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 #include <linux/syscalls.h> /* killme */
  46
  47 #define P9_PORT 564
  48 #define MAX_SOCK_BUF (64*1024)
  49 #define MAXPOLLWADDR    2
  50
  51 /**
  52  * struct p9_fd_opts - per-transport options
  53  * @rfd: file descriptor for reading (trans=fd)
  54  * @wfd: file descriptor for writing (trans=fd)
  55  * @port: port to connect to (trans=tcp)
  56  *
  57  */
  58
  59 struct p9_fd_opts {
  60         int rfd;
  61         int wfd;
  62         u16 port;
  63 };
  64
  65 /**
  66  * struct p9_trans_fd - transport state
  67  * @rd: reference to file to read from
  68  * @wr: reference of file to write to
  69  * @conn: connection state reference
  70  *
  71  */
  72
  73 struct p9_trans_fd {
  74         struct file *rd;
  75         struct file *wr;
  76         struct p9_conn *conn;
  77 };
  78
  79 /*
  80   * Option Parsing (code inspired by NFS code)
  81   *  - a little lazy - parse all fd-transport options
  82   */
  83
  84 enum {
  85         /* Options that take integer arguments */
  86         Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
  87 };
  88
  89 static const match_table_t tokens = {
  90         {Opt_port, "port=%u"},
  91         {Opt_rfdno, "rfdno=%u"},
  92         {Opt_wfdno, "wfdno=%u"},
  93         {Opt_err, NULL},
  94 };
  95
  96 enum {
  97         Rworksched = 1,         /* read work scheduled or running */
  98         Rpending = 2,           /* can read */
  99         Wworksched = 4,         /* write work scheduled or running */
 100         Wpending = 8,           /* can write */
 101 };
 102
 103 struct p9_poll_wait {
 104         struct p9_conn *conn;
 105         wait_queue_t wait;
 106         wait_queue_head_t *wait_addr;
 107 };
 108
 109 /**
 110  * struct p9_conn - fd mux connection state information
 111  * @mux_list: list link for mux to manage multiple connections (?)
 112  * @client: reference to client instance for this connection
 113  * @err: error state
 114  * @req_list: accounting for requests which have been sent
 115  * @unsent_req_list: accounting for requests that haven't been sent
 116  * @req: current request being processed (if any)
 117  * @tmp_buf: temporary buffer to read in header
 118  * @rsize: amount to read for current frame
 119  * @rpos: read position in current frame
 120  * @rbuf: current read buffer
 121  * @wpos: write position for current frame
 122  * @wsize: amount of data to write for current frame
 123  * @wbuf: current write buffer
 124  * @poll_pending_link: pending links to be polled per conn
 125  * @poll_wait: array of wait_q's for various worker threads
 126  * @pt: poll state
 127  * @rq: current read work
 128  * @wq: current write work
 129  * @wsched: ????
 130  *
 131  */
 132
 133 struct p9_conn {
 134         struct list_head mux_list;
 135         struct p9_client *client;
 136         int err;
 137         struct list_head req_list;
 138         struct list_head unsent_req_list;
 139         struct p9_req_t *req;
 140         char tmp_buf[7];
 141         int rsize;
 142         int rpos;
 143         char *rbuf;
 144         int wpos;
 145         int wsize;
 146         char *wbuf;
 147         struct list_head poll_pending_link;
 148         struct p9_poll_wait poll_wait[MAXPOLLWADDR];
 149         poll_table pt;
 150         struct work_struct rq;
 151         struct work_struct wq;
 152         unsigned long wsched;
 153 };
 154
 155 static DEFINE_SPINLOCK(p9_poll_lock);
 156 static LIST_HEAD(p9_poll_pending_list);
 157 static struct workqueue_struct *p9_mux_wq;
 158 static struct task_struct *p9_poll_task;
 159
 160 static void p9_mux_poll_stop(struct p9_conn *m)
 161 {
 162         unsigned long flags;
 163         int i;
 164
 165         for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
 166                 struct p9_poll_wait *pwait = &m->poll_wait[i];
 167
 168                 if (pwait->wait_addr) {
 169                         remove_wait_queue(pwait->wait_addr, &pwait->wait);
 170                         pwait->wait_addr = NULL;
 171                 }
 172         }
 173
 174         spin_lock_irqsave(&p9_poll_lock, flags);
 175         list_del_init(&m->poll_pending_link);
 176         spin_unlock_irqrestore(&p9_poll_lock, flags);
 177 }
 178
 179 /**
 180  * p9_conn_cancel - cancel all pending requests with error
 181  * @m: mux data
 182  * @err: error code
 183  *
 184  */
 185
 186 static void p9_conn_cancel(struct p9_conn *m, int err)
 187 {
 188         struct p9_req_t *req, *rtmp;
 189         unsigned long flags;
 190         LIST_HEAD(cancel_list);
 191
 192         P9_DPRINTK(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
 193
 194         spin_lock_irqsave(&m->client->lock, flags);
 195
 196         if (m->err) {
 197                 spin_unlock_irqrestore(&m->client->lock, flags);
 198                 return;
 199         }
 200
 201         m->err = err;
 202
 203         list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
 204                 req->status = REQ_STATUS_ERROR;
 205                 if (!req->t_err)
 206                         req->t_err = err;
 207                 list_move(&req->req_list, &cancel_list);
 208         }
 209         list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
 210                 req->status = REQ_STATUS_ERROR;
 211                 if (!req->t_err)
 212                         req->t_err = err;
 213                 list_move(&req->req_list, &cancel_list);
 214         }
 215         spin_unlock_irqrestore(&m->client->lock, flags);
 216
 217         list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
 218                 P9_DPRINTK(P9_DEBUG_ERROR, "call back req %p\n", req);
 219                 list_del(&req->req_list);
 220                 p9_client_cb(m->client, req);
 221         }
 222 }
 223
 224 static unsigned int
 225 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt)
 226 {
 227         int ret, n;
 228         struct p9_trans_fd *ts = NULL;
 229
 230         if (client && client->status == Connected)
 231                 ts = client->trans;
 232
 233         if (!ts)
 234                 return -EREMOTEIO;
 235
 236         if (!ts->rd->f_op || !ts->rd->f_op->poll)
 237                 return -EIO;
 238
 239         if (!ts->wr->f_op || !ts->wr->f_op->poll)
 240                 return -EIO;
 241
 242         ret = ts->rd->f_op->poll(ts->rd, pt);
 243         if (ret < 0)
 244                 return ret;
 245
 246         if (ts->rd != ts->wr) {
 247                 n = ts->wr->f_op->poll(ts->wr, pt);
 248                 if (n < 0)
 249                         return n;
 250                 ret = (ret & ~POLLOUT) | (n & ~POLLIN);
 251         }
 252
 253         return ret;
 254 }
 255
 256 /**
 257  * p9_fd_read- read from a fd
 258  * @client: client instance
 259  * @v: buffer to receive data into
 260  * @len: size of receive buffer
 261  *
 262  */
 263
 264 static int p9_fd_read(struct p9_client *client, void *v, int len)
 265 {
 266         int ret;
 267         struct p9_trans_fd *ts = NULL;
 268
 269         if (client && client->status != Disconnected)
 270                 ts = client->trans;
 271
 272         if (!ts)
 273                 return -EREMOTEIO;
 274
 275         if (!(ts->rd->f_flags & O_NONBLOCK))
 276                 P9_DPRINTK(P9_DEBUG_ERROR, "blocking read ...\n");
 277
 278         ret = kernel_read(ts->rd, ts->rd->f_pos, v, len);
 279         if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
 280                 client->status = Disconnected;
 281         return ret;
 282 }
 283
 284 /**
 285  * p9_read_work - called when there is some data to be read from a transport
 286  * @work: container of work to be done
 287  *
 288  */
 289
 290 static void p9_read_work(struct work_struct *work)
 291 {
 292         int n, err;
 293         struct p9_conn *m;
 294
 295         m = container_of(work, struct p9_conn, rq);
 296
 297         if (m->err < 0)
 298                 return;
 299
 300         P9_DPRINTK(P9_DEBUG_TRANS, "start mux %p pos %d\n", m, m->rpos);
 301
 302         if (!m->rbuf) {
 303                 m->rbuf = m->tmp_buf;
 304                 m->rpos = 0;
 305                 m->rsize = 7; /* start by reading header */
 306         }
 307
 308         clear_bit(Rpending, &m->wsched);
 309         P9_DPRINTK(P9_DEBUG_TRANS, "read mux %p pos %d size: %d = %d\n", m,
 310                                         m->rpos, m->rsize, m->rsize-m->rpos);
 311         err = p9_fd_read(m->client, m->rbuf + m->rpos,
 312                                                 m->rsize - m->rpos);
 313         P9_DPRINTK(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
 314         if (err == -EAGAIN) {
 315                 clear_bit(Rworksched, &m->wsched);
 316                 return;
 317         }
 318
 319         if (err <= 0)
 320                 goto error;
 321
 322         m->rpos += err;
 323
 324         if ((!m->req) && (m->rpos == m->rsize)) { /* header read in */
 325                 u16 tag;
 326                 P9_DPRINTK(P9_DEBUG_TRANS, "got new header\n");
 327
 328                 n = le32_to_cpu(*(__le32 *) m->rbuf); /* read packet size */
 329                 if (n >= m->client->msize) {
 330                         P9_DPRINTK(P9_DEBUG_ERROR,
 331                                 "requested packet size too big: %d\n", n);
 332                         err = -EIO;
 333                         goto error;
 334                 }
 335
 336                 tag = le16_to_cpu(*(__le16 *) (m->rbuf+5)); /* read tag */
 337                 P9_DPRINTK(P9_DEBUG_TRANS,
 338                         "mux %p pkt: size: %d bytes tag: %d\n", m, n, tag);
 339
 340                 m->req = p9_tag_lookup(m->client, tag);
 341                 if (!m->req || (m->req->status != REQ_STATUS_SENT &&
 342                                         m->req->status != REQ_STATUS_FLSH)) {
 343                         P9_DPRINTK(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
 344                                                                  tag);
 345                         err = -EIO;
 346                         goto error;
 347                 }
 348
 349                 if (m->req->rc == NULL) {
 350                         m->req->rc = kmalloc(sizeof(struct p9_fcall) +
 351                                                 m->client->msize, GFP_KERNEL);
 352                         if (!m->req->rc) {
 353                                 m->req = NULL;
 354                                 err = -ENOMEM;
 355                                 goto error;
 356                         }
 357                 }
 358                 m->rbuf = (char *)m->req->rc + sizeof(struct p9_fcall);
 359                 memcpy(m->rbuf, m->tmp_buf, m->rsize);
 360                 m->rsize = n;
 361         }
 362
 363         /* not an else because some packets (like clunk) have no payload */
 364         if ((m->req) && (m->rpos == m->rsize)) { /* packet is read in */
 365                 P9_DPRINTK(P9_DEBUG_TRANS, "got new packet\n");
 366                 spin_lock(&m->client->lock);
 367                 if (m->req->status != REQ_STATUS_ERROR)
 368                         m->req->status = REQ_STATUS_RCVD;
 369                 list_del(&m->req->req_list);
 370                 spin_unlock(&m->client->lock);
 371                 p9_client_cb(m->client, m->req);
 372                 m->rbuf = NULL;
 373                 m->rpos = 0;
 374                 m->rsize = 0;
 375                 m->req = NULL;
 376         }
 377
 378         if (!list_empty(&m->req_list)) {
 379                 if (test_and_clear_bit(Rpending, &m->wsched))
 380                         n = POLLIN;
 381                 else
 382                         n = p9_fd_poll(m->client, NULL);
 383
 384                 if (n & POLLIN) {
 385                         P9_DPRINTK(P9_DEBUG_TRANS, "sched read work %p\n", m);
 386                         queue_work(p9_mux_wq, &m->rq);
 387                 } else
 388                         clear_bit(Rworksched, &m->wsched);
 389         } else
 390                 clear_bit(Rworksched, &m->wsched);
 391
 392         return;
 393 error:
 394         p9_conn_cancel(m, err);
 395         clear_bit(Rworksched, &m->wsched);
 396 }
 397
 398 /**
 399  * p9_fd_write - write to a socket
 400  * @client: client instance
 401  * @v: buffer to send data from
 402  * @len: size of send buffer
 403  *
 404  */
 405
 406 static int p9_fd_write(struct p9_client *client, void *v, int len)
 407 {
 408         int ret;
 409         mm_segment_t oldfs;
 410         struct p9_trans_fd *ts = NULL;
 411
 412         if (client && client->status != Disconnected)
 413                 ts = client->trans;
 414
 415         if (!ts)
 416                 return -EREMOTEIO;
 417
 418         if (!(ts->wr->f_flags & O_NONBLOCK))
 419                 P9_DPRINTK(P9_DEBUG_ERROR, "blocking write ...\n");
 420
 421         oldfs = get_fs();
 422         set_fs(get_ds());
 423         /* The cast to a user pointer is valid due to the set_fs() */
 424         ret = vfs_write(ts->wr, (__force void __user *)v, len, &ts->wr->f_pos);
 425         set_fs(oldfs);
 426
 427         if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
 428                 client->status = Disconnected;
 429         return ret;
 430 }
 431
 432 /**
 433  * p9_write_work - called when a transport can send some data
 434  * @work: container for work to be done
 435  *
 436  */
 437
 438 static void p9_write_work(struct work_struct *work)
 439 {
 440         int n, err;
 441         struct p9_conn *m;
 442         struct p9_req_t *req;
 443
 444         m = container_of(work, struct p9_conn, wq);
 445
 446         if (m->err < 0) {
 447                 clear_bit(Wworksched, &m->wsched);
 448                 return;
 449         }
 450
 451         if (!m->wsize) {
 452                 if (list_empty(&m->unsent_req_list)) {
 453                         clear_bit(Wworksched, &m->wsched);
 454                         return;
 455                 }
 456
 457                 spin_lock(&m->client->lock);
 458                 req = list_entry(m->unsent_req_list.next, struct p9_req_t,
 459                                req_list);
 460                 req->status = REQ_STATUS_SENT;
 461                 P9_DPRINTK(P9_DEBUG_TRANS, "move req %p\n", req);
 462                 list_move_tail(&req->req_list, &m->req_list);
 463
 464                 m->wbuf = req->tc->sdata;
 465                 m->wsize = req->tc->size;
 466                 m->wpos = 0;
 467                 spin_unlock(&m->client->lock);
 468         }
 469
 470         P9_DPRINTK(P9_DEBUG_TRANS, "mux %p pos %d size %d\n", m, m->wpos,
 471                                                                 m->wsize);
 472         clear_bit(Wpending, &m->wsched);
 473         err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
 474         P9_DPRINTK(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
 475         if (err == -EAGAIN) {
 476                 clear_bit(Wworksched, &m->wsched);
 477                 return;
 478         }
 479
 480         if (err < 0)
 481                 goto error;
 482         else if (err == 0) {
 483                 err = -EREMOTEIO;
 484                 goto error;
 485         }
 486
 487         m->wpos += err;
 488         if (m->wpos == m->wsize)
 489                 m->wpos = m->wsize = 0;
 490
 491         if (m->wsize == 0 && !list_empty(&m->unsent_req_list)) {
 492                 if (test_and_clear_bit(Wpending, &m->wsched))
 493                         n = POLLOUT;
 494                 else
 495                         n = p9_fd_poll(m->client, NULL);
 496
 497                 if (n & POLLOUT) {
 498                         P9_DPRINTK(P9_DEBUG_TRANS, "sched write work %p\n", m);
 499                         queue_work(p9_mux_wq, &m->wq);
 500                 } else
 501                         clear_bit(Wworksched, &m->wsched);
 502         } else
 503                 clear_bit(Wworksched, &m->wsched);
 504
 505         return;
 506
 507 error:
 508         p9_conn_cancel(m, err);
 509         clear_bit(Wworksched, &m->wsched);
 510 }
 511
 512 static int p9_pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
 513 {
 514         struct p9_poll_wait *pwait =
 515                 container_of(wait, struct p9_poll_wait, wait);
 516         struct p9_conn *m = pwait->conn;
 517         unsigned long flags;
 518         DECLARE_WAITQUEUE(dummy_wait, p9_poll_task);
 519
 520         spin_lock_irqsave(&p9_poll_lock, flags);
 521         if (list_empty(&m->poll_pending_link))
 522                 list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
 523         spin_unlock_irqrestore(&p9_poll_lock, flags);
 524
 525         /* perform the default wake up operation */
 526         return default_wake_function(&dummy_wait, mode, sync, key);
 527 }
 528
 529 /**
 530  * p9_pollwait - add poll task to the wait queue
 531  * @filp: file pointer being polled
 532  * @wait_address: wait_q to block on
 533  * @p: poll state
 534  *
 535  * called by files poll operation to add v9fs-poll task to files wait queue
 536  */
 537
 538 static void
 539 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
 540 {
 541         struct p9_conn *m = container_of(p, struct p9_conn, pt);
 542         struct p9_poll_wait *pwait = NULL;
 543         int i;
 544
 545         for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
 546                 if (m->poll_wait[i].wait_addr == NULL) {
 547                         pwait = &m->poll_wait[i];
 548                         break;
 549                 }
 550         }
 551
 552         if (!pwait) {
 553                 P9_DPRINTK(P9_DEBUG_ERROR, "not enough wait_address slots\n");
 554                 return;
 555         }
 556
 557         pwait->conn = m;
 558         pwait->wait_addr = wait_address;
 559         init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
 560         add_wait_queue(wait_address, &pwait->wait);
 561 }
 562
 563 /**
 564  * p9_conn_create - allocate and initialize the per-session mux data
 565  * @client: client instance
 566  *
 567  * Note: Creates the polling task if this is the first session.
 568  */
 569
 570 static struct p9_conn *p9_conn_create(struct p9_client *client)
 571 {
 572         int n;
 573         struct p9_conn *m;
 574
 575         P9_DPRINTK(P9_DEBUG_TRANS, "client %p msize %d\n", client,
 576                                                                 client->msize);
 577         m = kzalloc(sizeof(struct p9_conn), GFP_KERNEL);
 578         if (!m)
 579                 return ERR_PTR(-ENOMEM);
 580
 581         INIT_LIST_HEAD(&m->mux_list);
 582         m->client = client;
 583
 584         INIT_LIST_HEAD(&m->req_list);
 585         INIT_LIST_HEAD(&m->unsent_req_list);
 586         INIT_WORK(&m->rq, p9_read_work);
 587         INIT_WORK(&m->wq, p9_write_work);
 588         INIT_LIST_HEAD(&m->poll_pending_link);
 589         init_poll_funcptr(&m->pt, p9_pollwait);
 590
 591         n = p9_fd_poll(client, &m->pt);
 592         if (n & POLLIN) {
 593                 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can read\n", m);
 594                 set_bit(Rpending, &m->wsched);
 595         }
 596
 597         if (n & POLLOUT) {
 598                 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can write\n", m);
 599                 set_bit(Wpending, &m->wsched);
 600         }
 601
 602         return m;
 603 }
 604
 605 /**
 606  * p9_poll_mux - polls a mux and schedules read or write works if necessary
 607  * @m: connection to poll
 608  *
 609  */
 610
 611 static void p9_poll_mux(struct p9_conn *m)
 612 {
 613         int n;
 614
 615         if (m->err < 0)
 616                 return;
 617
 618         n = p9_fd_poll(m->client, NULL);
 619         if (n < 0 || n & (POLLERR | POLLHUP | POLLNVAL)) {
 620                 P9_DPRINTK(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
 621                 if (n >= 0)
 622                         n = -ECONNRESET;
 623                 p9_conn_cancel(m, n);
 624         }
 625
 626         if (n & POLLIN) {
 627                 set_bit(Rpending, &m->wsched);
 628                 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can read\n", m);
 629                 if (!test_and_set_bit(Rworksched, &m->wsched)) {
 630                         P9_DPRINTK(P9_DEBUG_TRANS, "sched read work %p\n", m);
 631                         queue_work(p9_mux_wq, &m->rq);
 632                 }
 633         }
 634
 635         if (n & POLLOUT) {
 636                 set_bit(Wpending, &m->wsched);
 637                 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can write\n", m);
 638                 if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
 639                     !test_and_set_bit(Wworksched, &m->wsched)) {
 640                         P9_DPRINTK(P9_DEBUG_TRANS, "sched write work %p\n", m);
 641                         queue_work(p9_mux_wq, &m->wq);
 642                 }
 643         }
 644 }
 645
 646 /**
 647  * p9_fd_request - send 9P request
 648  * The function can sleep until the request is scheduled for sending.
 649  * The function can be interrupted. Return from the function is not
 650  * a guarantee that the request is sent successfully.
 651  *
 652  * @client: client instance
 653  * @req: request to be sent
 654  *
 655  */
 656
 657 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
 658 {
 659         int n;
 660         struct p9_trans_fd *ts = client->trans;
 661         struct p9_conn *m = ts->conn;
 662
 663         P9_DPRINTK(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n", m,
 664                                                 current, req->tc, req->tc->id);
 665         if (m->err < 0)
 666                 return m->err;
 667
 668         spin_lock(&client->lock);
 669         req->status = REQ_STATUS_UNSENT;
 670         list_add_tail(&req->req_list, &m->unsent_req_list);
 671         spin_unlock(&client->lock);
 672
 673         if (test_and_clear_bit(Wpending, &m->wsched))
 674                 n = POLLOUT;
 675         else
 676                 n = p9_fd_poll(m->client, NULL);
 677
 678         if (n & POLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
 679                 queue_work(p9_mux_wq, &m->wq);
 680
 681         return 0;
 682 }
 683
 684 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
 685 {
 686         int ret = 1;
 687
 688         P9_DPRINTK(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
 689
 690         spin_lock(&client->lock);
 691
 692         if (req->status == REQ_STATUS_UNSENT) {
 693                 list_del(&req->req_list);
 694                 req->status = REQ_STATUS_FLSHD;
 695                 ret = 0;
 696         } else if (req->status == REQ_STATUS_SENT)
 697                 req->status = REQ_STATUS_FLSH;
 698
 699         spin_unlock(&client->lock);
 700
 701         return ret;
 702 }
 703
 704 /**
 705  * parse_opts - parse mount options into p9_fd_opts structure
 706  * @params: options string passed from mount
 707  * @opts: fd transport-specific structure to parse options into
 708  *
 709  * Returns 0 upon success, -ERRNO upon failure
 710  */
 711
 712 static int parse_opts(char *params, struct p9_fd_opts *opts)
 713 {
 714         char *p;
 715         substring_t args[MAX_OPT_ARGS];
 716         int option;
 717         char *options, *tmp_options;
 718         int ret;
 719
 720         opts->port = P9_PORT;
 721         opts->rfd = ~0;
 722         opts->wfd = ~0;
 723
 724         if (!params)
 725                 return 0;
 726
 727         tmp_options = kstrdup(params, GFP_KERNEL);
 728         if (!tmp_options) {
 729                 P9_DPRINTK(P9_DEBUG_ERROR,
 730                                 "failed to allocate copy of option string\n");
 731                 return -ENOMEM;
 732         }
 733         options = tmp_options;
 734
 735         while ((p = strsep(&options, ",")) != NULL) {
 736                 int token;
 737                 int r;
 738                 if (!*p)
 739                         continue;
 740                 token = match_token(p, tokens, args);
 741                 if (token != Opt_err) {
 742                         r = match_int(&args[0], &option);
 743                         if (r < 0) {
 744                                 P9_DPRINTK(P9_DEBUG_ERROR,
 745                                 "integer field, but no integer?\n");
 746                                 ret = r;
 747                                 continue;
 748                         }
 749                 }
 750                 switch (token) {
 751                 case Opt_port:
 752                         opts->port = option;
 753                         break;
 754                 case Opt_rfdno:
 755                         opts->rfd = option;
 756                         break;
 757                 case Opt_wfdno:
 758                         opts->wfd = option;
 759                         break;
 760                 default:
 761                         continue;
 762                 }
 763         }
 764
 765         kfree(tmp_options);
 766         return 0;
 767 }
 768
 769 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
 770 {
 771         struct p9_trans_fd *ts = kmalloc(sizeof(struct p9_trans_fd),
 772                                            GFP_KERNEL);
 773         if (!ts)
 774                 return -ENOMEM;
 775
 776         ts->rd = fget(rfd);
 777         ts->wr = fget(wfd);
 778         if (!ts->rd || !ts->wr) {
 779                 if (ts->rd)
 780                         fput(ts->rd);
 781                 if (ts->wr)
 782                         fput(ts->wr);
 783                 kfree(ts);
 784                 return -EIO;
 785         }
 786
 787         client->trans = ts;
 788         client->status = Connected;
 789
 790         return 0;
 791 }
 792
 793 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
 794 {
 795         struct p9_trans_fd *p;
 796         int ret, fd;
 797
 798         p = kmalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
 799         if (!p)
 800                 return -ENOMEM;
 801
 802         csocket->sk->sk_allocation = GFP_NOIO;
 803         fd = sock_map_fd(csocket, 0);
 804         if (fd < 0) {
 805                 P9_EPRINTK(KERN_ERR, "p9_socket_open: failed to map fd\n");
 806                 sock_release(csocket);
 807                 kfree(p);
 808                 return fd;
 809         }
 810
 811         get_file(csocket->file);
 812         get_file(csocket->file);
 813         p->wr = p->rd = csocket->file;
 814         client->trans = p;
 815         client->status = Connected;
 816
 817         sys_close(fd);  /* still racy */
 818
 819         p->rd->f_flags |= O_NONBLOCK;
 820
 821         p->conn = p9_conn_create(client);
 822         if (IS_ERR(p->conn)) {
 823                 ret = PTR_ERR(p->conn);
 824                 p->conn = NULL;
 825                 kfree(p);
 826                 sockfd_put(csocket);
 827                 sockfd_put(csocket);
 828                 return ret;
 829         }
 830         return 0;
 831 }
 832
 833 /**
 834  * p9_mux_destroy - cancels all pending requests and frees mux resources
 835  * @m: mux to destroy
 836  *
 837  */
 838
 839 static void p9_conn_destroy(struct p9_conn *m)
 840 {
 841         P9_DPRINTK(P9_DEBUG_TRANS, "mux %p prev %p next %p\n", m,
 842                 m->mux_list.prev, m->mux_list.next);
 843
 844         p9_mux_poll_stop(m);
 845         cancel_work_sync(&m->rq);
 846         cancel_work_sync(&m->wq);
 847
 848         p9_conn_cancel(m, -ECONNRESET);
 849
 850         m->client = NULL;
 851         kfree(m);
 852 }
 853
 854 /**
 855  * p9_fd_close - shutdown file descriptor transport
 856  * @client: client instance
 857  *
 858  */
 859
 860 static void p9_fd_close(struct p9_client *client)
 861 {
 862         struct p9_trans_fd *ts;
 863
 864         if (!client)
 865                 return;
 866
 867         ts = client->trans;
 868         if (!ts)
 869                 return;
 870
 871         client->status = Disconnected;
 872
 873         p9_conn_destroy(ts->conn);
 874
 875         if (ts->rd)
 876                 fput(ts->rd);
 877         if (ts->wr)
 878                 fput(ts->wr);
 879
 880         kfree(ts);
 881 }
 882
 883 /*
 884  * stolen from NFS - maybe should be made a generic function?
 885  */
 886 static inline int valid_ipaddr4(const char *buf)
 887 {
 888         int rc, count, in[4];
 889
 890         rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
 891         if (rc != 4)
 892                 return -EINVAL;
 893         for (count = 0; count < 4; count++) {
 894                 if (in[count] > 255)
 895                         return -EINVAL;
 896         }
 897         return 0;
 898 }
 899
 900 static int
 901 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
 902 {
 903         int err;
 904         struct socket *csocket;
 905         struct sockaddr_in sin_server;
 906         struct p9_fd_opts opts;
 907
 908         err = parse_opts(args, &opts);
 909         if (err < 0)
 910                 return err;
 911
 912         if (valid_ipaddr4(addr) < 0)
 913                 return -EINVAL;
 914
 915         csocket = NULL;
 916
 917         sin_server.sin_family = AF_INET;
 918         sin_server.sin_addr.s_addr = in_aton(addr);
 919         sin_server.sin_port = htons(opts.port);
 920         err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &csocket);
 921
 922         if (err) {
 923                 P9_EPRINTK(KERN_ERR, "p9_trans_tcp: problem creating socket\n");
 924                 return err;
 925         }
 926
 927         err = csocket->ops->connect(csocket,
 928                                     (struct sockaddr *)&sin_server,
 929                                     sizeof(struct sockaddr_in), 0);
 930         if (err < 0) {
 931                 P9_EPRINTK(KERN_ERR,
 932                         "p9_trans_tcp: problem connecting socket to %s\n",
 933                         addr);
 934                 sock_release(csocket);
 935                 return err;
 936         }
 937
 938         return p9_socket_open(client, csocket);
 939 }
 940
 941 static int
 942 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
 943 {
 944         int err;
 945         struct socket *csocket;
 946         struct sockaddr_un sun_server;
 947
 948         csocket = NULL;
 949
 950         if (strlen(addr) > UNIX_PATH_MAX) {
 951                 P9_EPRINTK(KERN_ERR, "p9_trans_unix: address too long: %s\n",
 952                         addr);
 953                 return -ENAMETOOLONG;
 954         }
 955
 956         sun_server.sun_family = PF_UNIX;
 957         strcpy(sun_server.sun_path, addr);
 958         err = sock_create_kern(PF_UNIX, SOCK_STREAM, 0, &csocket);
 959         if (err < 0) {
 960                 P9_EPRINTK(KERN_ERR, "p9_trans_unix: problem creating socket\n");
 961                 return err;
 962         }
 963         err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
 964                         sizeof(struct sockaddr_un) - 1, 0);
 965         if (err < 0) {
 966                 P9_EPRINTK(KERN_ERR,
 967                         "p9_trans_unix: problem connecting socket: %s: %d\n",
 968                         addr, err);
 969                 sock_release(csocket);
 970                 return err;
 971         }
 972
 973         return p9_socket_open(client, csocket);
 974 }
 975
 976 static int
 977 p9_fd_create(struct p9_client *client, const char *addr, char *args)
 978 {
 979         int err;
 980         struct p9_fd_opts opts;
 981         struct p9_trans_fd *p;
 982
 983         parse_opts(args, &opts);
 984
 985         if (opts.rfd == ~0 || opts.wfd == ~0) {
 986                 printk(KERN_ERR "v9fs: Insufficient options for proto=fd\n");
 987                 return -ENOPROTOOPT;
 988         }
 989
 990         err = p9_fd_open(client, opts.rfd, opts.wfd);
 991         if (err < 0)
 992                 return err;
 993
 994         p = (struct p9_trans_fd *) client->trans;
 995         p->conn = p9_conn_create(client);
 996         if (IS_ERR(p->conn)) {
 997                 err = PTR_ERR(p->conn);
 998                 p->conn = NULL;
 999                 fput(p->rd);
1000                 fput(p->wr);
1001                 return err;
1002         }
1003
1004         return 0;
1005 }
1006
1007 static struct p9_trans_module p9_tcp_trans = {
1008         .name = "tcp",
1009         .maxsize = MAX_SOCK_BUF,
1010         .def = 1,
1011         .create = p9_fd_create_tcp,
1012         .close = p9_fd_close,
1013         .request = p9_fd_request,
1014         .cancel = p9_fd_cancel,
1015         .owner = THIS_MODULE,
1016 };
1017
1018 static struct p9_trans_module p9_unix_trans = {
1019         .name = "unix",
1020         .maxsize = MAX_SOCK_BUF,
1021         .def = 0,
1022         .create = p9_fd_create_unix,
1023         .close = p9_fd_close,
1024         .request = p9_fd_request,
1025         .cancel = p9_fd_cancel,
1026         .owner = THIS_MODULE,
1027 };
1028
1029 static struct p9_trans_module p9_fd_trans = {
1030         .name = "fd",
1031         .maxsize = MAX_SOCK_BUF,
1032         .def = 0,
1033         .create = p9_fd_create,
1034         .close = p9_fd_close,
1035         .request = p9_fd_request,
1036         .cancel = p9_fd_cancel,
1037         .owner = THIS_MODULE,
1038 };
1039
1040 /**
1041  * p9_poll_proc - poll worker thread
1042  * @a: thread state and arguments
1043  *
1044  * polls all v9fs transports for new events and queues the appropriate
1045  * work to the work queue
1046  *
1047  */
1048
1049 static int p9_poll_proc(void *a)
1050 {
1051         unsigned long flags;
1052
1053         P9_DPRINTK(P9_DEBUG_TRANS, "start %p\n", current);
1054  repeat:
1055         spin_lock_irqsave(&p9_poll_lock, flags);
1056         while (!list_empty(&p9_poll_pending_list)) {
1057                 struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1058                                                         struct p9_conn,
1059                                                         poll_pending_link);
1060                 list_del_init(&conn->poll_pending_link);
1061                 spin_unlock_irqrestore(&p9_poll_lock, flags);
1062
1063                 p9_poll_mux(conn);
1064
1065                 spin_lock_irqsave(&p9_poll_lock, flags);
1066         }
1067         spin_unlock_irqrestore(&p9_poll_lock, flags);
1068
1069         set_current_state(TASK_INTERRUPTIBLE);
1070         if (list_empty(&p9_poll_pending_list)) {
1071                 P9_DPRINTK(P9_DEBUG_TRANS, "sleeping...\n");
1072                 schedule();
1073         }
1074         __set_current_state(TASK_RUNNING);
1075
1076         if (!kthread_should_stop())
1077                 goto repeat;
1078
1079         P9_DPRINTK(P9_DEBUG_TRANS, "finish\n");
1080         return 0;
1081 }
1082
1083 int p9_trans_fd_init(void)
1084 {
1085         p9_mux_wq = create_workqueue("v9fs");
1086         if (!p9_mux_wq) {
1087                 printk(KERN_WARNING "v9fs: mux: creating workqueue failed\n");
1088                 return -ENOMEM;
1089         }
1090
1091         p9_poll_task = kthread_run(p9_poll_proc, NULL, "v9fs-poll");
1092         if (IS_ERR(p9_poll_task)) {
1093                 destroy_workqueue(p9_mux_wq);
1094                 printk(KERN_WARNING "v9fs: mux: creating poll task failed\n");
1095                 return PTR_ERR(p9_poll_task);
1096         }
1097
1098         v9fs_register_trans(&p9_tcp_trans);
1099         v9fs_register_trans(&p9_unix_trans);
1100         v9fs_register_trans(&p9_fd_trans);
1101
1102         return 0;
1103 }
1104
1105 void p9_trans_fd_exit(void)
1106 {
1107         kthread_stop(p9_poll_task);
1108         v9fs_unregister_trans(&p9_tcp_trans);
1109         v9fs_unregister_trans(&p9_unix_trans);
1110         v9fs_unregister_trans(&p9_fd_trans);
1111
1112         destroy_workqueue(p9_mux_wq);
1113 }