xref: /linux/net/9p/trans_fd.c (revision f79e4d5f92a129a1159c973735007d4ddc8541f3)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 
30 #include <linux/in.h>
31 #include <linux/module.h>
32 #include <linux/net.h>
33 #include <linux/ipv6.h>
34 #include <linux/kthread.h>
35 #include <linux/errno.h>
36 #include <linux/kernel.h>
37 #include <linux/un.h>
38 #include <linux/uaccess.h>
39 #include <linux/inet.h>
40 #include <linux/idr.h>
41 #include <linux/file.h>
42 #include <linux/parser.h>
43 #include <linux/slab.h>
44 #include <linux/seq_file.h>
45 #include <net/9p/9p.h>
46 #include <net/9p/client.h>
47 #include <net/9p/transport.h>
48 
49 #include <linux/syscalls.h> /* killme */
50 
51 #define P9_PORT 564
52 #define MAX_SOCK_BUF (64*1024)
53 #define MAXPOLLWADDR	2
54 
55 static struct p9_trans_module p9_tcp_trans;
56 static struct p9_trans_module p9_fd_trans;
57 
58 /**
59  * struct p9_fd_opts - per-transport options
60  * @rfd: file descriptor for reading (trans=fd)
61  * @wfd: file descriptor for writing (trans=fd)
62  * @port: port to connect to (trans=tcp)
63  *
64  */
65 
66 struct p9_fd_opts {
67 	int rfd;
68 	int wfd;
69 	u16 port;
70 	bool privport;
71 };
72 
73 /*
74   * Option Parsing (code inspired by NFS code)
75   *  - a little lazy - parse all fd-transport options
76   */
77 
78 enum {
79 	/* Options that take integer arguments */
80 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
81 	/* Options that take no arguments */
82 	Opt_privport,
83 };
84 
85 static const match_table_t tokens = {
86 	{Opt_port, "port=%u"},
87 	{Opt_rfdno, "rfdno=%u"},
88 	{Opt_wfdno, "wfdno=%u"},
89 	{Opt_privport, "privport"},
90 	{Opt_err, NULL},
91 };
92 
93 enum {
94 	Rworksched = 1,		/* read work scheduled or running */
95 	Rpending = 2,		/* can read */
96 	Wworksched = 4,		/* write work scheduled or running */
97 	Wpending = 8,		/* can write */
98 };
99 
100 struct p9_poll_wait {
101 	struct p9_conn *conn;
102 	wait_queue_entry_t wait;
103 	wait_queue_head_t *wait_addr;
104 };
105 
106 /**
107  * struct p9_conn - fd mux connection state information
108  * @mux_list: list link for mux to manage multiple connections (?)
109  * @client: reference to client instance for this connection
110  * @err: error state
111  * @req_list: accounting for requests which have been sent
112  * @unsent_req_list: accounting for requests that haven't been sent
113  * @req: current request being processed (if any)
114  * @tmp_buf: temporary buffer to read in header
115  * @rc: temporary fcall for reading current frame
116  * @wpos: write position for current frame
117  * @wsize: amount of data to write for current frame
118  * @wbuf: current write buffer
119  * @poll_pending_link: pending links to be polled per conn
120  * @poll_wait: array of wait_q's for various worker threads
121  * @pt: poll state
122  * @rq: current read work
123  * @wq: current write work
124  * @wsched: ????
125  *
126  */
127 
128 struct p9_conn {
129 	struct list_head mux_list;
130 	struct p9_client *client;
131 	int err;
132 	struct list_head req_list;
133 	struct list_head unsent_req_list;
134 	struct p9_req_t *req;
135 	char tmp_buf[7];
136 	struct p9_fcall rc;
137 	int wpos;
138 	int wsize;
139 	char *wbuf;
140 	struct list_head poll_pending_link;
141 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
142 	poll_table pt;
143 	struct work_struct rq;
144 	struct work_struct wq;
145 	unsigned long wsched;
146 };
147 
148 /**
149  * struct p9_trans_fd - transport state
150  * @rd: reference to file to read from
151  * @wr: reference of file to write to
152  * @conn: connection state reference
153  *
154  */
155 
156 struct p9_trans_fd {
157 	struct file *rd;
158 	struct file *wr;
159 	struct p9_conn conn;
160 };
161 
162 static void p9_poll_workfn(struct work_struct *work);
163 
164 static DEFINE_SPINLOCK(p9_poll_lock);
165 static LIST_HEAD(p9_poll_pending_list);
166 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
167 
168 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
169 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
170 
171 static void p9_mux_poll_stop(struct p9_conn *m)
172 {
173 	unsigned long flags;
174 	int i;
175 
176 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
177 		struct p9_poll_wait *pwait = &m->poll_wait[i];
178 
179 		if (pwait->wait_addr) {
180 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
181 			pwait->wait_addr = NULL;
182 		}
183 	}
184 
185 	spin_lock_irqsave(&p9_poll_lock, flags);
186 	list_del_init(&m->poll_pending_link);
187 	spin_unlock_irqrestore(&p9_poll_lock, flags);
188 }
189 
190 /**
191  * p9_conn_cancel - cancel all pending requests with error
192  * @m: mux data
193  * @err: error code
194  *
195  */
196 
197 static void p9_conn_cancel(struct p9_conn *m, int err)
198 {
199 	struct p9_req_t *req, *rtmp;
200 	unsigned long flags;
201 	LIST_HEAD(cancel_list);
202 
203 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
204 
205 	spin_lock_irqsave(&m->client->lock, flags);
206 
207 	if (m->err) {
208 		spin_unlock_irqrestore(&m->client->lock, flags);
209 		return;
210 	}
211 
212 	m->err = err;
213 
214 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
215 		list_move(&req->req_list, &cancel_list);
216 	}
217 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
218 		list_move(&req->req_list, &cancel_list);
219 	}
220 	spin_unlock_irqrestore(&m->client->lock, flags);
221 
222 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
223 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
224 		list_del(&req->req_list);
225 		if (!req->t_err)
226 			req->t_err = err;
227 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
228 	}
229 }
230 
231 static __poll_t
232 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
233 {
234 	__poll_t ret;
235 	struct p9_trans_fd *ts = NULL;
236 
237 	if (client && client->status == Connected)
238 		ts = client->trans;
239 
240 	if (!ts) {
241 		if (err)
242 			*err = -EREMOTEIO;
243 		return EPOLLERR;
244 	}
245 
246 	ret = vfs_poll(ts->rd, pt);
247 	if (ts->rd != ts->wr)
248 		ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN);
249 	return ret;
250 }
251 
252 /**
253  * p9_fd_read- read from a fd
254  * @client: client instance
255  * @v: buffer to receive data into
256  * @len: size of receive buffer
257  *
258  */
259 
260 static int p9_fd_read(struct p9_client *client, void *v, int len)
261 {
262 	int ret;
263 	struct p9_trans_fd *ts = NULL;
264 	loff_t pos;
265 
266 	if (client && client->status != Disconnected)
267 		ts = client->trans;
268 
269 	if (!ts)
270 		return -EREMOTEIO;
271 
272 	if (!(ts->rd->f_flags & O_NONBLOCK))
273 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
274 
275 	pos = ts->rd->f_pos;
276 	ret = kernel_read(ts->rd, v, len, &pos);
277 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
278 		client->status = Disconnected;
279 	return ret;
280 }
281 
282 /**
283  * p9_read_work - called when there is some data to be read from a transport
284  * @work: container of work to be done
285  *
286  */
287 
288 static void p9_read_work(struct work_struct *work)
289 {
290 	__poll_t n;
291 	int err;
292 	struct p9_conn *m;
293 	int status = REQ_STATUS_ERROR;
294 
295 	m = container_of(work, struct p9_conn, rq);
296 
297 	if (m->err < 0)
298 		return;
299 
300 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
301 
302 	if (!m->rc.sdata) {
303 		m->rc.sdata = m->tmp_buf;
304 		m->rc.offset = 0;
305 		m->rc.capacity = 7; /* start by reading header */
306 	}
307 
308 	clear_bit(Rpending, &m->wsched);
309 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
310 		 m, m->rc.offset, m->rc.capacity,
311 		 m->rc.capacity - m->rc.offset);
312 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
313 			 m->rc.capacity - m->rc.offset);
314 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
315 	if (err == -EAGAIN)
316 		goto end_clear;
317 
318 	if (err <= 0)
319 		goto error;
320 
321 	m->rc.offset += err;
322 
323 	/* header read in */
324 	if ((!m->req) && (m->rc.offset == m->rc.capacity)) {
325 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
326 
327 		err = p9_parse_header(&m->rc, NULL, NULL, NULL, 0);
328 		if (err) {
329 			p9_debug(P9_DEBUG_ERROR,
330 				 "error parsing header: %d\n", err);
331 			goto error;
332 		}
333 
334 		if (m->rc.size >= m->client->msize) {
335 			p9_debug(P9_DEBUG_ERROR,
336 				 "requested packet size too big: %d\n",
337 				 m->rc.size);
338 			err = -EIO;
339 			goto error;
340 		}
341 
342 		p9_debug(P9_DEBUG_TRANS,
343 			 "mux %p pkt: size: %d bytes tag: %d\n",
344 			 m, m->rc.size, m->rc.tag);
345 
346 		m->req = p9_tag_lookup(m->client, m->rc.tag);
347 		if (!m->req || (m->req->status != REQ_STATUS_SENT)) {
348 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
349 				 m->rc.tag);
350 			err = -EIO;
351 			goto error;
352 		}
353 
354 		if (m->req->rc == NULL) {
355 			p9_debug(P9_DEBUG_ERROR,
356 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
357 				 m->rc.tag, m->req);
358 			m->req = NULL;
359 			err = -EIO;
360 			goto error;
361 		}
362 		m->rc.sdata = (char *)m->req->rc + sizeof(struct p9_fcall);
363 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
364 		m->rc.capacity = m->rc.size;
365 	}
366 
367 	/* packet is read in
368 	 * not an else because some packets (like clunk) have no payload
369 	 */
370 	if ((m->req) && (m->rc.offset == m->rc.capacity)) {
371 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
372 		spin_lock(&m->client->lock);
373 		if (m->req->status != REQ_STATUS_ERROR)
374 			status = REQ_STATUS_RCVD;
375 		list_del(&m->req->req_list);
376 		spin_unlock(&m->client->lock);
377 		p9_client_cb(m->client, m->req, status);
378 		m->rc.sdata = NULL;
379 		m->rc.offset = 0;
380 		m->rc.capacity = 0;
381 		m->req = NULL;
382 	}
383 
384 end_clear:
385 	clear_bit(Rworksched, &m->wsched);
386 
387 	if (!list_empty(&m->req_list)) {
388 		if (test_and_clear_bit(Rpending, &m->wsched))
389 			n = EPOLLIN;
390 		else
391 			n = p9_fd_poll(m->client, NULL, NULL);
392 
393 		if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
394 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
395 			schedule_work(&m->rq);
396 		}
397 	}
398 
399 	return;
400 error:
401 	p9_conn_cancel(m, err);
402 	clear_bit(Rworksched, &m->wsched);
403 }
404 
405 /**
406  * p9_fd_write - write to a socket
407  * @client: client instance
408  * @v: buffer to send data from
409  * @len: size of send buffer
410  *
411  */
412 
413 static int p9_fd_write(struct p9_client *client, void *v, int len)
414 {
415 	ssize_t ret;
416 	struct p9_trans_fd *ts = NULL;
417 
418 	if (client && client->status != Disconnected)
419 		ts = client->trans;
420 
421 	if (!ts)
422 		return -EREMOTEIO;
423 
424 	if (!(ts->wr->f_flags & O_NONBLOCK))
425 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
426 
427 	ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
428 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
429 		client->status = Disconnected;
430 	return ret;
431 }
432 
433 /**
434  * p9_write_work - called when a transport can send some data
435  * @work: container for work to be done
436  *
437  */
438 
439 static void p9_write_work(struct work_struct *work)
440 {
441 	__poll_t n;
442 	int err;
443 	struct p9_conn *m;
444 	struct p9_req_t *req;
445 
446 	m = container_of(work, struct p9_conn, wq);
447 
448 	if (m->err < 0) {
449 		clear_bit(Wworksched, &m->wsched);
450 		return;
451 	}
452 
453 	if (!m->wsize) {
454 		spin_lock(&m->client->lock);
455 		if (list_empty(&m->unsent_req_list)) {
456 			clear_bit(Wworksched, &m->wsched);
457 			spin_unlock(&m->client->lock);
458 			return;
459 		}
460 
461 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
462 			       req_list);
463 		req->status = REQ_STATUS_SENT;
464 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
465 		list_move_tail(&req->req_list, &m->req_list);
466 
467 		m->wbuf = req->tc->sdata;
468 		m->wsize = req->tc->size;
469 		m->wpos = 0;
470 		spin_unlock(&m->client->lock);
471 	}
472 
473 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
474 		 m, m->wpos, m->wsize);
475 	clear_bit(Wpending, &m->wsched);
476 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
477 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
478 	if (err == -EAGAIN)
479 		goto end_clear;
480 
481 
482 	if (err < 0)
483 		goto error;
484 	else if (err == 0) {
485 		err = -EREMOTEIO;
486 		goto error;
487 	}
488 
489 	m->wpos += err;
490 	if (m->wpos == m->wsize)
491 		m->wpos = m->wsize = 0;
492 
493 end_clear:
494 	clear_bit(Wworksched, &m->wsched);
495 
496 	if (m->wsize || !list_empty(&m->unsent_req_list)) {
497 		if (test_and_clear_bit(Wpending, &m->wsched))
498 			n = EPOLLOUT;
499 		else
500 			n = p9_fd_poll(m->client, NULL, NULL);
501 
502 		if ((n & EPOLLOUT) &&
503 		   !test_and_set_bit(Wworksched, &m->wsched)) {
504 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
505 			schedule_work(&m->wq);
506 		}
507 	}
508 
509 	return;
510 
511 error:
512 	p9_conn_cancel(m, err);
513 	clear_bit(Wworksched, &m->wsched);
514 }
515 
516 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
517 {
518 	struct p9_poll_wait *pwait =
519 		container_of(wait, struct p9_poll_wait, wait);
520 	struct p9_conn *m = pwait->conn;
521 	unsigned long flags;
522 
523 	spin_lock_irqsave(&p9_poll_lock, flags);
524 	if (list_empty(&m->poll_pending_link))
525 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
526 	spin_unlock_irqrestore(&p9_poll_lock, flags);
527 
528 	schedule_work(&p9_poll_work);
529 	return 1;
530 }
531 
532 /**
533  * p9_pollwait - add poll task to the wait queue
534  * @filp: file pointer being polled
535  * @wait_address: wait_q to block on
536  * @p: poll state
537  *
538  * called by files poll operation to add v9fs-poll task to files wait queue
539  */
540 
541 static void
542 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
543 {
544 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
545 	struct p9_poll_wait *pwait = NULL;
546 	int i;
547 
548 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
549 		if (m->poll_wait[i].wait_addr == NULL) {
550 			pwait = &m->poll_wait[i];
551 			break;
552 		}
553 	}
554 
555 	if (!pwait) {
556 		p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
557 		return;
558 	}
559 
560 	pwait->conn = m;
561 	pwait->wait_addr = wait_address;
562 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
563 	add_wait_queue(wait_address, &pwait->wait);
564 }
565 
566 /**
567  * p9_conn_create - initialize the per-session mux data
568  * @client: client instance
569  *
570  * Note: Creates the polling task if this is the first session.
571  */
572 
573 static void p9_conn_create(struct p9_client *client)
574 {
575 	__poll_t n;
576 	struct p9_trans_fd *ts = client->trans;
577 	struct p9_conn *m = &ts->conn;
578 
579 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
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, NULL);
592 	if (n & EPOLLIN) {
593 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
594 		set_bit(Rpending, &m->wsched);
595 	}
596 
597 	if (n & EPOLLOUT) {
598 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
599 		set_bit(Wpending, &m->wsched);
600 	}
601 }
602 
603 /**
604  * p9_poll_mux - polls a mux and schedules read or write works if necessary
605  * @m: connection to poll
606  *
607  */
608 
609 static void p9_poll_mux(struct p9_conn *m)
610 {
611 	__poll_t n;
612 	int err = -ECONNRESET;
613 
614 	if (m->err < 0)
615 		return;
616 
617 	n = p9_fd_poll(m->client, NULL, &err);
618 	if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
619 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
620 		p9_conn_cancel(m, err);
621 	}
622 
623 	if (n & EPOLLIN) {
624 		set_bit(Rpending, &m->wsched);
625 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
626 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
627 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
628 			schedule_work(&m->rq);
629 		}
630 	}
631 
632 	if (n & EPOLLOUT) {
633 		set_bit(Wpending, &m->wsched);
634 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
635 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
636 		    !test_and_set_bit(Wworksched, &m->wsched)) {
637 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
638 			schedule_work(&m->wq);
639 		}
640 	}
641 }
642 
643 /**
644  * p9_fd_request - send 9P request
645  * The function can sleep until the request is scheduled for sending.
646  * The function can be interrupted. Return from the function is not
647  * a guarantee that the request is sent successfully.
648  *
649  * @client: client instance
650  * @req: request to be sent
651  *
652  */
653 
654 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
655 {
656 	__poll_t n;
657 	struct p9_trans_fd *ts = client->trans;
658 	struct p9_conn *m = &ts->conn;
659 
660 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
661 		 m, current, req->tc, req->tc->id);
662 	if (m->err < 0)
663 		return m->err;
664 
665 	spin_lock(&client->lock);
666 	req->status = REQ_STATUS_UNSENT;
667 	list_add_tail(&req->req_list, &m->unsent_req_list);
668 	spin_unlock(&client->lock);
669 
670 	if (test_and_clear_bit(Wpending, &m->wsched))
671 		n = EPOLLOUT;
672 	else
673 		n = p9_fd_poll(m->client, NULL, NULL);
674 
675 	if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
676 		schedule_work(&m->wq);
677 
678 	return 0;
679 }
680 
681 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
682 {
683 	int ret = 1;
684 
685 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
686 
687 	spin_lock(&client->lock);
688 
689 	if (req->status == REQ_STATUS_UNSENT) {
690 		list_del(&req->req_list);
691 		req->status = REQ_STATUS_FLSHD;
692 		ret = 0;
693 	}
694 	spin_unlock(&client->lock);
695 
696 	return ret;
697 }
698 
699 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
700 {
701 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
702 
703 	/* we haven't received a response for oldreq,
704 	 * remove it from the list.
705 	 */
706 	spin_lock(&client->lock);
707 	list_del(&req->req_list);
708 	spin_unlock(&client->lock);
709 
710 	return 0;
711 }
712 
713 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
714 {
715 	if (clnt->trans_mod == &p9_tcp_trans) {
716 		if (clnt->trans_opts.tcp.port != P9_PORT)
717 			seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
718 	} else if (clnt->trans_mod == &p9_fd_trans) {
719 		if (clnt->trans_opts.fd.rfd != ~0)
720 			seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
721 		if (clnt->trans_opts.fd.wfd != ~0)
722 			seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
723 	}
724 	return 0;
725 }
726 
727 /**
728  * parse_opts - parse mount options into p9_fd_opts structure
729  * @params: options string passed from mount
730  * @opts: fd transport-specific structure to parse options into
731  *
732  * Returns 0 upon success, -ERRNO upon failure
733  */
734 
735 static int parse_opts(char *params, struct p9_fd_opts *opts)
736 {
737 	char *p;
738 	substring_t args[MAX_OPT_ARGS];
739 	int option;
740 	char *options, *tmp_options;
741 
742 	opts->port = P9_PORT;
743 	opts->rfd = ~0;
744 	opts->wfd = ~0;
745 	opts->privport = false;
746 
747 	if (!params)
748 		return 0;
749 
750 	tmp_options = kstrdup(params, GFP_KERNEL);
751 	if (!tmp_options) {
752 		p9_debug(P9_DEBUG_ERROR,
753 			 "failed to allocate copy of option string\n");
754 		return -ENOMEM;
755 	}
756 	options = tmp_options;
757 
758 	while ((p = strsep(&options, ",")) != NULL) {
759 		int token;
760 		int r;
761 		if (!*p)
762 			continue;
763 		token = match_token(p, tokens, args);
764 		if ((token != Opt_err) && (token != Opt_privport)) {
765 			r = match_int(&args[0], &option);
766 			if (r < 0) {
767 				p9_debug(P9_DEBUG_ERROR,
768 					 "integer field, but no integer?\n");
769 				continue;
770 			}
771 		}
772 		switch (token) {
773 		case Opt_port:
774 			opts->port = option;
775 			break;
776 		case Opt_rfdno:
777 			opts->rfd = option;
778 			break;
779 		case Opt_wfdno:
780 			opts->wfd = option;
781 			break;
782 		case Opt_privport:
783 			opts->privport = true;
784 			break;
785 		default:
786 			continue;
787 		}
788 	}
789 
790 	kfree(tmp_options);
791 	return 0;
792 }
793 
794 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
795 {
796 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
797 					   GFP_KERNEL);
798 	if (!ts)
799 		return -ENOMEM;
800 
801 	ts->rd = fget(rfd);
802 	ts->wr = fget(wfd);
803 	if (!ts->rd || !ts->wr) {
804 		if (ts->rd)
805 			fput(ts->rd);
806 		if (ts->wr)
807 			fput(ts->wr);
808 		kfree(ts);
809 		return -EIO;
810 	}
811 
812 	client->trans = ts;
813 	client->status = Connected;
814 
815 	return 0;
816 }
817 
818 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
819 {
820 	struct p9_trans_fd *p;
821 	struct file *file;
822 
823 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
824 	if (!p)
825 		return -ENOMEM;
826 
827 	csocket->sk->sk_allocation = GFP_NOIO;
828 	file = sock_alloc_file(csocket, 0, NULL);
829 	if (IS_ERR(file)) {
830 		pr_err("%s (%d): failed to map fd\n",
831 		       __func__, task_pid_nr(current));
832 		kfree(p);
833 		return PTR_ERR(file);
834 	}
835 
836 	get_file(file);
837 	p->wr = p->rd = file;
838 	client->trans = p;
839 	client->status = Connected;
840 
841 	p->rd->f_flags |= O_NONBLOCK;
842 
843 	p9_conn_create(client);
844 	return 0;
845 }
846 
847 /**
848  * p9_mux_destroy - cancels all pending requests of mux
849  * @m: mux to destroy
850  *
851  */
852 
853 static void p9_conn_destroy(struct p9_conn *m)
854 {
855 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
856 		 m, m->mux_list.prev, m->mux_list.next);
857 
858 	p9_mux_poll_stop(m);
859 	cancel_work_sync(&m->rq);
860 	cancel_work_sync(&m->wq);
861 
862 	p9_conn_cancel(m, -ECONNRESET);
863 
864 	m->client = NULL;
865 }
866 
867 /**
868  * p9_fd_close - shutdown file descriptor transport
869  * @client: client instance
870  *
871  */
872 
873 static void p9_fd_close(struct p9_client *client)
874 {
875 	struct p9_trans_fd *ts;
876 
877 	if (!client)
878 		return;
879 
880 	ts = client->trans;
881 	if (!ts)
882 		return;
883 
884 	client->status = Disconnected;
885 
886 	p9_conn_destroy(&ts->conn);
887 
888 	if (ts->rd)
889 		fput(ts->rd);
890 	if (ts->wr)
891 		fput(ts->wr);
892 
893 	kfree(ts);
894 }
895 
896 /*
897  * stolen from NFS - maybe should be made a generic function?
898  */
899 static inline int valid_ipaddr4(const char *buf)
900 {
901 	int rc, count, in[4];
902 
903 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
904 	if (rc != 4)
905 		return -EINVAL;
906 	for (count = 0; count < 4; count++) {
907 		if (in[count] > 255)
908 			return -EINVAL;
909 	}
910 	return 0;
911 }
912 
913 static int p9_bind_privport(struct socket *sock)
914 {
915 	struct sockaddr_in cl;
916 	int port, err = -EINVAL;
917 
918 	memset(&cl, 0, sizeof(cl));
919 	cl.sin_family = AF_INET;
920 	cl.sin_addr.s_addr = INADDR_ANY;
921 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
922 		cl.sin_port = htons((ushort)port);
923 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
924 		if (err != -EADDRINUSE)
925 			break;
926 	}
927 	return err;
928 }
929 
930 
931 static int
932 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
933 {
934 	int err;
935 	struct socket *csocket;
936 	struct sockaddr_in sin_server;
937 	struct p9_fd_opts opts;
938 
939 	err = parse_opts(args, &opts);
940 	if (err < 0)
941 		return err;
942 
943 	if (valid_ipaddr4(addr) < 0)
944 		return -EINVAL;
945 
946 	csocket = NULL;
947 
948 	client->trans_opts.tcp.port = opts.port;
949 	client->trans_opts.tcp.privport = opts.privport;
950 	sin_server.sin_family = AF_INET;
951 	sin_server.sin_addr.s_addr = in_aton(addr);
952 	sin_server.sin_port = htons(opts.port);
953 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
954 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
955 	if (err) {
956 		pr_err("%s (%d): problem creating socket\n",
957 		       __func__, task_pid_nr(current));
958 		return err;
959 	}
960 
961 	if (opts.privport) {
962 		err = p9_bind_privport(csocket);
963 		if (err < 0) {
964 			pr_err("%s (%d): problem binding to privport\n",
965 			       __func__, task_pid_nr(current));
966 			sock_release(csocket);
967 			return err;
968 		}
969 	}
970 
971 	err = csocket->ops->connect(csocket,
972 				    (struct sockaddr *)&sin_server,
973 				    sizeof(struct sockaddr_in), 0);
974 	if (err < 0) {
975 		pr_err("%s (%d): problem connecting socket to %s\n",
976 		       __func__, task_pid_nr(current), addr);
977 		sock_release(csocket);
978 		return err;
979 	}
980 
981 	return p9_socket_open(client, csocket);
982 }
983 
984 static int
985 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
986 {
987 	int err;
988 	struct socket *csocket;
989 	struct sockaddr_un sun_server;
990 
991 	csocket = NULL;
992 
993 	if (strlen(addr) >= UNIX_PATH_MAX) {
994 		pr_err("%s (%d): address too long: %s\n",
995 		       __func__, task_pid_nr(current), addr);
996 		return -ENAMETOOLONG;
997 	}
998 
999 	sun_server.sun_family = PF_UNIX;
1000 	strcpy(sun_server.sun_path, addr);
1001 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1002 			    SOCK_STREAM, 0, &csocket, 1);
1003 	if (err < 0) {
1004 		pr_err("%s (%d): problem creating socket\n",
1005 		       __func__, task_pid_nr(current));
1006 
1007 		return err;
1008 	}
1009 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1010 			sizeof(struct sockaddr_un) - 1, 0);
1011 	if (err < 0) {
1012 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1013 		       __func__, task_pid_nr(current), addr, err);
1014 		sock_release(csocket);
1015 		return err;
1016 	}
1017 
1018 	return p9_socket_open(client, csocket);
1019 }
1020 
1021 static int
1022 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1023 {
1024 	int err;
1025 	struct p9_fd_opts opts;
1026 
1027 	parse_opts(args, &opts);
1028 	client->trans_opts.fd.rfd = opts.rfd;
1029 	client->trans_opts.fd.wfd = opts.wfd;
1030 
1031 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1032 		pr_err("Insufficient options for proto=fd\n");
1033 		return -ENOPROTOOPT;
1034 	}
1035 
1036 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1037 	if (err < 0)
1038 		return err;
1039 
1040 	p9_conn_create(client);
1041 
1042 	return 0;
1043 }
1044 
1045 static struct p9_trans_module p9_tcp_trans = {
1046 	.name = "tcp",
1047 	.maxsize = MAX_SOCK_BUF,
1048 	.def = 0,
1049 	.create = p9_fd_create_tcp,
1050 	.close = p9_fd_close,
1051 	.request = p9_fd_request,
1052 	.cancel = p9_fd_cancel,
1053 	.cancelled = p9_fd_cancelled,
1054 	.show_options = p9_fd_show_options,
1055 	.owner = THIS_MODULE,
1056 };
1057 
1058 static struct p9_trans_module p9_unix_trans = {
1059 	.name = "unix",
1060 	.maxsize = MAX_SOCK_BUF,
1061 	.def = 0,
1062 	.create = p9_fd_create_unix,
1063 	.close = p9_fd_close,
1064 	.request = p9_fd_request,
1065 	.cancel = p9_fd_cancel,
1066 	.cancelled = p9_fd_cancelled,
1067 	.show_options = p9_fd_show_options,
1068 	.owner = THIS_MODULE,
1069 };
1070 
1071 static struct p9_trans_module p9_fd_trans = {
1072 	.name = "fd",
1073 	.maxsize = MAX_SOCK_BUF,
1074 	.def = 0,
1075 	.create = p9_fd_create,
1076 	.close = p9_fd_close,
1077 	.request = p9_fd_request,
1078 	.cancel = p9_fd_cancel,
1079 	.cancelled = p9_fd_cancelled,
1080 	.show_options = p9_fd_show_options,
1081 	.owner = THIS_MODULE,
1082 };
1083 
1084 /**
1085  * p9_poll_workfn - poll worker thread
1086  * @work: work queue
1087  *
1088  * polls all v9fs transports for new events and queues the appropriate
1089  * work to the work queue
1090  *
1091  */
1092 
1093 static void p9_poll_workfn(struct work_struct *work)
1094 {
1095 	unsigned long flags;
1096 
1097 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1098 
1099 	spin_lock_irqsave(&p9_poll_lock, flags);
1100 	while (!list_empty(&p9_poll_pending_list)) {
1101 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1102 							struct p9_conn,
1103 							poll_pending_link);
1104 		list_del_init(&conn->poll_pending_link);
1105 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1106 
1107 		p9_poll_mux(conn);
1108 
1109 		spin_lock_irqsave(&p9_poll_lock, flags);
1110 	}
1111 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1112 
1113 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1114 }
1115 
1116 int p9_trans_fd_init(void)
1117 {
1118 	v9fs_register_trans(&p9_tcp_trans);
1119 	v9fs_register_trans(&p9_unix_trans);
1120 	v9fs_register_trans(&p9_fd_trans);
1121 
1122 	return 0;
1123 }
1124 
1125 void p9_trans_fd_exit(void)
1126 {
1127 	flush_work(&p9_poll_work);
1128 	v9fs_unregister_trans(&p9_tcp_trans);
1129 	v9fs_unregister_trans(&p9_unix_trans);
1130 	v9fs_unregister_trans(&p9_fd_trans);
1131 }
1132