drivers/target/target_core_user.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
   4  * Copyright (C) 2014 Red Hat, Inc.
   5  * Copyright (C) 2015 Arrikto, Inc.
   6  * Copyright (C) 2017 Chinamobile, Inc.
   7  */
   8
   9 #include <linux/spinlock.h>
  10 #include <linux/module.h>
  11 #include <linux/kernel.h>
  12 #include <linux/timer.h>
  13 #include <linux/parser.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/uio_driver.h>
  16 #include <linux/xarray.h>
  17 #include <linux/stringify.h>
  18 #include <linux/bitops.h>
  19 #include <linux/highmem.h>
  20 #include <linux/configfs.h>
  21 #include <linux/mutex.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/pagemap.h>
  24 #include <net/genetlink.h>
  25 #include <scsi/scsi_common.h>
  26 #include <scsi/scsi_proto.h>
  27 #include <target/target_core_base.h>
  28 #include <target/target_core_fabric.h>
  29 #include <target/target_core_backend.h>
  30
  31 #include <linux/target_core_user.h>
  32
  33 /**
  34  * DOC: Userspace I/O
  35  * Userspace I/O
  36  * -------------
  37  *
  38  * Define a shared-memory interface for LIO to pass SCSI commands and
  39  * data to userspace for processing. This is to allow backends that
  40  * are too complex for in-kernel support to be possible.
  41  *
  42  * It uses the UIO framework to do a lot of the device-creation and
  43  * introspection work for us.
  44  *
  45  * See the .h file for how the ring is laid out. Note that while the
  46  * command ring is defined, the particulars of the data area are
  47  * not. Offset values in the command entry point to other locations
  48  * internal to the mmap-ed area. There is separate space outside the
  49  * command ring for data buffers. This leaves maximum flexibility for
  50  * moving buffer allocations, or even page flipping or other
  51  * allocation techniques, without altering the command ring layout.
  52  *
  53  * SECURITY:
  54  * The user process must be assumed to be malicious. There's no way to
  55  * prevent it breaking the command ring protocol if it wants, but in
  56  * order to prevent other issues we must only ever read *data* from
  57  * the shared memory area, not offsets or sizes. This applies to
  58  * command ring entries as well as the mailbox. Extra code needed for
  59  * this may have a 'UAM' comment.
  60  */
  61
  62 #define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
  63
  64 /* For mailbox plus cmd ring, the size is fixed 8MB */
  65 #define MB_CMDR_SIZE (8 * 1024 * 1024)
  66 /* Offset of cmd ring is size of mailbox */
  67 #define CMDR_OFF sizeof(struct tcmu_mailbox)
  68 #define CMDR_SIZE (MB_CMDR_SIZE - CMDR_OFF)
  69
  70 /*
  71  * For data area, the default block size is PAGE_SIZE and
  72  * the default total size is 256K * PAGE_SIZE.
  73  */
  74 #define DATA_PAGES_PER_BLK_DEF 1
  75 #define DATA_AREA_PAGES_DEF (256 * 1024)
  76
  77 #define TCMU_MBS_TO_PAGES(_mbs) ((size_t)_mbs << (20 - PAGE_SHIFT))
  78 #define TCMU_PAGES_TO_MBS(_pages) (_pages >> (20 - PAGE_SHIFT))
  79
  80 /*
  81  * Default number of global data blocks(512K * PAGE_SIZE)
  82  * when the unmap thread will be started.
  83  */
  84 #define TCMU_GLOBAL_MAX_PAGES_DEF (512 * 1024)
  85
  86 static u8 tcmu_kern_cmd_reply_supported;
  87 static u8 tcmu_netlink_blocked;
  88
  89 static struct device *tcmu_root_device;
  90
  91 struct tcmu_hba {
  92         u32 host_id;
  93 };
  94
  95 #define TCMU_CONFIG_LEN 256
  96
  97 static DEFINE_MUTEX(tcmu_nl_cmd_mutex);
  98 static LIST_HEAD(tcmu_nl_cmd_list);
  99
 100 struct tcmu_dev;
 101
 102 struct tcmu_nl_cmd {
 103         /* wake up thread waiting for reply */
 104         struct completion complete;
 105         struct list_head nl_list;
 106         struct tcmu_dev *udev;
 107         int cmd;
 108         int status;
 109 };
 110
 111 struct tcmu_dev {
 112         struct list_head node;
 113         struct kref kref;
 114
 115         struct se_device se_dev;
 116         struct se_dev_plug se_plug;
 117
 118         char *name;
 119         struct se_hba *hba;
 120
 121 #define TCMU_DEV_BIT_OPEN 0
 122 #define TCMU_DEV_BIT_BROKEN 1
 123 #define TCMU_DEV_BIT_BLOCKED 2
 124 #define TCMU_DEV_BIT_TMR_NOTIFY 3
 125 #define TCMU_DEV_BIT_PLUGGED 4
 126         unsigned long flags;
 127
 128         struct uio_info uio_info;
 129
 130         struct inode *inode;
 131
 132         uint64_t dev_size;
 133
 134         struct tcmu_mailbox *mb_addr;
 135         void *cmdr;
 136         u32 cmdr_size;
 137         u32 cmdr_last_cleaned;
 138         /* Offset of data area from start of mb */
 139         /* Must add data_off and mb_addr to get the address */
 140         size_t data_off;
 141         int data_area_mb;
 142         uint32_t max_blocks;
 143         size_t mmap_pages;
 144
 145         struct mutex cmdr_lock;
 146         struct list_head qfull_queue;
 147         struct list_head tmr_queue;
 148
 149         uint32_t dbi_max;
 150         uint32_t dbi_thresh;
 151         unsigned long *data_bitmap;
 152         struct xarray data_pages;
 153         uint32_t data_pages_per_blk;
 154         uint32_t data_blk_size;
 155
 156         struct xarray commands;
 157
 158         struct timer_list cmd_timer;
 159         unsigned int cmd_time_out;
 160         struct list_head inflight_queue;
 161
 162         struct timer_list qfull_timer;
 163         int qfull_time_out;
 164
 165         struct list_head timedout_entry;
 166
 167         struct tcmu_nl_cmd curr_nl_cmd;
 168
 169         char dev_config[TCMU_CONFIG_LEN];
 170
 171         int nl_reply_supported;
 172 };
 173
 174 #define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
 175
 176 struct tcmu_cmd {
 177         struct se_cmd *se_cmd;
 178         struct tcmu_dev *tcmu_dev;
 179         struct list_head queue_entry;
 180
 181         uint16_t cmd_id;
 182
 183         /* Can't use se_cmd when cleaning up expired cmds, because if
 184            cmd has been completed then accessing se_cmd is off limits */
 185         uint32_t dbi_cnt;
 186         uint32_t dbi_bidi_cnt;
 187         uint32_t dbi_cur;
 188         uint32_t *dbi;
 189
 190         uint32_t data_len_bidi;
 191
 192         unsigned long deadline;
 193
 194 #define TCMU_CMD_BIT_EXPIRED 0
 195 #define TCMU_CMD_BIT_KEEP_BUF 1
 196         unsigned long flags;
 197 };
 198
 199 struct tcmu_tmr {
 200         struct list_head queue_entry;
 201
 202         uint8_t tmr_type;
 203         uint32_t tmr_cmd_cnt;
 204         int16_t tmr_cmd_ids[];
 205 };
 206
 207 /*
 208  * To avoid dead lock the mutex lock order should always be:
 209  *
 210  * mutex_lock(&root_udev_mutex);
 211  * ...
 212  * mutex_lock(&tcmu_dev->cmdr_lock);
 213  * mutex_unlock(&tcmu_dev->cmdr_lock);
 214  * ...
 215  * mutex_unlock(&root_udev_mutex);
 216  */
 217 static DEFINE_MUTEX(root_udev_mutex);
 218 static LIST_HEAD(root_udev);
 219
 220 static DEFINE_SPINLOCK(timed_out_udevs_lock);
 221 static LIST_HEAD(timed_out_udevs);
 222
 223 static struct kmem_cache *tcmu_cmd_cache;
 224
 225 static atomic_t global_page_count = ATOMIC_INIT(0);
 226 static struct delayed_work tcmu_unmap_work;
 227 static int tcmu_global_max_pages = TCMU_GLOBAL_MAX_PAGES_DEF;
 228
 229 static int tcmu_set_global_max_data_area(const char *str,
 230                                          const struct kernel_param *kp)
 231 {
 232         int ret, max_area_mb;
 233
 234         ret = kstrtoint(str, 10, &max_area_mb);
 235         if (ret)
 236                 return -EINVAL;
 237
 238         if (max_area_mb <= 0) {
 239                 pr_err("global_max_data_area must be larger than 0.\n");
 240                 return -EINVAL;
 241         }
 242
 243         tcmu_global_max_pages = TCMU_MBS_TO_PAGES(max_area_mb);
 244         if (atomic_read(&global_page_count) > tcmu_global_max_pages)
 245                 schedule_delayed_work(&tcmu_unmap_work, 0);
 246         else
 247                 cancel_delayed_work_sync(&tcmu_unmap_work);
 248
 249         return 0;
 250 }
 251
 252 static int tcmu_get_global_max_data_area(char *buffer,
 253                                          const struct kernel_param *kp)
 254 {
 255         return sprintf(buffer, "%d\n", TCMU_PAGES_TO_MBS(tcmu_global_max_pages));
 256 }
 257
 258 static const struct kernel_param_ops tcmu_global_max_data_area_op = {
 259         .set = tcmu_set_global_max_data_area,
 260         .get = tcmu_get_global_max_data_area,
 261 };
 262
 263 module_param_cb(global_max_data_area_mb, &tcmu_global_max_data_area_op, NULL,
 264                 S_IWUSR | S_IRUGO);
 265 MODULE_PARM_DESC(global_max_data_area_mb,
 266                  "Max MBs allowed to be allocated to all the tcmu device's "
 267                  "data areas.");
 268
 269 static int tcmu_get_block_netlink(char *buffer,
 270                                   const struct kernel_param *kp)
 271 {
 272         return sprintf(buffer, "%s\n", tcmu_netlink_blocked ?
 273                        "blocked" : "unblocked");
 274 }
 275
 276 static int tcmu_set_block_netlink(const char *str,
 277                                   const struct kernel_param *kp)
 278 {
 279         int ret;
 280         u8 val;
 281
 282         ret = kstrtou8(str, 0, &val);
 283         if (ret < 0)
 284                 return ret;
 285
 286         if (val > 1) {
 287                 pr_err("Invalid block netlink value %u\n", val);
 288                 return -EINVAL;
 289         }
 290
 291         tcmu_netlink_blocked = val;
 292         return 0;
 293 }
 294
 295 static const struct kernel_param_ops tcmu_block_netlink_op = {
 296         .set = tcmu_set_block_netlink,
 297         .get = tcmu_get_block_netlink,
 298 };
 299
 300 module_param_cb(block_netlink, &tcmu_block_netlink_op, NULL, S_IWUSR | S_IRUGO);
 301 MODULE_PARM_DESC(block_netlink, "Block new netlink commands.");
 302
 303 static int tcmu_fail_netlink_cmd(struct tcmu_nl_cmd *nl_cmd)
 304 {
 305         struct tcmu_dev *udev = nl_cmd->udev;
 306
 307         if (!tcmu_netlink_blocked) {
 308                 pr_err("Could not reset device's netlink interface. Netlink is not blocked.\n");
 309                 return -EBUSY;
 310         }
 311
 312         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
 313                 pr_debug("Aborting nl cmd %d on %s\n", nl_cmd->cmd, udev->name);
 314                 nl_cmd->status = -EINTR;
 315                 list_del(&nl_cmd->nl_list);
 316                 complete(&nl_cmd->complete);
 317         }
 318         return 0;
 319 }
 320
 321 static int tcmu_set_reset_netlink(const char *str,
 322                                   const struct kernel_param *kp)
 323 {
 324         struct tcmu_nl_cmd *nl_cmd, *tmp_cmd;
 325         int ret;
 326         u8 val;
 327
 328         ret = kstrtou8(str, 0, &val);
 329         if (ret < 0)
 330                 return ret;
 331
 332         if (val != 1) {
 333                 pr_err("Invalid reset netlink value %u\n", val);
 334                 return -EINVAL;
 335         }
 336
 337         mutex_lock(&tcmu_nl_cmd_mutex);
 338         list_for_each_entry_safe(nl_cmd, tmp_cmd, &tcmu_nl_cmd_list, nl_list) {
 339                 ret = tcmu_fail_netlink_cmd(nl_cmd);
 340                 if (ret)
 341                         break;
 342         }
 343         mutex_unlock(&tcmu_nl_cmd_mutex);
 344
 345         return ret;
 346 }
 347
 348 static const struct kernel_param_ops tcmu_reset_netlink_op = {
 349         .set = tcmu_set_reset_netlink,
 350 };
 351
 352 module_param_cb(reset_netlink, &tcmu_reset_netlink_op, NULL, S_IWUSR);
 353 MODULE_PARM_DESC(reset_netlink, "Reset netlink commands.");
 354
 355 /* multicast group */
 356 enum tcmu_multicast_groups {
 357         TCMU_MCGRP_CONFIG,
 358 };
 359
 360 static const struct genl_multicast_group tcmu_mcgrps[] = {
 361         [TCMU_MCGRP_CONFIG] = { .name = "config", },
 362 };
 363
 364 static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
 365         [TCMU_ATTR_DEVICE]      = { .type = NLA_STRING },
 366         [TCMU_ATTR_MINOR]       = { .type = NLA_U32 },
 367         [TCMU_ATTR_CMD_STATUS]  = { .type = NLA_S32 },
 368         [TCMU_ATTR_DEVICE_ID]   = { .type = NLA_U32 },
 369         [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
 370 };
 371
 372 static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
 373 {
 374         struct tcmu_dev *udev = NULL;
 375         struct tcmu_nl_cmd *nl_cmd;
 376         int dev_id, rc, ret = 0;
 377
 378         if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
 379             !info->attrs[TCMU_ATTR_DEVICE_ID]) {
 380                 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
 381                 return -EINVAL;
 382         }
 383
 384         dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
 385         rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
 386
 387         mutex_lock(&tcmu_nl_cmd_mutex);
 388         list_for_each_entry(nl_cmd, &tcmu_nl_cmd_list, nl_list) {
 389                 if (nl_cmd->udev->se_dev.dev_index == dev_id) {
 390                         udev = nl_cmd->udev;
 391                         break;
 392                 }
 393         }
 394
 395         if (!udev) {
 396                 pr_err("tcmu nl cmd %u/%d completion could not find device with dev id %u.\n",
 397                        completed_cmd, rc, dev_id);
 398                 ret = -ENODEV;
 399                 goto unlock;
 400         }
 401         list_del(&nl_cmd->nl_list);
 402
 403         pr_debug("%s genl cmd done got id %d curr %d done %d rc %d stat %d\n",
 404                  udev->name, dev_id, nl_cmd->cmd, completed_cmd, rc,
 405                  nl_cmd->status);
 406
 407         if (nl_cmd->cmd != completed_cmd) {
 408                 pr_err("Mismatched commands on %s (Expecting reply for %d. Current %d).\n",
 409                        udev->name, completed_cmd, nl_cmd->cmd);
 410                 ret = -EINVAL;
 411                 goto unlock;
 412         }
 413
 414         nl_cmd->status = rc;
 415         complete(&nl_cmd->complete);
 416 unlock:
 417         mutex_unlock(&tcmu_nl_cmd_mutex);
 418         return ret;
 419 }
 420
 421 static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
 422 {
 423         return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
 424 }
 425
 426 static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
 427 {
 428         return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
 429 }
 430
 431 static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
 432                                        struct genl_info *info)
 433 {
 434         return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
 435 }
 436
 437 static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
 438 {
 439         if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
 440                 tcmu_kern_cmd_reply_supported  =
 441                         nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
 442                 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
 443                        tcmu_kern_cmd_reply_supported);
 444         }
 445
 446         return 0;
 447 }
 448
 449 static const struct genl_small_ops tcmu_genl_ops[] = {
 450         {
 451                 .cmd    = TCMU_CMD_SET_FEATURES,
 452                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 453                 .flags  = GENL_ADMIN_PERM,
 454                 .doit   = tcmu_genl_set_features,
 455         },
 456         {
 457                 .cmd    = TCMU_CMD_ADDED_DEVICE_DONE,
 458                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 459                 .flags  = GENL_ADMIN_PERM,
 460                 .doit   = tcmu_genl_add_dev_done,
 461         },
 462         {
 463                 .cmd    = TCMU_CMD_REMOVED_DEVICE_DONE,
 464                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 465                 .flags  = GENL_ADMIN_PERM,
 466                 .doit   = tcmu_genl_rm_dev_done,
 467         },
 468         {
 469                 .cmd    = TCMU_CMD_RECONFIG_DEVICE_DONE,
 470                 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 471                 .flags  = GENL_ADMIN_PERM,
 472                 .doit   = tcmu_genl_reconfig_dev_done,
 473         },
 474 };
 475
 476 /* Our generic netlink family */
 477 static struct genl_family tcmu_genl_family __ro_after_init = {
 478         .module = THIS_MODULE,
 479         .hdrsize = 0,
 480         .name = "TCM-USER",
 481         .version = 2,
 482         .maxattr = TCMU_ATTR_MAX,
 483         .policy = tcmu_attr_policy,
 484         .mcgrps = tcmu_mcgrps,
 485         .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
 486         .netnsok = true,
 487         .small_ops = tcmu_genl_ops,
 488         .n_small_ops = ARRAY_SIZE(tcmu_genl_ops),
 489 };
 490
 491 #define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
 492 #define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
 493 #define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
 494 #define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
 495
 496 static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
 497 {
 498         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 499         uint32_t i;
 500
 501         for (i = 0; i < len; i++)
 502                 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
 503 }
 504
 505 static inline int tcmu_get_empty_block(struct tcmu_dev *udev,
 506                                        struct tcmu_cmd *tcmu_cmd,
 507                                        int prev_dbi, int length, int *iov_cnt)
 508 {
 509         XA_STATE(xas, &udev->data_pages, 0);
 510         struct page *page;
 511         int i, cnt, dbi, dpi;
 512         int page_cnt = DIV_ROUND_UP(length, PAGE_SIZE);
 513
 514         dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
 515         if (dbi == udev->dbi_thresh)
 516                 return -1;
 517
 518         dpi = dbi * udev->data_pages_per_blk;
 519         /* Count the number of already allocated pages */
 520         xas_set(&xas, dpi);
 521         rcu_read_lock();
 522         for (cnt = 0; xas_next(&xas) && cnt < page_cnt;)
 523                 cnt++;
 524         rcu_read_unlock();
 525
 526         for (i = cnt; i < page_cnt; i++) {
 527                 /* try to get new page from the mm */
 528                 page = alloc_page(GFP_NOIO);
 529                 if (!page)
 530                         break;
 531
 532                 if (xa_store(&udev->data_pages, dpi + i, page, GFP_NOIO)) {
 533                         __free_page(page);
 534                         break;
 535                 }
 536         }
 537         if (atomic_add_return(i - cnt, &global_page_count) >
 538                               tcmu_global_max_pages)
 539                 schedule_delayed_work(&tcmu_unmap_work, 0);
 540
 541         if (i && dbi > udev->dbi_max)
 542                 udev->dbi_max = dbi;
 543
 544         set_bit(dbi, udev->data_bitmap);
 545         tcmu_cmd_set_dbi(tcmu_cmd, dbi);
 546
 547         if (dbi != prev_dbi + 1)
 548                 *iov_cnt += 1;
 549
 550         return i == page_cnt ? dbi : -1;
 551 }
 552
 553 static int tcmu_get_empty_blocks(struct tcmu_dev *udev,
 554                                  struct tcmu_cmd *tcmu_cmd, int length)
 555 {
 556         /* start value of dbi + 1 must not be a valid dbi */
 557         int dbi = -2;
 558         int blk_data_len, iov_cnt = 0;
 559         uint32_t blk_size = udev->data_blk_size;
 560
 561         for (; length > 0; length -= blk_size) {
 562                 blk_data_len = min_t(uint32_t, length, blk_size);
 563                 dbi = tcmu_get_empty_block(udev, tcmu_cmd, dbi, blk_data_len,
 564                                            &iov_cnt);
 565                 if (dbi < 0)
 566                         return -1;
 567         }
 568         return iov_cnt;
 569 }
 570
 571 static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
 572 {
 573         kfree(tcmu_cmd->dbi);
 574         kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 575 }
 576
 577 static inline void tcmu_cmd_set_block_cnts(struct tcmu_cmd *cmd)
 578 {
 579         int i, len;
 580         struct se_cmd *se_cmd = cmd->se_cmd;
 581         uint32_t blk_size = cmd->tcmu_dev->data_blk_size;
 582
 583         cmd->dbi_cnt = DIV_ROUND_UP(se_cmd->data_length, blk_size);
 584
 585         if (se_cmd->se_cmd_flags & SCF_BIDI) {
 586                 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
 587                 for (i = 0, len = 0; i < se_cmd->t_bidi_data_nents; i++)
 588                         len += se_cmd->t_bidi_data_sg[i].length;
 589                 cmd->dbi_bidi_cnt = DIV_ROUND_UP(len, blk_size);
 590                 cmd->dbi_cnt += cmd->dbi_bidi_cnt;
 591                 cmd->data_len_bidi = len;
 592         }
 593 }
 594
 595 static int new_block_to_iov(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 596                             struct iovec **iov, int prev_dbi, int len)
 597 {
 598         /* Get the next dbi */
 599         int dbi = tcmu_cmd_get_dbi(cmd);
 600
 601         /* Do not add more than udev->data_blk_size to iov */
 602         len = min_t(int,  len, udev->data_blk_size);
 603
 604         /*
 605          * The following code will gather and map the blocks to the same iovec
 606          * when the blocks are all next to each other.
 607          */
 608         if (dbi != prev_dbi + 1) {
 609                 /* dbi is not next to previous dbi, so start new iov */
 610                 if (prev_dbi >= 0)
 611                         (*iov)++;
 612                 /* write offset relative to mb_addr */
 613                 (*iov)->iov_base = (void __user *)
 614                                    (udev->data_off + dbi * udev->data_blk_size);
 615         }
 616         (*iov)->iov_len += len;
 617
 618         return dbi;
 619 }
 620
 621 static void tcmu_setup_iovs(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 622                             struct iovec **iov, int data_length)
 623 {
 624         /* start value of dbi + 1 must not be a valid dbi */
 625         int dbi = -2;
 626
 627         /* We prepare the IOVs for DMA_FROM_DEVICE transfer direction */
 628         for (; data_length > 0; data_length -= udev->data_blk_size)
 629                 dbi = new_block_to_iov(udev, cmd, iov, dbi, data_length);
 630 }
 631
 632 static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
 633 {
 634         struct se_device *se_dev = se_cmd->se_dev;
 635         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 636         struct tcmu_cmd *tcmu_cmd;
 637
 638         tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_NOIO);
 639         if (!tcmu_cmd)
 640                 return NULL;
 641
 642         INIT_LIST_HEAD(&tcmu_cmd->queue_entry);
 643         tcmu_cmd->se_cmd = se_cmd;
 644         tcmu_cmd->tcmu_dev = udev;
 645
 646         tcmu_cmd_set_block_cnts(tcmu_cmd);
 647         tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
 648                                 GFP_NOIO);
 649         if (!tcmu_cmd->dbi) {
 650                 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
 651                 return NULL;
 652         }
 653
 654         return tcmu_cmd;
 655 }
 656
 657 static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
 658 {
 659         unsigned long offset = offset_in_page(vaddr);
 660         void *start = vaddr - offset;
 661
 662         size = round_up(size+offset, PAGE_SIZE);
 663
 664         while (size) {
 665                 flush_dcache_page(vmalloc_to_page(start));
 666                 start += PAGE_SIZE;
 667                 size -= PAGE_SIZE;
 668         }
 669 }
 670
 671 /*
 672  * Some ring helper functions. We don't assume size is a power of 2 so
 673  * we can't use circ_buf.h.
 674  */
 675 static inline size_t spc_used(size_t head, size_t tail, size_t size)
 676 {
 677         int diff = head - tail;
 678
 679         if (diff >= 0)
 680                 return diff;
 681         else
 682                 return size + diff;
 683 }
 684
 685 static inline size_t spc_free(size_t head, size_t tail, size_t size)
 686 {
 687         /* Keep 1 byte unused or we can't tell full from empty */
 688         return (size - spc_used(head, tail, size) - 1);
 689 }
 690
 691 static inline size_t head_to_end(size_t head, size_t size)
 692 {
 693         return size - head;
 694 }
 695
 696 #define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
 697
 698 #define TCMU_SG_TO_DATA_AREA 1
 699 #define TCMU_DATA_AREA_TO_SG 2
 700
 701 static inline void tcmu_copy_data(struct tcmu_dev *udev,
 702                                   struct tcmu_cmd *tcmu_cmd, uint32_t direction,
 703                                   struct scatterlist *sg, unsigned int sg_nents,
 704                                   struct iovec **iov, size_t data_len)
 705 {
 706         /* start value of dbi + 1 must not be a valid dbi */
 707         int dbi = -2;
 708         size_t page_remaining, cp_len;
 709         int page_cnt, page_inx, dpi;
 710         struct sg_mapping_iter sg_iter;
 711         unsigned int sg_flags;
 712         struct page *page;
 713         void *data_page_start, *data_addr;
 714
 715         if (direction == TCMU_SG_TO_DATA_AREA)
 716                 sg_flags = SG_MITER_ATOMIC | SG_MITER_FROM_SG;
 717         else
 718                 sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG;
 719         sg_miter_start(&sg_iter, sg, sg_nents, sg_flags);
 720
 721         while (data_len) {
 722                 if (direction == TCMU_SG_TO_DATA_AREA)
 723                         dbi = new_block_to_iov(udev, tcmu_cmd, iov, dbi,
 724                                                data_len);
 725                 else
 726                         dbi = tcmu_cmd_get_dbi(tcmu_cmd);
 727
 728                 page_cnt = DIV_ROUND_UP(data_len, PAGE_SIZE);
 729                 if (page_cnt > udev->data_pages_per_blk)
 730                         page_cnt = udev->data_pages_per_blk;
 731
 732                 dpi = dbi * udev->data_pages_per_blk;
 733                 for (page_inx = 0; page_inx < page_cnt && data_len;
 734                      page_inx++, dpi++) {
 735                         page = xa_load(&udev->data_pages, dpi);
 736
 737                         if (direction == TCMU_DATA_AREA_TO_SG)
 738                                 flush_dcache_page(page);
 739                         data_page_start = kmap_atomic(page);
 740                         page_remaining = PAGE_SIZE;
 741
 742                         while (page_remaining && data_len) {
 743                                 if (!sg_miter_next(&sg_iter)) {
 744                                         /* set length to 0 to abort outer loop */
 745                                         data_len = 0;
 746                                         pr_debug("%s: aborting data copy due to exhausted sg_list\n",
 747                                                  __func__);
 748                                         break;
 749                                 }
 750                                 cp_len = min3(sg_iter.length, page_remaining,
 751                                               data_len);
 752
 753                                 data_addr = data_page_start +
 754                                             PAGE_SIZE - page_remaining;
 755                                 if (direction == TCMU_SG_TO_DATA_AREA)
 756                                         memcpy(data_addr, sg_iter.addr, cp_len);
 757                                 else
 758                                         memcpy(sg_iter.addr, data_addr, cp_len);
 759
 760                                 data_len -= cp_len;
 761                                 page_remaining -= cp_len;
 762                                 sg_iter.consumed = cp_len;
 763                         }
 764                         sg_miter_stop(&sg_iter);
 765
 766                         kunmap_atomic(data_page_start);
 767                         if (direction == TCMU_SG_TO_DATA_AREA)
 768                                 flush_dcache_page(page);
 769                 }
 770         }
 771 }
 772
 773 static void scatter_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
 774                               struct iovec **iov)
 775 {
 776         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 777
 778         tcmu_copy_data(udev, tcmu_cmd, TCMU_SG_TO_DATA_AREA, se_cmd->t_data_sg,
 779                        se_cmd->t_data_nents, iov, se_cmd->data_length);
 780 }
 781
 782 static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
 783                              bool bidi, uint32_t read_len)
 784 {
 785         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 786         struct scatterlist *data_sg;
 787         unsigned int data_nents;
 788
 789         if (!bidi) {
 790                 data_sg = se_cmd->t_data_sg;
 791                 data_nents = se_cmd->t_data_nents;
 792         } else {
 793                 /*
 794                  * For bidi case, the first count blocks are for Data-Out
 795                  * buffer blocks, and before gathering the Data-In buffer
 796                  * the Data-Out buffer blocks should be skipped.
 797                  */
 798                 tcmu_cmd_set_dbi_cur(tcmu_cmd,
 799                                      tcmu_cmd->dbi_cnt - tcmu_cmd->dbi_bidi_cnt);
 800
 801                 data_sg = se_cmd->t_bidi_data_sg;
 802                 data_nents = se_cmd->t_bidi_data_nents;
 803         }
 804
 805         tcmu_copy_data(udev, tcmu_cmd, TCMU_DATA_AREA_TO_SG, data_sg,
 806                        data_nents, NULL, read_len);
 807 }
 808
 809 static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
 810 {
 811         return thresh - bitmap_weight(bitmap, thresh);
 812 }
 813
 814 /*
 815  * We can't queue a command until we have space available on the cmd ring.
 816  *
 817  * Called with ring lock held.
 818  */
 819 static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size)
 820 {
 821         struct tcmu_mailbox *mb = udev->mb_addr;
 822         size_t space, cmd_needed;
 823         u32 cmd_head;
 824
 825         tcmu_flush_dcache_range(mb, sizeof(*mb));
 826
 827         cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 828
 829         /*
 830          * If cmd end-of-ring space is too small then we need space for a NOP plus
 831          * original cmd - cmds are internally contiguous.
 832          */
 833         if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
 834                 cmd_needed = cmd_size;
 835         else
 836                 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
 837
 838         space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
 839         if (space < cmd_needed) {
 840                 pr_debug("no cmd space: %u %u %u\n", cmd_head,
 841                        udev->cmdr_last_cleaned, udev->cmdr_size);
 842                 return false;
 843         }
 844         return true;
 845 }
 846
 847 /*
 848  * We have to allocate data buffers before we can queue a command.
 849  * Returns -1 on error (not enough space) or number of needed iovs on success
 850  *
 851  * Called with ring lock held.
 852  */
 853 static int tcmu_alloc_data_space(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 854                                   int *iov_bidi_cnt)
 855 {
 856         int space, iov_cnt = 0, ret = 0;
 857
 858         if (!cmd->dbi_cnt)
 859                 goto wr_iov_cnts;
 860
 861         /* try to check and get the data blocks as needed */
 862         space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
 863         if (space < cmd->dbi_cnt) {
 864                 unsigned long blocks_left =
 865                                 (udev->max_blocks - udev->dbi_thresh) + space;
 866
 867                 if (blocks_left < cmd->dbi_cnt) {
 868                         pr_debug("no data space: only %lu available, but ask for %u\n",
 869                                         blocks_left * udev->data_blk_size,
 870                                         cmd->dbi_cnt * udev->data_blk_size);
 871                         return -1;
 872                 }
 873
 874                 udev->dbi_thresh += cmd->dbi_cnt;
 875                 if (udev->dbi_thresh > udev->max_blocks)
 876                         udev->dbi_thresh = udev->max_blocks;
 877         }
 878
 879         iov_cnt = tcmu_get_empty_blocks(udev, cmd, cmd->se_cmd->data_length);
 880         if (iov_cnt < 0)
 881                 return -1;
 882
 883         if (cmd->dbi_bidi_cnt) {
 884                 ret = tcmu_get_empty_blocks(udev, cmd, cmd->data_len_bidi);
 885                 if (ret < 0)
 886                         return -1;
 887         }
 888 wr_iov_cnts:
 889         *iov_bidi_cnt = ret;
 890         return iov_cnt + ret;
 891 }
 892
 893 static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
 894 {
 895         return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 896                         sizeof(struct tcmu_cmd_entry));
 897 }
 898
 899 static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
 900                                            size_t base_command_size)
 901 {
 902         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
 903         size_t command_size;
 904
 905         command_size = base_command_size +
 906                 round_up(scsi_command_size(se_cmd->t_task_cdb),
 907                                 TCMU_OP_ALIGN_SIZE);
 908
 909         WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
 910
 911         return command_size;
 912 }
 913
 914 static void tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
 915                                  struct timer_list *timer)
 916 {
 917         if (!tmo)
 918                 return;
 919
 920         tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
 921         if (!timer_pending(timer))
 922                 mod_timer(timer, tcmu_cmd->deadline);
 923
 924         pr_debug("Timeout set up for cmd %p, dev = %s, tmo = %lu\n", tcmu_cmd,
 925                  tcmu_cmd->tcmu_dev->name, tmo / MSEC_PER_SEC);
 926 }
 927
 928 static int add_to_qfull_queue(struct tcmu_cmd *tcmu_cmd)
 929 {
 930         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
 931         unsigned int tmo;
 932
 933         /*
 934          * For backwards compat if qfull_time_out is not set use
 935          * cmd_time_out and if that's not set use the default time out.
 936          */
 937         if (!udev->qfull_time_out)
 938                 return -ETIMEDOUT;
 939         else if (udev->qfull_time_out > 0)
 940                 tmo = udev->qfull_time_out;
 941         else if (udev->cmd_time_out)
 942                 tmo = udev->cmd_time_out;
 943         else
 944                 tmo = TCMU_TIME_OUT;
 945
 946         tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
 947
 948         list_add_tail(&tcmu_cmd->queue_entry, &udev->qfull_queue);
 949         pr_debug("adding cmd %p on dev %s to ring space wait queue\n",
 950                  tcmu_cmd, udev->name);
 951         return 0;
 952 }
 953
 954 static uint32_t ring_insert_padding(struct tcmu_dev *udev, size_t cmd_size)
 955 {
 956         struct tcmu_cmd_entry_hdr *hdr;
 957         struct tcmu_mailbox *mb = udev->mb_addr;
 958         uint32_t cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 959
 960         /* Insert a PAD if end-of-ring space is too small */
 961         if (head_to_end(cmd_head, udev->cmdr_size) < cmd_size) {
 962                 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
 963
 964                 hdr = udev->cmdr + cmd_head;
 965                 tcmu_hdr_set_op(&hdr->len_op, TCMU_OP_PAD);
 966                 tcmu_hdr_set_len(&hdr->len_op, pad_size);
 967                 hdr->cmd_id = 0; /* not used for PAD */
 968                 hdr->kflags = 0;
 969                 hdr->uflags = 0;
 970                 tcmu_flush_dcache_range(hdr, sizeof(*hdr));
 971
 972                 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
 973                 tcmu_flush_dcache_range(mb, sizeof(*mb));
 974
 975                 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
 976                 WARN_ON(cmd_head != 0);
 977         }
 978
 979         return cmd_head;
 980 }
 981
 982 static void tcmu_unplug_device(struct se_dev_plug *se_plug)
 983 {
 984         struct se_device *se_dev = se_plug->se_dev;
 985         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 986
 987         clear_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags);
 988         uio_event_notify(&udev->uio_info);
 989 }
 990
 991 static struct se_dev_plug *tcmu_plug_device(struct se_device *se_dev)
 992 {
 993         struct tcmu_dev *udev = TCMU_DEV(se_dev);
 994
 995         if (!test_and_set_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags))
 996                 return &udev->se_plug;
 997
 998         return NULL;
 999 }
1000
1001 /**
1002  * queue_cmd_ring - queue cmd to ring or internally
1003  * @tcmu_cmd: cmd to queue
1004  * @scsi_err: TCM error code if failure (-1) returned.
1005  *
1006  * Returns:
1007  * -1 we cannot queue internally or to the ring.
1008  *  0 success
1009  *  1 internally queued to wait for ring memory to free.
1010  */
1011 static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
1012 {
1013         struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
1014         struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
1015         size_t base_command_size, command_size;
1016         struct tcmu_mailbox *mb = udev->mb_addr;
1017         struct tcmu_cmd_entry *entry;
1018         struct iovec *iov;
1019         int iov_cnt, iov_bidi_cnt;
1020         uint32_t cmd_id, cmd_head;
1021         uint64_t cdb_off;
1022         uint32_t blk_size = udev->data_blk_size;
1023         /* size of data buffer needed */
1024         size_t data_length = (size_t)tcmu_cmd->dbi_cnt * blk_size;
1025
1026         *scsi_err = TCM_NO_SENSE;
1027
1028         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
1029                 *scsi_err = TCM_LUN_BUSY;
1030                 return -1;
1031         }
1032
1033         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1034                 *scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1035                 return -1;
1036         }
1037
1038         if (!list_empty(&udev->qfull_queue))
1039                 goto queue;
1040
1041         if (data_length > (size_t)udev->max_blocks * blk_size) {
1042                 pr_warn("TCMU: Request of size %zu is too big for %zu data area\n",
1043                         data_length, (size_t)udev->max_blocks * blk_size);
1044                 *scsi_err = TCM_INVALID_CDB_FIELD;
1045                 return -1;
1046         }
1047
1048         iov_cnt = tcmu_alloc_data_space(udev, tcmu_cmd, &iov_bidi_cnt);
1049         if (iov_cnt < 0)
1050                 goto free_and_queue;
1051
1052         /*
1053          * Must be a certain minimum size for response sense info, but
1054          * also may be larger if the iov array is large.
1055          */
1056         base_command_size = tcmu_cmd_get_base_cmd_size(iov_cnt);
1057         command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
1058
1059         if (command_size > (udev->cmdr_size / 2)) {
1060                 pr_warn("TCMU: Request of size %zu is too big for %u cmd ring\n",
1061                         command_size, udev->cmdr_size);
1062                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1063                 *scsi_err = TCM_INVALID_CDB_FIELD;
1064                 return -1;
1065         }
1066
1067         if (!is_ring_space_avail(udev, command_size))
1068                 /*
1069                  * Don't leave commands partially setup because the unmap
1070                  * thread might need the blocks to make forward progress.
1071                  */
1072                 goto free_and_queue;
1073
1074         if (xa_alloc(&udev->commands, &cmd_id, tcmu_cmd, XA_LIMIT(1, 0xffff),
1075                      GFP_NOWAIT) < 0) {
1076                 pr_err("tcmu: Could not allocate cmd id.\n");
1077
1078                 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
1079                 *scsi_err = TCM_OUT_OF_RESOURCES;
1080                 return -1;
1081         }
1082         tcmu_cmd->cmd_id = cmd_id;
1083
1084         pr_debug("allocated cmd id %u for cmd %p dev %s\n", tcmu_cmd->cmd_id,
1085                  tcmu_cmd, udev->name);
1086
1087         cmd_head = ring_insert_padding(udev, command_size);
1088
1089         entry = udev->cmdr + cmd_head;
1090         memset(entry, 0, command_size);
1091         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
1092
1093         /* prepare iov list and copy data to data area if necessary */
1094         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1095         iov = &entry->req.iov[0];
1096
1097         if (se_cmd->data_direction == DMA_TO_DEVICE ||
1098             se_cmd->se_cmd_flags & SCF_BIDI)
1099                 scatter_data_area(udev, tcmu_cmd, &iov);
1100         else
1101                 tcmu_setup_iovs(udev, tcmu_cmd, &iov, se_cmd->data_length);
1102
1103         entry->req.iov_cnt = iov_cnt - iov_bidi_cnt;
1104
1105         /* Handle BIDI commands */
1106         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1107                 iov++;
1108                 tcmu_setup_iovs(udev, tcmu_cmd, &iov, tcmu_cmd->data_len_bidi);
1109                 entry->req.iov_bidi_cnt = iov_bidi_cnt;
1110         }
1111
1112         tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out, &udev->cmd_timer);
1113
1114         entry->hdr.cmd_id = tcmu_cmd->cmd_id;
1115
1116         tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
1117
1118         /* All offsets relative to mb_addr, not start of entry! */
1119         cdb_off = CMDR_OFF + cmd_head + base_command_size;
1120         memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
1121         entry->req.cdb_off = cdb_off;
1122         tcmu_flush_dcache_range(entry, command_size);
1123
1124         UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1125         tcmu_flush_dcache_range(mb, sizeof(*mb));
1126
1127         list_add_tail(&tcmu_cmd->queue_entry, &udev->inflight_queue);
1128
1129         if (!test_bit(TCMU_DEV_BIT_PLUGGED, &udev->flags))
1130                 uio_event_notify(&udev->uio_info);
1131
1132         return 0;
1133
1134 free_and_queue:
1135         tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
1136         tcmu_cmd_reset_dbi_cur(tcmu_cmd);
1137
1138 queue:
1139         if (add_to_qfull_queue(tcmu_cmd)) {
1140                 *scsi_err = TCM_OUT_OF_RESOURCES;
1141                 return -1;
1142         }
1143
1144         return 1;
1145 }
1146
1147 /**
1148  * queue_tmr_ring - queue tmr info to ring or internally
1149  * @udev: related tcmu_dev
1150  * @tmr: tcmu_tmr containing tmr info to queue
1151  *
1152  * Returns:
1153  *  0 success
1154  *  1 internally queued to wait for ring memory to free.
1155  */
1156 static int
1157 queue_tmr_ring(struct tcmu_dev *udev, struct tcmu_tmr *tmr)
1158 {
1159         struct tcmu_tmr_entry *entry;
1160         int cmd_size;
1161         int id_list_sz;
1162         struct tcmu_mailbox *mb = udev->mb_addr;
1163         uint32_t cmd_head;
1164
1165         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
1166                 goto out_free;
1167
1168         id_list_sz = sizeof(tmr->tmr_cmd_ids[0]) * tmr->tmr_cmd_cnt;
1169         cmd_size = round_up(sizeof(*entry) + id_list_sz, TCMU_OP_ALIGN_SIZE);
1170
1171         if (!list_empty(&udev->tmr_queue) ||
1172             !is_ring_space_avail(udev, cmd_size)) {
1173                 list_add_tail(&tmr->queue_entry, &udev->tmr_queue);
1174                 pr_debug("adding tmr %p on dev %s to TMR ring space wait queue\n",
1175                          tmr, udev->name);
1176                 return 1;
1177         }
1178
1179         cmd_head = ring_insert_padding(udev, cmd_size);
1180
1181         entry = udev->cmdr + cmd_head;
1182         memset(entry, 0, cmd_size);
1183         tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_TMR);
1184         tcmu_hdr_set_len(&entry->hdr.len_op, cmd_size);
1185         entry->tmr_type = tmr->tmr_type;
1186         entry->cmd_cnt = tmr->tmr_cmd_cnt;
1187         memcpy(&entry->cmd_ids[0], &tmr->tmr_cmd_ids[0], id_list_sz);
1188         tcmu_flush_dcache_range(entry, cmd_size);
1189
1190         UPDATE_HEAD(mb->cmd_head, cmd_size, udev->cmdr_size);
1191         tcmu_flush_dcache_range(mb, sizeof(*mb));
1192
1193         uio_event_notify(&udev->uio_info);
1194
1195 out_free:
1196         kfree(tmr);
1197
1198         return 0;
1199 }
1200
1201 static sense_reason_t
1202 tcmu_queue_cmd(struct se_cmd *se_cmd)
1203 {
1204         struct se_device *se_dev = se_cmd->se_dev;
1205         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1206         struct tcmu_cmd *tcmu_cmd;
1207         sense_reason_t scsi_ret = TCM_CHECK_CONDITION_ABORT_CMD;
1208         int ret = -1;
1209
1210         tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1211         if (!tcmu_cmd)
1212                 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1213
1214         mutex_lock(&udev->cmdr_lock);
1215         if (!(se_cmd->transport_state & CMD_T_ABORTED))
1216                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1217         if (ret < 0)
1218                 tcmu_free_cmd(tcmu_cmd);
1219         else
1220                 se_cmd->priv = tcmu_cmd;
1221         mutex_unlock(&udev->cmdr_lock);
1222         return scsi_ret;
1223 }
1224
1225 static void tcmu_set_next_deadline(struct list_head *queue,
1226                                    struct timer_list *timer)
1227 {
1228         struct tcmu_cmd *cmd;
1229
1230         if (!list_empty(queue)) {
1231                 cmd = list_first_entry(queue, struct tcmu_cmd, queue_entry);
1232                 mod_timer(timer, cmd->deadline);
1233         } else
1234                 del_timer(timer);
1235 }
1236
1237 static int
1238 tcmu_tmr_type(enum tcm_tmreq_table tmf)
1239 {
1240         switch (tmf) {
1241         case TMR_ABORT_TASK:            return TCMU_TMR_ABORT_TASK;
1242         case TMR_ABORT_TASK_SET:        return TCMU_TMR_ABORT_TASK_SET;
1243         case TMR_CLEAR_ACA:             return TCMU_TMR_CLEAR_ACA;
1244         case TMR_CLEAR_TASK_SET:        return TCMU_TMR_CLEAR_TASK_SET;
1245         case TMR_LUN_RESET:             return TCMU_TMR_LUN_RESET;
1246         case TMR_TARGET_WARM_RESET:     return TCMU_TMR_TARGET_WARM_RESET;
1247         case TMR_TARGET_COLD_RESET:     return TCMU_TMR_TARGET_COLD_RESET;
1248         case TMR_LUN_RESET_PRO:         return TCMU_TMR_LUN_RESET_PRO;
1249         default:                        return TCMU_TMR_UNKNOWN;
1250         }
1251 }
1252
1253 static void
1254 tcmu_tmr_notify(struct se_device *se_dev, enum tcm_tmreq_table tmf,
1255                 struct list_head *cmd_list)
1256 {
1257         int i = 0, cmd_cnt = 0;
1258         bool unqueued = false;
1259         uint16_t *cmd_ids = NULL;
1260         struct tcmu_cmd *cmd;
1261         struct se_cmd *se_cmd;
1262         struct tcmu_tmr *tmr;
1263         struct tcmu_dev *udev = TCMU_DEV(se_dev);
1264
1265         mutex_lock(&udev->cmdr_lock);
1266
1267         /* First we check for aborted commands in qfull_queue */
1268         list_for_each_entry(se_cmd, cmd_list, state_list) {
1269                 i++;
1270                 if (!se_cmd->priv)
1271                         continue;
1272                 cmd = se_cmd->priv;
1273                 /* Commands on qfull queue have no id yet */
1274                 if (cmd->cmd_id) {
1275                         cmd_cnt++;
1276                         continue;
1277                 }
1278                 pr_debug("Removing aborted command %p from queue on dev %s.\n",
1279                          cmd, udev->name);
1280
1281                 list_del_init(&cmd->queue_entry);
1282                 tcmu_free_cmd(cmd);
1283                 se_cmd->priv = NULL;
1284                 target_complete_cmd(se_cmd, SAM_STAT_TASK_ABORTED);
1285                 unqueued = true;
1286         }
1287         if (unqueued)
1288                 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1289
1290         if (!test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags))
1291                 goto unlock;
1292
1293         pr_debug("TMR event %d on dev %s, aborted cmds %d, afflicted cmd_ids %d\n",
1294                  tcmu_tmr_type(tmf), udev->name, i, cmd_cnt);
1295
1296         tmr = kmalloc(sizeof(*tmr) + cmd_cnt * sizeof(*cmd_ids), GFP_NOIO);
1297         if (!tmr)
1298                 goto unlock;
1299
1300         tmr->tmr_type = tcmu_tmr_type(tmf);
1301         tmr->tmr_cmd_cnt = cmd_cnt;
1302
1303         if (cmd_cnt != 0) {
1304                 cmd_cnt = 0;
1305                 list_for_each_entry(se_cmd, cmd_list, state_list) {
1306                         if (!se_cmd->priv)
1307                                 continue;
1308                         cmd = se_cmd->priv;
1309                         if (cmd->cmd_id)
1310                                 tmr->tmr_cmd_ids[cmd_cnt++] = cmd->cmd_id;
1311                 }
1312         }
1313
1314         queue_tmr_ring(udev, tmr);
1315
1316 unlock:
1317         mutex_unlock(&udev->cmdr_lock);
1318 }
1319
1320 static bool tcmu_handle_completion(struct tcmu_cmd *cmd,
1321                                    struct tcmu_cmd_entry *entry, bool keep_buf)
1322 {
1323         struct se_cmd *se_cmd = cmd->se_cmd;
1324         struct tcmu_dev *udev = cmd->tcmu_dev;
1325         bool read_len_valid = false;
1326         bool ret = true;
1327         uint32_t read_len;
1328
1329         /*
1330          * cmd has been completed already from timeout, just reclaim
1331          * data area space and free cmd
1332          */
1333         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1334                 WARN_ON_ONCE(se_cmd);
1335                 goto out;
1336         }
1337         if (test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags)) {
1338                 pr_err("cmd_id %u already completed with KEEP_BUF, ring is broken\n",
1339                        entry->hdr.cmd_id);
1340                 set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1341                 ret = false;
1342                 goto out;
1343         }
1344
1345         list_del_init(&cmd->queue_entry);
1346
1347         tcmu_cmd_reset_dbi_cur(cmd);
1348
1349         if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1350                 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1351                         cmd->se_cmd);
1352                 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1353                 goto done;
1354         }
1355
1356         read_len = se_cmd->data_length;
1357         if (se_cmd->data_direction == DMA_FROM_DEVICE &&
1358             (entry->hdr.uflags & TCMU_UFLAG_READ_LEN) && entry->rsp.read_len) {
1359                 read_len_valid = true;
1360                 if (entry->rsp.read_len < read_len)
1361                         read_len = entry->rsp.read_len;
1362         }
1363
1364         if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1365                 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1366                 if (!read_len_valid )
1367                         goto done;
1368                 else
1369                         se_cmd->se_cmd_flags |= SCF_TREAT_READ_AS_NORMAL;
1370         }
1371         if (se_cmd->se_cmd_flags & SCF_BIDI) {
1372                 /* Get Data-In buffer before clean up */
1373                 gather_data_area(udev, cmd, true, read_len);
1374         } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1375                 gather_data_area(udev, cmd, false, read_len);
1376         } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1377                 /* TODO: */
1378         } else if (se_cmd->data_direction != DMA_NONE) {
1379                 pr_warn("TCMU: data direction was %d!\n",
1380                         se_cmd->data_direction);
1381         }
1382
1383 done:
1384         se_cmd->priv = NULL;
1385         if (read_len_valid) {
1386                 pr_debug("read_len = %d\n", read_len);
1387                 target_complete_cmd_with_length(cmd->se_cmd,
1388                                         entry->rsp.scsi_status, read_len);
1389         } else
1390                 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1391
1392 out:
1393         if (!keep_buf) {
1394                 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1395                 tcmu_free_cmd(cmd);
1396         } else {
1397                 /*
1398                  * Keep this command after completion, since userspace still
1399                  * needs the data buffer. Mark it with TCMU_CMD_BIT_KEEP_BUF
1400                  * and reset potential TCMU_CMD_BIT_EXPIRED, so we don't accept
1401                  * a second completion later.
1402                  * Userspace can free the buffer later by writing the cmd_id
1403                  * to new action attribute free_kept_buf.
1404                  */
1405                 clear_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1406                 set_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags);
1407         }
1408         return ret;
1409 }
1410
1411 static int tcmu_run_tmr_queue(struct tcmu_dev *udev)
1412 {
1413         struct tcmu_tmr *tmr, *tmp;
1414         LIST_HEAD(tmrs);
1415
1416         if (list_empty(&udev->tmr_queue))
1417                 return 1;
1418
1419         pr_debug("running %s's tmr queue\n", udev->name);
1420
1421         list_splice_init(&udev->tmr_queue, &tmrs);
1422
1423         list_for_each_entry_safe(tmr, tmp, &tmrs, queue_entry) {
1424                 list_del_init(&tmr->queue_entry);
1425
1426                 pr_debug("removing tmr %p on dev %s from queue\n",
1427                          tmr, udev->name);
1428
1429                 if (queue_tmr_ring(udev, tmr)) {
1430                         pr_debug("ran out of space during tmr queue run\n");
1431                         /*
1432                          * tmr was requeued, so just put all tmrs back in
1433                          * the queue
1434                          */
1435                         list_splice_tail(&tmrs, &udev->tmr_queue);
1436                         return 0;
1437                 }
1438         }
1439
1440         return 1;
1441 }
1442
1443 static bool tcmu_handle_completions(struct tcmu_dev *udev)
1444 {
1445         struct tcmu_mailbox *mb;
1446         struct tcmu_cmd *cmd;
1447         bool free_space = false;
1448
1449         if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1450                 pr_err("ring broken, not handling completions\n");
1451                 return false;
1452         }
1453
1454         mb = udev->mb_addr;
1455         tcmu_flush_dcache_range(mb, sizeof(*mb));
1456
1457         while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1458
1459                 struct tcmu_cmd_entry *entry = udev->cmdr + udev->cmdr_last_cleaned;
1460                 bool keep_buf;
1461
1462                 /*
1463                  * Flush max. up to end of cmd ring since current entry might
1464                  * be a padding that is shorter than sizeof(*entry)
1465                  */
1466                 size_t ring_left = head_to_end(udev->cmdr_last_cleaned,
1467                                                udev->cmdr_size);
1468                 tcmu_flush_dcache_range(entry, ring_left < sizeof(*entry) ?
1469                                         ring_left : sizeof(*entry));
1470
1471                 free_space = true;
1472
1473                 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD ||
1474                     tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_TMR) {
1475                         UPDATE_HEAD(udev->cmdr_last_cleaned,
1476                                     tcmu_hdr_get_len(entry->hdr.len_op),
1477                                     udev->cmdr_size);
1478                         continue;
1479                 }
1480                 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1481
1482                 keep_buf = !!(entry->hdr.uflags & TCMU_UFLAG_KEEP_BUF);
1483                 if (keep_buf)
1484                         cmd = xa_load(&udev->commands, entry->hdr.cmd_id);
1485                 else
1486                         cmd = xa_erase(&udev->commands, entry->hdr.cmd_id);
1487                 if (!cmd) {
1488                         pr_err("cmd_id %u not found, ring is broken\n",
1489                                entry->hdr.cmd_id);
1490                         set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1491                         return false;
1492                 }
1493
1494                 if (!tcmu_handle_completion(cmd, entry, keep_buf))
1495                         break;
1496
1497                 UPDATE_HEAD(udev->cmdr_last_cleaned,
1498                             tcmu_hdr_get_len(entry->hdr.len_op),
1499                             udev->cmdr_size);
1500         }
1501         if (free_space)
1502                 free_space = tcmu_run_tmr_queue(udev);
1503
1504         if (atomic_read(&global_page_count) > tcmu_global_max_pages &&
1505             xa_empty(&udev->commands) && list_empty(&udev->qfull_queue)) {
1506                 /*
1507                  * Allocated blocks exceeded global block limit, currently no
1508                  * more pending or waiting commands so try to reclaim blocks.
1509                  */
1510                 schedule_delayed_work(&tcmu_unmap_work, 0);
1511         }
1512         if (udev->cmd_time_out)
1513                 tcmu_set_next_deadline(&udev->inflight_queue, &udev->cmd_timer);
1514
1515         return free_space;
1516 }
1517
1518 static void tcmu_check_expired_ring_cmd(struct tcmu_cmd *cmd)
1519 {
1520         struct se_cmd *se_cmd;
1521
1522         if (!time_after_eq(jiffies, cmd->deadline))
1523                 return;
1524
1525         set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1526         list_del_init(&cmd->queue_entry);
1527         se_cmd = cmd->se_cmd;
1528         se_cmd->priv = NULL;
1529         cmd->se_cmd = NULL;
1530
1531         pr_debug("Timing out inflight cmd %u on dev %s.\n",
1532                  cmd->cmd_id, cmd->tcmu_dev->name);
1533
1534         target_complete_cmd(se_cmd, SAM_STAT_CHECK_CONDITION);
1535 }
1536
1537 static void tcmu_check_expired_queue_cmd(struct tcmu_cmd *cmd)
1538 {
1539         struct se_cmd *se_cmd;
1540
1541         if (!time_after_eq(jiffies, cmd->deadline))
1542                 return;
1543
1544         pr_debug("Timing out queued cmd %p on dev %s.\n",
1545                   cmd, cmd->tcmu_dev->name);
1546
1547         list_del_init(&cmd->queue_entry);
1548         se_cmd = cmd->se_cmd;
1549         tcmu_free_cmd(cmd);
1550
1551         se_cmd->priv = NULL;
1552         target_complete_cmd(se_cmd, SAM_STAT_TASK_SET_FULL);
1553 }
1554
1555 static void tcmu_device_timedout(struct tcmu_dev *udev)
1556 {
1557         spin_lock(&timed_out_udevs_lock);
1558         if (list_empty(&udev->timedout_entry))
1559                 list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1560         spin_unlock(&timed_out_udevs_lock);
1561
1562         schedule_delayed_work(&tcmu_unmap_work, 0);
1563 }
1564
1565 static void tcmu_cmd_timedout(struct timer_list *t)
1566 {
1567         struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1568
1569         pr_debug("%s cmd timeout has expired\n", udev->name);
1570         tcmu_device_timedout(udev);
1571 }
1572
1573 static void tcmu_qfull_timedout(struct timer_list *t)
1574 {
1575         struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1576
1577         pr_debug("%s qfull timeout has expired\n", udev->name);
1578         tcmu_device_timedout(udev);
1579 }
1580
1581 static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1582 {
1583         struct tcmu_hba *tcmu_hba;
1584
1585         tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1586         if (!tcmu_hba)
1587                 return -ENOMEM;
1588
1589         tcmu_hba->host_id = host_id;
1590         hba->hba_ptr = tcmu_hba;
1591
1592         return 0;
1593 }
1594
1595 static void tcmu_detach_hba(struct se_hba *hba)
1596 {
1597         kfree(hba->hba_ptr);
1598         hba->hba_ptr = NULL;
1599 }
1600
1601 static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1602 {
1603         struct tcmu_dev *udev;
1604
1605         udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1606         if (!udev)
1607                 return NULL;
1608         kref_init(&udev->kref);
1609
1610         udev->name = kstrdup(name, GFP_KERNEL);
1611         if (!udev->name) {
1612                 kfree(udev);
1613                 return NULL;
1614         }
1615
1616         udev->hba = hba;
1617         udev->cmd_time_out = TCMU_TIME_OUT;
1618         udev->qfull_time_out = -1;
1619
1620         udev->data_pages_per_blk = DATA_PAGES_PER_BLK_DEF;
1621         udev->max_blocks = DATA_AREA_PAGES_DEF / udev->data_pages_per_blk;
1622         udev->data_area_mb = TCMU_PAGES_TO_MBS(DATA_AREA_PAGES_DEF);
1623
1624         mutex_init(&udev->cmdr_lock);
1625
1626         INIT_LIST_HEAD(&udev->node);
1627         INIT_LIST_HEAD(&udev->timedout_entry);
1628         INIT_LIST_HEAD(&udev->qfull_queue);
1629         INIT_LIST_HEAD(&udev->tmr_queue);
1630         INIT_LIST_HEAD(&udev->inflight_queue);
1631         xa_init_flags(&udev->commands, XA_FLAGS_ALLOC1);
1632
1633         timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1634         timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1635
1636         xa_init(&udev->data_pages);
1637
1638         return &udev->se_dev;
1639 }
1640
1641 static void tcmu_dev_call_rcu(struct rcu_head *p)
1642 {
1643         struct se_device *dev = container_of(p, struct se_device, rcu_head);
1644         struct tcmu_dev *udev = TCMU_DEV(dev);
1645
1646         kfree(udev->uio_info.name);
1647         kfree(udev->name);
1648         kfree(udev);
1649 }
1650
1651 static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1652 {
1653         if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags) ||
1654             test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags)) {
1655                 kmem_cache_free(tcmu_cmd_cache, cmd);
1656                 return 0;
1657         }
1658         return -EINVAL;
1659 }
1660
1661 static u32 tcmu_blocks_release(struct tcmu_dev *udev, unsigned long first,
1662                                 unsigned long last)
1663 {
1664         struct page *page;
1665         unsigned long dpi;
1666         u32 pages_freed = 0;
1667
1668         first = first * udev->data_pages_per_blk;
1669         last = (last + 1) * udev->data_pages_per_blk - 1;
1670         xa_for_each_range(&udev->data_pages, dpi, page, first, last) {
1671                 xa_erase(&udev->data_pages, dpi);
1672                 /*
1673                  * While reaching here there may be page faults occurring on
1674                  * the to-be-released pages. A race condition may occur if
1675                  * unmap_mapping_range() is called before page faults on these
1676                  * pages have completed; a valid but stale map is created.
1677                  *
1678                  * If another command subsequently runs and needs to extend
1679                  * dbi_thresh, it may reuse the slot corresponding to the
1680                  * previous page in data_bitmap. Though we will allocate a new
1681                  * page for the slot in data_area, no page fault will happen
1682                  * because we have a valid map. Therefore the command's data
1683                  * will be lost.
1684                  *
1685                  * We lock and unlock pages that are to be released to ensure
1686                  * all page faults have completed. This way
1687                  * unmap_mapping_range() can ensure stale maps are cleanly
1688                  * removed.
1689                  */
1690                 lock_page(page);
1691                 unlock_page(page);
1692                 __free_page(page);
1693                 pages_freed++;
1694         }
1695
1696         atomic_sub(pages_freed, &global_page_count);
1697
1698         return pages_freed;
1699 }
1700
1701 static void tcmu_remove_all_queued_tmr(struct tcmu_dev *udev)
1702 {
1703         struct tcmu_tmr *tmr, *tmp;
1704
1705         list_for_each_entry_safe(tmr, tmp, &udev->tmr_queue, queue_entry) {
1706                 list_del_init(&tmr->queue_entry);
1707                 kfree(tmr);
1708         }
1709 }
1710
1711 static void tcmu_dev_kref_release(struct kref *kref)
1712 {
1713         struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1714         struct se_device *dev = &udev->se_dev;
1715         struct tcmu_cmd *cmd;
1716         bool all_expired = true;
1717         unsigned long i;
1718
1719         vfree(udev->mb_addr);
1720         udev->mb_addr = NULL;
1721
1722         spin_lock_bh(&timed_out_udevs_lock);
1723         if (!list_empty(&udev->timedout_entry))
1724                 list_del(&udev->timedout_entry);
1725         spin_unlock_bh(&timed_out_udevs_lock);
1726
1727         /* Upper layer should drain all requests before calling this */
1728         mutex_lock(&udev->cmdr_lock);
1729         xa_for_each(&udev->commands, i, cmd) {
1730                 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1731                         all_expired = false;
1732         }
1733         /* There can be left over TMR cmds. Remove them. */
1734         tcmu_remove_all_queued_tmr(udev);
1735         if (!list_empty(&udev->qfull_queue))
1736                 all_expired = false;
1737         xa_destroy(&udev->commands);
1738         WARN_ON(!all_expired);
1739
1740         tcmu_blocks_release(udev, 0, udev->dbi_max);
1741         bitmap_free(udev->data_bitmap);
1742         mutex_unlock(&udev->cmdr_lock);
1743
1744         pr_debug("dev_kref_release\n");
1745
1746         call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1747 }
1748
1749 static void run_qfull_queue(struct tcmu_dev *udev, bool fail)
1750 {
1751         struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1752         LIST_HEAD(cmds);
1753         sense_reason_t scsi_ret;
1754         int ret;
1755
1756         if (list_empty(&udev->qfull_queue))
1757                 return;
1758
1759         pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1760
1761         list_splice_init(&udev->qfull_queue, &cmds);
1762
1763         list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, queue_entry) {
1764                 list_del_init(&tcmu_cmd->queue_entry);
1765
1766                 pr_debug("removing cmd %p on dev %s from queue\n",
1767                          tcmu_cmd, udev->name);
1768
1769                 if (fail) {
1770                         /*
1771                          * We were not able to even start the command, so
1772                          * fail with busy to allow a retry in case runner
1773                          * was only temporarily down. If the device is being
1774                          * removed then LIO core will do the right thing and
1775                          * fail the retry.
1776                          */
1777                         tcmu_cmd->se_cmd->priv = NULL;
1778                         target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1779                         tcmu_free_cmd(tcmu_cmd);
1780                         continue;
1781                 }
1782
1783                 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1784                 if (ret < 0) {
1785                         pr_debug("cmd %p on dev %s failed with %u\n",
1786                                  tcmu_cmd, udev->name, scsi_ret);
1787                         /*
1788                          * Ignore scsi_ret for now. target_complete_cmd
1789                          * drops it.
1790                          */
1791                         tcmu_cmd->se_cmd->priv = NULL;
1792                         target_complete_cmd(tcmu_cmd->se_cmd,
1793                                             SAM_STAT_CHECK_CONDITION);
1794                         tcmu_free_cmd(tcmu_cmd);
1795                 } else if (ret > 0) {
1796                         pr_debug("ran out of space during cmdr queue run\n");
1797                         /*
1798                          * cmd was requeued, so just put all cmds back in
1799                          * the queue
1800                          */
1801                         list_splice_tail(&cmds, &udev->qfull_queue);
1802                         break;
1803                 }
1804         }
1805
1806         tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
1807 }
1808
1809 static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1810 {
1811         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1812
1813         mutex_lock(&udev->cmdr_lock);
1814         if (tcmu_handle_completions(udev))
1815                 run_qfull_queue(udev, false);
1816         mutex_unlock(&udev->cmdr_lock);
1817
1818         return 0;
1819 }
1820
1821 /*
1822  * mmap code from uio.c. Copied here because we want to hook mmap()
1823  * and this stuff must come along.
1824  */
1825 static int tcmu_find_mem_index(struct vm_area_struct *vma)
1826 {
1827         struct tcmu_dev *udev = vma->vm_private_data;
1828         struct uio_info *info = &udev->uio_info;
1829
1830         if (vma->vm_pgoff < MAX_UIO_MAPS) {
1831                 if (info->mem[vma->vm_pgoff].size == 0)
1832                         return -1;
1833                 return (int)vma->vm_pgoff;
1834         }
1835         return -1;
1836 }
1837
1838 static struct page *tcmu_try_get_data_page(struct tcmu_dev *udev, uint32_t dpi)
1839 {
1840         struct page *page;
1841
1842         mutex_lock(&udev->cmdr_lock);
1843         page = xa_load(&udev->data_pages, dpi);
1844         if (likely(page)) {
1845                 get_page(page);
1846                 lock_page(page);
1847                 mutex_unlock(&udev->cmdr_lock);
1848                 return page;
1849         }
1850
1851         /*
1852          * Userspace messed up and passed in a address not in the
1853          * data iov passed to it.
1854          */
1855         pr_err("Invalid addr to data page mapping (dpi %u) on device %s\n",
1856                dpi, udev->name);
1857         mutex_unlock(&udev->cmdr_lock);
1858
1859         return NULL;
1860 }
1861
1862 static void tcmu_vma_open(struct vm_area_struct *vma)
1863 {
1864         struct tcmu_dev *udev = vma->vm_private_data;
1865
1866         pr_debug("vma_open\n");
1867
1868         kref_get(&udev->kref);
1869 }
1870
1871 static void tcmu_vma_close(struct vm_area_struct *vma)
1872 {
1873         struct tcmu_dev *udev = vma->vm_private_data;
1874
1875         pr_debug("vma_close\n");
1876
1877         /* release ref from tcmu_vma_open */
1878         kref_put(&udev->kref, tcmu_dev_kref_release);
1879 }
1880
1881 static vm_fault_t tcmu_vma_fault(struct vm_fault *vmf)
1882 {
1883         struct tcmu_dev *udev = vmf->vma->vm_private_data;
1884         struct uio_info *info = &udev->uio_info;
1885         struct page *page;
1886         unsigned long offset;
1887         void *addr;
1888         vm_fault_t ret = 0;
1889
1890         int mi = tcmu_find_mem_index(vmf->vma);
1891         if (mi < 0)
1892                 return VM_FAULT_SIGBUS;
1893
1894         /*
1895          * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1896          * to use mem[N].
1897          */
1898         offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1899
1900         if (offset < udev->data_off) {
1901                 /* For the vmalloc()ed cmd area pages */
1902                 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1903                 page = vmalloc_to_page(addr);
1904                 get_page(page);
1905         } else {
1906                 uint32_t dpi;
1907
1908                 /* For the dynamically growing data area pages */
1909                 dpi = (offset - udev->data_off) / PAGE_SIZE;
1910                 page = tcmu_try_get_data_page(udev, dpi);
1911                 if (!page)
1912                         return VM_FAULT_SIGBUS;
1913                 ret = VM_FAULT_LOCKED;
1914         }
1915
1916         vmf->page = page;
1917         return ret;
1918 }
1919
1920 static const struct vm_operations_struct tcmu_vm_ops = {
1921         .open = tcmu_vma_open,
1922         .close = tcmu_vma_close,
1923         .fault = tcmu_vma_fault,
1924 };
1925
1926 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1927 {
1928         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1929
1930         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1931         vma->vm_ops = &tcmu_vm_ops;
1932
1933         vma->vm_private_data = udev;
1934
1935         /* Ensure the mmap is exactly the right size */
1936         if (vma_pages(vma) != udev->mmap_pages)
1937                 return -EINVAL;
1938
1939         tcmu_vma_open(vma);
1940
1941         return 0;
1942 }
1943
1944 static int tcmu_open(struct uio_info *info, struct inode *inode)
1945 {
1946         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1947
1948         /* O_EXCL not supported for char devs, so fake it? */
1949         if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1950                 return -EBUSY;
1951
1952         udev->inode = inode;
1953
1954         pr_debug("open\n");
1955
1956         return 0;
1957 }
1958
1959 static int tcmu_release(struct uio_info *info, struct inode *inode)
1960 {
1961         struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1962         struct tcmu_cmd *cmd;
1963         unsigned long i;
1964         bool freed = false;
1965
1966         mutex_lock(&udev->cmdr_lock);
1967
1968         xa_for_each(&udev->commands, i, cmd) {
1969                 /* Cmds with KEEP_BUF set are no longer on the ring, but
1970                  * userspace still holds the data buffer. If userspace closes
1971                  * we implicitly free these cmds and buffers, since after new
1972                  * open the (new ?) userspace cannot find the cmd in the ring
1973                  * and thus never will release the buffer by writing cmd_id to
1974                  * free_kept_buf action attribute.
1975                  */
1976                 if (!test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags))
1977                         continue;
1978                 pr_debug("removing KEEP_BUF cmd %u on dev %s from ring\n",
1979                          cmd->cmd_id, udev->name);
1980                 freed = true;
1981
1982                 xa_erase(&udev->commands, i);
1983                 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1984                 tcmu_free_cmd(cmd);
1985         }
1986         /*
1987          * We only freed data space, not ring space. Therefore we dont call
1988          * run_tmr_queue, but call run_qfull_queue if tmr_list is empty.
1989          */
1990         if (freed && list_empty(&udev->tmr_queue))
1991                 run_qfull_queue(udev, false);
1992
1993         mutex_unlock(&udev->cmdr_lock);
1994
1995         clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1996
1997         pr_debug("close\n");
1998
1999         return 0;
2000 }
2001
2002 static int tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
2003 {
2004         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
2005
2006         if (!tcmu_kern_cmd_reply_supported)
2007                 return 0;
2008
2009         if (udev->nl_reply_supported <= 0)
2010                 return 0;
2011
2012         mutex_lock(&tcmu_nl_cmd_mutex);
2013
2014         if (tcmu_netlink_blocked) {
2015                 mutex_unlock(&tcmu_nl_cmd_mutex);
2016                 pr_warn("Failing nl cmd %d on %s. Interface is blocked.\n", cmd,
2017                         udev->name);
2018                 return -EAGAIN;
2019         }
2020
2021         if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
2022                 mutex_unlock(&tcmu_nl_cmd_mutex);
2023                 pr_warn("netlink cmd %d already executing on %s\n",
2024                          nl_cmd->cmd, udev->name);
2025                 return -EBUSY;
2026         }
2027
2028         memset(nl_cmd, 0, sizeof(*nl_cmd));
2029         nl_cmd->cmd = cmd;
2030         nl_cmd->udev = udev;
2031         init_completion(&nl_cmd->complete);
2032         INIT_LIST_HEAD(&nl_cmd->nl_list);
2033
2034         list_add_tail(&nl_cmd->nl_list, &tcmu_nl_cmd_list);
2035
2036         mutex_unlock(&tcmu_nl_cmd_mutex);
2037         return 0;
2038 }
2039
2040 static void tcmu_destroy_genl_cmd_reply(struct tcmu_dev *udev)
2041 {
2042         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
2043
2044         if (!tcmu_kern_cmd_reply_supported)
2045                 return;
2046
2047         if (udev->nl_reply_supported <= 0)
2048                 return;
2049
2050         mutex_lock(&tcmu_nl_cmd_mutex);
2051
2052         list_del(&nl_cmd->nl_list);
2053         memset(nl_cmd, 0, sizeof(*nl_cmd));
2054
2055         mutex_unlock(&tcmu_nl_cmd_mutex);
2056 }
2057
2058 static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
2059 {
2060         struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
2061         int ret;
2062
2063         if (!tcmu_kern_cmd_reply_supported)
2064                 return 0;
2065
2066         if (udev->nl_reply_supported <= 0)
2067                 return 0;
2068
2069         pr_debug("sleeping for nl reply\n");
2070         wait_for_completion(&nl_cmd->complete);
2071
2072         mutex_lock(&tcmu_nl_cmd_mutex);
2073         nl_cmd->cmd = TCMU_CMD_UNSPEC;
2074         ret = nl_cmd->status;
2075         mutex_unlock(&tcmu_nl_cmd_mutex);
2076
2077         return ret;
2078 }
2079
2080 static int tcmu_netlink_event_init(struct tcmu_dev *udev,
2081                                    enum tcmu_genl_cmd cmd,
2082                                    struct sk_buff **buf, void **hdr)
2083 {
2084         struct sk_buff *skb;
2085         void *msg_header;
2086         int ret = -ENOMEM;
2087
2088         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
2089         if (!skb)
2090                 return ret;
2091
2092         msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
2093         if (!msg_header)
2094                 goto free_skb;
2095
2096         ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
2097         if (ret < 0)
2098                 goto free_skb;
2099
2100         ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
2101         if (ret < 0)
2102                 goto free_skb;
2103
2104         ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
2105         if (ret < 0)
2106                 goto free_skb;
2107
2108         *buf = skb;
2109         *hdr = msg_header;
2110         return ret;
2111
2112 free_skb:
2113         nlmsg_free(skb);
2114         return ret;
2115 }
2116
2117 static int tcmu_netlink_event_send(struct tcmu_dev *udev,
2118                                    enum tcmu_genl_cmd cmd,
2119                                    struct sk_buff *skb, void *msg_header)
2120 {
2121         int ret;
2122
2123         genlmsg_end(skb, msg_header);
2124
2125         ret = tcmu_init_genl_cmd_reply(udev, cmd);
2126         if (ret) {
2127                 nlmsg_free(skb);
2128                 return ret;
2129         }
2130
2131         ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
2132                                       TCMU_MCGRP_CONFIG, GFP_KERNEL);
2133
2134         /* Wait during an add as the listener may not be up yet */
2135         if (ret == 0 ||
2136            (ret == -ESRCH && cmd == TCMU_CMD_ADDED_DEVICE))
2137                 return tcmu_wait_genl_cmd_reply(udev);
2138         else
2139                 tcmu_destroy_genl_cmd_reply(udev);
2140
2141         return ret;
2142 }
2143
2144 static int tcmu_send_dev_add_event(struct tcmu_dev *udev)
2145 {
2146         struct sk_buff *skb = NULL;
2147         void *msg_header = NULL;
2148         int ret = 0;
2149
2150         ret = tcmu_netlink_event_init(udev, TCMU_CMD_ADDED_DEVICE, &skb,
2151                                       &msg_header);
2152         if (ret < 0)
2153                 return ret;
2154         return tcmu_netlink_event_send(udev, TCMU_CMD_ADDED_DEVICE, skb,
2155                                        msg_header);
2156 }
2157
2158 static int tcmu_send_dev_remove_event(struct tcmu_dev *udev)
2159 {
2160         struct sk_buff *skb = NULL;
2161         void *msg_header = NULL;
2162         int ret = 0;
2163
2164         ret = tcmu_netlink_event_init(udev, TCMU_CMD_REMOVED_DEVICE,
2165                                       &skb, &msg_header);
2166         if (ret < 0)
2167                 return ret;
2168         return tcmu_netlink_event_send(udev, TCMU_CMD_REMOVED_DEVICE,
2169                                        skb, msg_header);
2170 }
2171
2172 static int tcmu_update_uio_info(struct tcmu_dev *udev)
2173 {
2174         struct tcmu_hba *hba = udev->hba->hba_ptr;
2175         struct uio_info *info;
2176         char *str;
2177
2178         info = &udev->uio_info;
2179
2180         if (udev->dev_config[0])
2181                 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s/%s", hba->host_id,
2182                                 udev->name, udev->dev_config);
2183         else
2184                 str = kasprintf(GFP_KERNEL, "tcm-user/%u/%s", hba->host_id,
2185                                 udev->name);
2186         if (!str)
2187                 return -ENOMEM;
2188
2189         /* If the old string exists, free it */
2190         kfree(info->name);
2191         info->name = str;
2192
2193         return 0;
2194 }
2195
2196 static int tcmu_configure_device(struct se_device *dev)
2197 {
2198         struct tcmu_dev *udev = TCMU_DEV(dev);
2199         struct uio_info *info;
2200         struct tcmu_mailbox *mb;
2201         size_t data_size;
2202         int ret = 0;
2203
2204         ret = tcmu_update_uio_info(udev);
2205         if (ret)
2206                 return ret;
2207
2208         info = &udev->uio_info;
2209
2210         mutex_lock(&udev->cmdr_lock);
2211         udev->data_bitmap = bitmap_zalloc(udev->max_blocks, GFP_KERNEL);
2212         mutex_unlock(&udev->cmdr_lock);
2213         if (!udev->data_bitmap) {
2214                 ret = -ENOMEM;
2215                 goto err_bitmap_alloc;
2216         }
2217
2218         mb = vzalloc(MB_CMDR_SIZE);
2219         if (!mb) {
2220                 ret = -ENOMEM;
2221                 goto err_vzalloc;
2222         }
2223
2224         /* mailbox fits in first part of CMDR space */
2225         udev->mb_addr = mb;
2226         udev->cmdr = (void *)mb + CMDR_OFF;
2227         udev->cmdr_size = CMDR_SIZE;
2228         udev->data_off = MB_CMDR_SIZE;
2229         data_size = TCMU_MBS_TO_PAGES(udev->data_area_mb) << PAGE_SHIFT;
2230         udev->mmap_pages = (data_size + MB_CMDR_SIZE) >> PAGE_SHIFT;
2231         udev->data_blk_size = udev->data_pages_per_blk * PAGE_SIZE;
2232         udev->dbi_thresh = 0; /* Default in Idle state */
2233
2234         /* Initialise the mailbox of the ring buffer */
2235         mb->version = TCMU_MAILBOX_VERSION;
2236         mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC |
2237                     TCMU_MAILBOX_FLAG_CAP_READ_LEN |
2238                     TCMU_MAILBOX_FLAG_CAP_TMR |
2239                     TCMU_MAILBOX_FLAG_CAP_KEEP_BUF;
2240         mb->cmdr_off = CMDR_OFF;
2241         mb->cmdr_size = udev->cmdr_size;
2242
2243         WARN_ON(!PAGE_ALIGNED(udev->data_off));
2244         WARN_ON(data_size % PAGE_SIZE);
2245
2246         info->version = __stringify(TCMU_MAILBOX_VERSION);
2247
2248         info->mem[0].name = "tcm-user command & data buffer";
2249         info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
2250         info->mem[0].size = data_size + MB_CMDR_SIZE;
2251         info->mem[0].memtype = UIO_MEM_NONE;
2252
2253         info->irqcontrol = tcmu_irqcontrol;
2254         info->irq = UIO_IRQ_CUSTOM;
2255
2256         info->mmap = tcmu_mmap;
2257         info->open = tcmu_open;
2258         info->release = tcmu_release;
2259
2260         ret = uio_register_device(tcmu_root_device, info);
2261         if (ret)
2262                 goto err_register;
2263
2264         /* User can set hw_block_size before enable the device */
2265         if (dev->dev_attrib.hw_block_size == 0)
2266                 dev->dev_attrib.hw_block_size = 512;
2267         /* Other attributes can be configured in userspace */
2268         if (!dev->dev_attrib.hw_max_sectors)
2269                 dev->dev_attrib.hw_max_sectors = 128;
2270         if (!dev->dev_attrib.emulate_write_cache)
2271                 dev->dev_attrib.emulate_write_cache = 0;
2272         dev->dev_attrib.hw_queue_depth = 128;
2273
2274         /* If user didn't explicitly disable netlink reply support, use
2275          * module scope setting.
2276          */
2277         if (udev->nl_reply_supported >= 0)
2278                 udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
2279
2280         /*
2281          * Get a ref incase userspace does a close on the uio device before
2282          * LIO has initiated tcmu_free_device.
2283          */
2284         kref_get(&udev->kref);
2285
2286         ret = tcmu_send_dev_add_event(udev);
2287         if (ret)
2288                 goto err_netlink;
2289
2290         mutex_lock(&root_udev_mutex);
2291         list_add(&udev->node, &root_udev);
2292         mutex_unlock(&root_udev_mutex);
2293
2294         return 0;
2295
2296 err_netlink:
2297         kref_put(&udev->kref, tcmu_dev_kref_release);
2298         uio_unregister_device(&udev->uio_info);
2299 err_register:
2300         vfree(udev->mb_addr);
2301         udev->mb_addr = NULL;
2302 err_vzalloc:
2303         bitmap_free(udev->data_bitmap);
2304         udev->data_bitmap = NULL;
2305 err_bitmap_alloc:
2306         kfree(info->name);
2307         info->name = NULL;
2308
2309         return ret;
2310 }
2311
2312 static void tcmu_free_device(struct se_device *dev)
2313 {
2314         struct tcmu_dev *udev = TCMU_DEV(dev);
2315
2316         /* release ref from init */
2317         kref_put(&udev->kref, tcmu_dev_kref_release);
2318 }
2319
2320 static void tcmu_destroy_device(struct se_device *dev)
2321 {
2322         struct tcmu_dev *udev = TCMU_DEV(dev);
2323
2324         del_timer_sync(&udev->cmd_timer);
2325         del_timer_sync(&udev->qfull_timer);
2326
2327         mutex_lock(&root_udev_mutex);
2328         list_del(&udev->node);
2329         mutex_unlock(&root_udev_mutex);
2330
2331         tcmu_send_dev_remove_event(udev);
2332
2333         uio_unregister_device(&udev->uio_info);
2334
2335         /* release ref from configure */
2336         kref_put(&udev->kref, tcmu_dev_kref_release);
2337 }
2338
2339 static void tcmu_unblock_dev(struct tcmu_dev *udev)
2340 {
2341         mutex_lock(&udev->cmdr_lock);
2342         clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
2343         mutex_unlock(&udev->cmdr_lock);
2344 }
2345
2346 static void tcmu_block_dev(struct tcmu_dev *udev)
2347 {
2348         mutex_lock(&udev->cmdr_lock);
2349
2350         if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2351                 goto unlock;
2352
2353         /* complete IO that has executed successfully */
2354         tcmu_handle_completions(udev);
2355         /* fail IO waiting to be queued */
2356         run_qfull_queue(udev, true);
2357
2358 unlock:
2359         mutex_unlock(&udev->cmdr_lock);
2360 }
2361
2362 static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
2363 {
2364         struct tcmu_mailbox *mb;
2365         struct tcmu_cmd *cmd;
2366         unsigned long i;
2367
2368         mutex_lock(&udev->cmdr_lock);
2369
2370         xa_for_each(&udev->commands, i, cmd) {
2371                 pr_debug("removing cmd %u on dev %s from ring %s\n",
2372                          cmd->cmd_id, udev->name,
2373                          test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags) ?
2374                          "(is expired)" :
2375                          (test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags) ?
2376                          "(is keep buffer)" : ""));
2377
2378                 xa_erase(&udev->commands, i);
2379                 if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags) &&
2380                     !test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags)) {
2381                         WARN_ON(!cmd->se_cmd);
2382                         list_del_init(&cmd->queue_entry);
2383                         cmd->se_cmd->priv = NULL;
2384                         if (err_level == 1) {
2385                                 /*
2386                                  * Userspace was not able to start the
2387                                  * command or it is retryable.
2388                                  */
2389                                 target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
2390                         } else {
2391                                 /* hard failure */
2392                                 target_complete_cmd(cmd->se_cmd,
2393                                                     SAM_STAT_CHECK_CONDITION);
2394                         }
2395                 }
2396                 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
2397                 tcmu_free_cmd(cmd);
2398         }
2399
2400         mb = udev->mb_addr;
2401         tcmu_flush_dcache_range(mb, sizeof(*mb));
2402         pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
2403                  mb->cmd_tail, mb->cmd_head);
2404
2405         udev->cmdr_last_cleaned = 0;
2406         mb->cmd_tail = 0;
2407         mb->cmd_head = 0;
2408         tcmu_flush_dcache_range(mb, sizeof(*mb));
2409         clear_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
2410
2411         del_timer(&udev->cmd_timer);
2412
2413         /*
2414          * ring is empty and qfull queue never contains aborted commands.
2415          * So TMRs in tmr queue do not contain relevant cmd_ids.
2416          * After a ring reset userspace should do a fresh start, so
2417          * even LUN RESET message is no longer relevant.
2418          * Therefore remove all TMRs from qfull queue
2419          */
2420         tcmu_remove_all_queued_tmr(udev);
2421
2422         run_qfull_queue(udev, false);
2423
2424         mutex_unlock(&udev->cmdr_lock);
2425 }
2426
2427 enum {
2428         Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
2429         Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_data_pages_per_blk,
2430         Opt_err,
2431 };
2432
2433 static match_table_t tokens = {
2434         {Opt_dev_config, "dev_config=%s"},
2435         {Opt_dev_size, "dev_size=%s"},
2436         {Opt_hw_block_size, "hw_block_size=%d"},
2437         {Opt_hw_max_sectors, "hw_max_sectors=%d"},
2438         {Opt_nl_reply_supported, "nl_reply_supported=%d"},
2439         {Opt_max_data_area_mb, "max_data_area_mb=%d"},
2440         {Opt_data_pages_per_blk, "data_pages_per_blk=%d"},
2441         {Opt_err, NULL}
2442 };
2443
2444 static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
2445 {
2446         int val, ret;
2447
2448         ret = match_int(arg, &val);
2449         if (ret < 0) {
2450                 pr_err("match_int() failed for dev attrib. Error %d.\n",
2451                        ret);
2452                 return ret;
2453         }
2454
2455         if (val <= 0) {
2456                 pr_err("Invalid dev attrib value %d. Must be greater than zero.\n",
2457                        val);
2458                 return -EINVAL;
2459         }
2460         *dev_attrib = val;
2461         return 0;
2462 }
2463
2464 static int tcmu_set_max_blocks_param(struct tcmu_dev *udev, substring_t *arg)
2465 {
2466         int val, ret;
2467         uint32_t pages_per_blk = udev->data_pages_per_blk;
2468
2469         ret = match_int(arg, &val);
2470         if (ret < 0) {
2471                 pr_err("match_int() failed for max_data_area_mb=. Error %d.\n",
2472                        ret);
2473                 return ret;
2474         }
2475         if (val <= 0) {
2476                 pr_err("Invalid max_data_area %d.\n", val);
2477                 return -EINVAL;
2478         }
2479         if (val > TCMU_PAGES_TO_MBS(tcmu_global_max_pages)) {
2480                 pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2481                        val, TCMU_PAGES_TO_MBS(tcmu_global_max_pages));
2482                 val = TCMU_PAGES_TO_MBS(tcmu_global_max_pages);
2483         }
2484         if (TCMU_MBS_TO_PAGES(val) < pages_per_blk) {
2485                 pr_err("Invalid max_data_area %d (%zu pages): smaller than data_pages_per_blk (%u pages).\n",
2486                        val, TCMU_MBS_TO_PAGES(val), pages_per_blk);
2487                 return -EINVAL;
2488         }
2489
2490         mutex_lock(&udev->cmdr_lock);
2491         if (udev->data_bitmap) {
2492                 pr_err("Cannot set max_data_area_mb after it has been enabled.\n");
2493                 ret = -EINVAL;
2494                 goto unlock;
2495         }
2496
2497         udev->data_area_mb = val;
2498         udev->max_blocks = TCMU_MBS_TO_PAGES(val) / pages_per_blk;
2499
2500 unlock:
2501         mutex_unlock(&udev->cmdr_lock);
2502         return ret;
2503 }
2504
2505 static int tcmu_set_data_pages_per_blk(struct tcmu_dev *udev, substring_t *arg)
2506 {
2507         int val, ret;
2508
2509         ret = match_int(arg, &val);
2510         if (ret < 0) {
2511                 pr_err("match_int() failed for data_pages_per_blk=. Error %d.\n",
2512                        ret);
2513                 return ret;
2514         }
2515
2516         if (val > TCMU_MBS_TO_PAGES(udev->data_area_mb)) {
2517                 pr_err("Invalid data_pages_per_blk %d: greater than max_data_area_mb %d -> %zd pages).\n",
2518                        val, udev->data_area_mb,
2519                        TCMU_MBS_TO_PAGES(udev->data_area_mb));
2520                 return -EINVAL;
2521         }
2522
2523         mutex_lock(&udev->cmdr_lock);
2524         if (udev->data_bitmap) {
2525                 pr_err("Cannot set data_pages_per_blk after it has been enabled.\n");
2526                 ret = -EINVAL;
2527                 goto unlock;
2528         }
2529
2530         udev->data_pages_per_blk = val;
2531         udev->max_blocks = TCMU_MBS_TO_PAGES(udev->data_area_mb) / val;
2532
2533 unlock:
2534         mutex_unlock(&udev->cmdr_lock);
2535         return ret;
2536 }
2537
2538 static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
2539                 const char *page, ssize_t count)
2540 {
2541         struct tcmu_dev *udev = TCMU_DEV(dev);
2542         char *orig, *ptr, *opts;
2543         substring_t args[MAX_OPT_ARGS];
2544         int ret = 0, token;
2545
2546         opts = kstrdup(page, GFP_KERNEL);
2547         if (!opts)
2548                 return -ENOMEM;
2549
2550         orig = opts;
2551
2552         while ((ptr = strsep(&opts, ",\n")) != NULL) {
2553                 if (!*ptr)
2554                         continue;
2555
2556                 token = match_token(ptr, tokens, args);
2557                 switch (token) {
2558                 case Opt_dev_config:
2559                         if (match_strlcpy(udev->dev_config, &args[0],
2560                                           TCMU_CONFIG_LEN) == 0) {
2561                                 ret = -EINVAL;
2562                                 break;
2563                         }
2564                         pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
2565                         break;
2566                 case Opt_dev_size:
2567                         ret = match_u64(&args[0], &udev->dev_size);
2568                         if (ret < 0)
2569                                 pr_err("match_u64() failed for dev_size=. Error %d.\n",
2570                                        ret);
2571                         break;
2572                 case Opt_hw_block_size:
2573                         ret = tcmu_set_dev_attrib(&args[0],
2574                                         &(dev->dev_attrib.hw_block_size));
2575                         break;
2576                 case Opt_hw_max_sectors:
2577                         ret = tcmu_set_dev_attrib(&args[0],
2578                                         &(dev->dev_attrib.hw_max_sectors));
2579                         break;
2580                 case Opt_nl_reply_supported:
2581                         ret = match_int(&args[0], &udev->nl_reply_supported);
2582                         if (ret < 0)
2583                                 pr_err("match_int() failed for nl_reply_supported=. Error %d.\n",
2584                                        ret);
2585                         break;
2586                 case Opt_max_data_area_mb:
2587                         ret = tcmu_set_max_blocks_param(udev, &args[0]);
2588                         break;
2589                 case Opt_data_pages_per_blk:
2590                         ret = tcmu_set_data_pages_per_blk(udev, &args[0]);
2591                         break;
2592                 default:
2593                         break;
2594                 }
2595
2596                 if (ret)
2597                         break;
2598         }
2599
2600         kfree(orig);
2601         return (!ret) ? count : ret;
2602 }
2603
2604 static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2605 {
2606         struct tcmu_dev *udev = TCMU_DEV(dev);
2607         ssize_t bl = 0;
2608
2609         bl = sprintf(b + bl, "Config: %s ",
2610                      udev->dev_config[0] ? udev->dev_config : "NULL");
2611         bl += sprintf(b + bl, "Size: %llu ", udev->dev_size);
2612         bl += sprintf(b + bl, "MaxDataAreaMB: %u ", udev->data_area_mb);
2613         bl += sprintf(b + bl, "DataPagesPerBlk: %u\n", udev->data_pages_per_blk);
2614
2615         return bl;
2616 }
2617
2618 static sector_t tcmu_get_blocks(struct se_device *dev)
2619 {
2620         struct tcmu_dev *udev = TCMU_DEV(dev);
2621
2622         return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2623                        dev->dev_attrib.block_size);
2624 }
2625
2626 static sense_reason_t
2627 tcmu_parse_cdb(struct se_cmd *cmd)
2628 {
2629         return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2630 }
2631
2632 static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2633 {
2634         struct se_dev_attrib *da = container_of(to_config_group(item),
2635                                         struct se_dev_attrib, da_group);
2636         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2637
2638         return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2639 }
2640
2641 static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2642                                        size_t count)
2643 {
2644         struct se_dev_attrib *da = container_of(to_config_group(item),
2645                                         struct se_dev_attrib, da_group);
2646         struct tcmu_dev *udev = container_of(da->da_dev,
2647                                         struct tcmu_dev, se_dev);
2648         u32 val;
2649         int ret;
2650
2651         if (da->da_dev->export_count) {
2652                 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2653                 return -EINVAL;
2654         }
2655
2656         ret = kstrtou32(page, 0, &val);
2657         if (ret < 0)
2658                 return ret;
2659
2660         udev->cmd_time_out = val * MSEC_PER_SEC;
2661         return count;
2662 }
2663 CONFIGFS_ATTR(tcmu_, cmd_time_out);
2664
2665 static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2666 {
2667         struct se_dev_attrib *da = container_of(to_config_group(item),
2668                                                 struct se_dev_attrib, da_group);
2669         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2670
2671         return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2672                         udev->qfull_time_out :
2673                         udev->qfull_time_out / MSEC_PER_SEC);
2674 }
2675
2676 static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2677                                          const char *page, size_t count)
2678 {
2679         struct se_dev_attrib *da = container_of(to_config_group(item),
2680                                         struct se_dev_attrib, da_group);
2681         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2682         s32 val;
2683         int ret;
2684
2685         ret = kstrtos32(page, 0, &val);
2686         if (ret < 0)
2687                 return ret;
2688
2689         if (val >= 0) {
2690                 udev->qfull_time_out = val * MSEC_PER_SEC;
2691         } else if (val == -1) {
2692                 udev->qfull_time_out = val;
2693         } else {
2694                 printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2695                 return -EINVAL;
2696         }
2697         return count;
2698 }
2699 CONFIGFS_ATTR(tcmu_, qfull_time_out);
2700
2701 static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2702 {
2703         struct se_dev_attrib *da = container_of(to_config_group(item),
2704                                                 struct se_dev_attrib, da_group);
2705         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2706
2707         return snprintf(page, PAGE_SIZE, "%u\n", udev->data_area_mb);
2708 }
2709 CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2710
2711 static ssize_t tcmu_data_pages_per_blk_show(struct config_item *item,
2712                                             char *page)
2713 {
2714         struct se_dev_attrib *da = container_of(to_config_group(item),
2715                                                 struct se_dev_attrib, da_group);
2716         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2717
2718         return snprintf(page, PAGE_SIZE, "%u\n", udev->data_pages_per_blk);
2719 }
2720 CONFIGFS_ATTR_RO(tcmu_, data_pages_per_blk);
2721
2722 static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2723 {
2724         struct se_dev_attrib *da = container_of(to_config_group(item),
2725                                                 struct se_dev_attrib, da_group);
2726         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2727
2728         return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2729 }
2730
2731 static int tcmu_send_dev_config_event(struct tcmu_dev *udev,
2732                                       const char *reconfig_data)
2733 {
2734         struct sk_buff *skb = NULL;
2735         void *msg_header = NULL;
2736         int ret = 0;
2737
2738         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2739                                       &skb, &msg_header);
2740         if (ret < 0)
2741                 return ret;
2742         ret = nla_put_string(skb, TCMU_ATTR_DEV_CFG, reconfig_data);
2743         if (ret < 0) {
2744                 nlmsg_free(skb);
2745                 return ret;
2746         }
2747         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2748                                        skb, msg_header);
2749 }
2750
2751
2752 static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2753                                      size_t count)
2754 {
2755         struct se_dev_attrib *da = container_of(to_config_group(item),
2756                                                 struct se_dev_attrib, da_group);
2757         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2758         int ret, len;
2759
2760         len = strlen(page);
2761         if (!len || len > TCMU_CONFIG_LEN - 1)
2762                 return -EINVAL;
2763
2764         /* Check if device has been configured before */
2765         if (target_dev_configured(&udev->se_dev)) {
2766                 ret = tcmu_send_dev_config_event(udev, page);
2767                 if (ret) {
2768                         pr_err("Unable to reconfigure device\n");
2769                         return ret;
2770                 }
2771                 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2772
2773                 ret = tcmu_update_uio_info(udev);
2774                 if (ret)
2775                         return ret;
2776                 return count;
2777         }
2778         strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2779
2780         return count;
2781 }
2782 CONFIGFS_ATTR(tcmu_, dev_config);
2783
2784 static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2785 {
2786         struct se_dev_attrib *da = container_of(to_config_group(item),
2787                                                 struct se_dev_attrib, da_group);
2788         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2789
2790         return snprintf(page, PAGE_SIZE, "%llu\n", udev->dev_size);
2791 }
2792
2793 static int tcmu_send_dev_size_event(struct tcmu_dev *udev, u64 size)
2794 {
2795         struct sk_buff *skb = NULL;
2796         void *msg_header = NULL;
2797         int ret = 0;
2798
2799         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2800                                       &skb, &msg_header);
2801         if (ret < 0)
2802                 return ret;
2803         ret = nla_put_u64_64bit(skb, TCMU_ATTR_DEV_SIZE,
2804                                 size, TCMU_ATTR_PAD);
2805         if (ret < 0) {
2806                 nlmsg_free(skb);
2807                 return ret;
2808         }
2809         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2810                                        skb, msg_header);
2811 }
2812
2813 static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2814                                    size_t count)
2815 {
2816         struct se_dev_attrib *da = container_of(to_config_group(item),
2817                                                 struct se_dev_attrib, da_group);
2818         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2819         u64 val;
2820         int ret;
2821
2822         ret = kstrtou64(page, 0, &val);
2823         if (ret < 0)
2824                 return ret;
2825
2826         /* Check if device has been configured before */
2827         if (target_dev_configured(&udev->se_dev)) {
2828                 ret = tcmu_send_dev_size_event(udev, val);
2829                 if (ret) {
2830                         pr_err("Unable to reconfigure device\n");
2831                         return ret;
2832                 }
2833         }
2834         udev->dev_size = val;
2835         return count;
2836 }
2837 CONFIGFS_ATTR(tcmu_, dev_size);
2838
2839 static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
2840                 char *page)
2841 {
2842         struct se_dev_attrib *da = container_of(to_config_group(item),
2843                                                 struct se_dev_attrib, da_group);
2844         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2845
2846         return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2847 }
2848
2849 static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2850                 const char *page, size_t count)
2851 {
2852         struct se_dev_attrib *da = container_of(to_config_group(item),
2853                                                 struct se_dev_attrib, da_group);
2854         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2855         s8 val;
2856         int ret;
2857
2858         ret = kstrtos8(page, 0, &val);
2859         if (ret < 0)
2860                 return ret;
2861
2862         udev->nl_reply_supported = val;
2863         return count;
2864 }
2865 CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2866
2867 static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2868                                              char *page)
2869 {
2870         struct se_dev_attrib *da = container_of(to_config_group(item),
2871                                         struct se_dev_attrib, da_group);
2872
2873         return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2874 }
2875
2876 static int tcmu_send_emulate_write_cache(struct tcmu_dev *udev, u8 val)
2877 {
2878         struct sk_buff *skb = NULL;
2879         void *msg_header = NULL;
2880         int ret = 0;
2881
2882         ret = tcmu_netlink_event_init(udev, TCMU_CMD_RECONFIG_DEVICE,
2883                                       &skb, &msg_header);
2884         if (ret < 0)
2885                 return ret;
2886         ret = nla_put_u8(skb, TCMU_ATTR_WRITECACHE, val);
2887         if (ret < 0) {
2888                 nlmsg_free(skb);
2889                 return ret;
2890         }
2891         return tcmu_netlink_event_send(udev, TCMU_CMD_RECONFIG_DEVICE,
2892                                        skb, msg_header);
2893 }
2894
2895 static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2896                                               const char *page, size_t count)
2897 {
2898         struct se_dev_attrib *da = container_of(to_config_group(item),
2899                                         struct se_dev_attrib, da_group);
2900         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2901         u8 val;
2902         int ret;
2903
2904         ret = kstrtou8(page, 0, &val);
2905         if (ret < 0)
2906                 return ret;
2907
2908         /* Check if device has been configured before */
2909         if (target_dev_configured(&udev->se_dev)) {
2910                 ret = tcmu_send_emulate_write_cache(udev, val);
2911                 if (ret) {
2912                         pr_err("Unable to reconfigure device\n");
2913                         return ret;
2914                 }
2915         }
2916
2917         da->emulate_write_cache = val;
2918         return count;
2919 }
2920 CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2921
2922 static ssize_t tcmu_tmr_notification_show(struct config_item *item, char *page)
2923 {
2924         struct se_dev_attrib *da = container_of(to_config_group(item),
2925                                         struct se_dev_attrib, da_group);
2926         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2927
2928         return snprintf(page, PAGE_SIZE, "%i\n",
2929                         test_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags));
2930 }
2931
2932 static ssize_t tcmu_tmr_notification_store(struct config_item *item,
2933                                            const char *page, size_t count)
2934 {
2935         struct se_dev_attrib *da = container_of(to_config_group(item),
2936                                         struct se_dev_attrib, da_group);
2937         struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2938         u8 val;
2939         int ret;
2940
2941         ret = kstrtou8(page, 0, &val);
2942         if (ret < 0)
2943                 return ret;
2944         if (val > 1)
2945                 return -EINVAL;
2946
2947         if (val)
2948                 set_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
2949         else
2950                 clear_bit(TCMU_DEV_BIT_TMR_NOTIFY, &udev->flags);
2951         return count;
2952 }
2953 CONFIGFS_ATTR(tcmu_, tmr_notification);
2954
2955 static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
2956 {
2957         struct se_device *se_dev = container_of(to_config_group(item),
2958                                                 struct se_device,
2959                                                 dev_action_group);
2960         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2961
2962         if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2963                 return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
2964         else
2965                 return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
2966 }
2967
2968 static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
2969                                     size_t count)
2970 {
2971         struct se_device *se_dev = container_of(to_config_group(item),
2972                                                 struct se_device,
2973                                                 dev_action_group);
2974         struct tcmu_dev *udev = TCMU_DEV(se_dev);
2975         u8 val;
2976         int ret;
2977
2978         if (!target_dev_configured(&udev->se_dev)) {
2979                 pr_err("Device is not configured.\n");
2980                 return -EINVAL;
2981         }
2982
2983         ret = kstrtou8(page, 0, &val);
2984         if (ret < 0)
2985                 return ret;
2986
2987         if (val > 1) {
2988                 pr_err("Invalid block value %d\n", val);
2989                 return -EINVAL;
2990         }
2991
2992         if (!val)
2993                 tcmu_unblock_dev(udev);
2994         else
2995                 tcmu_block_dev(udev);
2996         return count;
2997 }
2998 CONFIGFS_ATTR(tcmu_, block_dev);
2999
3000 static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
3001                                      size_t count)
3002 {
3003         struct se_device *se_dev = container_of(to_config_group(item),
3004                                                 struct se_device,
3005                                                 dev_action_group);
3006         struct tcmu_dev *udev = TCMU_DEV(se_dev);
3007         u8 val;
3008         int ret;
3009
3010         if (!target_dev_configured(&udev->se_dev)) {
3011                 pr_err("Device is not configured.\n");
3012                 return -EINVAL;
3013         }
3014
3015         ret = kstrtou8(page, 0, &val);
3016         if (ret < 0)
3017                 return ret;
3018
3019         if (val != 1 && val != 2) {
3020                 pr_err("Invalid reset ring value %d\n", val);
3021                 return -EINVAL;
3022         }
3023
3024         tcmu_reset_ring(udev, val);
3025         return count;
3026 }
3027 CONFIGFS_ATTR_WO(tcmu_, reset_ring);
3028
3029 static ssize_t tcmu_free_kept_buf_store(struct config_item *item, const char *page,
3030                                         size_t count)
3031 {
3032         struct se_device *se_dev = container_of(to_config_group(item),
3033                                                 struct se_device,
3034                                                 dev_action_group);
3035         struct tcmu_dev *udev = TCMU_DEV(se_dev);
3036         struct tcmu_cmd *cmd;
3037         u16 cmd_id;
3038         int ret;
3039
3040         if (!target_dev_configured(&udev->se_dev)) {
3041                 pr_err("Device is not configured.\n");
3042                 return -EINVAL;
3043         }
3044
3045         ret = kstrtou16(page, 0, &cmd_id);
3046         if (ret < 0)
3047                 return ret;
3048
3049         mutex_lock(&udev->cmdr_lock);
3050
3051         {
3052                 XA_STATE(xas, &udev->commands, cmd_id);
3053
3054                 xas_lock(&xas);
3055                 cmd = xas_load(&xas);
3056                 if (!cmd) {
3057                         pr_err("free_kept_buf: cmd_id %d not found\n", cmd_id);
3058                         count = -EINVAL;
3059                         xas_unlock(&xas);
3060                         goto out_unlock;
3061                 }
3062                 if (!test_bit(TCMU_CMD_BIT_KEEP_BUF, &cmd->flags)) {
3063                         pr_err("free_kept_buf: cmd_id %d was not completed with KEEP_BUF\n",
3064                                cmd_id);
3065                         count = -EINVAL;
3066                         xas_unlock(&xas);
3067                         goto out_unlock;
3068                 }
3069                 xas_store(&xas, NULL);
3070                 xas_unlock(&xas);
3071         }
3072
3073         tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
3074         tcmu_free_cmd(cmd);
3075         /*
3076          * We only freed data space, not ring space. Therefore we dont call
3077          * run_tmr_queue, but call run_qfull_queue if tmr_list is empty.
3078          */
3079         if (list_empty(&udev->tmr_queue))
3080                 run_qfull_queue(udev, false);
3081
3082 out_unlock:
3083         mutex_unlock(&udev->cmdr_lock);
3084         return count;
3085 }
3086 CONFIGFS_ATTR_WO(tcmu_, free_kept_buf);
3087
3088 static struct configfs_attribute *tcmu_attrib_attrs[] = {
3089         &tcmu_attr_cmd_time_out,
3090         &tcmu_attr_qfull_time_out,
3091         &tcmu_attr_max_data_area_mb,
3092         &tcmu_attr_data_pages_per_blk,
3093         &tcmu_attr_dev_config,
3094         &tcmu_attr_dev_size,
3095         &tcmu_attr_emulate_write_cache,
3096         &tcmu_attr_tmr_notification,
3097         &tcmu_attr_nl_reply_supported,
3098         NULL,
3099 };
3100
3101 static struct configfs_attribute **tcmu_attrs;
3102
3103 static struct configfs_attribute *tcmu_action_attrs[] = {
3104         &tcmu_attr_block_dev,
3105         &tcmu_attr_reset_ring,
3106         &tcmu_attr_free_kept_buf,
3107         NULL,
3108 };
3109
3110 static struct target_backend_ops tcmu_ops = {
3111         .name                   = "user",
3112         .owner                  = THIS_MODULE,
3113         .transport_flags_default = TRANSPORT_FLAG_PASSTHROUGH,
3114         .transport_flags_changeable = TRANSPORT_FLAG_PASSTHROUGH_PGR |
3115                                       TRANSPORT_FLAG_PASSTHROUGH_ALUA,
3116         .attach_hba             = tcmu_attach_hba,
3117         .detach_hba             = tcmu_detach_hba,
3118         .alloc_device           = tcmu_alloc_device,
3119         .configure_device       = tcmu_configure_device,
3120         .destroy_device         = tcmu_destroy_device,
3121         .free_device            = tcmu_free_device,
3122         .unplug_device          = tcmu_unplug_device,
3123         .plug_device            = tcmu_plug_device,
3124         .parse_cdb              = tcmu_parse_cdb,
3125         .tmr_notify             = tcmu_tmr_notify,
3126         .set_configfs_dev_params = tcmu_set_configfs_dev_params,
3127         .show_configfs_dev_params = tcmu_show_configfs_dev_params,
3128         .get_device_type        = sbc_get_device_type,
3129         .get_blocks             = tcmu_get_blocks,
3130         .tb_dev_action_attrs    = tcmu_action_attrs,
3131 };
3132
3133 static void find_free_blocks(void)
3134 {
3135         struct tcmu_dev *udev;
3136         loff_t off;
3137         u32 pages_freed, total_pages_freed = 0;
3138         u32 start, end, block, total_blocks_freed = 0;
3139
3140         if (atomic_read(&global_page_count) <= tcmu_global_max_pages)
3141                 return;
3142
3143         mutex_lock(&root_udev_mutex);
3144         list_for_each_entry(udev, &root_udev, node) {
3145                 mutex_lock(&udev->cmdr_lock);
3146
3147                 if (!target_dev_configured(&udev->se_dev)) {
3148                         mutex_unlock(&udev->cmdr_lock);
3149                         continue;
3150                 }
3151
3152                 /* Try to complete the finished commands first */
3153                 if (tcmu_handle_completions(udev))
3154                         run_qfull_queue(udev, false);
3155
3156                 /* Skip the udevs in idle */
3157                 if (!udev->dbi_thresh) {
3158                         mutex_unlock(&udev->cmdr_lock);
3159                         continue;
3160                 }
3161
3162                 end = udev->dbi_max + 1;
3163                 block = find_last_bit(udev->data_bitmap, end);
3164                 if (block == udev->dbi_max) {
3165                         /*
3166                          * The last bit is dbi_max, so it is not possible
3167                          * reclaim any blocks.
3168                          */
3169                         mutex_unlock(&udev->cmdr_lock);
3170                         continue;
3171                 } else if (block == end) {
3172                         /* The current udev will goto idle state */
3173                         udev->dbi_thresh = start = 0;
3174                         udev->dbi_max = 0;
3175                 } else {
3176                         udev->dbi_thresh = start = block + 1;
3177                         udev->dbi_max = block;
3178                 }
3179
3180                 /*
3181                  * Release the block pages.
3182                  *
3183                  * Also note that since tcmu_vma_fault() gets an extra page
3184                  * refcount, tcmu_blocks_release() won't free pages if pages
3185                  * are mapped. This means it is safe to call
3186                  * tcmu_blocks_release() before unmap_mapping_range() which
3187                  * drops the refcount of any pages it unmaps and thus releases
3188                  * them.
3189                  */
3190                 pages_freed = tcmu_blocks_release(udev, start, end - 1);
3191
3192                 /* Here will truncate the data area from off */
3193                 off = udev->data_off + (loff_t)start * udev->data_blk_size;
3194                 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
3195
3196                 mutex_unlock(&udev->cmdr_lock);
3197
3198                 total_pages_freed += pages_freed;
3199                 total_blocks_freed += end - start;
3200                 pr_debug("Freed %u pages (total %u) from %u blocks (total %u) from %s.\n",
3201                          pages_freed, total_pages_freed, end - start,
3202                          total_blocks_freed, udev->name);
3203         }
3204         mutex_unlock(&root_udev_mutex);
3205
3206         if (atomic_read(&global_page_count) > tcmu_global_max_pages)
3207                 schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
3208 }
3209
3210 static void check_timedout_devices(void)
3211 {
3212         struct tcmu_dev *udev, *tmp_dev;
3213         struct tcmu_cmd *cmd, *tmp_cmd;
3214         LIST_HEAD(devs);
3215
3216         spin_lock_bh(&timed_out_udevs_lock);
3217         list_splice_init(&timed_out_udevs, &devs);
3218
3219         list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
3220                 list_del_init(&udev->timedout_entry);
3221                 spin_unlock_bh(&timed_out_udevs_lock);
3222
3223                 mutex_lock(&udev->cmdr_lock);
3224
3225                 /*
3226                  * If cmd_time_out is disabled but qfull is set deadline
3227                  * will only reflect the qfull timeout. Ignore it.
3228                  */
3229                 if (udev->cmd_time_out) {
3230                         list_for_each_entry_safe(cmd, tmp_cmd,
3231                                                  &udev->inflight_queue,
3232                                                  queue_entry) {
3233                                 tcmu_check_expired_ring_cmd(cmd);
3234                         }
3235                         tcmu_set_next_deadline(&udev->inflight_queue,
3236                                                &udev->cmd_timer);
3237                 }
3238                 list_for_each_entry_safe(cmd, tmp_cmd, &udev->qfull_queue,
3239                                          queue_entry) {
3240                         tcmu_check_expired_queue_cmd(cmd);
3241                 }
3242                 tcmu_set_next_deadline(&udev->qfull_queue, &udev->qfull_timer);
3243
3244                 mutex_unlock(&udev->cmdr_lock);
3245
3246                 spin_lock_bh(&timed_out_udevs_lock);
3247         }
3248
3249         spin_unlock_bh(&timed_out_udevs_lock);
3250 }
3251
3252 static void tcmu_unmap_work_fn(struct work_struct *work)
3253 {
3254         check_timedout_devices();
3255         find_free_blocks();
3256 }
3257
3258 static int __init tcmu_module_init(void)
3259 {
3260         int ret, i, k, len = 0;
3261
3262         BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
3263
3264         INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
3265
3266         tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
3267                                 sizeof(struct tcmu_cmd),
3268                                 __alignof__(struct tcmu_cmd),
3269                                 0, NULL);
3270         if (!tcmu_cmd_cache)
3271                 return -ENOMEM;
3272
3273         tcmu_root_device = root_device_register("tcm_user");
3274         if (IS_ERR(tcmu_root_device)) {
3275                 ret = PTR_ERR(tcmu_root_device);
3276                 goto out_free_cache;
3277         }
3278
3279         ret = genl_register_family(&tcmu_genl_family);
3280         if (ret < 0) {
3281                 goto out_unreg_device;
3282         }
3283
3284         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3285                 len += sizeof(struct configfs_attribute *);
3286         for (i = 0; passthrough_pr_attrib_attrs[i] != NULL; i++)
3287                 len += sizeof(struct configfs_attribute *);
3288         for (i = 0; tcmu_attrib_attrs[i] != NULL; i++)
3289                 len += sizeof(struct configfs_attribute *);
3290         len += sizeof(struct configfs_attribute *);
3291
3292         tcmu_attrs = kzalloc(len, GFP_KERNEL);
3293         if (!tcmu_attrs) {
3294                 ret = -ENOMEM;
3295                 goto out_unreg_genl;
3296         }
3297
3298         for (i = 0; passthrough_attrib_attrs[i] != NULL; i++)
3299                 tcmu_attrs[i] = passthrough_attrib_attrs[i];
3300         for (k = 0; passthrough_pr_attrib_attrs[k] != NULL; k++)
3301                 tcmu_attrs[i++] = passthrough_pr_attrib_attrs[k];
3302         for (k = 0; tcmu_attrib_attrs[k] != NULL; k++)
3303                 tcmu_attrs[i++] = tcmu_attrib_attrs[k];
3304         tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
3305
3306         ret = transport_backend_register(&tcmu_ops);
3307         if (ret)
3308                 goto out_attrs;
3309
3310         return 0;
3311
3312 out_attrs:
3313         kfree(tcmu_attrs);
3314 out_unreg_genl:
3315         genl_unregister_family(&tcmu_genl_family);
3316 out_unreg_device:
3317         root_device_unregister(tcmu_root_device);
3318 out_free_cache:
3319         kmem_cache_destroy(tcmu_cmd_cache);
3320
3321         return ret;
3322 }
3323
3324 static void __exit tcmu_module_exit(void)
3325 {
3326         cancel_delayed_work_sync(&tcmu_unmap_work);
3327         target_backend_unregister(&tcmu_ops);
3328         kfree(tcmu_attrs);
3329         genl_unregister_family(&tcmu_genl_family);
3330         root_device_unregister(tcmu_root_device);
3331         kmem_cache_destroy(tcmu_cmd_cache);
3332 }
3333
3334 MODULE_DESCRIPTION("TCM USER subsystem plugin");
3335 MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
3336 MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
3337 MODULE_LICENSE("GPL");
3338
3339 module_init(tcmu_module_init);
3340 module_exit(tcmu_module_exit);