1// SPDX-License-Identifier: GPL-2.0-only 2/* 3 * linux/fs/9p/trans_fd.c 4 * 5 * Fd transport layer. Includes deprecated socket layer. 6 * 7 * Copyright (C) 2006 by Russ Cox <rsc@swtch.com> 8 * Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net> 9 * Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com> 10 * Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com> 11 */ 12 13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 14 15#include <linux/in.h> 16#include <linux/module.h> 17#include <linux/net.h> 18#include <linux/ipv6.h> 19#include <linux/kthread.h> 20#include <linux/errno.h> 21#include <linux/kernel.h> 22#include <linux/un.h> 23#include <linux/uaccess.h> 24#include <linux/inet.h> 25#include <linux/idr.h> 26#include <linux/file.h> 27#include <linux/parser.h> 28#include <linux/slab.h> 29#include <linux/seq_file.h> 30#include <net/9p/9p.h> 31#include <net/9p/client.h> 32#include <net/9p/transport.h> 33 34#include <linux/syscalls.h> /* killme */ 35 36#define P9_PORT 564 37#define MAX_SOCK_BUF (64*1024) 38#define MAXPOLLWADDR 2 39 40static struct p9_trans_module p9_tcp_trans; 41static struct p9_trans_module p9_fd_trans; 42 43/** 44 * struct p9_fd_opts - per-transport options 45 * @rfd: file descriptor for reading (trans=fd) 46 * @wfd: file descriptor for writing (trans=fd) 47 * @port: port to connect to (trans=tcp) 48 * 49 */ 50 51struct p9_fd_opts { 52 int rfd; 53 int wfd; 54 u16 port; 55 bool privport; 56}; 57 58/* 59 * Option Parsing (code inspired by NFS code) 60 * - a little lazy - parse all fd-transport options 61 */ 62 63enum { 64 /* Options that take integer arguments */ 65 Opt_port, Opt_rfdno, Opt_wfdno, Opt_err, 66 /* Options that take no arguments */ 67 Opt_privport, 68}; 69 70static const match_table_t tokens = { 71 {Opt_port, "port=%u"}, 72 {Opt_rfdno, "rfdno=%u"}, 73 {Opt_wfdno, "wfdno=%u"}, 74 {Opt_privport, "privport"}, 75 {Opt_err, NULL}, 76}; 77 78enum { 79 Rworksched = 1, /* read work scheduled or running */ 80 Rpending = 2, /* can read */ 81 Wworksched = 4, /* write work scheduled or running */ 82 Wpending = 8, /* can write */ 83}; 84 85struct p9_poll_wait { 86 struct p9_conn *conn; 87 wait_queue_entry_t wait; 88 wait_queue_head_t *wait_addr; 89}; 90 91/** 92 * struct p9_conn - fd mux connection state information 93 * @mux_list: list link for mux to manage multiple connections (?) 94 * @client: reference to client instance for this connection 95 * @err: error state 96 * @req_list: accounting for requests which have been sent 97 * @unsent_req_list: accounting for requests that haven't been sent 98 * @req: current request being processed (if any) 99 * @tmp_buf: temporary buffer to read in header 100 * @rc: temporary fcall for reading current frame 101 * @wpos: write position for current frame 102 * @wsize: amount of data to write for current frame 103 * @wbuf: current write buffer 104 * @poll_pending_link: pending links to be polled per conn 105 * @poll_wait: array of wait_q's for various worker threads 106 * @pt: poll state 107 * @rq: current read work 108 * @wq: current write work 109 * @wsched: ???? 110 * 111 */ 112 113struct p9_conn { 114 struct list_head mux_list; 115 struct p9_client *client; 116 int err; 117 struct list_head req_list; 118 struct list_head unsent_req_list; 119 struct p9_req_t *rreq; 120 struct p9_req_t *wreq; 121 char tmp_buf[P9_HDRSZ]; 122 struct p9_fcall rc; 123 int wpos; 124 int wsize; 125 char *wbuf; 126 struct list_head poll_pending_link; 127 struct p9_poll_wait poll_wait[MAXPOLLWADDR]; 128 poll_table pt; 129 struct work_struct rq; 130 struct work_struct wq; 131 unsigned long wsched; 132}; 133 134/** 135 * struct p9_trans_fd - transport state 136 * @rd: reference to file to read from 137 * @wr: reference of file to write to 138 * @conn: connection state reference 139 * 140 */ 141 142struct p9_trans_fd { 143 struct file *rd; 144 struct file *wr; 145 struct p9_conn conn; 146}; 147 148static void p9_poll_workfn(struct work_struct *work); 149 150static DEFINE_SPINLOCK(p9_poll_lock); 151static LIST_HEAD(p9_poll_pending_list); 152static DECLARE_WORK(p9_poll_work, p9_poll_workfn); 153 154static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT; 155static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT; 156 157static void p9_mux_poll_stop(struct p9_conn *m) 158{ 159 unsigned long flags; 160 int i; 161 162 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) { 163 struct p9_poll_wait *pwait = &m->poll_wait[i]; 164 165 if (pwait->wait_addr) { 166 remove_wait_queue(pwait->wait_addr, &pwait->wait); 167 pwait->wait_addr = NULL; 168 } 169 } 170 171 spin_lock_irqsave(&p9_poll_lock, flags); 172 list_del_init(&m->poll_pending_link); 173 spin_unlock_irqrestore(&p9_poll_lock, flags); 174 175 flush_work(&p9_poll_work); 176} 177 178/** 179 * p9_conn_cancel - cancel all pending requests with error 180 * @m: mux data 181 * @err: error code 182 * 183 */ 184 185static void p9_conn_cancel(struct p9_conn *m, int err) 186{ 187 struct p9_req_t *req, *rtmp; 188 LIST_HEAD(cancel_list); 189 190 p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err); 191 192 spin_lock(&m->client->lock); 193 194 if (m->err) { 195 spin_unlock(&m->client->lock); 196 return; 197 } 198 199 m->err = err; 200 201 list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) { 202 list_move(&req->req_list, &cancel_list); 203 req->status = REQ_STATUS_ERROR; 204 } 205 list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) { 206 list_move(&req->req_list, &cancel_list); 207 req->status = REQ_STATUS_ERROR; 208 } 209 210 spin_unlock(&m->client->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 221static __poll_t 222p9_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 250static 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 278static 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 = P9_HDRSZ; /* 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 = P9_HDRSZ; 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 m->rreq = NULL; 350 err = -EIO; 351 goto error; 352 } 353 m->rc.sdata = m->rreq->rc.sdata; 354 memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity); 355 m->rc.capacity = m->rc.size; 356 } 357 358 /* packet is read in 359 * not an else because some packets (like clunk) have no payload 360 */ 361 if ((m->rreq) && (m->rc.offset == m->rc.capacity)) { 362 p9_debug(P9_DEBUG_TRANS, "got new packet\n"); 363 m->rreq->rc.size = m->rc.offset; 364 spin_lock(&m->client->lock); 365 if (m->rreq->status == REQ_STATUS_SENT) { 366 list_del(&m->rreq->req_list); 367 p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD); 368 } else if (m->rreq->status == REQ_STATUS_FLSHD) { 369 /* Ignore replies associated with a cancelled request. */ 370 p9_debug(P9_DEBUG_TRANS, 371 "Ignore replies associated with a cancelled request\n"); 372 } else { 373 spin_unlock(&m->client->lock); 374 p9_debug(P9_DEBUG_ERROR, 375 "Request tag %d errored out while we were reading the reply\n", 376 m->rc.tag); 377 err = -EIO; 378 goto error; 379 } 380 spin_unlock(&m->client->lock); 381 m->rc.sdata = NULL; 382 m->rc.offset = 0; 383 m->rc.capacity = 0; 384 p9_req_put(m->rreq); 385 m->rreq = NULL; 386 } 387 388end_clear: 389 clear_bit(Rworksched, &m->wsched); 390 391 if (!list_empty(&m->req_list)) { 392 if (test_and_clear_bit(Rpending, &m->wsched)) 393 n = EPOLLIN; 394 else 395 n = p9_fd_poll(m->client, NULL, NULL); 396 397 if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) { 398 p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m); 399 schedule_work(&m->rq); 400 } 401 } 402 403 return; 404error: 405 p9_conn_cancel(m, err); 406 clear_bit(Rworksched, &m->wsched); 407} 408 409/** 410 * p9_fd_write - write to a socket 411 * @client: client instance 412 * @v: buffer to send data from 413 * @len: size of send buffer 414 * 415 */ 416 417static int p9_fd_write(struct p9_client *client, void *v, int len) 418{ 419 ssize_t ret; 420 struct p9_trans_fd *ts = NULL; 421 422 if (client && client->status != Disconnected) 423 ts = client->trans; 424 425 if (!ts) 426 return -EREMOTEIO; 427 428 if (!(ts->wr->f_flags & O_NONBLOCK)) 429 p9_debug(P9_DEBUG_ERROR, "blocking write ...\n"); 430 431 ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos); 432 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN) 433 client->status = Disconnected; 434 return ret; 435} 436 437/** 438 * p9_write_work - called when a transport can send some data 439 * @work: container for work to be done 440 * 441 */ 442 443static void p9_write_work(struct work_struct *work) 444{ 445 __poll_t n; 446 int err; 447 struct p9_conn *m; 448 struct p9_req_t *req; 449 450 m = container_of(work, struct p9_conn, wq); 451 452 if (m->err < 0) { 453 clear_bit(Wworksched, &m->wsched); 454 return; 455 } 456 457 if (!m->wsize) { 458 spin_lock(&m->client->lock); 459 if (list_empty(&m->unsent_req_list)) { 460 clear_bit(Wworksched, &m->wsched); 461 spin_unlock(&m->client->lock); 462 return; 463 } 464 465 req = list_entry(m->unsent_req_list.next, struct p9_req_t, 466 req_list); 467 req->status = REQ_STATUS_SENT; 468 p9_debug(P9_DEBUG_TRANS, "move req %p\n", req); 469 list_move_tail(&req->req_list, &m->req_list); 470 471 m->wbuf = req->tc.sdata; 472 m->wsize = req->tc.size; 473 m->wpos = 0; 474 p9_req_get(req); 475 m->wreq = req; 476 spin_unlock(&m->client->lock); 477 } 478 479 p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n", 480 m, m->wpos, m->wsize); 481 clear_bit(Wpending, &m->wsched); 482 err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos); 483 p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err); 484 if (err == -EAGAIN) 485 goto end_clear; 486 487 488 if (err < 0) 489 goto error; 490 else if (err == 0) { 491 err = -EREMOTEIO; 492 goto error; 493 } 494 495 m->wpos += err; 496 if (m->wpos == m->wsize) { 497 m->wpos = m->wsize = 0; 498 p9_req_put(m->wreq); 499 m->wreq = NULL; 500 } 501 502end_clear: 503 clear_bit(Wworksched, &m->wsched); 504 505 if (m->wsize || !list_empty(&m->unsent_req_list)) { 506 if (test_and_clear_bit(Wpending, &m->wsched)) 507 n = EPOLLOUT; 508 else 509 n = p9_fd_poll(m->client, NULL, NULL); 510 511 if ((n & EPOLLOUT) && 512 !test_and_set_bit(Wworksched, &m->wsched)) { 513 p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m); 514 schedule_work(&m->wq); 515 } 516 } 517 518 return; 519 520error: 521 p9_conn_cancel(m, err); 522 clear_bit(Wworksched, &m->wsched); 523} 524 525static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key) 526{ 527 struct p9_poll_wait *pwait = 528 container_of(wait, struct p9_poll_wait, wait); 529 struct p9_conn *m = pwait->conn; 530 unsigned long flags; 531 532 spin_lock_irqsave(&p9_poll_lock, flags); 533 if (list_empty(&m->poll_pending_link)) 534 list_add_tail(&m->poll_pending_link, &p9_poll_pending_list); 535 spin_unlock_irqrestore(&p9_poll_lock, flags); 536 537 schedule_work(&p9_poll_work); 538 return 1; 539} 540 541/** 542 * p9_pollwait - add poll task to the wait queue 543 * @filp: file pointer being polled 544 * @wait_address: wait_q to block on 545 * @p: poll state 546 * 547 * called by files poll operation to add v9fs-poll task to files wait queue 548 */ 549 550static void 551p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p) 552{ 553 struct p9_conn *m = container_of(p, struct p9_conn, pt); 554 struct p9_poll_wait *pwait = NULL; 555 int i; 556 557 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) { 558 if (m->poll_wait[i].wait_addr == NULL) { 559 pwait = &m->poll_wait[i]; 560 break; 561 } 562 } 563 564 if (!pwait) { 565 p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n"); 566 return; 567 } 568 569 pwait->conn = m; 570 pwait->wait_addr = wait_address; 571 init_waitqueue_func_entry(&pwait->wait, p9_pollwake); 572 add_wait_queue(wait_address, &pwait->wait); 573} 574 575/** 576 * p9_conn_create - initialize the per-session mux data 577 * @client: client instance 578 * 579 * Note: Creates the polling task if this is the first session. 580 */ 581 582static void p9_conn_create(struct p9_client *client) 583{ 584 __poll_t n; 585 struct p9_trans_fd *ts = client->trans; 586 struct p9_conn *m = &ts->conn; 587 588 p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize); 589 590 INIT_LIST_HEAD(&m->mux_list); 591 m->client = client; 592 593 INIT_LIST_HEAD(&m->req_list); 594 INIT_LIST_HEAD(&m->unsent_req_list); 595 INIT_WORK(&m->rq, p9_read_work); 596 INIT_WORK(&m->wq, p9_write_work); 597 INIT_LIST_HEAD(&m->poll_pending_link); 598 init_poll_funcptr(&m->pt, p9_pollwait); 599 600 n = p9_fd_poll(client, &m->pt, NULL); 601 if (n & EPOLLIN) { 602 p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m); 603 set_bit(Rpending, &m->wsched); 604 } 605 606 if (n & EPOLLOUT) { 607 p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m); 608 set_bit(Wpending, &m->wsched); 609 } 610} 611 612/** 613 * p9_poll_mux - polls a mux and schedules read or write works if necessary 614 * @m: connection to poll 615 * 616 */ 617 618static void p9_poll_mux(struct p9_conn *m) 619{ 620 __poll_t n; 621 int err = -ECONNRESET; 622 623 if (m->err < 0) 624 return; 625 626 n = p9_fd_poll(m->client, NULL, &err); 627 if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) { 628 p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n); 629 p9_conn_cancel(m, err); 630 } 631 632 if (n & EPOLLIN) { 633 set_bit(Rpending, &m->wsched); 634 p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m); 635 if (!test_and_set_bit(Rworksched, &m->wsched)) { 636 p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m); 637 schedule_work(&m->rq); 638 } 639 } 640 641 if (n & EPOLLOUT) { 642 set_bit(Wpending, &m->wsched); 643 p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m); 644 if ((m->wsize || !list_empty(&m->unsent_req_list)) && 645 !test_and_set_bit(Wworksched, &m->wsched)) { 646 p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m); 647 schedule_work(&m->wq); 648 } 649 } 650} 651 652/** 653 * p9_fd_request - send 9P request 654 * The function can sleep until the request is scheduled for sending. 655 * The function can be interrupted. Return from the function is not 656 * a guarantee that the request is sent successfully. 657 * 658 * @client: client instance 659 * @req: request to be sent 660 * 661 */ 662 663static int p9_fd_request(struct p9_client *client, struct p9_req_t *req) 664{ 665 __poll_t n; 666 struct p9_trans_fd *ts = client->trans; 667 struct p9_conn *m = &ts->conn; 668 669 p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n", 670 m, current, &req->tc, req->tc.id); 671 if (m->err < 0) 672 return m->err; 673 674 spin_lock(&client->lock); 675 req->status = REQ_STATUS_UNSENT; 676 list_add_tail(&req->req_list, &m->unsent_req_list); 677 spin_unlock(&client->lock); 678 679 if (test_and_clear_bit(Wpending, &m->wsched)) 680 n = EPOLLOUT; 681 else 682 n = p9_fd_poll(m->client, NULL, NULL); 683 684 if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched)) 685 schedule_work(&m->wq); 686 687 return 0; 688} 689 690static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req) 691{ 692 int ret = 1; 693 694 p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req); 695 696 spin_lock(&client->lock); 697 698 if (req->status == REQ_STATUS_UNSENT) { 699 list_del(&req->req_list); 700 req->status = REQ_STATUS_FLSHD; 701 p9_req_put(req); 702 ret = 0; 703 } 704 spin_unlock(&client->lock); 705 706 return ret; 707} 708 709static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req) 710{ 711 p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req); 712 713 spin_lock(&client->lock); 714 /* Ignore cancelled request if message has been received 715 * before lock. 716 */ 717 if (req->status == REQ_STATUS_RCVD) { 718 spin_unlock(&client->lock); 719 return 0; 720 } 721 722 /* we haven't received a response for oldreq, 723 * remove it from the list. 724 */ 725 list_del(&req->req_list); 726 req->status = REQ_STATUS_FLSHD; 727 spin_unlock(&client->lock); 728 p9_req_put(req); 729 730 return 0; 731} 732 733static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt) 734{ 735 if (clnt->trans_mod == &p9_tcp_trans) { 736 if (clnt->trans_opts.tcp.port != P9_PORT) 737 seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port); 738 } else if (clnt->trans_mod == &p9_fd_trans) { 739 if (clnt->trans_opts.fd.rfd != ~0) 740 seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd); 741 if (clnt->trans_opts.fd.wfd != ~0) 742 seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd); 743 } 744 return 0; 745} 746 747/** 748 * parse_opts - parse mount options into p9_fd_opts structure 749 * @params: options string passed from mount 750 * @opts: fd transport-specific structure to parse options into 751 * 752 * Returns 0 upon success, -ERRNO upon failure 753 */ 754 755static int parse_opts(char *params, struct p9_fd_opts *opts) 756{ 757 char *p; 758 substring_t args[MAX_OPT_ARGS]; 759 int option; 760 char *options, *tmp_options; 761 762 opts->port = P9_PORT; 763 opts->rfd = ~0; 764 opts->wfd = ~0; 765 opts->privport = false; 766 767 if (!params) 768 return 0; 769 770 tmp_options = kstrdup(params, GFP_KERNEL); 771 if (!tmp_options) { 772 p9_debug(P9_DEBUG_ERROR, 773 "failed to allocate copy of option string\n"); 774 return -ENOMEM; 775 } 776 options = tmp_options; 777 778 while ((p = strsep(&options, ",")) != NULL) { 779 int token; 780 int r; 781 if (!*p) 782 continue; 783 token = match_token(p, tokens, args); 784 if ((token != Opt_err) && (token != Opt_privport)) { 785 r = match_int(&args[0], &option); 786 if (r < 0) { 787 p9_debug(P9_DEBUG_ERROR, 788 "integer field, but no integer?\n"); 789 continue; 790 } 791 } 792 switch (token) { 793 case Opt_port: 794 opts->port = option; 795 break; 796 case Opt_rfdno: 797 opts->rfd = option; 798 break; 799 case Opt_wfdno: 800 opts->wfd = option; 801 break; 802 case Opt_privport: 803 opts->privport = true; 804 break; 805 default: 806 continue; 807 } 808 } 809 810 kfree(tmp_options); 811 return 0; 812} 813 814static int p9_fd_open(struct p9_client *client, int rfd, int wfd) 815{ 816 struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd), 817 GFP_KERNEL); 818 if (!ts) 819 return -ENOMEM; 820 821 ts->rd = fget(rfd); 822 if (!ts->rd) 823 goto out_free_ts; 824 if (!(ts->rd->f_mode & FMODE_READ)) 825 goto out_put_rd; 826 /* Prevent workers from hanging on IO when fd is a pipe. 827 * It's technically possible for userspace or concurrent mounts to 828 * modify this flag concurrently, which will likely result in a 829 * broken filesystem. However, just having bad flags here should 830 * not crash the kernel or cause any other sort of bug, so mark this 831 * particular data race as intentional so that tooling (like KCSAN) 832 * can allow it and detect further problems. 833 */ 834 data_race(ts->rd->f_flags |= O_NONBLOCK); 835 ts->wr = fget(wfd); 836 if (!ts->wr) 837 goto out_put_rd; 838 if (!(ts->wr->f_mode & FMODE_WRITE)) 839 goto out_put_wr; 840 data_race(ts->wr->f_flags |= O_NONBLOCK); 841 842 client->trans = ts; 843 client->status = Connected; 844 845 return 0; 846 847out_put_wr: 848 fput(ts->wr); 849out_put_rd: 850 fput(ts->rd); 851out_free_ts: 852 kfree(ts); 853 return -EIO; 854} 855 856static int p9_socket_open(struct p9_client *client, struct socket *csocket) 857{ 858 struct p9_trans_fd *p; 859 struct file *file; 860 861 p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL); 862 if (!p) { 863 sock_release(csocket); 864 return -ENOMEM; 865 } 866 867 csocket->sk->sk_allocation = GFP_NOIO; 868 file = sock_alloc_file(csocket, 0, NULL); 869 if (IS_ERR(file)) { 870 pr_err("%s (%d): failed to map fd\n", 871 __func__, task_pid_nr(current)); 872 kfree(p); 873 return PTR_ERR(file); 874 } 875 876 get_file(file); 877 p->wr = p->rd = file; 878 client->trans = p; 879 client->status = Connected; 880 881 p->rd->f_flags |= O_NONBLOCK; 882 883 p9_conn_create(client); 884 return 0; 885} 886 887/** 888 * p9_mux_destroy - cancels all pending requests of mux 889 * @m: mux to destroy 890 * 891 */ 892 893static void p9_conn_destroy(struct p9_conn *m) 894{ 895 p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n", 896 m, m->mux_list.prev, m->mux_list.next); 897 898 p9_mux_poll_stop(m); 899 cancel_work_sync(&m->rq); 900 if (m->rreq) { 901 p9_req_put(m->rreq); 902 m->rreq = NULL; 903 } 904 cancel_work_sync(&m->wq); 905 if (m->wreq) { 906 p9_req_put(m->wreq); 907 m->wreq = NULL; 908 } 909 910 p9_conn_cancel(m, -ECONNRESET); 911 912 m->client = NULL; 913} 914 915/** 916 * p9_fd_close - shutdown file descriptor transport 917 * @client: client instance 918 * 919 */ 920 921static void p9_fd_close(struct p9_client *client) 922{ 923 struct p9_trans_fd *ts; 924 925 if (!client) 926 return; 927 928 ts = client->trans; 929 if (!ts) 930 return; 931 932 client->status = Disconnected; 933 934 p9_conn_destroy(&ts->conn); 935 936 if (ts->rd) 937 fput(ts->rd); 938 if (ts->wr) 939 fput(ts->wr); 940 941 kfree(ts); 942} 943 944/* 945 * stolen from NFS - maybe should be made a generic function? 946 */ 947static inline int valid_ipaddr4(const char *buf) 948{ 949 int rc, count, in[4]; 950 951 rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]); 952 if (rc != 4) 953 return -EINVAL; 954 for (count = 0; count < 4; count++) { 955 if (in[count] > 255) 956 return -EINVAL; 957 } 958 return 0; 959} 960 961static int p9_bind_privport(struct socket *sock) 962{ 963 struct sockaddr_in cl; 964 int port, err = -EINVAL; 965 966 memset(&cl, 0, sizeof(cl)); 967 cl.sin_family = AF_INET; 968 cl.sin_addr.s_addr = htonl(INADDR_ANY); 969 for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) { 970 cl.sin_port = htons((ushort)port); 971 err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl)); 972 if (err != -EADDRINUSE) 973 break; 974 } 975 return err; 976} 977 978 979static int 980p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args) 981{ 982 int err; 983 struct socket *csocket; 984 struct sockaddr_in sin_server; 985 struct p9_fd_opts opts; 986 987 err = parse_opts(args, &opts); 988 if (err < 0) 989 return err; 990 991 if (addr == NULL || valid_ipaddr4(addr) < 0) 992 return -EINVAL; 993 994 csocket = NULL; 995 996 client->trans_opts.tcp.port = opts.port; 997 client->trans_opts.tcp.privport = opts.privport; 998 sin_server.sin_family = AF_INET; 999 sin_server.sin_addr.s_addr = in_aton(addr); 1000 sin_server.sin_port = htons(opts.port); 1001 err = __sock_create(current->nsproxy->net_ns, PF_INET, 1002 SOCK_STREAM, IPPROTO_TCP, &csocket, 1); 1003 if (err) { 1004 pr_err("%s (%d): problem creating socket\n", 1005 __func__, task_pid_nr(current)); 1006 return err; 1007 } 1008 1009 if (opts.privport) { 1010 err = p9_bind_privport(csocket); 1011 if (err < 0) { 1012 pr_err("%s (%d): problem binding to privport\n", 1013 __func__, task_pid_nr(current)); 1014 sock_release(csocket); 1015 return err; 1016 } 1017 } 1018 1019 err = csocket->ops->connect(csocket, 1020 (struct sockaddr *)&sin_server, 1021 sizeof(struct sockaddr_in), 0); 1022 if (err < 0) { 1023 pr_err("%s (%d): problem connecting socket to %s\n", 1024 __func__, task_pid_nr(current), addr); 1025 sock_release(csocket); 1026 return err; 1027 } 1028 1029 return p9_socket_open(client, csocket); 1030} 1031 1032static int 1033p9_fd_create_unix(struct p9_client *client, const char *addr, char *args) 1034{ 1035 int err; 1036 struct socket *csocket; 1037 struct sockaddr_un sun_server; 1038 1039 csocket = NULL; 1040 1041 if (!addr || !strlen(addr)) 1042 return -EINVAL; 1043 1044 if (strlen(addr) >= UNIX_PATH_MAX) { 1045 pr_err("%s (%d): address too long: %s\n", 1046 __func__, task_pid_nr(current), addr); 1047 return -ENAMETOOLONG; 1048 } 1049 1050 sun_server.sun_family = PF_UNIX; 1051 strcpy(sun_server.sun_path, addr); 1052 err = __sock_create(current->nsproxy->net_ns, PF_UNIX, 1053 SOCK_STREAM, 0, &csocket, 1); 1054 if (err < 0) { 1055 pr_err("%s (%d): problem creating socket\n", 1056 __func__, task_pid_nr(current)); 1057 1058 return err; 1059 } 1060 err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server, 1061 sizeof(struct sockaddr_un) - 1, 0); 1062 if (err < 0) { 1063 pr_err("%s (%d): problem connecting socket: %s: %d\n", 1064 __func__, task_pid_nr(current), addr, err); 1065 sock_release(csocket); 1066 return err; 1067 } 1068 1069 return p9_socket_open(client, csocket); 1070} 1071 1072static int 1073p9_fd_create(struct p9_client *client, const char *addr, char *args) 1074{ 1075 int err; 1076 struct p9_fd_opts opts; 1077 1078 parse_opts(args, &opts); 1079 client->trans_opts.fd.rfd = opts.rfd; 1080 client->trans_opts.fd.wfd = opts.wfd; 1081 1082 if (opts.rfd == ~0 || opts.wfd == ~0) { 1083 pr_err("Insufficient options for proto=fd\n"); 1084 return -ENOPROTOOPT; 1085 } 1086 1087 err = p9_fd_open(client, opts.rfd, opts.wfd); 1088 if (err < 0) 1089 return err; 1090 1091 p9_conn_create(client); 1092 1093 return 0; 1094} 1095 1096static struct p9_trans_module p9_tcp_trans = { 1097 .name = "tcp", 1098 .maxsize = MAX_SOCK_BUF, 1099 .def = 0, 1100 .create = p9_fd_create_tcp, 1101 .close = p9_fd_close, 1102 .request = p9_fd_request, 1103 .cancel = p9_fd_cancel, 1104 .cancelled = p9_fd_cancelled, 1105 .show_options = p9_fd_show_options, 1106 .owner = THIS_MODULE, 1107}; 1108 1109static struct p9_trans_module p9_unix_trans = { 1110 .name = "unix", 1111 .maxsize = MAX_SOCK_BUF, 1112 .def = 0, 1113 .create = p9_fd_create_unix, 1114 .close = p9_fd_close, 1115 .request = p9_fd_request, 1116 .cancel = p9_fd_cancel, 1117 .cancelled = p9_fd_cancelled, 1118 .show_options = p9_fd_show_options, 1119 .owner = THIS_MODULE, 1120}; 1121 1122static struct p9_trans_module p9_fd_trans = { 1123 .name = "fd", 1124 .maxsize = MAX_SOCK_BUF, 1125 .def = 0, 1126 .create = p9_fd_create, 1127 .close = p9_fd_close, 1128 .request = p9_fd_request, 1129 .cancel = p9_fd_cancel, 1130 .cancelled = p9_fd_cancelled, 1131 .show_options = p9_fd_show_options, 1132 .owner = THIS_MODULE, 1133}; 1134 1135/** 1136 * p9_poll_workfn - poll worker thread 1137 * @work: work queue 1138 * 1139 * polls all v9fs transports for new events and queues the appropriate 1140 * work to the work queue 1141 * 1142 */ 1143 1144static void p9_poll_workfn(struct work_struct *work) 1145{ 1146 unsigned long flags; 1147 1148 p9_debug(P9_DEBUG_TRANS, "start %p\n", current); 1149 1150 spin_lock_irqsave(&p9_poll_lock, flags); 1151 while (!list_empty(&p9_poll_pending_list)) { 1152 struct p9_conn *conn = list_first_entry(&p9_poll_pending_list, 1153 struct p9_conn, 1154 poll_pending_link); 1155 list_del_init(&conn->poll_pending_link); 1156 spin_unlock_irqrestore(&p9_poll_lock, flags); 1157 1158 p9_poll_mux(conn); 1159 1160 spin_lock_irqsave(&p9_poll_lock, flags); 1161 } 1162 spin_unlock_irqrestore(&p9_poll_lock, flags); 1163 1164 p9_debug(P9_DEBUG_TRANS, "finish\n"); 1165} 1166 1167int p9_trans_fd_init(void) 1168{ 1169 v9fs_register_trans(&p9_tcp_trans); 1170 v9fs_register_trans(&p9_unix_trans); 1171 v9fs_register_trans(&p9_fd_trans); 1172 1173 return 0; 1174} 1175 1176void p9_trans_fd_exit(void) 1177{ 1178 flush_work(&p9_poll_work); 1179 v9fs_unregister_trans(&p9_tcp_trans); 1180 v9fs_unregister_trans(&p9_unix_trans); 1181 v9fs_unregister_trans(&p9_fd_trans); 1182} 1183