1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Fd transport layer. Includes deprecated socket layer. 4 * 5 * Copyright (C) 2006 by Russ Cox <rsc@swtch.com> 6 * Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net> 7 * Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com> 8 * Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com> 9 */ 10 11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 12 13 #include <linux/in.h> 14 #include <linux/module.h> 15 #include <linux/net.h> 16 #include <linux/ipv6.h> 17 #include <linux/kthread.h> 18 #include <linux/errno.h> 19 #include <linux/kernel.h> 20 #include <linux/un.h> 21 #include <linux/uaccess.h> 22 #include <linux/inet.h> 23 #include <linux/idr.h> 24 #include <linux/file.h> 25 #include <linux/parser.h> 26 #include <linux/slab.h> 27 #include <linux/seq_file.h> 28 #include <net/9p/9p.h> 29 #include <net/9p/client.h> 30 #include <net/9p/transport.h> 31 32 #include <linux/syscalls.h> /* killme */ 33 34 #define P9_PORT 564 35 #define MAX_SOCK_BUF (1024*1024) 36 #define MAXPOLLWADDR 2 37 38 static struct p9_trans_module p9_tcp_trans; 39 static struct p9_trans_module p9_fd_trans; 40 41 /** 42 * struct p9_fd_opts - per-transport options 43 * @rfd: file descriptor for reading (trans=fd) 44 * @wfd: file descriptor for writing (trans=fd) 45 * @port: port to connect to (trans=tcp) 46 * @privport: port is privileged 47 */ 48 49 struct p9_fd_opts { 50 int rfd; 51 int wfd; 52 u16 port; 53 bool privport; 54 }; 55 56 /* 57 * Option Parsing (code inspired by NFS code) 58 * - a little lazy - parse all fd-transport options 59 */ 60 61 enum { 62 /* Options that take integer arguments */ 63 Opt_port, Opt_rfdno, Opt_wfdno, Opt_err, 64 /* Options that take no arguments */ 65 Opt_privport, 66 }; 67 68 static const match_table_t tokens = { 69 {Opt_port, "port=%u"}, 70 {Opt_rfdno, "rfdno=%u"}, 71 {Opt_wfdno, "wfdno=%u"}, 72 {Opt_privport, "privport"}, 73 {Opt_err, NULL}, 74 }; 75 76 enum { 77 Rworksched = 1, /* read work scheduled or running */ 78 Rpending = 2, /* can read */ 79 Wworksched = 4, /* write work scheduled or running */ 80 Wpending = 8, /* can write */ 81 }; 82 83 struct p9_poll_wait { 84 struct p9_conn *conn; 85 wait_queue_entry_t wait; 86 wait_queue_head_t *wait_addr; 87 }; 88 89 /** 90 * struct p9_conn - fd mux connection state information 91 * @mux_list: list link for mux to manage multiple connections (?) 92 * @client: reference to client instance for this connection 93 * @err: error state 94 * @req_lock: lock protecting req_list and requests statuses 95 * @req_list: accounting for requests which have been sent 96 * @unsent_req_list: accounting for requests that haven't been sent 97 * @rreq: read request 98 * @wreq: write request 99 * @req: current request being processed (if any) 100 * @tmp_buf: temporary buffer to read in header 101 * @rc: temporary fcall for reading current frame 102 * @wpos: write position for current frame 103 * @wsize: amount of data to write for current frame 104 * @wbuf: current write buffer 105 * @poll_pending_link: pending links to be polled per conn 106 * @poll_wait: array of wait_q's for various worker threads 107 * @pt: poll state 108 * @rq: current read work 109 * @wq: current write work 110 * @wsched: ???? 111 * 112 */ 113 114 struct p9_conn { 115 struct list_head mux_list; 116 struct p9_client *client; 117 int err; 118 spinlock_t req_lock; 119 struct list_head req_list; 120 struct list_head unsent_req_list; 121 struct p9_req_t *rreq; 122 struct p9_req_t *wreq; 123 char tmp_buf[7]; 124 struct p9_fcall rc; 125 int wpos; 126 int wsize; 127 char *wbuf; 128 struct list_head poll_pending_link; 129 struct p9_poll_wait poll_wait[MAXPOLLWADDR]; 130 poll_table pt; 131 struct work_struct rq; 132 struct work_struct wq; 133 unsigned long wsched; 134 }; 135 136 /** 137 * struct p9_trans_fd - transport state 138 * @rd: reference to file to read from 139 * @wr: reference of file to write to 140 * @conn: connection state reference 141 * 142 */ 143 144 struct p9_trans_fd { 145 struct file *rd; 146 struct file *wr; 147 struct p9_conn conn; 148 }; 149 150 static void p9_poll_workfn(struct work_struct *work); 151 152 static DEFINE_SPINLOCK(p9_poll_lock); 153 static LIST_HEAD(p9_poll_pending_list); 154 static DECLARE_WORK(p9_poll_work, p9_poll_workfn); 155 156 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT; 157 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT; 158 159 static void p9_mux_poll_stop(struct p9_conn *m) 160 { 161 unsigned long flags; 162 int i; 163 164 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) { 165 struct p9_poll_wait *pwait = &m->poll_wait[i]; 166 167 if (pwait->wait_addr) { 168 remove_wait_queue(pwait->wait_addr, &pwait->wait); 169 pwait->wait_addr = NULL; 170 } 171 } 172 173 spin_lock_irqsave(&p9_poll_lock, flags); 174 list_del_init(&m->poll_pending_link); 175 spin_unlock_irqrestore(&p9_poll_lock, flags); 176 177 flush_work(&p9_poll_work); 178 } 179 180 /** 181 * p9_conn_cancel - cancel all pending requests with error 182 * @m: mux data 183 * @err: error code 184 * 185 */ 186 187 static void p9_conn_cancel(struct p9_conn *m, int err) 188 { 189 struct p9_req_t *req, *rtmp; 190 LIST_HEAD(cancel_list); 191 192 p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err); 193 194 spin_lock(&m->req_lock); 195 196 if (m->err) { 197 spin_unlock(&m->req_lock); 198 return; 199 } 200 201 m->err = err; 202 203 list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) { 204 list_move(&req->req_list, &cancel_list); 205 } 206 list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) { 207 list_move(&req->req_list, &cancel_list); 208 } 209 210 spin_unlock(&m->req_lock); 211 212 list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) { 213 p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req); 214 list_del(&req->req_list); 215 if (!req->t_err) 216 req->t_err = err; 217 p9_client_cb(m->client, req, REQ_STATUS_ERROR); 218 } 219 } 220 221 static __poll_t 222 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err) 223 { 224 __poll_t ret; 225 struct p9_trans_fd *ts = NULL; 226 227 if (client && client->status == Connected) 228 ts = client->trans; 229 230 if (!ts) { 231 if (err) 232 *err = -EREMOTEIO; 233 return EPOLLERR; 234 } 235 236 ret = vfs_poll(ts->rd, pt); 237 if (ts->rd != ts->wr) 238 ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN); 239 return ret; 240 } 241 242 /** 243 * p9_fd_read- read from a fd 244 * @client: client instance 245 * @v: buffer to receive data into 246 * @len: size of receive buffer 247 * 248 */ 249 250 static int p9_fd_read(struct p9_client *client, void *v, int len) 251 { 252 int ret; 253 struct p9_trans_fd *ts = NULL; 254 loff_t pos; 255 256 if (client && client->status != Disconnected) 257 ts = client->trans; 258 259 if (!ts) 260 return -EREMOTEIO; 261 262 if (!(ts->rd->f_flags & O_NONBLOCK)) 263 p9_debug(P9_DEBUG_ERROR, "blocking read ...\n"); 264 265 pos = ts->rd->f_pos; 266 ret = kernel_read(ts->rd, v, len, &pos); 267 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN) 268 client->status = Disconnected; 269 return ret; 270 } 271 272 /** 273 * p9_read_work - called when there is some data to be read from a transport 274 * @work: container of work to be done 275 * 276 */ 277 278 static void p9_read_work(struct work_struct *work) 279 { 280 __poll_t n; 281 int err; 282 struct p9_conn *m; 283 284 m = container_of(work, struct p9_conn, rq); 285 286 if (m->err < 0) 287 return; 288 289 p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset); 290 291 if (!m->rc.sdata) { 292 m->rc.sdata = m->tmp_buf; 293 m->rc.offset = 0; 294 m->rc.capacity = 7; /* start by reading header */ 295 } 296 297 clear_bit(Rpending, &m->wsched); 298 p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n", 299 m, m->rc.offset, m->rc.capacity, 300 m->rc.capacity - m->rc.offset); 301 err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset, 302 m->rc.capacity - m->rc.offset); 303 p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err); 304 if (err == -EAGAIN) 305 goto end_clear; 306 307 if (err <= 0) 308 goto error; 309 310 m->rc.offset += err; 311 312 /* header read in */ 313 if ((!m->rreq) && (m->rc.offset == m->rc.capacity)) { 314 p9_debug(P9_DEBUG_TRANS, "got new header\n"); 315 316 /* Header size */ 317 m->rc.size = 7; 318 err = p9_parse_header(&m->rc, &m->rc.size, NULL, NULL, 0); 319 if (err) { 320 p9_debug(P9_DEBUG_ERROR, 321 "error parsing header: %d\n", err); 322 goto error; 323 } 324 325 if (m->rc.size >= m->client->msize) { 326 p9_debug(P9_DEBUG_ERROR, 327 "requested packet size too big: %d\n", 328 m->rc.size); 329 err = -EIO; 330 goto error; 331 } 332 333 p9_debug(P9_DEBUG_TRANS, 334 "mux %p pkt: size: %d bytes tag: %d\n", 335 m, m->rc.size, m->rc.tag); 336 337 m->rreq = p9_tag_lookup(m->client, m->rc.tag); 338 if (!m->rreq || (m->rreq->status != REQ_STATUS_SENT)) { 339 p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n", 340 m->rc.tag); 341 err = -EIO; 342 goto error; 343 } 344 345 if (!m->rreq->rc.sdata) { 346 p9_debug(P9_DEBUG_ERROR, 347 "No recv fcall for tag %d (req %p), disconnecting!\n", 348 m->rc.tag, m->rreq); 349 p9_req_put(m->client, m->rreq); 350 m->rreq = NULL; 351 err = -EIO; 352 goto error; 353 } 354 m->rc.sdata = m->rreq->rc.sdata; 355 memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity); 356 m->rc.capacity = m->rc.size; 357 } 358 359 /* packet is read in 360 * not an else because some packets (like clunk) have no payload 361 */ 362 if ((m->rreq) && (m->rc.offset == m->rc.capacity)) { 363 p9_debug(P9_DEBUG_TRANS, "got new packet\n"); 364 m->rreq->rc.size = m->rc.offset; 365 spin_lock(&m->req_lock); 366 if (m->rreq->status == REQ_STATUS_SENT) { 367 list_del(&m->rreq->req_list); 368 p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD); 369 } else if (m->rreq->status == REQ_STATUS_FLSHD) { 370 /* Ignore replies associated with a cancelled request. */ 371 p9_debug(P9_DEBUG_TRANS, 372 "Ignore replies associated with a cancelled request\n"); 373 } else { 374 spin_unlock(&m->req_lock); 375 p9_debug(P9_DEBUG_ERROR, 376 "Request tag %d errored out while we were reading the reply\n", 377 m->rc.tag); 378 err = -EIO; 379 goto error; 380 } 381 spin_unlock(&m->req_lock); 382 m->rc.sdata = NULL; 383 m->rc.offset = 0; 384 m->rc.capacity = 0; 385 p9_req_put(m->client, m->rreq); 386 m->rreq = NULL; 387 } 388 389 end_clear: 390 clear_bit(Rworksched, &m->wsched); 391 392 if (!list_empty(&m->req_list)) { 393 if (test_and_clear_bit(Rpending, &m->wsched)) 394 n = EPOLLIN; 395 else 396 n = p9_fd_poll(m->client, NULL, NULL); 397 398 if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) { 399 p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m); 400 schedule_work(&m->rq); 401 } 402 } 403 404 return; 405 error: 406 p9_conn_cancel(m, err); 407 clear_bit(Rworksched, &m->wsched); 408 } 409 410 /** 411 * p9_fd_write - write to a socket 412 * @client: client instance 413 * @v: buffer to send data from 414 * @len: size of send buffer 415 * 416 */ 417 418 static int p9_fd_write(struct p9_client *client, void *v, int len) 419 { 420 ssize_t ret; 421 struct p9_trans_fd *ts = NULL; 422 423 if (client && client->status != Disconnected) 424 ts = client->trans; 425 426 if (!ts) 427 return -EREMOTEIO; 428 429 if (!(ts->wr->f_flags & O_NONBLOCK)) 430 p9_debug(P9_DEBUG_ERROR, "blocking write ...\n"); 431 432 ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos); 433 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN) 434 client->status = Disconnected; 435 return ret; 436 } 437 438 /** 439 * p9_write_work - called when a transport can send some data 440 * @work: container for work to be done 441 * 442 */ 443 444 static void p9_write_work(struct work_struct *work) 445 { 446 __poll_t n; 447 int err; 448 struct p9_conn *m; 449 struct p9_req_t *req; 450 451 m = container_of(work, struct p9_conn, wq); 452 453 if (m->err < 0) { 454 clear_bit(Wworksched, &m->wsched); 455 return; 456 } 457 458 if (!m->wsize) { 459 spin_lock(&m->req_lock); 460 if (list_empty(&m->unsent_req_list)) { 461 clear_bit(Wworksched, &m->wsched); 462 spin_unlock(&m->req_lock); 463 return; 464 } 465 466 req = list_entry(m->unsent_req_list.next, struct p9_req_t, 467 req_list); 468 req->status = REQ_STATUS_SENT; 469 p9_debug(P9_DEBUG_TRANS, "move req %p\n", req); 470 list_move_tail(&req->req_list, &m->req_list); 471 472 m->wbuf = req->tc.sdata; 473 m->wsize = req->tc.size; 474 m->wpos = 0; 475 p9_req_get(req); 476 m->wreq = req; 477 spin_unlock(&m->req_lock); 478 } 479 480 p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n", 481 m, m->wpos, m->wsize); 482 clear_bit(Wpending, &m->wsched); 483 err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos); 484 p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err); 485 if (err == -EAGAIN) 486 goto end_clear; 487 488 489 if (err < 0) 490 goto error; 491 else if (err == 0) { 492 err = -EREMOTEIO; 493 goto error; 494 } 495 496 m->wpos += err; 497 if (m->wpos == m->wsize) { 498 m->wpos = m->wsize = 0; 499 p9_req_put(m->client, m->wreq); 500 m->wreq = NULL; 501 } 502 503 end_clear: 504 clear_bit(Wworksched, &m->wsched); 505 506 if (m->wsize || !list_empty(&m->unsent_req_list)) { 507 if (test_and_clear_bit(Wpending, &m->wsched)) 508 n = EPOLLOUT; 509 else 510 n = p9_fd_poll(m->client, NULL, NULL); 511 512 if ((n & EPOLLOUT) && 513 !test_and_set_bit(Wworksched, &m->wsched)) { 514 p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m); 515 schedule_work(&m->wq); 516 } 517 } 518 519 return; 520 521 error: 522 p9_conn_cancel(m, err); 523 clear_bit(Wworksched, &m->wsched); 524 } 525 526 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key) 527 { 528 struct p9_poll_wait *pwait = 529 container_of(wait, struct p9_poll_wait, wait); 530 struct p9_conn *m = pwait->conn; 531 unsigned long flags; 532 533 spin_lock_irqsave(&p9_poll_lock, flags); 534 if (list_empty(&m->poll_pending_link)) 535 list_add_tail(&m->poll_pending_link, &p9_poll_pending_list); 536 spin_unlock_irqrestore(&p9_poll_lock, flags); 537 538 schedule_work(&p9_poll_work); 539 return 1; 540 } 541 542 /** 543 * p9_pollwait - add poll task to the wait queue 544 * @filp: file pointer being polled 545 * @wait_address: wait_q to block on 546 * @p: poll state 547 * 548 * called by files poll operation to add v9fs-poll task to files wait queue 549 */ 550 551 static void 552 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p) 553 { 554 struct p9_conn *m = container_of(p, struct p9_conn, pt); 555 struct p9_poll_wait *pwait = NULL; 556 int i; 557 558 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) { 559 if (m->poll_wait[i].wait_addr == NULL) { 560 pwait = &m->poll_wait[i]; 561 break; 562 } 563 } 564 565 if (!pwait) { 566 p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n"); 567 return; 568 } 569 570 pwait->conn = m; 571 pwait->wait_addr = wait_address; 572 init_waitqueue_func_entry(&pwait->wait, p9_pollwake); 573 add_wait_queue(wait_address, &pwait->wait); 574 } 575 576 /** 577 * p9_conn_create - initialize the per-session mux data 578 * @client: client instance 579 * 580 * Note: Creates the polling task if this is the first session. 581 */ 582 583 static void p9_conn_create(struct p9_client *client) 584 { 585 __poll_t n; 586 struct p9_trans_fd *ts = client->trans; 587 struct p9_conn *m = &ts->conn; 588 589 p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize); 590 591 INIT_LIST_HEAD(&m->mux_list); 592 m->client = client; 593 594 spin_lock_init(&m->req_lock); 595 INIT_LIST_HEAD(&m->req_list); 596 INIT_LIST_HEAD(&m->unsent_req_list); 597 INIT_WORK(&m->rq, p9_read_work); 598 INIT_WORK(&m->wq, p9_write_work); 599 INIT_LIST_HEAD(&m->poll_pending_link); 600 init_poll_funcptr(&m->pt, p9_pollwait); 601 602 n = p9_fd_poll(client, &m->pt, NULL); 603 if (n & EPOLLIN) { 604 p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m); 605 set_bit(Rpending, &m->wsched); 606 } 607 608 if (n & EPOLLOUT) { 609 p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m); 610 set_bit(Wpending, &m->wsched); 611 } 612 } 613 614 /** 615 * p9_poll_mux - polls a mux and schedules read or write works if necessary 616 * @m: connection to poll 617 * 618 */ 619 620 static void p9_poll_mux(struct p9_conn *m) 621 { 622 __poll_t n; 623 int err = -ECONNRESET; 624 625 if (m->err < 0) 626 return; 627 628 n = p9_fd_poll(m->client, NULL, &err); 629 if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) { 630 p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n); 631 p9_conn_cancel(m, err); 632 } 633 634 if (n & EPOLLIN) { 635 set_bit(Rpending, &m->wsched); 636 p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m); 637 if (!test_and_set_bit(Rworksched, &m->wsched)) { 638 p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m); 639 schedule_work(&m->rq); 640 } 641 } 642 643 if (n & EPOLLOUT) { 644 set_bit(Wpending, &m->wsched); 645 p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m); 646 if ((m->wsize || !list_empty(&m->unsent_req_list)) && 647 !test_and_set_bit(Wworksched, &m->wsched)) { 648 p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m); 649 schedule_work(&m->wq); 650 } 651 } 652 } 653 654 /** 655 * p9_fd_request - send 9P request 656 * The function can sleep until the request is scheduled for sending. 657 * The function can be interrupted. Return from the function is not 658 * a guarantee that the request is sent successfully. 659 * 660 * @client: client instance 661 * @req: request to be sent 662 * 663 */ 664 665 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req) 666 { 667 __poll_t n; 668 struct p9_trans_fd *ts = client->trans; 669 struct p9_conn *m = &ts->conn; 670 671 p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n", 672 m, current, &req->tc, req->tc.id); 673 if (m->err < 0) 674 return m->err; 675 676 spin_lock(&m->req_lock); 677 req->status = REQ_STATUS_UNSENT; 678 list_add_tail(&req->req_list, &m->unsent_req_list); 679 spin_unlock(&m->req_lock); 680 681 if (test_and_clear_bit(Wpending, &m->wsched)) 682 n = EPOLLOUT; 683 else 684 n = p9_fd_poll(m->client, NULL, NULL); 685 686 if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched)) 687 schedule_work(&m->wq); 688 689 return 0; 690 } 691 692 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req) 693 { 694 struct p9_trans_fd *ts = client->trans; 695 struct p9_conn *m = &ts->conn; 696 int ret = 1; 697 698 p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req); 699 700 spin_lock(&m->req_lock); 701 702 if (req->status == REQ_STATUS_UNSENT) { 703 list_del(&req->req_list); 704 req->status = REQ_STATUS_FLSHD; 705 p9_req_put(client, req); 706 ret = 0; 707 } 708 spin_unlock(&m->req_lock); 709 710 return ret; 711 } 712 713 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req) 714 { 715 struct p9_trans_fd *ts = client->trans; 716 struct p9_conn *m = &ts->conn; 717 718 p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req); 719 720 spin_lock(&m->req_lock); 721 /* Ignore cancelled request if message has been received 722 * before lock. 723 */ 724 if (req->status == REQ_STATUS_RCVD) { 725 spin_unlock(&m->req_lock); 726 return 0; 727 } 728 729 /* we haven't received a response for oldreq, 730 * remove it from the list. 731 */ 732 list_del(&req->req_list); 733 req->status = REQ_STATUS_FLSHD; 734 spin_unlock(&m->req_lock); 735 736 p9_req_put(client, req); 737 738 return 0; 739 } 740 741 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt) 742 { 743 if (clnt->trans_mod == &p9_tcp_trans) { 744 if (clnt->trans_opts.tcp.port != P9_PORT) 745 seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port); 746 } else if (clnt->trans_mod == &p9_fd_trans) { 747 if (clnt->trans_opts.fd.rfd != ~0) 748 seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd); 749 if (clnt->trans_opts.fd.wfd != ~0) 750 seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd); 751 } 752 return 0; 753 } 754 755 /** 756 * parse_opts - parse mount options into p9_fd_opts structure 757 * @params: options string passed from mount 758 * @opts: fd transport-specific structure to parse options into 759 * 760 * Returns 0 upon success, -ERRNO upon failure 761 */ 762 763 static int parse_opts(char *params, struct p9_fd_opts *opts) 764 { 765 char *p; 766 substring_t args[MAX_OPT_ARGS]; 767 int option; 768 char *options, *tmp_options; 769 770 opts->port = P9_PORT; 771 opts->rfd = ~0; 772 opts->wfd = ~0; 773 opts->privport = false; 774 775 if (!params) 776 return 0; 777 778 tmp_options = kstrdup(params, GFP_KERNEL); 779 if (!tmp_options) { 780 p9_debug(P9_DEBUG_ERROR, 781 "failed to allocate copy of option string\n"); 782 return -ENOMEM; 783 } 784 options = tmp_options; 785 786 while ((p = strsep(&options, ",")) != NULL) { 787 int token; 788 int r; 789 if (!*p) 790 continue; 791 token = match_token(p, tokens, args); 792 if ((token != Opt_err) && (token != Opt_privport)) { 793 r = match_int(&args[0], &option); 794 if (r < 0) { 795 p9_debug(P9_DEBUG_ERROR, 796 "integer field, but no integer?\n"); 797 continue; 798 } 799 } 800 switch (token) { 801 case Opt_port: 802 opts->port = option; 803 break; 804 case Opt_rfdno: 805 opts->rfd = option; 806 break; 807 case Opt_wfdno: 808 opts->wfd = option; 809 break; 810 case Opt_privport: 811 opts->privport = true; 812 break; 813 default: 814 continue; 815 } 816 } 817 818 kfree(tmp_options); 819 return 0; 820 } 821 822 static int p9_fd_open(struct p9_client *client, int rfd, int wfd) 823 { 824 struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd), 825 GFP_KERNEL); 826 if (!ts) 827 return -ENOMEM; 828 829 ts->rd = fget(rfd); 830 if (!ts->rd) 831 goto out_free_ts; 832 if (!(ts->rd->f_mode & FMODE_READ)) 833 goto out_put_rd; 834 /* prevent workers from hanging on IO when fd is a pipe */ 835 ts->rd->f_flags |= O_NONBLOCK; 836 ts->wr = fget(wfd); 837 if (!ts->wr) 838 goto out_put_rd; 839 if (!(ts->wr->f_mode & FMODE_WRITE)) 840 goto out_put_wr; 841 ts->wr->f_flags |= O_NONBLOCK; 842 843 client->trans = ts; 844 client->status = Connected; 845 846 return 0; 847 848 out_put_wr: 849 fput(ts->wr); 850 out_put_rd: 851 fput(ts->rd); 852 out_free_ts: 853 kfree(ts); 854 return -EIO; 855 } 856 857 static int p9_socket_open(struct p9_client *client, struct socket *csocket) 858 { 859 struct p9_trans_fd *p; 860 struct file *file; 861 862 p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL); 863 if (!p) { 864 sock_release(csocket); 865 return -ENOMEM; 866 } 867 868 csocket->sk->sk_allocation = GFP_NOIO; 869 file = sock_alloc_file(csocket, 0, NULL); 870 if (IS_ERR(file)) { 871 pr_err("%s (%d): failed to map fd\n", 872 __func__, task_pid_nr(current)); 873 kfree(p); 874 return PTR_ERR(file); 875 } 876 877 get_file(file); 878 p->wr = p->rd = file; 879 client->trans = p; 880 client->status = Connected; 881 882 p->rd->f_flags |= O_NONBLOCK; 883 884 p9_conn_create(client); 885 return 0; 886 } 887 888 /** 889 * p9_conn_destroy - cancels all pending requests of mux 890 * @m: mux to destroy 891 * 892 */ 893 894 static void p9_conn_destroy(struct p9_conn *m) 895 { 896 p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n", 897 m, m->mux_list.prev, m->mux_list.next); 898 899 p9_mux_poll_stop(m); 900 cancel_work_sync(&m->rq); 901 if (m->rreq) { 902 p9_req_put(m->client, m->rreq); 903 m->rreq = NULL; 904 } 905 cancel_work_sync(&m->wq); 906 if (m->wreq) { 907 p9_req_put(m->client, m->wreq); 908 m->wreq = NULL; 909 } 910 911 p9_conn_cancel(m, -ECONNRESET); 912 913 m->client = NULL; 914 } 915 916 /** 917 * p9_fd_close - shutdown file descriptor transport 918 * @client: client instance 919 * 920 */ 921 922 static void p9_fd_close(struct p9_client *client) 923 { 924 struct p9_trans_fd *ts; 925 926 if (!client) 927 return; 928 929 ts = client->trans; 930 if (!ts) 931 return; 932 933 client->status = Disconnected; 934 935 p9_conn_destroy(&ts->conn); 936 937 if (ts->rd) 938 fput(ts->rd); 939 if (ts->wr) 940 fput(ts->wr); 941 942 kfree(ts); 943 } 944 945 /* 946 * stolen from NFS - maybe should be made a generic function? 947 */ 948 static inline int valid_ipaddr4(const char *buf) 949 { 950 int rc, count, in[4]; 951 952 rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]); 953 if (rc != 4) 954 return -EINVAL; 955 for (count = 0; count < 4; count++) { 956 if (in[count] > 255) 957 return -EINVAL; 958 } 959 return 0; 960 } 961 962 static int p9_bind_privport(struct socket *sock) 963 { 964 struct sockaddr_in cl; 965 int port, err = -EINVAL; 966 967 memset(&cl, 0, sizeof(cl)); 968 cl.sin_family = AF_INET; 969 cl.sin_addr.s_addr = htonl(INADDR_ANY); 970 for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) { 971 cl.sin_port = htons((ushort)port); 972 err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl)); 973 if (err != -EADDRINUSE) 974 break; 975 } 976 return err; 977 } 978 979 980 static int 981 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args) 982 { 983 int err; 984 struct socket *csocket; 985 struct sockaddr_in sin_server; 986 struct p9_fd_opts opts; 987 988 err = parse_opts(args, &opts); 989 if (err < 0) 990 return err; 991 992 if (addr == NULL || valid_ipaddr4(addr) < 0) 993 return -EINVAL; 994 995 csocket = NULL; 996 997 client->trans_opts.tcp.port = opts.port; 998 client->trans_opts.tcp.privport = opts.privport; 999 sin_server.sin_family = AF_INET; 1000 sin_server.sin_addr.s_addr = in_aton(addr); 1001 sin_server.sin_port = htons(opts.port); 1002 err = __sock_create(current->nsproxy->net_ns, PF_INET, 1003 SOCK_STREAM, IPPROTO_TCP, &csocket, 1); 1004 if (err) { 1005 pr_err("%s (%d): problem creating socket\n", 1006 __func__, task_pid_nr(current)); 1007 return err; 1008 } 1009 1010 if (opts.privport) { 1011 err = p9_bind_privport(csocket); 1012 if (err < 0) { 1013 pr_err("%s (%d): problem binding to privport\n", 1014 __func__, task_pid_nr(current)); 1015 sock_release(csocket); 1016 return err; 1017 } 1018 } 1019 1020 err = csocket->ops->connect(csocket, 1021 (struct sockaddr *)&sin_server, 1022 sizeof(struct sockaddr_in), 0); 1023 if (err < 0) { 1024 pr_err("%s (%d): problem connecting socket to %s\n", 1025 __func__, task_pid_nr(current), addr); 1026 sock_release(csocket); 1027 return err; 1028 } 1029 1030 return p9_socket_open(client, csocket); 1031 } 1032 1033 static int 1034 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args) 1035 { 1036 int err; 1037 struct socket *csocket; 1038 struct sockaddr_un sun_server; 1039 1040 csocket = NULL; 1041 1042 if (!addr || !strlen(addr)) 1043 return -EINVAL; 1044 1045 if (strlen(addr) >= UNIX_PATH_MAX) { 1046 pr_err("%s (%d): address too long: %s\n", 1047 __func__, task_pid_nr(current), addr); 1048 return -ENAMETOOLONG; 1049 } 1050 1051 sun_server.sun_family = PF_UNIX; 1052 strcpy(sun_server.sun_path, addr); 1053 err = __sock_create(current->nsproxy->net_ns, PF_UNIX, 1054 SOCK_STREAM, 0, &csocket, 1); 1055 if (err < 0) { 1056 pr_err("%s (%d): problem creating socket\n", 1057 __func__, task_pid_nr(current)); 1058 1059 return err; 1060 } 1061 err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server, 1062 sizeof(struct sockaddr_un) - 1, 0); 1063 if (err < 0) { 1064 pr_err("%s (%d): problem connecting socket: %s: %d\n", 1065 __func__, task_pid_nr(current), addr, err); 1066 sock_release(csocket); 1067 return err; 1068 } 1069 1070 return p9_socket_open(client, csocket); 1071 } 1072 1073 static int 1074 p9_fd_create(struct p9_client *client, const char *addr, char *args) 1075 { 1076 int err; 1077 struct p9_fd_opts opts; 1078 1079 err = parse_opts(args, &opts); 1080 if (err < 0) 1081 return err; 1082 client->trans_opts.fd.rfd = opts.rfd; 1083 client->trans_opts.fd.wfd = opts.wfd; 1084 1085 if (opts.rfd == ~0 || opts.wfd == ~0) { 1086 pr_err("Insufficient options for proto=fd\n"); 1087 return -ENOPROTOOPT; 1088 } 1089 1090 err = p9_fd_open(client, opts.rfd, opts.wfd); 1091 if (err < 0) 1092 return err; 1093 1094 p9_conn_create(client); 1095 1096 return 0; 1097 } 1098 1099 static struct p9_trans_module p9_tcp_trans = { 1100 .name = "tcp", 1101 .maxsize = MAX_SOCK_BUF, 1102 .pooled_rbuffers = false, 1103 .def = 0, 1104 .create = p9_fd_create_tcp, 1105 .close = p9_fd_close, 1106 .request = p9_fd_request, 1107 .cancel = p9_fd_cancel, 1108 .cancelled = p9_fd_cancelled, 1109 .show_options = p9_fd_show_options, 1110 .owner = THIS_MODULE, 1111 }; 1112 MODULE_ALIAS_9P("tcp"); 1113 1114 static struct p9_trans_module p9_unix_trans = { 1115 .name = "unix", 1116 .maxsize = MAX_SOCK_BUF, 1117 .def = 0, 1118 .create = p9_fd_create_unix, 1119 .close = p9_fd_close, 1120 .request = p9_fd_request, 1121 .cancel = p9_fd_cancel, 1122 .cancelled = p9_fd_cancelled, 1123 .show_options = p9_fd_show_options, 1124 .owner = THIS_MODULE, 1125 }; 1126 MODULE_ALIAS_9P("unix"); 1127 1128 static struct p9_trans_module p9_fd_trans = { 1129 .name = "fd", 1130 .maxsize = MAX_SOCK_BUF, 1131 .def = 0, 1132 .create = p9_fd_create, 1133 .close = p9_fd_close, 1134 .request = p9_fd_request, 1135 .cancel = p9_fd_cancel, 1136 .cancelled = p9_fd_cancelled, 1137 .show_options = p9_fd_show_options, 1138 .owner = THIS_MODULE, 1139 }; 1140 MODULE_ALIAS_9P("fd"); 1141 1142 /** 1143 * p9_poll_workfn - poll worker thread 1144 * @work: work queue 1145 * 1146 * polls all v9fs transports for new events and queues the appropriate 1147 * work to the work queue 1148 * 1149 */ 1150 1151 static void p9_poll_workfn(struct work_struct *work) 1152 { 1153 unsigned long flags; 1154 1155 p9_debug(P9_DEBUG_TRANS, "start %p\n", current); 1156 1157 spin_lock_irqsave(&p9_poll_lock, flags); 1158 while (!list_empty(&p9_poll_pending_list)) { 1159 struct p9_conn *conn = list_first_entry(&p9_poll_pending_list, 1160 struct p9_conn, 1161 poll_pending_link); 1162 list_del_init(&conn->poll_pending_link); 1163 spin_unlock_irqrestore(&p9_poll_lock, flags); 1164 1165 p9_poll_mux(conn); 1166 1167 spin_lock_irqsave(&p9_poll_lock, flags); 1168 } 1169 spin_unlock_irqrestore(&p9_poll_lock, flags); 1170 1171 p9_debug(P9_DEBUG_TRANS, "finish\n"); 1172 } 1173 1174 static int __init p9_trans_fd_init(void) 1175 { 1176 v9fs_register_trans(&p9_tcp_trans); 1177 v9fs_register_trans(&p9_unix_trans); 1178 v9fs_register_trans(&p9_fd_trans); 1179 1180 return 0; 1181 } 1182 1183 static void __exit p9_trans_fd_exit(void) 1184 { 1185 flush_work(&p9_poll_work); 1186 v9fs_unregister_trans(&p9_tcp_trans); 1187 v9fs_unregister_trans(&p9_unix_trans); 1188 v9fs_unregister_trans(&p9_fd_trans); 1189 } 1190 1191 module_init(p9_trans_fd_init); 1192 module_exit(p9_trans_fd_exit); 1193 1194 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>"); 1195 MODULE_DESCRIPTION("Filedescriptor Transport for 9P"); 1196 MODULE_LICENSE("GPL"); 1197