info, struct inode *inode)
1306 {
1307         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1308
1309         clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1310
1311         pr_debug("close\n");
1312         /* release ref from open */
1313         kref_put(&udev->kref, tcmu_dev_kref_release);
1314         return 0;
1315 }
1316
1317 static void tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1318 {
1319         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1320
1321         if (!tcmu_kern_cmd_reply_supported)
1322                 return;
1323 relock:
1324         spin_lock(&udev->nl_cmd_lock);
1325
1326         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1327                 spin_unlock(&udev->nl_cmd_lock);
1328                 pr_debug("sleeping for open nl cmd\n");
1329                 wait_event(udev->nl_cmd_wq, (nl_cmd->cmd == TCMU_CMD_UNSPEC));
1330                 goto relock;
1331         }
1332
1333         memset(nl_cmd, 0, sizeof(*nl_cmd));
1334         nl_cmd->cmd = cmd;
1335         init_completion(&nl_cmd->complete);
1336
1337         spin_unlock(&udev->nl_cmd_lock);
1338 }
1339
1340 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1341 {
1342         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1343         int ret;
1344         DEFINE_WAIT(__wait);
1345
1346         if (!tcmu_kern_cmd_reply_supported)
1347                 return 0;
1348
1349         pr_debug("sleeping for nl reply\n");
1350         wait_for_completion(&nl_cmd->complete);
1351
1352         spin_lock(&udev->nl_cmd_lock);
1353         nl_cmd->cmd = TCMU_CMD_UNSPEC;
1354         ret = nl_cmd->status;
1355         nl_cmd->status = 0;
1356         spin_unlock(&udev->nl_cmd_lock);
1357
1358         wake_up_all(&udev->nl_cmd_wq);
1359
1360         return ret;;
1361 }
1362
1363 static int tcmu_netlink_event(struct tcmu_dev *udev, enum tcmu_genl_cmd cmd,
1364                               int reconfig_attr, const void *reconfig_data)
1365 {
1366         struct sk_buff *skb;
1367         void *msg_header;
1368         int ret = -ENOMEM;
1369
1370         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1371         if (!skb)
1372                 return ret;
1373
1374         msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1375         if (!msg_header)
1376                 goto free_skb;
1377
1378         ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1379         if (ret < 0)
1380                 goto free_skb;
1381
1382         ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1383         if (ret < 0)
1384                 goto free_skb;
1385
1386         ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1387         if (ret < 0)
1388                 goto free_skb;
1389
1390         if (cmd == TCMU_CMD_RECONFIG_DEVICE) {
1391                 switch (reconfig_attr) {
1392                 case TCMU_ATTR_DEV_CFG:
1393                         ret = nla_put_string(skb, reconfig_attr, reconfig_data);
1394                         break;
1395                 case TCMU_ATTR_DEV_SIZE:
1396                         ret = nla_put_u64_64bit(skb, reconfig_attr,
1397                                                 *((u64 *)reconfig_data),
1398                                                 TCMU_ATTR_PAD);
1399                         break;
1400                 case TCMU_ATTR_WRITECACHE:
1401                         ret = nla_put_u8(skb, reconfig_attr,
1402                                           *((u8 *)reconfig_data));
1403                         break;
1404                 default:
1405                         BUG();
1406                 }
1407
1408                 if (ret < 0)
1409                         goto free_skb;
1410         }
1411
1412         genlmsg_end(skb, msg_header);
1413
1414         tcmu_init_genl_cmd_reply(udev, cmd);
1415
1416         ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1417                                 TCMU_MCGRP_CONFIG, GFP_KERNEL);
1418         /* We don't care if no one is listening */
1419         if (ret == -ESRCH)
1420                 ret = 0;
1421         if (!ret)
1422                 ret = tcmu_wait_genl_cmd_reply(udev);
1423
1424         return ret;
1425 free_skb:
1426         nlmsg_free(skb);
1427         return ret;
1428 }
1429
1430 static int tcmu_update_uio_info(struct tcmu_dev *udev)
1431 {
1432         struct tcmu_hba *hba = udev->hba->hba_ptr;
1433         struct uio_info *info;
1434         size_t size, used;
1435         char *str;
1436
1437         info = &udev->uio_info;
1438         size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
1439                         udev->dev_config);
1440         size += 1; /* for \0 */
1441         str = kmalloc(size, GFP_KERNEL);
1442         if (!str)
1443                 return -ENOMEM;
1444
1445         used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
1446         if (udev->dev_config[0])
1447                 snprintf(str + used, size - used, "/%s", udev->dev_config);
1448
1449         /* If the old string exists, free it */
1450         kfree(info->name);
1451         info->name = str;
1452
1453         return 0;
1454 }
1455
1456 static int tcmu_configure_device(struct se_device *dev)
1457 {
1458         struct tcmu_dev *udev = TCMU_DEV(dev);
1459         struct uio_info *info;
1460         struct tcmu_mailbox *mb;
1461         int ret = 0;
1462
1463         ret = tcmu_update_uio_info(udev);
1464         if (ret)
1465                 return ret;
1466
1467         info = &udev->uio_info;
1468
1469         udev->mb_addr = vzalloc(CMDR_SIZE);
1470         if (!udev->mb_addr) {
1471                 ret = -ENOMEM;
1472                 goto err_vzalloc;
1473         }
1474
1475         /* mailbox fits in first part of CMDR space */
1476         udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1477         udev->data_off = CMDR_SIZE;
1478         udev->data_size = DATA_SIZE;
1479         udev->dbi_thresh = 0; /* Default in Idle state */
1480         udev->waiting_global = false;
1481
1482         /* Initialise the mailbox of the ring buffer */
1483         mb = udev->mb_addr;
1484         mb->version = TCMU_MAILBOX_VERSION;
1485         mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
1486         mb->cmdr_off = CMDR_OFF;
1487         mb->cmdr_size = udev->cmdr_size;
1488
1489         WARN_ON(!PAGE_ALIGNED(udev->data_off));
1490         WARN_ON(udev->data_size % PAGE_SIZE);
1491         WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1492
1493         INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1494
1495         info->version = __stringify(TCMU_MAILBOX_VERSION);
1496
1497         info->mem[0].name = "tcm-user command & data buffer";
1498         info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1499         info->mem[0].size = TCMU_RING_SIZE;
1500         info->mem[0].memtype = UIO_MEM_NONE;
1501
1502         info->irqcontrol = tcmu_irqcontrol;
1503         info->irq = UIO_IRQ_CUSTOM;
1504
1505         info->mmap = tcmu_mmap;
1506         info->open = tcmu_open;
1507         info->release = tcmu_release;
1508
1509         ret = uio_register_device(tcmu_root_device, info);
1510         if (ret)
1511                 goto err_register;
1512
1513         /* User can set hw_block_size before enable the device */
1514         if (dev->dev_attrib.hw_block_size == 0)
1515                 dev->dev_attrib.hw_block_size = 512;
1516         /* Other attributes can be configured in userspace */
1517         if (!dev->dev_attrib.hw_max_sectors)
1518                 dev->dev_attrib.hw_max_sectors = 128;
1519         if (!dev->dev_attrib.emulate_write_cache)
1520                 dev->dev_attrib.emulate_write_cache = 0;
1521         dev->dev_attrib.hw_queue_depth = 128;
1522
1523         /*
1524          * Get a ref incase userspace does a close on the uio device before
1525          * LIO has initiated tcmu_free_device.
1526          */
1527         kref_get(&udev->kref);
1528
1529         ret = tcmu_netlink_event(udev, TCMU_CMD_ADDED_DEVICE, 0, NULL);
1530         if (ret)
1531                 goto err_netlink;
1532
1533         mutex_lock(&root_udev_mutex);
1534         list_add(&udev->node, &root_udev);
1535         mutex_unlock(&root_udev_mutex);
1536
1537         return 0;
1538
1539 err_netlink:
1540         kref_put(&udev->kref, tcmu_dev_kref_release);
1541         uio_unregister_device(&udev->uio_info);
1542 err_register:
1543         vfree(udev->mb_addr);
1544 err_vzalloc:
1545         kfree(info->name);
1546         info->name = NULL;
1547
1548         return ret;
1549 }
1550
1551 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1552 {
1553         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1554                 kmem_cache_free(tcmu_cmd_cache, cmd);
1555                 return 0;
1556         }
1557         return -EINVAL;
1558 }
1559
1560 static bool tcmu_dev_configured(struct tcmu_dev *udev)
1561 {
1562         return udev->uio_info.uio_dev ? true : false;
1563 }
1564
1565 static void tcmu_blocks_release(struct tcmu_dev *udev)
1566 {
1567         int i;
1568         struct page *page;
1569
1570         /* Try to release all block pages */
1571         mutex_lock(&udev->cmdr_lock);
1572         for (i = 0; i <= udev->dbi_max; i++) {
1573                 page = radix_tree_delete(&udev->data_blocks, i);
1574                 if (page) {
1575                         __free_page(page);
1576                         atomic_dec(&global_db_count);
1577                 }
1578         }
1579         mutex_unlock(&udev->cmdr_lock);
1580 }
1581
1582 static void tcmu_free_device(struct se_device *dev)
1583 {
1584         struct tcmu_dev *udev = TCMU_DEV(dev);
1585
1586         /* release ref from init */
1587         kref_put(&udev->kref, tcmu_dev_kref_release);
1588 }
1589
1590 static void tcmu_destroy_device(struct se_device *dev)
1591 {
1592         struct tcmu_dev *udev = TCMU_DEV(dev);
1593         struct tcmu_cmd *cmd;
1594         bool all_expired = true;
1595         int i;
1596
1597         del_timer_sync(&udev->timeout);
1598
1599         mutex_lock(&root_udev_mutex);
1600         list_del(&udev->node);
1601         mutex_unlock(&root_udev_mutex);
1602
1603         vfree(udev->mb_addr);
1604
1605         /* Upper layer should drain all requests before calling this */
1606         spin_lock_irq(&udev->commands_lock);
1607         idr_for_each_entry(&udev->commands, cmd, i) {
1608                 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1609                         all_expired = false;
1610         }
1611         idr_destroy(&udev->commands);
1612         spin_unlock_irq(&udev->commands_lock);
1613         WARN_ON(!all_expired);
1614
1615         tcmu_blocks_release(udev);
1616
1617         tcmu_netlink_event(udev, TCMU_CMD_REMOVED_DEVICE, 0, NULL);
1618
1619         uio_unregister_device(&udev->uio_info);
1620
1621         /* release ref from configure */
1622         kref_put(&udev->kref, tcmu_dev_kref_release);
1623 }
1624
1625 enum {
1626         Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
1627         Opt_err,
1628 };
1629
1630 static match_table_t tokens = {
1631         {Opt_dev_config, "dev_config=%s"},
1632         {Opt_dev_size, "dev_size=%u"},
1633         {Opt_hw_block_size, "hw_block_size=%u"},
1634         {Opt_hw_max_sectors, "hw_max_sectors=%u"},
1635         {Opt_err, NULL}
1636 };
1637
1638 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
1639 {
1640         unsigned long tmp_ul;
1641         char *arg_p;
1642         int ret;
1643
1644         arg_p = match_strdup(arg);
1645         if (!arg_p)
1646                 return -ENOMEM;
1647
1648         ret = kstrtoul(arg_p, 0, &tmp_ul);
1649         kfree(arg_p);
1650         if (ret < 0) {
1651                 pr_err("kstrtoul() failed for dev attrib\n");
1652                 return ret;
1653         }
1654         if (!tmp_ul) {
1655                 pr_err("dev attrib must be nonzero\n");
1656                 return -EINVAL;
1657         }
1658         *dev_attrib = tmp_ul;
1659         return 0;
1660 }
1661
1662 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1663                 const char *page, ssize_t count)
1664 {
1665         struct tcmu_dev *udev = TCMU_DEV(dev);
1666         char *orig, *ptr, *opts, *arg_p;
1667         substring_t args[MAX_OPT_ARGS];
1668         int ret = 0, token;
1669
1670         opts = kstrdup(page, GFP_KERNEL);
1671         if (!opts)
1672                 return -ENOMEM;
1673
1674         orig = opts;
1675
1676         while ((ptr = strsep(&opts, ",\n")) != NULL) {
1677                 if (!*ptr)
1678                         continue;
1679
1680                 token = match_token(ptr, tokens, args);
1681                 switch (token) {
1682                 case Opt_dev_config:
1683                         if (match_strlcpy(udev->dev_config, &args[0],
1684                                           TCMU_CONFIG_LEN) == 0) {
1685                                 ret = -EINVAL;
1686                                 break;
1687                         }
1688                         pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
1689                         break;
1690                 case Opt_dev_size:
1691                         arg_p = match_strdup(&args[0]);
1692                         if (!arg_p) {
1693                                 ret = -ENOMEM;
1694                                 break;
1695                         }
1696                         ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
1697                         kfree(arg_p);
1698                         if (ret < 0)
1699                                 pr_err("kstrtoul() failed for dev_size=\n");
1700                         break;
1701                 case Opt_hw_block_size:
1702                         ret = tcmu_set_dev_attrib(&args[0],
1703                                         &(dev->dev_attrib.hw_block_size));
1704                         break;
1705                 case Opt_hw_max_sectors:
1706                         ret = tcmu_set_dev_attrib(&args[0],
1707                                         &(dev->dev_attrib.hw_max_sectors));
1708                         break;
1709                 default:
1710                         break;
1711                 }
1712
1713                 if (ret)
1714                         break;
1715         }
1716
1717         kfree(orig);
1718         return (!ret) ? count : ret;
1719 }
1720
1721 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
1722 {
1723         struct tcmu_dev *udev = TCMU_DEV(dev);
1724         ssize_t bl = 0;
1725
1726         bl = sprintf(b + bl, "Config: %s ",
1727                      udev->dev_config[0] ? udev->dev_config : "NULL");
1728         bl += sprintf(b + bl, "Size: %zu\n", udev->dev_size);
1729
1730         return bl;
1731 }
1732
1733 static sector_t tcmu_get_blocks(struct se_device *dev)
1734 {
1735         struct tcmu_dev *udev = TCMU_DEV(dev);
1736
1737         return div_u64(udev->dev_size - dev->dev_attrib.block_size,
1738                        dev->dev_attrib.block_size);
1739 }
1740
1741 static sense_reason_t
1742 tcmu_parse_cdb(struct se_cmd *cmd)
1743 {
1744         return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
1745 }
1746
1747 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
1748 {
1749         struct se_dev_attrib *da = container_of(to_config_group(item),
1750                                         struct se_dev_attrib, da_group);
1751         struct tcmu_dev *udev = container_of(da->da_dev,
1752                                         struct tcmu_dev, se_dev);
1753
1754         return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
1755 }
1756
1757 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
1758                                        size_t count)
1759 {
1760         struct se_dev_attrib *da = container_of(to_config_group(item),
1761                                         struct se_dev_attrib, da_group);
1762         struct tcmu_dev *udev = container_of(da->da_dev,
1763                                         struct tcmu_dev, se_dev);
1764         u32 val;
1765         int ret;
1766
1767         if (da->da_dev->export_count) {
1768                 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
1769                 return -EINVAL;
1770         }
1771
1772         ret = kstrtou32(page, 0, &val);
1773         if (ret < 0)
1774                 return ret;
1775
1776         udev->cmd_time_out = val * MSEC_PER_SEC;
1777         return count;
1778 }
1779 CONFIGFS_ATTR(tcmu_, cmd_time_out);
1780
1781 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
1782 {
1783         struct se_dev_attrib *da = container_of(to_config_group(item),
1784                                                 struct se_dev_attrib, da_group);
1785         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1786
1787         return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
1788 }
1789
1790 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
1791                                      size_t count)
1792 {
1793         struct se_dev_attrib *da = container_of(to_config_group(item),
1794                                                 struct se_dev_attrib, da_group);
1795         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1796         int ret, len;
1797
1798         len = strlen(page);
1799         if (!len || len > TCMU_CONFIG_LEN - 1)
1800                 return -EINVAL;
1801
1802         /* Check if device has been configured before */
1803         if (tcmu_dev_configured(udev)) {
1804                 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1805                                          TCMU_ATTR_DEV_CFG, page);
1806                 if (ret) {
1807                         pr_err("Unable to reconfigure device\n");
1808                         return ret;
1809                 }
1810                 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1811
1812                 ret = tcmu_update_uio_info(udev);
1813                 if (ret)
1814                         return ret;
1815                 return count;
1816         }
1817         strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
1818
1819         return count;
1820 }
1821 CONFIGFS_ATTR(tcmu_, dev_config);
1822
1823 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
1824 {
1825         struct se_dev_attrib *da = container_of(to_config_group(item),
1826                                                 struct se_dev_attrib, da_group);
1827         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1828
1829         return snprintf(page, PAGE_SIZE, "%zu\n", udev->dev_size);
1830 }
1831
1832 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
1833                                    size_t count)
1834 {
1835         struct se_dev_attrib *da = container_of(to_config_group(item),
1836                                                 struct se_dev_attrib, da_group);
1837         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1838         u64 val;
1839         int ret;
1840
1841         ret = kstrtou64(page, 0, &val);
1842         if (ret < 0)
1843                 return ret;
1844
1845         /* Check if device has been configured before */
1846         if (tcmu_dev_configured(udev)) {
1847                 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1848                                          TCMU_ATTR_DEV_SIZE, &val);
1849                 if (ret) {
1850                         pr_err("Unable to reconfigure device\n");
1851                         return ret;
1852                 }
1853         }
1854         udev->dev_size = val;
1855         return count;
1856 }
1857 CONFIGFS_ATTR(tcmu_, dev_size);
1858
1859 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
1860                                              char *page)
1861 {
1862         struct se_dev_attrib *da = container_of(to_config_group(item),
1863                                         struct se_dev_attrib, da_group);
1864
1865         return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
1866 }
1867
1868 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
1869                                               const char *page, size_t count)
1870 {
1871         struct se_dev_attrib *da = container_of(to_config_group(item),
1872                                         struct se_dev_attrib, da_group);
1873         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
1874         u8 val;
1875         int ret;
1876
1877         ret = kstrtou8(page, 0, &val);
1878         if (ret < 0)
1879                 return ret;
1880
1881         /* Check if device has been configured before */
1882         if (tcmu_dev_configured(udev)) {
1883                 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
1884                                          TCMU_ATTR_WRITECACHE, &val);
1885                 if (ret) {
1886                         pr_err("Unable to reconfigure device\n");
1887                         return ret;
1888                 }
1889         }
1890
1891         da->emulate_write_cache = val;
1892         return count;
1893 }
1894 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
1895
1896 static struct configfs_attribute *tcmu_attrib_attrs[] = {
1897         &tcmu_attr_cmd_time_out,
1898         &tcmu_attr_dev_config,
1899         &tcmu_attr_dev_size,
1900         &tcmu_attr_emulate_write_cache,
1901         NULL,
1902 };
1903
1904 static struct configfs_attribute **tcmu_attrs;
1905
1906 static struct target_backend_ops tcmu_ops = {
1907         .name                   = "user",
1908         .owner                  = THIS_MODULE,
1909         .transport_flags        = TRANSPORT_FLAG_PASSTHROUGH,
1910         .attach_hba             = tcmu_attach_hba,
1911         .detach_hba             = tcmu_detach_hba,
1912         .alloc_device           = tcmu_alloc_device,
1913         .configure_device       = tcmu_configure_device,
1914         .destroy_device         = tcmu_destroy_device,
1915         .free_device            = tcmu_free_device,
1916         .parse_cdb              = tcmu_parse_cdb,
1917         .set_configfs_dev_params = tcmu_set_configfs_dev_params,
1918         .show_configfs_dev_params = tcmu_show_configfs_dev_params,
1919         .get_device_type        = sbc_get_device_type,
1920         .get_blocks             = tcmu_get_blocks,
1921         .tb_dev_attrib_attrs    = NULL,
1922 };
1923
1924 static int unmap_thread_fn(void *data)
1925 {
1926         struct tcmu_dev *udev;
1927         loff_t off;
1928         uint32_t start, end, block;
1929         struct page *page;
1930         int i;
1931
1932         while (!kthread_should_stop()) {
1933                 DEFINE_WAIT(__wait);
1934
1935                 prepare_to_wait(&unmap_wait, &__wait, TASK_INTERRUPTIBLE);
1936                 schedule();
1937                 finish_wait(&unmap_wait, &__wait);
1938
1939                 if (kthread_should_stop())
1940                         break;
1941
1942                 mutex_lock(&root_udev_mutex);
1943                 list_for_each_entry(udev, &root_udev, node) {
1944                         mutex_lock(&udev->cmdr_lock);
1945
1946                         /* Try to complete the finished commands first */
1947                         tcmu_handle_completions(udev);
1948
1949                         /* Skip the udevs waiting the global pool or in idle */
1950                         if (udev->waiting_global || !udev->dbi_thresh) {
1951                                 mutex_unlock(&udev->cmdr_lock);
1952                                 continue;
1953                         }
1954
1955                         end = udev->dbi_max + 1;
1956                         block = find_last_bit(udev->data_bitmap, end);
1957                         if (block == udev->dbi_max) {
1958                                 /*
1959                                  * The last bit is dbi_max, so there is
1960                                  * no need to shrink any blocks.
1961                                  */
1962                                 mutex_unlock(&udev->cmdr_lock);
1963                                 continue;
1964                         } else if (block == end) {
1965                                 /* The current udev will goto idle state */
1966                                 udev->dbi_thresh = start = 0;
1967                                 udev->dbi_max = 0;
1968                         } else {
1969                                 udev->dbi_thresh = start = block + 1;
1970                                 udev->dbi_max = block;
1971                         }
1972
1973                         /* Here will truncate the data area from off */
1974                         off = udev->data_off + start * DATA_BLOCK_SIZE;
1975                         unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
1976
1977                         /* Release the block pages */
1978                         for (i = start; i < end; i++) {
1979                                 page = radix_tree_delete(&udev->data_blocks, i);
1980                                 if (page) {
1981                                         __free_page(page);
1982                                         atomic_dec(&global_db_count);
1983                                 }
1984                         }
1985                         mutex_unlock(&udev->cmdr_lock);
1986                 }
1987
1988                 /*
1989                  * Try to wake up the udevs who are waiting
1990                  * for the global data pool.
1991                  */
1992                 list_for_each_entry(udev, &root_udev, node) {
1993                         if (udev->waiting_global)
1994                                 wake_up(&udev->wait_cmdr);
1995                 }
1996                 mutex_unlock(&root_udev_mutex);
1997         }
1998
1999         return 0;
2000 }
2001
2002 static int __init tcmu_module_init(void)
2003 {
2004         int ret, i, k, len = 0;
2005
2006         BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2007
2008         tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2009                                 sizeof(struct tcmu_cmd),
2010                                 __alignof__(struct tcmu_cmd),
2011                                 0, NULL);
2012         if (!tcmu_cmd_cache)
2013                 return -ENOMEM;
2014
2015         tcmu_root_device = root_device_register("tcm_user");
2016         if (IS_ERR(tcmu_root_device)) {
2017                 ret = PTR_ERR(tcmu_root_device);
2018                 goto out_free_cache;
2019         }
2020
2021         ret = genl_register_family(&tcmu_genl_family);
2022         if (ret < 0) {
2023                 goto out_unreg_device;
2024         }
2025
2026         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2027                 len += sizeof(struct configfs_attribute *);
2028         }
2029         for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2030                 len += sizeof(struct configfs_attribute *);
2031         }
2032         len += sizeof(struct configfs_attribute *);
2033
2034         tcmu_attrs = kzalloc(len, GFP_KERNEL);
2035         if (!tcmu_attrs) {
2036                 ret = -ENOMEM;
2037                 goto out_unreg_genl;
2038         }
2039
2040         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2041                 tcmu_attrs[i] = passthrough_attrib_attrs[i];
2042         }
2043         for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2044                 tcmu_attrs[i] = tcmu_attrib_attrs[k];
2045                 i++;
2046         }
2047         tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2048
2049         ret = transport_backend_register(&tcmu_ops);
2050         if (ret)
2051                 goto out_attrs;
2052
2053         init_waitqueue_head(&unmap_wait);
2054         unmap_thread = kthread_run(unmap_thread_fn, NULL, "tcmu_unmap");
2055         if (IS_ERR(unmap_thread)) {
2056                 ret = PTR_ERR(unmap_thread);
2057                 goto out_unreg_transport;
2058         }
2059
2060         return 0;
2061
2062 out_unreg_transport:
2063         target_backend_unregister(&tcmu_ops);
2064 out_attrs:
2065         kfree(tcmu_attrs);
2066 out_unreg_genl:
2067         genl_unregister_family(&tcmu_genl_family);
2068 out_unreg_device:
2069         root_device_unregister(tcmu_root_device);
2070 out_free_cache:
2071         kmem_cache_destroy(tcmu_cmd_cache);
2072
2073         return ret;
2074 }
2075
2076 static void __exit tcmu_module_exit(void)
2077 {
2078         kthread_stop(unmap_thread);
2079         target_backend_unregister(&tcmu_ops);
2080         kfree(tcmu_attrs);
2081         genl_unregister_family(&tcmu_genl_family);
2082         root_device_unregister(tcmu_root_device);
2083         kmem_cache_destroy(tcmu_cmd_cache);
2084 }
2085
2086 MODULE_DESCRIPTION("TCM USER subsystem plugin");
2087 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2088 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2089 MODULE_LICENSE("GPL");
2090
2091 module_init(tcmu_module_init);
2092 module_exit(tcmu_module_exit);