iscsi-scst/kernel/nthread.c

   1 /*
   2  *  Network threads.
   3  *
   4  *  Copyright (C) 2004 - 2005 FUJITA Tomonori <tomof@acm.org>
   5  *  Copyright (C) 2007 Vladislav Bolkhovitin
   6  *  Copyright (C) 2007 CMS Distribution Limited
   7  *
   8  *  This program is free software; you can redistribute it and/or
   9  *  modify it under the terms of the GNU General Public License
  10  *  as published by the Free Software Foundation.
  11  *
  12  *  This program is distributed in the hope that it will be useful,
  13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  15  *  GNU General Public License for more details.
  16  */
  17
  18 #include <linux/sched.h>
  19 #include <linux/file.h>
  20 #include <linux/kthread.h>
  21 #include <asm/ioctls.h>
  22 #include <linux/delay.h>
  23 #include <net/tcp.h>
  24
  25 #include "iscsi.h"
  26 #include "digest.h"
  27
  28 enum rx_state {
  29         RX_INIT_BHS, /* Must be zero. */
  30         RX_BHS,
  31
  32         RX_INIT_AHS,
  33         RX_AHS,
  34
  35         RX_INIT_HDIGEST,
  36         RX_HDIGEST,
  37         RX_CHECK_HDIGEST,
  38
  39         RX_INIT_DATA,
  40         RX_DATA,
  41
  42         RX_INIT_DDIGEST,
  43         RX_DDIGEST,
  44         RX_CHECK_DDIGEST,
  45
  46         RX_END,
  47 };
  48
  49 enum tx_state {
  50         TX_INIT, /* Must be zero. */
  51         TX_BHS_DATA,
  52         TX_INIT_DDIGEST,
  53         TX_DDIGEST,
  54         TX_END,
  55 };
  56
  57 /* No locks */
  58 static void close_conn(struct iscsi_conn *conn)
  59 {
  60         struct iscsi_session *session = conn->session;
  61         struct iscsi_target *target = conn->target;
  62
  63         TRACE_ENTRY();
  64
  65         TRACE_CONN_CLOSE("conn %p, conn_ref_cnt=%d", conn,
  66                 atomic_read(&conn->conn_ref_cnt));
  67
  68         iscsi_extracheck_is_rd_thread(conn);
  69
  70         /* We want all our already send operations to complete */
  71         conn->sock->ops->shutdown(conn->sock, RCV_SHUTDOWN);
  72
  73         conn_abort(conn);
  74
  75         if (conn->read_state != RX_INIT_BHS) {
  76                 req_cmnd_release_force(conn->read_cmnd, 0);
  77                 conn->read_cmnd = NULL;
  78                 conn->read_state = RX_INIT_BHS;
  79         }
  80
  81         /* ToDo: not the best way to wait */
  82         while(atomic_read(&conn->conn_ref_cnt) != 0) {
  83                 struct iscsi_cmnd *cmnd;
  84
  85                 if (!list_empty(&session->pending_list)) {
  86                         struct list_head *pending_list = &session->pending_list;
  87                         struct iscsi_cmnd *tmp;
  88
  89                         TRACE_CONN_CLOSE("Disposing pending commands on conn "
  90                                 "%p, conn_ref_cnt=%d", conn,
  91                                 atomic_read(&conn->conn_ref_cnt));
  92
  93                         list_for_each_entry_safe(cmnd, tmp, pending_list,
  94                                                 pending_list_entry) {
  95                                 if (cmnd->conn == conn) {
  96                                         TRACE_CONN_CLOSE("Freeing pending cmd %p",
  97                                                 cmnd);
  98                                         list_del(&cmnd->pending_list_entry);
  99                                         cmnd->pending = 0;
 100                                         req_cmnd_release_force(cmnd, 0);
 101                                 }
 102                         }
 103                 }
 104
 105                 iscsi_make_conn_wr_active(conn);
 106                 msleep(50);
 107
 108                 TRACE_CONN_CLOSE("conn %p, conn_ref_cnt %d left, wr_state %d",
 109                         conn, atomic_read(&conn->conn_ref_cnt), conn->wr_state);
 110 #ifdef DEBUG
 111                 {
 112 #ifdef NET_PAGE_CALLBACKS_DEFINED
 113                         struct iscsi_cmnd *rsp;
 114 #endif
 115                         spin_lock_bh(&conn->cmd_list_lock);
 116                         list_for_each_entry(cmnd, &conn->cmd_list, cmd_list_entry) {
 117                                 TRACE_DBG("cmd %p, scst_state %x, data_waiting "
 118                                         "%d, ref_cnt %d, parent_req %p", cmnd,
 119                                         cmnd->scst_state, cmnd->data_waiting,
 120                                         atomic_read(&cmnd->ref_cnt), cmnd->parent_req);
 121 #ifdef NET_PAGE_CALLBACKS_DEFINED
 122                                 TRACE_DBG("net_ref_cnt %d, sg %p",
 123                                         atomic_read(&cmnd->net_ref_cnt), cmnd->sg);
 124                                 if (cmnd->sg != NULL) {
 125                                         int sg_cnt, i;
 126                                         sg_cnt = get_pgcnt(cmnd->bufflen,
 127                                                 cmnd->sg[0].offset);
 128                                         for(i = 0; i < sg_cnt; i++) {
 129                                                 TRACE_DBG("page %p, net_priv %p, _count %d",
 130                                                         cmnd->sg[i].page, cmnd->sg[i].page->net_priv,
 131                                                         atomic_read(&cmnd->sg[i].page->_count));
 132                                         }
 133                                 }
 134
 135                                 sBUG_ON(cmnd->parent_req != NULL);
 136
 137                                 spin_lock_bh(&cmnd->rsp_cmd_lock);
 138                                 list_for_each_entry(rsp, &cmnd->rsp_cmd_list, rsp_cmd_list_entry) {
 139                                         TRACE_DBG("  rsp %p, ref_cnt %d, net_ref_cnt %d, "
 140                                                 "sg %p", rsp, atomic_read(&rsp->ref_cnt),
 141                                                 atomic_read(&rsp->net_ref_cnt), rsp->sg);
 142                                         if ((rsp->sg != cmnd->sg) && (rsp->sg != NULL)) {
 143                                                 int sg_cnt, i;
 144                                                 sg_cnt = get_pgcnt(rsp->bufflen,
 145                                                         rsp->sg[0].offset);
 146                                                 sBUG_ON(rsp->sg_cnt != sg_cnt);
 147                                                 for(i = 0; i < sg_cnt; i++) {
 148                                                         TRACE_DBG("    page %p, net_priv %p, "
 149                                                                 "_count %d", rsp->sg[i].page,
 150                                                                 rsp->sg[i].page->net_priv,
 151                                                                 atomic_read(&rsp->sg[i].page->_count));
 152                                                 }
 153                                         }
 154                                 }
 155                                 spin_unlock_bh(&cmnd->rsp_cmd_lock);
 156 #endif
 157                         }
 158                         spin_unlock_bh(&conn->cmd_list_lock);
 159                 }
 160 #endif
 161         }
 162
 163         write_lock_bh(&conn->sock->sk->sk_callback_lock);
 164         conn->sock->sk->sk_state_change = conn->old_state_change;
 165         conn->sock->sk->sk_data_ready = conn->old_data_ready;
 166         conn->sock->sk->sk_write_space = conn->old_write_space;
 167         write_unlock_bh(&conn->sock->sk->sk_callback_lock);
 168
 169         while(conn->wr_state != ISCSI_CONN_WR_STATE_IDLE) {
 170                 TRACE_CONN_CLOSE("Waiting for wr thread (conn %p), wr_state %x",
 171                         conn, conn->wr_state);
 172                 msleep(50);
 173         }
 174
 175         mutex_lock(&target->target_mutex);
 176         conn_free(conn);
 177         if (list_empty(&session->conn_list))
 178                 session_del(target, session->sid);
 179         mutex_unlock(&target->target_mutex);
 180
 181         TRACE_CONN_CLOSE("Notifying user space about closing conn %p", conn);
 182         event_send(target->tid, session->sid, conn->cid, E_CONN_CLOSE, 0);
 183
 184         TRACE_EXIT();
 185         return;
 186 }
 187
 188 static inline void iscsi_conn_init_read(struct iscsi_conn *conn, void *data, size_t len)
 189 {
 190         len = (len + 3) & -4; // XXX ???
 191         conn->read_iov[0].iov_base = data;
 192         conn->read_iov[0].iov_len = len;
 193         conn->read_msg.msg_iov = conn->read_iov;
 194         conn->read_msg.msg_iovlen = 1;
 195         conn->read_size = (len + 3) & -4;
 196 }
 197
 198 static void iscsi_conn_read_ahs(struct iscsi_conn *conn, struct iscsi_cmnd *cmnd)
 199 {
 200         /* ToDo: __GFP_NOFAIL ?? */
 201         cmnd->pdu.ahs = kmalloc(cmnd->pdu.ahssize, __GFP_NOFAIL|GFP_KERNEL);
 202         sBUG_ON(cmnd->pdu.ahs == NULL);
 203         iscsi_conn_init_read(conn, cmnd->pdu.ahs, cmnd->pdu.ahssize);
 204 }
 205
 206 static struct iscsi_cmnd *iscsi_get_send_cmnd(struct iscsi_conn *conn)
 207 {
 208         struct iscsi_cmnd *cmnd = NULL;
 209
 210         spin_lock(&conn->write_list_lock);
 211         if (!list_empty(&conn->write_list)) {
 212                 cmnd = list_entry(conn->write_list.next, struct iscsi_cmnd,
 213                                 write_list_entry);
 214                 cmd_del_from_write_list(cmnd);
 215                 cmnd->write_processing_started = 1;
 216         }
 217         spin_unlock(&conn->write_list_lock);
 218
 219         return cmnd;
 220 }
 221
 222 static int do_recv(struct iscsi_conn *conn, int state)
 223 {
 224         mm_segment_t oldfs;
 225         struct msghdr msg;
 226         int res, first_len;
 227
 228         if (unlikely(conn->closing)) {
 229                 res = -EIO;
 230                 goto out;
 231         }
 232
 233         memset(&msg, 0, sizeof(msg));
 234         msg.msg_iov = conn->read_msg.msg_iov;
 235         msg.msg_iovlen = conn->read_msg.msg_iovlen;
 236         first_len = msg.msg_iov->iov_len;
 237
 238         oldfs = get_fs();
 239         set_fs(get_ds());
 240         res = sock_recvmsg(conn->sock, &msg, conn->read_size, MSG_DONTWAIT | MSG_NOSIGNAL);
 241         set_fs(oldfs);
 242
 243         if (res <= 0) {
 244                 switch (res) {
 245                 case -EAGAIN:
 246                 case -ERESTARTSYS:
 247                         TRACE_DBG("EAGAIN or ERESTARTSYS (%d) received for "
 248                                 "conn %p", res, conn);
 249                         break;
 250                 default:
 251                         PRINT_ERROR_PR("sock_recvmsg() failed: %d", res);
 252                         mark_conn_closed(conn);
 253                         break;
 254                 }
 255         } else {
 256                 /*
 257                  * To save some considerable effort and CPU power we suppose
 258                  * that TCP functions adjust conn->read_msg.msg_iov and
 259                  * conn->read_msg.msg_iovlen on amount of copied data. This
 260                  * BUG_ON is intended to catch if it is changed in the future.
 261                  */
 262                 sBUG_ON((res >= first_len) &&
 263                         (conn->read_msg.msg_iov->iov_len != 0));
 264                 conn->read_size -= res;
 265                 if (conn->read_size) {
 266                         if (res >= first_len) {
 267                                 int done = 1 + ((res - first_len) >> PAGE_SHIFT);
 268                                 conn->read_msg.msg_iov += done;
 269                                 conn->read_msg.msg_iovlen -= done;
 270                         }
 271                 } else
 272                         conn->read_state = state;
 273         }
 274
 275 out:
 276         TRACE_EXIT_RES(res);
 277         return res;
 278 }
 279
 280 static int rx_hdigest(struct iscsi_conn *conn)
 281 {
 282         struct iscsi_cmnd *cmnd = conn->read_cmnd;
 283         int res = digest_rx_header(cmnd);
 284
 285         if (unlikely(res != 0)) {
 286                 PRINT_ERROR_PR("rx header digest for initiator %s failed "
 287                         "(%d)", conn->session->initiator_name, res);
 288                 mark_conn_closed(conn);
 289         }
 290         return res;
 291 }
 292
 293 static struct iscsi_cmnd *create_cmnd(struct iscsi_conn *conn)
 294 {
 295         struct iscsi_cmnd *cmnd;
 296
 297         cmnd = cmnd_alloc(conn, NULL);
 298         iscsi_conn_init_read(cmnd->conn, &cmnd->pdu.bhs, sizeof(cmnd->pdu.bhs));
 299         conn->read_state = RX_BHS;
 300
 301         return cmnd;
 302 }
 303
 304 /* Returns >0 for success, <=0 for error or successful finish */
 305 static int recv(struct iscsi_conn *conn)
 306 {
 307         struct iscsi_cmnd *cmnd = conn->read_cmnd;
 308         int hdigest, ddigest, res = 1, rc;
 309
 310         TRACE_ENTRY();
 311
 312         hdigest = conn->hdigest_type & DIGEST_NONE ? 0 : 1;
 313         ddigest = conn->ddigest_type & DIGEST_NONE ? 0 : 1;
 314
 315         switch (conn->read_state) {
 316         case RX_INIT_BHS:
 317                 sBUG_ON(cmnd != NULL);
 318                 cmnd = conn->read_cmnd = create_cmnd(conn);
 319         case RX_BHS:
 320                 res = do_recv(conn, RX_INIT_AHS);
 321                 if (res <= 0 || conn->read_state != RX_INIT_AHS)
 322                         break;
 323         case RX_INIT_AHS:
 324                 iscsi_cmnd_get_length(&cmnd->pdu);
 325                 if (cmnd->pdu.ahssize) {
 326                         iscsi_conn_read_ahs(conn, cmnd);
 327                         conn->read_state = RX_AHS;
 328                 } else
 329                         conn->read_state = hdigest ? RX_INIT_HDIGEST : RX_INIT_DATA;
 330
 331                 if (conn->read_state != RX_AHS)
 332                         break;
 333         case RX_AHS:
 334                 res = do_recv(conn, hdigest ? RX_INIT_HDIGEST : RX_INIT_DATA);
 335                 if (res <= 0 || conn->read_state != RX_INIT_HDIGEST)
 336                         break;
 337         case RX_INIT_HDIGEST:
 338                 iscsi_conn_init_read(conn, &cmnd->hdigest, sizeof(u32));
 339                 conn->read_state = RX_HDIGEST;
 340         case RX_HDIGEST:
 341                 res = do_recv(conn, RX_CHECK_HDIGEST);
 342                 if (res <= 0 || conn->read_state != RX_CHECK_HDIGEST)
 343                         break;
 344         case RX_CHECK_HDIGEST:
 345                 rc = rx_hdigest(conn);
 346                 if (likely(rc == 0))
 347                         conn->read_state = RX_INIT_DATA;
 348                 else {
 349                         res = rc;
 350                         break;
 351                 }
 352         case RX_INIT_DATA:
 353                 rc = cmnd_rx_start(cmnd);
 354                 if (unlikely(rc != 0)) {
 355                         sBUG_ON(!conn->closing);
 356                         conn->read_state = RX_END;
 357                         res = rc;
 358                         /* cmnd will be freed in close_conn() */
 359                         goto out;
 360                 }
 361                 conn->read_state = cmnd->pdu.datasize ? RX_DATA : RX_END;
 362                 if (conn->read_state != RX_DATA)
 363                         break;
 364         case RX_DATA:
 365                 res = do_recv(conn, ddigest ? RX_INIT_DDIGEST : RX_END);
 366                 if (res <= 0 || conn->read_state != RX_INIT_DDIGEST)
 367                         break;
 368         case RX_INIT_DDIGEST:
 369                 iscsi_conn_init_read(conn, &cmnd->ddigest, sizeof(u32));
 370                 conn->read_state = RX_DDIGEST;
 371         case RX_DDIGEST:
 372                 res = do_recv(conn, RX_CHECK_DDIGEST);
 373                 if (res <= 0 || conn->read_state != RX_CHECK_DDIGEST)
 374                         break;
 375         case RX_CHECK_DDIGEST:
 376                 conn->read_state = RX_END;
 377                 if (cmnd_opcode(cmnd) == ISCSI_OP_SCSI_CMD) {
 378                         TRACE_DBG("Adding RX ddigest cmd %p to digest list "
 379                                 "of self", cmnd);
 380                         list_add_tail(&cmnd->rx_ddigest_cmd_list_entry,
 381                                 &cmnd->rx_ddigest_cmd_list);
 382                         cmnd_get(cmnd);
 383                         conn->read_state = RX_END;
 384                 } else if (cmnd_opcode(cmnd) != ISCSI_OP_SCSI_DATA_OUT) {
 385                         /*
 386                          * We could get here only for NOP-Out. ISCSI RFC doesn't
 387                          * specify how to deal with digest errors in this case.
 388                          * Is closing connection correct?
 389                          */
 390                         TRACE_DBG("cmnd %p, opcode %x: checking RX "
 391                                 "ddigest inline", cmnd, cmnd_opcode(cmnd));
 392                         rc = digest_rx_data(cmnd);
 393                         if (unlikely(rc != 0)) {
 394                                 conn->read_state = RX_CHECK_DDIGEST;
 395                                 mark_conn_closed(conn);
 396                         }
 397                 }
 398                 break;
 399         default:
 400                 PRINT_ERROR_PR("%d %x", conn->read_state, cmnd_opcode(cmnd));
 401                 sBUG();
 402         }
 403
 404         if (res <= 0)
 405                 goto out;
 406
 407         if (conn->read_state != RX_END)
 408                 goto out;
 409
 410         if (conn->read_size) {
 411                 PRINT_ERROR_PR("%d %x %d", res, cmnd_opcode(cmnd), conn->read_size);
 412                 sBUG();
 413         }
 414
 415         cmnd_rx_end(cmnd);
 416
 417         sBUG_ON(conn->read_size != 0);
 418
 419         conn->read_cmnd = NULL;
 420         conn->read_state = RX_INIT_BHS;
 421         res = 0;
 422
 423 out:
 424         TRACE_EXIT_RES(res);
 425         return res;
 426 }
 427
 428 /* No locks, conn is rd processing */
 429 static int process_read_io(struct iscsi_conn *conn, int *closed)
 430 {
 431         int res;
 432
 433         do {
 434                 res = recv(conn);
 435                 if (unlikely(conn->closing)) {
 436                         close_conn(conn);
 437                         *closed = 1;
 438                         break;
 439                 }
 440         } while(res > 0);
 441
 442         TRACE_EXIT_RES(res);
 443         return res;
 444 }
 445
 446 /*
 447  * Called under iscsi_rd_lock and BHs disabled, but will drop it inside,
 448  * then reaquire.
 449  */
 450 static void scst_do_job_rd(void)
 451 {
 452         TRACE_ENTRY();
 453
 454         /* We delete/add to tail connections to maintain fairness between them */
 455
 456         while(!list_empty(&iscsi_rd_list)) {
 457                 int rc, closed = 0;
 458                 struct iscsi_conn *conn = list_entry(iscsi_rd_list.next,
 459                         typeof(*conn), rd_list_entry);
 460
 461                 list_del(&conn->rd_list_entry);
 462
 463                 sBUG_ON(conn->rd_state == ISCSI_CONN_RD_STATE_PROCESSING);
 464                 conn->rd_data_ready = 0;
 465                 conn->rd_state = ISCSI_CONN_RD_STATE_PROCESSING;
 466 #ifdef EXTRACHECKS
 467                 conn->rd_task = current;
 468 #endif
 469                 spin_unlock_bh(&iscsi_rd_lock);
 470
 471                 rc = process_read_io(conn, &closed);
 472
 473                 spin_lock_bh(&iscsi_rd_lock);
 474
 475                 if (closed)
 476                         continue;
 477
 478 #ifdef EXTRACHECKS
 479                 conn->rd_task = NULL;
 480 #endif
 481                 if ((rc == 0) || conn->rd_data_ready) {
 482                         list_add_tail(&conn->rd_list_entry, &iscsi_rd_list);
 483                         conn->rd_state = ISCSI_CONN_RD_STATE_IN_LIST;
 484                 } else
 485                         conn->rd_state = ISCSI_CONN_RD_STATE_IDLE;
 486         }
 487
 488         TRACE_EXIT();
 489         return;
 490 }
 491
 492 static inline int test_rd_list(void)
 493 {
 494         int res = !list_empty(&iscsi_rd_list) ||
 495                   unlikely(kthread_should_stop());
 496         return res;
 497 }
 498
 499 int istrd(void *arg)
 500 {
 501         TRACE_ENTRY();
 502
 503         current->flags |= PF_NOFREEZE;
 504
 505         spin_lock_bh(&iscsi_rd_lock);
 506         while(!kthread_should_stop()) {
 507                 wait_queue_t wait;
 508                 init_waitqueue_entry(&wait, current);
 509
 510                 if (!test_rd_list()) {
 511                         add_wait_queue_exclusive(&iscsi_rd_waitQ, &wait);
 512                         for (;;) {
 513                                 set_current_state(TASK_INTERRUPTIBLE);
 514                                 if (test_rd_list())
 515                                         break;
 516                                 spin_unlock_bh(&iscsi_rd_lock);
 517                                 schedule();
 518                                 spin_lock_bh(&iscsi_rd_lock);
 519                         }
 520                         set_current_state(TASK_RUNNING);
 521                         remove_wait_queue(&iscsi_rd_waitQ, &wait);
 522                 }
 523                 scst_do_job_rd();
 524         }
 525         spin_unlock_bh(&iscsi_rd_lock);
 526
 527         /*
 528          * If kthread_should_stop() is true, we are guaranteed to be
 529          * on the module unload, so iscsi_rd_list must be empty.
 530          */
 531         sBUG_ON(!list_empty(&iscsi_rd_list));
 532
 533         TRACE_EXIT();
 534         return 0;
 535 }
 536
 537 #ifdef NET_PAGE_CALLBACKS_DEFINED
 538 void iscsi_get_page_callback(struct page *page)
 539 {
 540         struct iscsi_cmnd *cmd = (struct iscsi_cmnd*)page->net_priv;
 541         int v;
 542
 543         TRACE_DBG("cmd %p, page %p, _count %d, new net_ref_cnt %d",
 544                 cmd, page, atomic_read(&page->_count),
 545                 atomic_read(&cmd->net_ref_cnt)+1);
 546
 547         v = atomic_inc_return(&cmd->net_ref_cnt);
 548         if (v == 1) {
 549                 TRACE_DBG("getting cmd %p for page %p", cmd, page);
 550                 cmnd_get(cmd);
 551         }
 552 }
 553
 554 void iscsi_put_page_callback(struct page *page)
 555 {
 556         struct iscsi_cmnd *cmd = (struct iscsi_cmnd*)page->net_priv;
 557
 558         TRACE_DBG("cmd %p, page %p, _count %d, new net_ref_cnt %d",
 559                 cmd, page, atomic_read(&page->_count),
 560                 atomic_read(&cmd->net_ref_cnt)-1);
 561
 562         if (atomic_dec_and_test(&cmd->net_ref_cnt)) {
 563                 int i, sg_cnt = get_pgcnt(cmd->bufflen, cmd->sg[0].offset);
 564                 for(i = 0; i < sg_cnt; i++) {
 565                         TRACE_DBG("Clearing page %p", cmd->sg[i].page);
 566                         cmd->sg[i].page->net_priv = NULL;
 567                 }
 568                 cmnd_put(cmd);
 569         }
 570 }
 571 #endif
 572
 573 /* This is partially taken from the Ardis code. */
 574 static int write_data(struct iscsi_conn *conn)
 575 {
 576         mm_segment_t oldfs;
 577         struct file *file;
 578         struct socket *sock;
 579         ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
 580         struct iscsi_cmnd *write_cmnd = conn->write_cmnd;
 581         struct iscsi_cmnd *ref_cmd;
 582         struct scatterlist *sg;
 583         struct iovec *iop;
 584         int saved_size, size, sendsize;
 585         int offset, idx;
 586         int flags, res, count;
 587
 588         iscsi_extracheck_is_wr_thread(conn);
 589
 590         if (write_cmnd->own_sg == 0)
 591                 ref_cmd = write_cmnd->parent_req;
 592         else
 593                 ref_cmd = write_cmnd;
 594
 595         file = conn->file;
 596         saved_size = size = conn->write_size;
 597         iop = conn->write_iop;
 598         count = conn->write_iop_used;
 599
 600         if (iop) while (1) {
 601                 loff_t off = 0;
 602                 int rest;
 603
 604                 sBUG_ON(count > sizeof(conn->write_iov)/sizeof(conn->write_iov[0]));
 605 retry:
 606                 oldfs = get_fs();
 607                 set_fs(KERNEL_DS);
 608                 res = vfs_writev(file, (struct iovec __user *)iop, count, &off);
 609                 set_fs(oldfs);
 610                 TRACE(TRACE_D_DATA, "%#Lx:%u: %d(%ld)",
 611                         (unsigned long long) conn->session->sid, conn->cid,
 612                         res, (long) iop->iov_len);
 613                 if (unlikely(res <= 0)) {
 614                         if (res == -EAGAIN) {
 615                                 conn->write_iop = iop;
 616                                 conn->write_iop_used = count;
 617                                 goto out_iov;
 618                         } else if (res == -EINTR)
 619                                 goto retry;
 620                         goto err;
 621                 }
 622
 623                 rest = res;
 624                 size -= res;
 625                 while (iop->iov_len <= rest && rest) {
 626                         rest -= iop->iov_len;
 627                         iop++;
 628                         count--;
 629                 }
 630                 if (count == 0) {
 631                         conn->write_iop = NULL;
 632                         conn->write_iop_used = 0;
 633                         if (size)
 634                                 break;
 635                         goto out_iov;
 636                 }
 637                 sBUG_ON(iop > conn->write_iov +
 638                         sizeof(conn->write_iov)/sizeof(conn->write_iov[0]));
 639                 iop->iov_base += rest;
 640                 iop->iov_len -= rest;
 641         }
 642
 643         sg = write_cmnd->sg;
 644         if (sg == NULL) {
 645                 PRINT_ERROR_PR("%s", "warning data missing!");
 646                 return 0;
 647         }
 648         offset = conn->write_offset;
 649         idx = offset >> PAGE_SHIFT;
 650         offset &= ~PAGE_MASK;
 651
 652         sock = conn->sock;
 653
 654 #ifdef NET_PAGE_CALLBACKS_DEFINED
 655         sendpage = sock->ops->sendpage;
 656 #else
 657         if ((write_cmnd->parent_req->scst_cmd != NULL) &&
 658             scst_cmd_get_data_buff_alloced(write_cmnd->parent_req->scst_cmd))
 659                 sendpage = sock_no_sendpage;
 660         else
 661                 sendpage = sock->ops->sendpage;
 662 #endif
 663
 664         flags = MSG_DONTWAIT;
 665
 666         while (1) {
 667 #ifdef NET_PAGE_CALLBACKS_DEFINED
 668                 if (unlikely((sg[idx].page->net_priv != NULL) &&
 669                                 (sg[idx].page->net_priv != ref_cmd))) {
 670                         PRINT_ERROR_PR("net_priv isn't NULL and != ref_cmd "
 671                                 "(write_cmnd %p, ref_cmd %p, sg %p, idx %d, "
 672                                 "net_priv %p)", write_cmnd, ref_cmd, sg, idx,
 673                                 sg[idx].page->net_priv);
 674                         sBUG();
 675                 }
 676                 sg[idx].page->net_priv = ref_cmd;
 677 #endif
 678                 sendsize = PAGE_SIZE - offset;
 679                 if (size <= sendsize) {
 680 retry2:
 681                         res = sendpage(sock, sg[idx].page, offset, size, flags);
 682                         TRACE(TRACE_D_DATA, "%s %#Lx:%u: %d(%lu,%u,%u)",
 683                                 sock->ops->sendpage ? "sendpage" : "sock_no_sendpage",
 684                                 (unsigned long long)conn->session->sid, conn->cid,
 685                                 res, sg[idx].page->index, offset, size);
 686                         if (unlikely(res <= 0)) {
 687                                 if (res == -EINTR)
 688                                         goto retry2;
 689                                 else
 690                                         goto out_res;
 691                         }
 692                         if (res == size) {
 693                                 conn->write_size = 0;
 694                                 return saved_size;
 695                         }
 696                         offset += res;
 697                         size -= res;
 698                         continue;
 699                 }
 700
 701 retry1:
 702                 res = sendpage(sock, sg[idx].page, offset, sendsize,
 703                         flags | MSG_MORE);
 704                 TRACE(TRACE_D_DATA, "%s %#Lx:%u: %d(%lu,%u,%u)",
 705                         sock->ops->sendpage ? "sendpage" : "sock_no_sendpage",
 706                         (unsigned long long ) conn->session->sid, conn->cid,
 707                         res, sg[idx].page->index, offset, sendsize);
 708                 if (unlikely(res <= 0)) {
 709                         if (res == -EINTR)
 710                                 goto retry1;
 711                         else
 712                                 goto out_res;
 713                 }
 714                 if (res == sendsize) {
 715                         idx++;
 716                         offset = 0;
 717                 } else
 718                         offset += res;
 719                 size -= res;
 720         }
 721 out:
 722         conn->write_offset = (idx << PAGE_SHIFT) + offset;
 723 out_iov:
 724         conn->write_size = size;
 725         if ((saved_size == size) && res == -EAGAIN)
 726                 return res;
 727
 728         return saved_size - size;
 729
 730 out_res:
 731 #ifdef NET_PAGE_CALLBACKS_DEFINED
 732         if (atomic_read(&ref_cmd->net_ref_cnt) == 0) {
 733                 TRACE_DBG("sendpage() returned %d, zeroing net_priv", res);
 734                 sg[idx].page->net_priv = NULL;
 735         }
 736 #endif
 737         if (res == -EAGAIN)
 738                 goto out;
 739         /* else go through */
 740
 741 err:
 742 #ifndef DEBUG
 743         if (!conn->closing)
 744 #endif
 745         {
 746                 PRINT_ERROR_PR("error %d at sid:cid %#Lx:%u, cmnd %p", res,
 747                         (unsigned long long)conn->session->sid, conn->cid,
 748                         conn->write_cmnd);
 749         }
 750         return res;
 751 }
 752
 753 static int exit_tx(struct iscsi_conn *conn, int res)
 754 {
 755         iscsi_extracheck_is_wr_thread(conn);
 756
 757         switch (res) {
 758         case -EAGAIN:
 759         case -ERESTARTSYS:
 760                 res = 0;
 761                 break;
 762         default:
 763 #ifndef DEBUG
 764                 if (!conn->closing)
 765 #endif
 766                 {
 767                         PRINT_ERROR_PR("Sending data failed: initiator %s, "
 768                                 "write_size %d, write_state %d, res %d",
 769                                 conn->session->initiator_name, conn->write_size,
 770                                 conn->write_state, res);
 771                 }
 772                 conn->write_state = TX_END;
 773                 conn->write_size = 0;
 774                 mark_conn_closed(conn);
 775                 break;
 776         }
 777         return res;
 778 }
 779
 780 static int tx_ddigest(struct iscsi_cmnd *cmnd, int state)
 781 {
 782         int res, rest = cmnd->conn->write_size;
 783         struct msghdr msg = {.msg_flags = MSG_NOSIGNAL | MSG_DONTWAIT};
 784         struct kvec iov;
 785
 786         iscsi_extracheck_is_wr_thread(cmnd->conn);
 787
 788         TRACE_DBG("Sending data digest %x (cmd %p)", cmnd->ddigest, cmnd);
 789
 790         iov.iov_base = (char *) (&cmnd->ddigest) + (sizeof(u32) - rest);
 791         iov.iov_len = rest;
 792
 793         res = kernel_sendmsg(cmnd->conn->sock, &msg, &iov, 1, rest);
 794         if (res > 0) {
 795                 cmnd->conn->write_size -= res;
 796                 if (!cmnd->conn->write_size)
 797                         cmnd->conn->write_state = state;
 798         } else
 799                 res = exit_tx(cmnd->conn, res);
 800
 801         return res;
 802 }
 803
 804 static void init_tx_hdigest(struct iscsi_cmnd *cmnd)
 805 {
 806         struct iscsi_conn *conn = cmnd->conn;
 807         struct iovec *iop;
 808
 809         iscsi_extracheck_is_wr_thread(conn);
 810
 811         digest_tx_header(cmnd);
 812
 813         sBUG_ON(conn->write_iop_used >= sizeof(conn->write_iov)/sizeof(conn->write_iov[0]));
 814         iop = &conn->write_iop[conn->write_iop_used];
 815         conn->write_iop_used++;
 816         iop->iov_base = &(cmnd->hdigest);
 817         iop->iov_len = sizeof(u32);
 818         conn->write_size += sizeof(u32);
 819
 820         return;
 821 }
 822
 823 static int iscsi_do_send(struct iscsi_conn *conn, int state)
 824 {
 825         int res;
 826
 827         iscsi_extracheck_is_wr_thread(conn);
 828
 829         res = write_data(conn);
 830         if (res > 0) {
 831                 if (!conn->write_size)
 832                         conn->write_state = state;
 833         } else
 834                 res = exit_tx(conn, res);
 835
 836         return res;
 837 }
 838
 839 /*
 840  * No locks, conn is wr processing.
 841  *
 842  * IMPORTANT! Connection conn must be protected by additional conn_get()
 843  * upon entrance in this function, because otherwise it could be destroyed
 844  * inside as a result of cmnd release.
 845  */
 846 int iscsi_send(struct iscsi_conn *conn)
 847 {
 848         struct iscsi_cmnd *cmnd = conn->write_cmnd;
 849         int ddigest, res = 0;
 850
 851         TRACE_ENTRY();
 852
 853         TRACE_DBG("conn %p, write_cmnd %p", conn, cmnd);
 854
 855         iscsi_extracheck_is_wr_thread(conn);
 856
 857         ddigest = conn->ddigest_type != DIGEST_NONE ? 1 : 0;
 858
 859         switch (conn->write_state) {
 860         case TX_INIT:
 861                 sBUG_ON(cmnd != NULL);
 862                 cmnd = conn->write_cmnd = iscsi_get_send_cmnd(conn);
 863                 if (!cmnd)
 864                         goto out;
 865                 cmnd_tx_start(cmnd);
 866                 if (!(conn->hdigest_type & DIGEST_NONE))
 867                     init_tx_hdigest(cmnd);
 868                 conn->write_state = TX_BHS_DATA;
 869         case TX_BHS_DATA:
 870                 res = iscsi_do_send(conn, ddigest && cmnd->pdu.datasize ?
 871                                         TX_INIT_DDIGEST : TX_END);
 872                 if (res <= 0 || conn->write_state != TX_INIT_DDIGEST)
 873                         break;
 874         case TX_INIT_DDIGEST:
 875                 cmnd->conn->write_size = sizeof(u32);
 876                 conn->write_state = TX_DDIGEST;
 877         case TX_DDIGEST:
 878                 res = tx_ddigest(cmnd, TX_END);
 879                 break;
 880         default:
 881                 PRINT_ERROR_PR("%d %d %x", res, conn->write_state,
 882                         cmnd_opcode(cmnd));
 883                 sBUG();
 884         }
 885
 886         if (res == 0)
 887                 goto out;
 888
 889         if (conn->write_state != TX_END)
 890                 goto out;
 891
 892         if (conn->write_size) {
 893                 PRINT_ERROR_PR("%d %x %u", res, cmnd_opcode(cmnd),
 894                         conn->write_size);
 895                 sBUG();
 896         }
 897         cmnd_tx_end(cmnd);
 898
 899         rsp_cmnd_release(cmnd);
 900
 901         conn->write_cmnd = NULL;
 902         conn->write_state = TX_INIT;
 903
 904 out:
 905         TRACE_EXIT_RES(res);
 906         return res;
 907 }
 908
 909 /* No locks, conn is wr processing.
 910  *
 911  * IMPORTANT! Connection conn must be protected by additional conn_get()
 912  * upon entrance in this function, because otherwise it could be destroyed
 913  * inside as a result of iscsi_send(), which releases sent commands.
 914  */
 915 static int process_write_queue(struct iscsi_conn *conn)
 916 {
 917         int res = 0;
 918
 919         TRACE_ENTRY();
 920
 921         if (likely(test_write_ready(conn)))
 922                 res = iscsi_send(conn);
 923
 924         TRACE_EXIT_RES(res);
 925         return res;
 926 }
 927
 928 /*
 929  * Called under iscsi_wr_lock and BHs disabled, but will drop it inside,
 930  * then reaquire.
 931  */
 932 static void scst_do_job_wr(void)
 933 {
 934         TRACE_ENTRY();
 935
 936         /* We delete/add to tail connections to maintain fairness between them */
 937
 938         while(!list_empty(&iscsi_wr_list)) {
 939                 int rc;
 940                 struct iscsi_conn *conn = list_entry(iscsi_wr_list.next,
 941                         typeof(*conn), wr_list_entry);
 942
 943                 TRACE_DBG("conn %p, wr_state %x, wr_space_ready %d, "
 944                         "write ready %d", conn, conn->wr_state,
 945                         conn->wr_space_ready, test_write_ready(conn));
 946
 947                 list_del(&conn->wr_list_entry);
 948
 949                 sBUG_ON(conn->wr_state == ISCSI_CONN_WR_STATE_PROCESSING);
 950
 951                 conn->wr_state = ISCSI_CONN_WR_STATE_PROCESSING;
 952                 conn->wr_space_ready = 0;
 953 #ifdef EXTRACHECKS
 954                 conn->wr_task = current;
 955 #endif
 956                 spin_unlock_bh(&iscsi_wr_lock);
 957
 958                 conn_get(conn);
 959
 960                 rc = process_write_queue(conn);
 961
 962                 spin_lock_bh(&iscsi_wr_lock);
 963 #ifdef EXTRACHECKS
 964                 conn->wr_task = NULL;
 965 #endif
 966                 if ((rc == -EAGAIN) && !conn->wr_space_ready) {
 967                         conn->wr_state = ISCSI_CONN_WR_STATE_SPACE_WAIT;
 968                         goto cont;
 969                 }
 970
 971                 if (test_write_ready(conn)) {
 972                         list_add_tail(&conn->wr_list_entry, &iscsi_wr_list);
 973                         conn->wr_state = ISCSI_CONN_WR_STATE_IN_LIST;
 974                 } else
 975                         conn->wr_state = ISCSI_CONN_WR_STATE_IDLE;
 976
 977 cont:
 978                 conn_put(conn);
 979         }
 980
 981         TRACE_EXIT();
 982         return;
 983 }
 984
 985 static inline int test_wr_list(void)
 986 {
 987         int res = !list_empty(&iscsi_wr_list) ||
 988                   unlikely(kthread_should_stop());
 989         return res;
 990 }
 991
 992 int istwr(void *arg)
 993 {
 994         TRACE_ENTRY();
 995
 996         current->flags |= PF_NOFREEZE;
 997
 998         spin_lock_bh(&iscsi_wr_lock);
 999         while(!kthread_should_stop()) {
1000                 wait_queue_t wait;
1001                 init_waitqueue_entry(&wait, current);
1002
1003                 if (!test_wr_list()) {
1004                         add_wait_queue_exclusive(&iscsi_wr_waitQ, &wait);
1005                         for (;;) {
1006                                 set_current_state(TASK_INTERRUPTIBLE);
1007                                 if (test_wr_list())
1008                                         break;
1009                                 spin_unlock_bh(&iscsi_wr_lock);
1010                                 schedule();
1011                                 spin_lock_bh(&iscsi_wr_lock);
1012                         }
1013                         set_current_state(TASK_RUNNING);
1014                         remove_wait_queue(&iscsi_wr_waitQ, &wait);
1015                 }
1016                 scst_do_job_wr();
1017         }
1018         spin_unlock_bh(&iscsi_wr_lock);
1019
1020         /*
1021          * If kthread_should_stop() is true, we are guaranteed to be
1022          * on the module unload, so iscsi_wr_list must be empty.
1023          */
1024         sBUG_ON(!list_empty(&iscsi_wr_list));
1025
1026         TRACE_EXIT();
1027         return 0;
1028 }