net/9p/trans_fd.c

   1 /*
   2  * linux/fs/9p/trans_fd.c
   3  *
   4  * Fd transport layer.  Includes deprecated socket layer.
   5  *
   6  *  Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
   7  *  Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
   8  *  Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
   9  *  Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
  10  *
  11  *  This program is free software; you can redistribute it and/or modify
  12  *  it under the terms of the GNU General Public License version 2
  13  *  as published by the Free Software Foundation.
  14  *
  15  *  This program is distributed in the hope that it will be useful,
  16  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  *  GNU General Public License for more details.
  19  *
  20  *  You should have received a copy of the GNU General Public License
  21  *  along with this program; if not, write to:
  22  *  Free Software Foundation
  23  *  51 Franklin Street, Fifth Floor
  24  *  Boston, MA  02111-1301  USA
  25  *
  26  */
  27
  28 #include <linux/in.h>
  29 #include <linux/module.h>
  30 #include <linux/net.h>
  31 #include <linux/ipv6.h>
  32 #include <linux/kthread.h>
  33 #include <linux/errno.h>
  34 #include <linux/kernel.h>
  35 #include <linux/un.h>
  36 #include <linux/uaccess.h>
  37 #include <linux/inet.h>
  38 #include <linux/idr.h>
  39 #include <linux/file.h>
  40 #include <linux/parser.h>
  41 #include <net/9p/9p.h>
  42 #include <net/9p/client.h>
  43 #include <net/9p/transport.h>
  44
  45 #define P9_PORT 564
  46 #define MAX_SOCK_BUF (64*1024)
  47 #define ERREQFLUSH      1
  48 #define MAXPOLLWADDR    2
  49
  50 /**
  51  * struct p9_fd_opts - per-transport options
  52  * @rfd: file descriptor for reading (trans=fd)
  53  * @wfd: file descriptor for writing (trans=fd)
  54  * @port: port to connect to (trans=tcp)
  55  *
  56  */
  57
  58 struct p9_fd_opts {
  59         int rfd;
  60         int wfd;
  61         u16 port;
  62 };
  63
  64 /**
  65  * struct p9_trans_fd - transport state
  66  * @rd: reference to file to read from
  67  * @wr: reference of file to write to
  68  * @conn: connection state reference
  69  *
  70  */
  71
  72 struct p9_trans_fd {
  73         struct file *rd;
  74         struct file *wr;
  75         struct p9_conn *conn;
  76 };
  77
  78 /*
  79   * Option Parsing (code inspired by NFS code)
  80   *  - a little lazy - parse all fd-transport options
  81   */
  82
  83 enum {
  84         /* Options that take integer arguments */
  85         Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
  86 };
  87
  88 static const match_table_t tokens = {
  89         {Opt_port, "port=%u"},
  90         {Opt_rfdno, "rfdno=%u"},
  91         {Opt_wfdno, "wfdno=%u"},
  92         {Opt_err, NULL},
  93 };
  94
  95 enum {
  96         Rworksched = 1,         /* read work scheduled or running */
  97         Rpending = 2,           /* can read */
  98         Wworksched = 4,         /* write work scheduled or running */
  99         Wpending = 8,           /* can write */
 100 };
 101
 102 enum {
 103         None,
 104         Flushing,
 105         Flushed,
 106 };
 107
 108 /**
 109  * struct p9_req - fd mux encoding of an rpc transaction
 110  * @lock: protects req_list
 111  * @tag: numeric tag for rpc transaction
 112  * @tcall: request &p9_fcall structure
 113  * @rcall: response &p9_fcall structure
 114  * @err: error state
 115  * @flush: flag to indicate RPC has been flushed
 116  * @req_list: list link for higher level objects to chain requests
 117  * @m: connection this request was issued on
 118  * @wqueue: wait queue that client is blocked on for this rpc
 119  *
 120  */
 121
 122 struct p9_req {
 123         spinlock_t lock;
 124         int tag;
 125         struct p9_fcall *tcall;
 126         struct p9_fcall *rcall;
 127         int err;
 128         int flush;
 129         struct list_head req_list;
 130         struct p9_conn *m;
 131         wait_queue_head_t wqueue;
 132 };
 133
 134 struct p9_poll_wait {
 135         struct p9_conn *conn;
 136         wait_queue_t wait;
 137         wait_queue_head_t *wait_addr;
 138 };
 139
 140 /**
 141  * struct p9_conn - fd mux connection state information
 142  * @lock: protects mux_list (?)
 143  * @mux_list: list link for mux to manage multiple connections (?)
 144  * @client: reference to client instance for this connection
 145  * @err: error state
 146  * @req_list: accounting for requests which have been sent
 147  * @unsent_req_list: accounting for requests that haven't been sent
 148  * @rcall: current response &p9_fcall structure
 149  * @rpos: read position in current frame
 150  * @rbuf: current read buffer
 151  * @wpos: write position for current frame
 152  * @wsize: amount of data to write for current frame
 153  * @wbuf: current write buffer
 154  * @poll_wait: array of wait_q's for various worker threads
 155  * @poll_waddr: ????
 156  * @pt: poll state
 157  * @rq: current read work
 158  * @wq: current write work
 159  * @wsched: ????
 160  *
 161  */
 162
 163 struct p9_conn {
 164         spinlock_t lock; /* protect lock structure */
 165         struct list_head mux_list;
 166         struct p9_client *client;
 167         int err;
 168         struct list_head req_list;
 169         struct list_head unsent_req_list;
 170         struct p9_fcall *rcall;
 171         int rpos;
 172         char *rbuf;
 173         int wpos;
 174         int wsize;
 175         char *wbuf;
 176         struct list_head poll_pending_link;
 177         struct p9_poll_wait poll_wait[MAXPOLLWADDR];
 178         poll_table pt;
 179         struct work_struct rq;
 180         struct work_struct wq;
 181         unsigned long wsched;
 182 };
 183
 184 static DEFINE_SPINLOCK(p9_poll_lock);
 185 static LIST_HEAD(p9_poll_pending_list);
 186 static struct workqueue_struct *p9_mux_wq;
 187 static struct task_struct *p9_poll_task;
 188
 189 static void p9_mux_poll_stop(struct p9_conn *m)
 190 {
 191         unsigned long flags;
 192         int i;
 193
 194         for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
 195                 struct p9_poll_wait *pwait = &m->poll_wait[i];
 196
 197                 if (pwait->wait_addr) {
 198                         remove_wait_queue(pwait->wait_addr, &pwait->wait);
 199                         pwait->wait_addr = NULL;
 200                 }
 201         }
 202
 203         spin_lock_irqsave(&p9_poll_lock, flags);
 204         list_del_init(&m->poll_pending_link);
 205         spin_unlock_irqrestore(&p9_poll_lock, flags);
 206 }
 207
 208 static void p9_mux_free_request(struct p9_conn *m, struct p9_req *req)
 209 {
 210         if (req->tag != P9_NOTAG &&
 211             p9_idpool_check(req->tag, m->client->tagpool))
 212                 p9_idpool_put(req->tag, m->client->tagpool);
 213         kfree(req);
 214 }
 215
 216 static void p9_conn_rpc_cb(struct p9_req *req);
 217
 218 static void p9_mux_flush_cb(struct p9_req *freq)
 219 {
 220         int tag;
 221         struct p9_conn *m = freq->m;
 222         struct p9_req *req, *rreq, *rptr;
 223
 224         P9_DPRINTK(P9_DEBUG_MUX, "mux %p tc %p rc %p err %d oldtag %d\n", m,
 225                 freq->tcall, freq->rcall, freq->err,
 226                 freq->tcall->params.tflush.oldtag);
 227
 228         spin_lock(&m->lock);
 229         tag = freq->tcall->params.tflush.oldtag;
 230         req = NULL;
 231         list_for_each_entry_safe(rreq, rptr, &m->req_list, req_list) {
 232                 if (rreq->tag == tag) {
 233                         req = rreq;
 234                         list_del(&req->req_list);
 235                         break;
 236                 }
 237         }
 238         spin_unlock(&m->lock);
 239
 240         if (req) {
 241                 spin_lock(&req->lock);
 242                 req->flush = Flushed;
 243                 spin_unlock(&req->lock);
 244
 245                 p9_conn_rpc_cb(req);
 246         }
 247
 248         kfree(freq->tcall);
 249         kfree(freq->rcall);
 250         p9_mux_free_request(m, freq);
 251 }
 252
 253 static void p9_conn_rpc_cb(struct p9_req *req)
 254 {
 255         P9_DPRINTK(P9_DEBUG_MUX, "req %p\n", req);
 256
 257         if (req->tcall->id == P9_TFLUSH) { /* flush callback */
 258                 P9_DPRINTK(P9_DEBUG_MUX, "flush req %p\n", req);
 259                 p9_mux_flush_cb(req);
 260         } else {                        /* normal wakeup path */
 261                 P9_DPRINTK(P9_DEBUG_MUX, "normal req %p\n", req);
 262                 if (req->flush != None && !req->err)
 263                         req->err = -ERESTARTSYS;
 264
 265                 wake_up(&req->wqueue);
 266         }
 267 }
 268
 269 /**
 270  * p9_conn_cancel - cancel all pending requests with error
 271  * @m: mux data
 272  * @err: error code
 273  *
 274  */
 275
 276 void p9_conn_cancel(struct p9_conn *m, int err)
 277 {
 278         struct p9_req *req, *rtmp;
 279         LIST_HEAD(cancel_list);
 280
 281         P9_DPRINTK(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
 282         m->err = err;
 283         spin_lock(&m->lock);
 284         list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
 285                 list_move(&req->req_list, &cancel_list);
 286         }
 287         list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
 288                 list_move(&req->req_list, &cancel_list);
 289         }
 290         spin_unlock(&m->lock);
 291
 292         list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
 293                 list_del(&req->req_list);
 294                 if (!req->err)
 295                         req->err = err;
 296
 297                 p9_conn_rpc_cb(req);
 298         }
 299 }
 300
 301 static void process_request(struct p9_conn *m, struct p9_req *req)
 302 {
 303         int ecode;
 304         struct p9_str *ename;
 305
 306         if (!req->err && req->rcall->id == P9_RERROR) {
 307                 ecode = req->rcall->params.rerror.errno;
 308                 ename = &req->rcall->params.rerror.error;
 309
 310                 P9_DPRINTK(P9_DEBUG_MUX, "Rerror %.*s\n", ename->len,
 311                                                                 ename->str);
 312
 313                 if (m->client->dotu)
 314                         req->err = -ecode;
 315
 316                 if (!req->err) {
 317                         req->err = p9_errstr2errno(ename->str, ename->len);
 318
 319                         /* string match failed */
 320                         if (!req->err) {
 321                                 PRINT_FCALL_ERROR("unknown error", req->rcall);
 322                                 req->err = -ESERVERFAULT;
 323                         }
 324                 }
 325         } else if (req->tcall && req->rcall->id != req->tcall->id + 1) {
 326                 P9_DPRINTK(P9_DEBUG_ERROR,
 327                                 "fcall mismatch: expected %d, got %d\n",
 328                                 req->tcall->id + 1, req->rcall->id);
 329                 if (!req->err)
 330                         req->err = -EIO;
 331         }
 332 }
 333
 334 static unsigned int
 335 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt)
 336 {
 337         int ret, n;
 338         struct p9_trans_fd *ts = NULL;
 339
 340         if (client && client->status == Connected)
 341                 ts = client->trans;
 342
 343         if (!ts)
 344                 return -EREMOTEIO;
 345
 346         if (!ts->rd->f_op || !ts->rd->f_op->poll)
 347                 return -EIO;
 348
 349         if (!ts->wr->f_op || !ts->wr->f_op->poll)
 350                 return -EIO;
 351
 352         ret = ts->rd->f_op->poll(ts->rd, pt);
 353         if (ret < 0)
 354                 return ret;
 355
 356         if (ts->rd != ts->wr) {
 357                 n = ts->wr->f_op->poll(ts->wr, pt);
 358                 if (n < 0)
 359                         return n;
 360                 ret = (ret & ~POLLOUT) | (n & ~POLLIN);
 361         }
 362
 363         return ret;
 364 }
 365
 366 /**
 367  * p9_fd_read- read from a fd
 368  * @client: client instance
 369  * @v: buffer to receive data into
 370  * @len: size of receive buffer
 371  *
 372  */
 373
 374 static int p9_fd_read(struct p9_client *client, void *v, int len)
 375 {
 376         int ret;
 377         struct p9_trans_fd *ts = NULL;
 378
 379         if (client && client->status != Disconnected)
 380                 ts = client->trans;
 381
 382         if (!ts)
 383                 return -EREMOTEIO;
 384
 385         if (!(ts->rd->f_flags & O_NONBLOCK))
 386                 P9_DPRINTK(P9_DEBUG_ERROR, "blocking read ...\n");
 387
 388         ret = kernel_read(ts->rd, ts->rd->f_pos, v, len);
 389         if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
 390                 client->status = Disconnected;
 391         return ret;
 392 }
 393
 394 /**
 395  * p9_read_work - called when there is some data to be read from a transport
 396  * @work: container of work to be done
 397  *
 398  */
 399
 400 static void p9_read_work(struct work_struct *work)
 401 {
 402         int n, err;
 403         struct p9_conn *m;
 404         struct p9_req *req, *rptr, *rreq;
 405         struct p9_fcall *rcall;
 406         char *rbuf;
 407
 408         m = container_of(work, struct p9_conn, rq);
 409
 410         if (m->err < 0)
 411                 return;
 412
 413         rcall = NULL;
 414         P9_DPRINTK(P9_DEBUG_MUX, "start mux %p pos %d\n", m, m->rpos);
 415
 416         if (!m->rcall) {
 417                 m->rcall =
 418                     kmalloc(sizeof(struct p9_fcall) + m->client->msize,
 419                                                                 GFP_KERNEL);
 420                 if (!m->rcall) {
 421                         err = -ENOMEM;
 422                         goto error;
 423                 }
 424
 425                 m->rbuf = (char *)m->rcall + sizeof(struct p9_fcall);
 426                 m->rpos = 0;
 427         }
 428
 429         clear_bit(Rpending, &m->wsched);
 430         err = p9_fd_read(m->client, m->rbuf + m->rpos,
 431                                                 m->client->msize - m->rpos);
 432         P9_DPRINTK(P9_DEBUG_MUX, "mux %p got %d bytes\n", m, err);
 433         if (err == -EAGAIN) {
 434                 clear_bit(Rworksched, &m->wsched);
 435                 return;
 436         }
 437
 438         if (err <= 0)
 439                 goto error;
 440
 441         m->rpos += err;
 442         while (m->rpos > 4) {
 443                 n = le32_to_cpu(*(__le32 *) m->rbuf);
 444                 if (n >= m->client->msize) {
 445                         P9_DPRINTK(P9_DEBUG_ERROR,
 446                                 "requested packet size too big: %d\n", n);
 447                         err = -EIO;
 448                         goto error;
 449                 }
 450
 451                 if (m->rpos < n)
 452                         break;
 453
 454                 err =
 455                     p9_deserialize_fcall(m->rbuf, n, m->rcall, m->client->dotu);
 456                 if (err < 0)
 457                         goto error;
 458
 459 #ifdef CONFIG_NET_9P_DEBUG
 460                 if ((p9_debug_level&P9_DEBUG_FCALL) == P9_DEBUG_FCALL) {
 461                         char buf[150];
 462
 463                         p9_printfcall(buf, sizeof(buf), m->rcall,
 464                                 m->client->dotu);
 465                         printk(KERN_NOTICE ">>> %p %s\n", m, buf);
 466                 }
 467 #endif
 468
 469                 rcall = m->rcall;
 470                 rbuf = m->rbuf;
 471                 if (m->rpos > n) {
 472                         m->rcall = kmalloc(sizeof(struct p9_fcall) +
 473                                                 m->client->msize, GFP_KERNEL);
 474                         if (!m->rcall) {
 475                                 err = -ENOMEM;
 476                                 goto error;
 477                         }
 478
 479                         m->rbuf = (char *)m->rcall + sizeof(struct p9_fcall);
 480                         memmove(m->rbuf, rbuf + n, m->rpos - n);
 481                         m->rpos -= n;
 482                 } else {
 483                         m->rcall = NULL;
 484                         m->rbuf = NULL;
 485                         m->rpos = 0;
 486                 }
 487
 488                 P9_DPRINTK(P9_DEBUG_MUX, "mux %p fcall id %d tag %d\n", m,
 489                                                         rcall->id, rcall->tag);
 490
 491                 req = NULL;
 492                 spin_lock(&m->lock);
 493                 list_for_each_entry_safe(rreq, rptr, &m->req_list, req_list) {
 494                         if (rreq->tag == rcall->tag) {
 495                                 req = rreq;
 496                                 if (req->flush != Flushing)
 497                                         list_del(&req->req_list);
 498                                 break;
 499                         }
 500                 }
 501                 spin_unlock(&m->lock);
 502
 503                 if (req) {
 504                         req->rcall = rcall;
 505                         process_request(m, req);
 506
 507                         if (req->flush != Flushing)
 508                                 p9_conn_rpc_cb(req);
 509                 } else {
 510                         if (err >= 0 && rcall->id != P9_RFLUSH)
 511                                 P9_DPRINTK(P9_DEBUG_ERROR,
 512                                   "unexpected response mux %p id %d tag %d\n",
 513                                   m, rcall->id, rcall->tag);
 514                         kfree(rcall);
 515                 }
 516         }
 517
 518         if (!list_empty(&m->req_list)) {
 519                 if (test_and_clear_bit(Rpending, &m->wsched))
 520                         n = POLLIN;
 521                 else
 522                         n = p9_fd_poll(m->client, NULL);
 523
 524                 if (n & POLLIN) {
 525                         P9_DPRINTK(P9_DEBUG_MUX, "schedule read work %p\n", m);
 526                         queue_work(p9_mux_wq, &m->rq);
 527                 } else
 528                         clear_bit(Rworksched, &m->wsched);
 529         } else
 530                 clear_bit(Rworksched, &m->wsched);
 531
 532         return;
 533
 534 error:
 535         p9_conn_cancel(m, err);
 536         clear_bit(Rworksched, &m->wsched);
 537 }
 538
 539 /**
 540  * p9_fd_write - write to a socket
 541  * @client: client instance
 542  * @v: buffer to send data from
 543  * @len: size of send buffer
 544  *
 545  */
 546
 547 static int p9_fd_write(struct p9_client *client, void *v, int len)
 548 {
 549         int ret;
 550         mm_segment_t oldfs;
 551         struct p9_trans_fd *ts = NULL;
 552
 553         if (client && client->status != Disconnected)
 554                 ts = client->trans;
 555
 556         if (!ts)
 557                 return -EREMOTEIO;
 558
 559         if (!(ts->wr->f_flags & O_NONBLOCK))
 560                 P9_DPRINTK(P9_DEBUG_ERROR, "blocking write ...\n");
 561
 562         oldfs = get_fs();
 563         set_fs(get_ds());
 564         /* The cast to a user pointer is valid due to the set_fs() */
 565         ret = vfs_write(ts->wr, (void __user *)v, len, &ts->wr->f_pos);
 566         set_fs(oldfs);
 567
 568         if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
 569                 client->status = Disconnected;
 570         return ret;
 571 }
 572
 573 /**
 574  * p9_write_work - called when a transport can send some data
 575  * @work: container for work to be done
 576  *
 577  */
 578
 579 static void p9_write_work(struct work_struct *work)
 580 {
 581         int n, err;
 582         struct p9_conn *m;
 583         struct p9_req *req;
 584
 585         m = container_of(work, struct p9_conn, wq);
 586
 587         if (m->err < 0) {
 588                 clear_bit(Wworksched, &m->wsched);
 589                 return;
 590         }
 591
 592         if (!m->wsize) {
 593                 if (list_empty(&m->unsent_req_list)) {
 594                         clear_bit(Wworksched, &m->wsched);
 595                         return;
 596                 }
 597
 598                 spin_lock(&m->lock);
 599 again:
 600                 req = list_entry(m->unsent_req_list.next, struct p9_req,
 601                                req_list);
 602                 list_move_tail(&req->req_list, &m->req_list);
 603                 if (req->err == ERREQFLUSH)
 604                         goto again;
 605
 606                 m->wbuf = req->tcall->sdata;
 607                 m->wsize = req->tcall->size;
 608                 m->wpos = 0;
 609                 spin_unlock(&m->lock);
 610         }
 611
 612         P9_DPRINTK(P9_DEBUG_MUX, "mux %p pos %d size %d\n", m, m->wpos,
 613                                                                 m->wsize);
 614         clear_bit(Wpending, &m->wsched);
 615         err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
 616         P9_DPRINTK(P9_DEBUG_MUX, "mux %p sent %d bytes\n", m, err);
 617         if (err == -EAGAIN) {
 618                 clear_bit(Wworksched, &m->wsched);
 619                 return;
 620         }
 621
 622         if (err < 0)
 623                 goto error;
 624         else if (err == 0) {
 625                 err = -EREMOTEIO;
 626                 goto error;
 627         }
 628
 629         m->wpos += err;
 630         if (m->wpos == m->wsize)
 631                 m->wpos = m->wsize = 0;
 632
 633         if (m->wsize == 0 && !list_empty(&m->unsent_req_list)) {
 634                 if (test_and_clear_bit(Wpending, &m->wsched))
 635                         n = POLLOUT;
 636                 else
 637                         n = p9_fd_poll(m->client, NULL);
 638
 639                 if (n & POLLOUT) {
 640                         P9_DPRINTK(P9_DEBUG_MUX, "schedule write work %p\n", m);
 641                         queue_work(p9_mux_wq, &m->wq);
 642                 } else
 643                         clear_bit(Wworksched, &m->wsched);
 644         } else
 645                 clear_bit(Wworksched, &m->wsched);
 646
 647         return;
 648
 649 error:
 650         p9_conn_cancel(m, err);
 651         clear_bit(Wworksched, &m->wsched);
 652 }
 653
 654 static int p9_pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
 655 {
 656         struct p9_poll_wait *pwait =
 657                 container_of(wait, struct p9_poll_wait, wait);
 658         struct p9_conn *m = pwait->conn;
 659         unsigned long flags;
 660         DECLARE_WAITQUEUE(dummy_wait, p9_poll_task);
 661
 662         spin_lock_irqsave(&p9_poll_lock, flags);
 663         if (list_empty(&m->poll_pending_link))
 664                 list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
 665         spin_unlock_irqrestore(&p9_poll_lock, flags);
 666
 667         /* perform the default wake up operation */
 668         return default_wake_function(&dummy_wait, mode, sync, key);
 669 }
 670
 671 /**
 672  * p9_pollwait - add poll task to the wait queue
 673  * @filp: file pointer being polled
 674  * @wait_address: wait_q to block on
 675  * @p: poll state
 676  *
 677  * called by files poll operation to add v9fs-poll task to files wait queue
 678  */
 679
 680 static void
 681 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
 682 {
 683         struct p9_conn *m = container_of(p, struct p9_conn, pt);
 684         struct p9_poll_wait *pwait = NULL;
 685         int i;
 686
 687         for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
 688                 if (m->poll_wait[i].wait_addr == NULL) {
 689                         pwait = &m->poll_wait[i];
 690                         break;
 691                 }
 692         }
 693
 694         if (!pwait) {
 695                 P9_DPRINTK(P9_DEBUG_ERROR, "not enough wait_address slots\n");
 696                 return;
 697         }
 698
 699         if (!wait_address) {
 700                 P9_DPRINTK(P9_DEBUG_ERROR, "no wait_address\n");
 701                 pwait->wait_addr = ERR_PTR(-EIO);
 702                 return;
 703         }
 704
 705         pwait->conn = m;
 706         pwait->wait_addr = wait_address;
 707         init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
 708         add_wait_queue(wait_address, &pwait->wait);
 709 }
 710
 711 /**
 712  * p9_conn_create - allocate and initialize the per-session mux data
 713  * @client: client instance
 714  *
 715  * Note: Creates the polling task if this is the first session.
 716  */
 717
 718 static struct p9_conn *p9_conn_create(struct p9_client *client)
 719 {
 720         int i, n;
 721         struct p9_conn *m;
 722
 723         P9_DPRINTK(P9_DEBUG_MUX, "client %p msize %d\n", client, client->msize);
 724         m = kzalloc(sizeof(struct p9_conn), GFP_KERNEL);
 725         if (!m)
 726                 return ERR_PTR(-ENOMEM);
 727
 728         spin_lock_init(&m->lock);
 729         INIT_LIST_HEAD(&m->mux_list);
 730         m->client = client;
 731
 732         INIT_LIST_HEAD(&m->req_list);
 733         INIT_LIST_HEAD(&m->unsent_req_list);
 734         INIT_WORK(&m->rq, p9_read_work);
 735         INIT_WORK(&m->wq, p9_write_work);
 736         INIT_LIST_HEAD(&m->poll_pending_link);
 737         init_poll_funcptr(&m->pt, p9_pollwait);
 738
 739         n = p9_fd_poll(client, &m->pt);
 740         if (n & POLLIN) {
 741                 P9_DPRINTK(P9_DEBUG_MUX, "mux %p can read\n", m);
 742                 set_bit(Rpending, &m->wsched);
 743         }
 744
 745         if (n & POLLOUT) {
 746                 P9_DPRINTK(P9_DEBUG_MUX, "mux %p can write\n", m);
 747                 set_bit(Wpending, &m->wsched);
 748         }
 749
 750         for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
 751                 if (IS_ERR(m->poll_wait[i].wait_addr)) {
 752                         p9_mux_poll_stop(m);
 753                         kfree(m);
 754                         /* return the error code */
 755                         return (void *)m->poll_wait[i].wait_addr;
 756                 }
 757         }
 758
 759         return m;
 760 }
 761
 762 /**
 763  * p9_poll_mux - polls a mux and schedules read or write works if necessary
 764  * @m: connection to poll
 765  *
 766  */
 767
 768 static void p9_poll_mux(struct p9_conn *m)
 769 {
 770         int n;
 771
 772         if (m->err < 0)
 773                 return;
 774
 775         n = p9_fd_poll(m->client, NULL);
 776         if (n < 0 || n & (POLLERR | POLLHUP | POLLNVAL)) {
 777                 P9_DPRINTK(P9_DEBUG_MUX, "error mux %p err %d\n", m, n);
 778                 if (n >= 0)
 779                         n = -ECONNRESET;
 780                 p9_conn_cancel(m, n);
 781         }
 782
 783         if (n & POLLIN) {
 784                 set_bit(Rpending, &m->wsched);
 785                 P9_DPRINTK(P9_DEBUG_MUX, "mux %p can read\n", m);
 786                 if (!test_and_set_bit(Rworksched, &m->wsched)) {
 787                         P9_DPRINTK(P9_DEBUG_MUX, "schedule read work %p\n", m);
 788                         queue_work(p9_mux_wq, &m->rq);
 789                 }
 790         }
 791
 792         if (n & POLLOUT) {
 793                 set_bit(Wpending, &m->wsched);
 794                 P9_DPRINTK(P9_DEBUG_MUX, "mux %p can write\n", m);
 795                 if ((m->wsize || !list_empty(&m->unsent_req_list))
 796                     && !test_and_set_bit(Wworksched, &m->wsched)) {
 797                         P9_DPRINTK(P9_DEBUG_MUX, "schedule write work %p\n", m);
 798                         queue_work(p9_mux_wq, &m->wq);
 799                 }
 800         }
 801 }
 802
 803 /**
 804  * p9_send_request - send 9P request
 805  * The function can sleep until the request is scheduled for sending.
 806  * The function can be interrupted. Return from the function is not
 807  * a guarantee that the request is sent successfully. Can return errors
 808  * that can be retrieved by PTR_ERR macros.
 809  *
 810  * @m: mux data
 811  * @tc: request to be sent
 812  *
 813  */
 814
 815 static struct p9_req *p9_send_request(struct p9_conn *m, struct p9_fcall *tc)
 816 {
 817         int n;
 818         struct p9_req *req;
 819
 820         P9_DPRINTK(P9_DEBUG_MUX, "mux %p task %p tcall %p id %d\n", m, current,
 821                 tc, tc->id);
 822         if (m->err < 0)
 823                 return ERR_PTR(m->err);
 824
 825         req = kmalloc(sizeof(struct p9_req), GFP_KERNEL);
 826         if (!req)
 827                 return ERR_PTR(-ENOMEM);
 828
 829         n = P9_NOTAG;
 830         if (tc->id != P9_TVERSION) {
 831                 n = p9_idpool_get(m->client->tagpool);
 832                 if (n < 0) {
 833                         kfree(req);
 834                         return ERR_PTR(-ENOMEM);
 835                 }
 836         }
 837
 838         p9_set_tag(tc, n);
 839
 840 #ifdef CONFIG_NET_9P_DEBUG
 841         if ((p9_debug_level&P9_DEBUG_FCALL) == P9_DEBUG_FCALL) {
 842                 char buf[150];
 843
 844                 p9_printfcall(buf, sizeof(buf), tc, m->client->dotu);
 845                 printk(KERN_NOTICE "<<< %p %s\n", m, buf);
 846         }
 847 #endif
 848
 849         spin_lock_init(&req->lock);
 850         req->m = m;
 851         init_waitqueue_head(&req->wqueue);
 852         req->tag = n;
 853         req->tcall = tc;
 854         req->rcall = NULL;
 855         req->err = 0;
 856         req->flush = None;
 857
 858         spin_lock(&m->lock);
 859         list_add_tail(&req->req_list, &m->unsent_req_list);
 860         spin_unlock(&m->lock);
 861
 862         if (test_and_clear_bit(Wpending, &m->wsched))
 863                 n = POLLOUT;
 864         else
 865                 n = p9_fd_poll(m->client, NULL);
 866
 867         if (n & POLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
 868                 queue_work(p9_mux_wq, &m->wq);
 869
 870         return req;
 871 }
 872
 873 static int
 874 p9_mux_flush_request(struct p9_conn *m, struct p9_req *req)
 875 {
 876         struct p9_fcall *fc;
 877         struct p9_req *rreq, *rptr;
 878
 879         P9_DPRINTK(P9_DEBUG_MUX, "mux %p req %p tag %d\n", m, req, req->tag);
 880
 881         /* if a response was received for a request, do nothing */
 882         spin_lock(&req->lock);
 883         if (req->rcall || req->err) {
 884                 spin_unlock(&req->lock);
 885                 P9_DPRINTK(P9_DEBUG_MUX,
 886                         "mux %p req %p response already received\n", m, req);
 887                 return 0;
 888         }
 889
 890         req->flush = Flushing;
 891         spin_unlock(&req->lock);
 892
 893         spin_lock(&m->lock);
 894         /* if the request is not sent yet, just remove it from the list */
 895         list_for_each_entry_safe(rreq, rptr, &m->unsent_req_list, req_list) {
 896                 if (rreq->tag == req->tag) {
 897                         P9_DPRINTK(P9_DEBUG_MUX,
 898                            "mux %p req %p request is not sent yet\n", m, req);
 899                         list_del(&rreq->req_list);
 900                         req->flush = Flushed;
 901                         spin_unlock(&m->lock);
 902                         p9_conn_rpc_cb(req);
 903                         return 0;
 904                 }
 905         }
 906         spin_unlock(&m->lock);
 907
 908         clear_thread_flag(TIF_SIGPENDING);
 909         fc = p9_create_tflush(req->tag);
 910         p9_send_request(m, fc);
 911         return 1;
 912 }
 913
 914 /**
 915  * p9_fd_rpc- sends 9P request and waits until a response is available.
 916  *      The function can be interrupted.
 917  * @client: client instance
 918  * @tc: request to be sent
 919  * @rc: pointer where a pointer to the response is stored
 920  *
 921  */
 922
 923 int
 924 p9_fd_rpc(struct p9_client *client, struct p9_fcall *tc, struct p9_fcall **rc)
 925 {
 926         struct p9_trans_fd *p = client->trans;
 927         struct p9_conn *m = p->conn;
 928         int err, sigpending;
 929         unsigned long flags;
 930         struct p9_req *req;
 931
 932         if (rc)
 933                 *rc = NULL;
 934
 935         sigpending = 0;
 936         if (signal_pending(current)) {
 937                 sigpending = 1;
 938                 clear_thread_flag(TIF_SIGPENDING);
 939         }
 940
 941         req = p9_send_request(m, tc);
 942         if (IS_ERR(req)) {
 943                 err = PTR_ERR(req);
 944                 P9_DPRINTK(P9_DEBUG_MUX, "error %d\n", err);
 945                 return err;
 946         }
 947
 948         err = wait_event_interruptible(req->wqueue, req->rcall != NULL ||
 949                                                                 req->err < 0);
 950         if (req->err < 0)
 951                 err = req->err;
 952
 953         if (err == -ERESTARTSYS && client->status == Connected
 954                                                         && m->err == 0) {
 955                 if (p9_mux_flush_request(m, req)) {
 956                         /* wait until we get response of the flush message */
 957                         do {
 958                                 clear_thread_flag(TIF_SIGPENDING);
 959                                 err = wait_event_interruptible(req->wqueue,
 960                                         req->rcall || req->err);
 961                         } while (!req->rcall && !req->err &&
 962                                         err == -ERESTARTSYS &&
 963                                         client->status == Connected && !m->err);
 964
 965                         err = -ERESTARTSYS;
 966                 }
 967                 sigpending = 1;
 968         }
 969
 970         if (sigpending) {
 971                 spin_lock_irqsave(&current->sighand->siglock, flags);
 972                 recalc_sigpending();
 973                 spin_unlock_irqrestore(&current->sighand->siglock, flags);
 974         }
 975
 976         if (rc)
 977                 *rc = req->rcall;
 978         else
 979                 kfree(req->rcall);
 980
 981         p9_mux_free_request(m, req);
 982         if (err > 0)
 983                 err = -EIO;
 984
 985         return err;
 986 }
 987
 988 /**
 989  * parse_options - parse mount options into session structure
 990  * @options: options string passed from mount
 991  * @opts: transport-specific structure to parse options into
 992  *
 993  * Returns 0 upon success, -ERRNO upon failure
 994  */
 995
 996 static int parse_opts(char *params, struct p9_fd_opts *opts)
 997 {
 998         char *p;
 999         substring_t args[MAX_OPT_ARGS];
1000         int option;
1001         char *options;
1002         int ret;
1003
1004         opts->port = P9_PORT;
1005         opts->rfd = ~0;
1006         opts->wfd = ~0;
1007
1008         if (!params)
1009                 return 0;
1010
1011         options = kstrdup(params, GFP_KERNEL);
1012         if (!options) {
1013                 P9_DPRINTK(P9_DEBUG_ERROR,
1014                                 "failed to allocate copy of option string\n");
1015                 return -ENOMEM;
1016         }
1017
1018         while ((p = strsep(&options, ",")) != NULL) {
1019                 int token;
1020                 int r;
1021                 if (!*p)
1022                         continue;
1023                 token = match_token(p, tokens, args);
1024                 r = match_int(&args[0], &option);
1025                 if (r < 0) {
1026                         P9_DPRINTK(P9_DEBUG_ERROR,
1027                          "integer field, but no integer?\n");
1028                         ret = r;
1029                         continue;
1030                 }
1031                 switch (token) {
1032                 case Opt_port:
1033                         opts->port = option;
1034                         break;
1035                 case Opt_rfdno:
1036                         opts->rfd = option;
1037                         break;
1038                 case Opt_wfdno:
1039                         opts->wfd = option;
1040                         break;
1041                 default:
1042                         continue;
1043                 }
1044         }
1045         kfree(options);
1046         return 0;
1047 }
1048
1049 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
1050 {
1051         struct p9_trans_fd *ts = kmalloc(sizeof(struct p9_trans_fd),
1052                                            GFP_KERNEL);
1053         if (!ts)
1054                 return -ENOMEM;
1055
1056         ts->rd = fget(rfd);
1057         ts->wr = fget(wfd);
1058         if (!ts->rd || !ts->wr) {
1059                 if (ts->rd)
1060                         fput(ts->rd);
1061                 if (ts->wr)
1062                         fput(ts->wr);
1063                 kfree(ts);
1064                 return -EIO;
1065         }
1066
1067         client->trans = ts;
1068         client->status = Connected;
1069
1070         return 0;
1071 }
1072
1073 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
1074 {
1075         int fd, ret;
1076
1077         csocket->sk->sk_allocation = GFP_NOIO;
1078         fd = sock_map_fd(csocket, 0);
1079         if (fd < 0) {
1080                 P9_EPRINTK(KERN_ERR, "p9_socket_open: failed to map fd\n");
1081                 return fd;
1082         }
1083
1084         ret = p9_fd_open(client, fd, fd);
1085         if (ret < 0) {
1086                 P9_EPRINTK(KERN_ERR, "p9_socket_open: failed to open fd\n");
1087                 sockfd_put(csocket);
1088                 return ret;
1089         }
1090
1091         ((struct p9_trans_fd *)client->trans)->rd->f_flags |= O_NONBLOCK;
1092
1093         return 0;
1094 }
1095
1096 /**
1097  * p9_mux_destroy - cancels all pending requests and frees mux resources
1098  * @m: mux to destroy
1099  *
1100  */
1101
1102 static void p9_conn_destroy(struct p9_conn *m)
1103 {
1104         P9_DPRINTK(P9_DEBUG_MUX, "mux %p prev %p next %p\n", m,
1105                 m->mux_list.prev, m->mux_list.next);
1106
1107         p9_mux_poll_stop(m);
1108         cancel_work_sync(&m->rq);
1109         cancel_work_sync(&m->wq);
1110
1111         p9_conn_cancel(m, -ECONNRESET);
1112
1113         m->client = NULL;
1114         kfree(m);
1115 }
1116
1117 /**
1118  * p9_fd_close - shutdown file descriptor transport
1119  * @client: client instance
1120  *
1121  */
1122
1123 static void p9_fd_close(struct p9_client *client)
1124 {
1125         struct p9_trans_fd *ts;
1126
1127         if (!client)
1128                 return;
1129
1130         ts = client->trans;
1131         if (!ts)
1132                 return;
1133
1134         client->status = Disconnected;
1135
1136         p9_conn_destroy(ts->conn);
1137
1138         if (ts->rd)
1139                 fput(ts->rd);
1140         if (ts->wr)
1141                 fput(ts->wr);
1142
1143         kfree(ts);
1144 }
1145
1146 /*
1147  * stolen from NFS - maybe should be made a generic function?
1148  */
1149 static inline int valid_ipaddr4(const char *buf)
1150 {
1151         int rc, count, in[4];
1152
1153         rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
1154         if (rc != 4)
1155                 return -EINVAL;
1156         for (count = 0; count < 4; count++) {
1157                 if (in[count] > 255)
1158                         return -EINVAL;
1159         }
1160         return 0;
1161 }
1162
1163 static int
1164 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
1165 {
1166         int err;
1167         struct socket *csocket;
1168         struct sockaddr_in sin_server;
1169         struct p9_fd_opts opts;
1170         struct p9_trans_fd *p = NULL; /* this gets allocated in p9_fd_open */
1171
1172         err = parse_opts(args, &opts);
1173         if (err < 0)
1174                 return err;
1175
1176         if (valid_ipaddr4(addr) < 0)
1177                 return -EINVAL;
1178
1179         csocket = NULL;
1180
1181         sin_server.sin_family = AF_INET;
1182         sin_server.sin_addr.s_addr = in_aton(addr);
1183         sin_server.sin_port = htons(opts.port);
1184         sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &csocket);
1185
1186         if (!csocket) {
1187                 P9_EPRINTK(KERN_ERR, "p9_trans_tcp: problem creating socket\n");
1188                 err = -EIO;
1189                 goto error;
1190         }
1191
1192         err = csocket->ops->connect(csocket,
1193                                     (struct sockaddr *)&sin_server,
1194                                     sizeof(struct sockaddr_in), 0);
1195         if (err < 0) {
1196                 P9_EPRINTK(KERN_ERR,
1197                         "p9_trans_tcp: problem connecting socket to %s\n",
1198                         addr);
1199                 goto error;
1200         }
1201
1202         err = p9_socket_open(client, csocket);
1203         if (err < 0)
1204                 goto error;
1205
1206         p = (struct p9_trans_fd *) client->trans;
1207         p->conn = p9_conn_create(client);
1208         if (IS_ERR(p->conn)) {
1209                 err = PTR_ERR(p->conn);
1210                 p->conn = NULL;
1211                 goto error;
1212         }
1213
1214         return 0;
1215
1216 error:
1217         if (csocket)
1218                 sock_release(csocket);
1219
1220         kfree(p);
1221
1222         return err;
1223 }
1224
1225 static int
1226 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
1227 {
1228         int err;
1229         struct socket *csocket;
1230         struct sockaddr_un sun_server;
1231         struct p9_trans_fd *p = NULL; /* this gets allocated in p9_fd_open */
1232
1233         csocket = NULL;
1234
1235         if (strlen(addr) > UNIX_PATH_MAX) {
1236                 P9_EPRINTK(KERN_ERR, "p9_trans_unix: address too long: %s\n",
1237                         addr);
1238                 err = -ENAMETOOLONG;
1239                 goto error;
1240         }
1241
1242         sun_server.sun_family = PF_UNIX;
1243         strcpy(sun_server.sun_path, addr);
1244         sock_create_kern(PF_UNIX, SOCK_STREAM, 0, &csocket);
1245         err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1246                         sizeof(struct sockaddr_un) - 1, 0);
1247         if (err < 0) {
1248                 P9_EPRINTK(KERN_ERR,
1249                         "p9_trans_unix: problem connecting socket: %s: %d\n",
1250                         addr, err);
1251                 goto error;
1252         }
1253
1254         err = p9_socket_open(client, csocket);
1255         if (err < 0)
1256                 goto error;
1257
1258         p = (struct p9_trans_fd *) client->trans;
1259         p->conn = p9_conn_create(client);
1260         if (IS_ERR(p->conn)) {
1261                 err = PTR_ERR(p->conn);
1262                 p->conn = NULL;
1263                 goto error;
1264         }
1265
1266         return 0;
1267
1268 error:
1269         if (csocket)
1270                 sock_release(csocket);
1271
1272         kfree(p);
1273         return err;
1274 }
1275
1276 static int
1277 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1278 {
1279         int err;
1280         struct p9_fd_opts opts;
1281         struct p9_trans_fd *p = NULL; /* this get allocated in p9_fd_open */
1282
1283         parse_opts(args, &opts);
1284
1285         if (opts.rfd == ~0 || opts.wfd == ~0) {
1286                 printk(KERN_ERR "v9fs: Insufficient options for proto=fd\n");
1287                 return -ENOPROTOOPT;
1288         }
1289
1290         err = p9_fd_open(client, opts.rfd, opts.wfd);
1291         if (err < 0)
1292                 goto error;
1293
1294         p = (struct p9_trans_fd *) client->trans;
1295         p->conn = p9_conn_create(client);
1296         if (IS_ERR(p->conn)) {
1297                 err = PTR_ERR(p->conn);
1298                 p->conn = NULL;
1299                 goto error;
1300         }
1301
1302         return 0;
1303
1304 error:
1305         kfree(p);
1306         return err;
1307 }
1308
1309 static struct p9_trans_module p9_tcp_trans = {
1310         .name = "tcp",
1311         .maxsize = MAX_SOCK_BUF,
1312         .def = 1,
1313         .create = p9_fd_create_tcp,
1314         .close = p9_fd_close,
1315         .rpc = p9_fd_rpc,
1316         .owner = THIS_MODULE,
1317 };
1318
1319 static struct p9_trans_module p9_unix_trans = {
1320         .name = "unix",
1321         .maxsize = MAX_SOCK_BUF,
1322         .def = 0,
1323         .create = p9_fd_create_unix,
1324         .close = p9_fd_close,
1325         .rpc = p9_fd_rpc,
1326         .owner = THIS_MODULE,
1327 };
1328
1329 static struct p9_trans_module p9_fd_trans = {
1330         .name = "fd",
1331         .maxsize = MAX_SOCK_BUF,
1332         .def = 0,
1333         .create = p9_fd_create,
1334         .close = p9_fd_close,
1335         .rpc = p9_fd_rpc,
1336         .owner = THIS_MODULE,
1337 };
1338
1339 /**
1340  * p9_poll_proc - poll worker thread
1341  * @a: thread state and arguments
1342  *
1343  * polls all v9fs transports for new events and queues the appropriate
1344  * work to the work queue
1345  *
1346  */
1347
1348 static int p9_poll_proc(void *a)
1349 {
1350         unsigned long flags;
1351
1352         P9_DPRINTK(P9_DEBUG_MUX, "start %p\n", current);
1353  repeat:
1354         spin_lock_irqsave(&p9_poll_lock, flags);
1355         while (!list_empty(&p9_poll_pending_list)) {
1356                 struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1357                                                         struct p9_conn,
1358                                                         poll_pending_link);
1359                 list_del_init(&conn->poll_pending_link);
1360                 spin_unlock_irqrestore(&p9_poll_lock, flags);
1361
1362                 p9_poll_mux(conn);
1363
1364                 spin_lock_irqsave(&p9_poll_lock, flags);
1365         }
1366         spin_unlock_irqrestore(&p9_poll_lock, flags);
1367
1368         set_current_state(TASK_INTERRUPTIBLE);
1369         if (list_empty(&p9_poll_pending_list)) {
1370                 P9_DPRINTK(P9_DEBUG_MUX, "sleeping...\n");
1371                 schedule();
1372         }
1373         __set_current_state(TASK_RUNNING);
1374
1375         if (!kthread_should_stop())
1376                 goto repeat;
1377
1378         P9_DPRINTK(P9_DEBUG_MUX, "finish\n");
1379         return 0;
1380 }
1381
1382 int p9_trans_fd_init(void)
1383 {
1384         p9_mux_wq = create_workqueue("v9fs");
1385         if (!p9_mux_wq) {
1386                 printk(KERN_WARNING "v9fs: mux: creating workqueue failed\n");
1387                 return -ENOMEM;
1388         }
1389
1390         p9_poll_task = kthread_run(p9_poll_proc, NULL, "v9fs-poll");
1391         if (IS_ERR(p9_poll_task)) {
1392                 destroy_workqueue(p9_mux_wq);
1393                 printk(KERN_WARNING "v9fs: mux: creating poll task failed\n");
1394                 return PTR_ERR(p9_poll_task);
1395         }
1396
1397         v9fs_register_trans(&p9_tcp_trans);
1398         v9fs_register_trans(&p9_unix_trans);
1399         v9fs_register_trans(&p9_fd_trans);
1400
1401         return 0;
1402 }
1403
1404 void p9_trans_fd_exit(void)
1405 {
1406         kthread_stop(p9_poll_task);
1407         v9fs_unregister_trans(&p9_tcp_trans);
1408         v9fs_unregister_trans(&p9_unix_trans);
1409         v9fs_unregister_trans(&p9_fd_trans);
1410
1411         destroy_workqueue(p9_mux_wq);
1412 }