xref: /openbmc/linux/net/9p/trans_fd.c (revision 4a3fad70)
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 int
232 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt)
233 {
234 	int ret, n;
235 	struct p9_trans_fd *ts = NULL;
236 
237 	if (client && client->status == Connected)
238 		ts = client->trans;
239 
240 	if (!ts)
241 		return -EREMOTEIO;
242 
243 	if (!ts->rd->f_op->poll)
244 		return -EIO;
245 
246 	if (!ts->wr->f_op->poll)
247 		return -EIO;
248 
249 	ret = ts->rd->f_op->poll(ts->rd, pt);
250 	if (ret < 0)
251 		return ret;
252 
253 	if (ts->rd != ts->wr) {
254 		n = ts->wr->f_op->poll(ts->wr, pt);
255 		if (n < 0)
256 			return n;
257 		ret = (ret & ~POLLOUT) | (n & ~POLLIN);
258 	}
259 
260 	return ret;
261 }
262 
263 /**
264  * p9_fd_read- read from a fd
265  * @client: client instance
266  * @v: buffer to receive data into
267  * @len: size of receive buffer
268  *
269  */
270 
271 static int p9_fd_read(struct p9_client *client, void *v, int len)
272 {
273 	int ret;
274 	struct p9_trans_fd *ts = NULL;
275 	loff_t pos;
276 
277 	if (client && client->status != Disconnected)
278 		ts = client->trans;
279 
280 	if (!ts)
281 		return -EREMOTEIO;
282 
283 	if (!(ts->rd->f_flags & O_NONBLOCK))
284 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
285 
286 	pos = ts->rd->f_pos;
287 	ret = kernel_read(ts->rd, v, len, &pos);
288 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
289 		client->status = Disconnected;
290 	return ret;
291 }
292 
293 /**
294  * p9_read_work - called when there is some data to be read from a transport
295  * @work: container of work to be done
296  *
297  */
298 
299 static void p9_read_work(struct work_struct *work)
300 {
301 	int n, err;
302 	struct p9_conn *m;
303 	int status = REQ_STATUS_ERROR;
304 
305 	m = container_of(work, struct p9_conn, rq);
306 
307 	if (m->err < 0)
308 		return;
309 
310 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
311 
312 	if (!m->rc.sdata) {
313 		m->rc.sdata = m->tmp_buf;
314 		m->rc.offset = 0;
315 		m->rc.capacity = 7; /* start by reading header */
316 	}
317 
318 	clear_bit(Rpending, &m->wsched);
319 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
320 		 m, m->rc.offset, m->rc.capacity,
321 		 m->rc.capacity - m->rc.offset);
322 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
323 			 m->rc.capacity - m->rc.offset);
324 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
325 	if (err == -EAGAIN)
326 		goto end_clear;
327 
328 	if (err <= 0)
329 		goto error;
330 
331 	m->rc.offset += err;
332 
333 	/* header read in */
334 	if ((!m->req) && (m->rc.offset == m->rc.capacity)) {
335 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
336 
337 		err = p9_parse_header(&m->rc, NULL, NULL, NULL, 0);
338 		if (err) {
339 			p9_debug(P9_DEBUG_ERROR,
340 				 "error parsing header: %d\n", err);
341 			goto error;
342 		}
343 
344 		if (m->rc.size >= m->client->msize) {
345 			p9_debug(P9_DEBUG_ERROR,
346 				 "requested packet size too big: %d\n",
347 				 m->rc.size);
348 			err = -EIO;
349 			goto error;
350 		}
351 
352 		p9_debug(P9_DEBUG_TRANS,
353 			 "mux %p pkt: size: %d bytes tag: %d\n",
354 			 m, m->rc.size, m->rc.tag);
355 
356 		m->req = p9_tag_lookup(m->client, m->rc.tag);
357 		if (!m->req || (m->req->status != REQ_STATUS_SENT)) {
358 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
359 				 m->rc.tag);
360 			err = -EIO;
361 			goto error;
362 		}
363 
364 		if (m->req->rc == NULL) {
365 			p9_debug(P9_DEBUG_ERROR,
366 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
367 				 m->rc.tag, m->req);
368 			m->req = NULL;
369 			err = -EIO;
370 			goto error;
371 		}
372 		m->rc.sdata = (char *)m->req->rc + sizeof(struct p9_fcall);
373 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
374 		m->rc.capacity = m->rc.size;
375 	}
376 
377 	/* packet is read in
378 	 * not an else because some packets (like clunk) have no payload
379 	 */
380 	if ((m->req) && (m->rc.offset == m->rc.capacity)) {
381 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
382 		spin_lock(&m->client->lock);
383 		if (m->req->status != REQ_STATUS_ERROR)
384 			status = REQ_STATUS_RCVD;
385 		list_del(&m->req->req_list);
386 		spin_unlock(&m->client->lock);
387 		p9_client_cb(m->client, m->req, status);
388 		m->rc.sdata = NULL;
389 		m->rc.offset = 0;
390 		m->rc.capacity = 0;
391 		m->req = NULL;
392 	}
393 
394 end_clear:
395 	clear_bit(Rworksched, &m->wsched);
396 
397 	if (!list_empty(&m->req_list)) {
398 		if (test_and_clear_bit(Rpending, &m->wsched))
399 			n = POLLIN;
400 		else
401 			n = p9_fd_poll(m->client, NULL);
402 
403 		if ((n & POLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
404 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
405 			schedule_work(&m->rq);
406 		}
407 	}
408 
409 	return;
410 error:
411 	p9_conn_cancel(m, err);
412 	clear_bit(Rworksched, &m->wsched);
413 }
414 
415 /**
416  * p9_fd_write - write to a socket
417  * @client: client instance
418  * @v: buffer to send data from
419  * @len: size of send buffer
420  *
421  */
422 
423 static int p9_fd_write(struct p9_client *client, void *v, int len)
424 {
425 	ssize_t ret;
426 	struct p9_trans_fd *ts = NULL;
427 
428 	if (client && client->status != Disconnected)
429 		ts = client->trans;
430 
431 	if (!ts)
432 		return -EREMOTEIO;
433 
434 	if (!(ts->wr->f_flags & O_NONBLOCK))
435 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
436 
437 	ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
438 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
439 		client->status = Disconnected;
440 	return ret;
441 }
442 
443 /**
444  * p9_write_work - called when a transport can send some data
445  * @work: container for work to be done
446  *
447  */
448 
449 static void p9_write_work(struct work_struct *work)
450 {
451 	int n, err;
452 	struct p9_conn *m;
453 	struct p9_req_t *req;
454 
455 	m = container_of(work, struct p9_conn, wq);
456 
457 	if (m->err < 0) {
458 		clear_bit(Wworksched, &m->wsched);
459 		return;
460 	}
461 
462 	if (!m->wsize) {
463 		spin_lock(&m->client->lock);
464 		if (list_empty(&m->unsent_req_list)) {
465 			clear_bit(Wworksched, &m->wsched);
466 			spin_unlock(&m->client->lock);
467 			return;
468 		}
469 
470 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
471 			       req_list);
472 		req->status = REQ_STATUS_SENT;
473 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
474 		list_move_tail(&req->req_list, &m->req_list);
475 
476 		m->wbuf = req->tc->sdata;
477 		m->wsize = req->tc->size;
478 		m->wpos = 0;
479 		spin_unlock(&m->client->lock);
480 	}
481 
482 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
483 		 m, m->wpos, m->wsize);
484 	clear_bit(Wpending, &m->wsched);
485 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
486 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
487 	if (err == -EAGAIN)
488 		goto end_clear;
489 
490 
491 	if (err < 0)
492 		goto error;
493 	else if (err == 0) {
494 		err = -EREMOTEIO;
495 		goto error;
496 	}
497 
498 	m->wpos += err;
499 	if (m->wpos == m->wsize)
500 		m->wpos = m->wsize = 0;
501 
502 end_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 = POLLOUT;
508 		else
509 			n = p9_fd_poll(m->client, NULL);
510 
511 		if ((n & POLLOUT) &&
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 
520 error:
521 	p9_conn_cancel(m, err);
522 	clear_bit(Wworksched, &m->wsched);
523 }
524 
525 static 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 
550 static void
551 p9_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 
582 static void p9_conn_create(struct p9_client *client)
583 {
584 	int 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);
601 	if (n & POLLIN) {
602 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
603 		set_bit(Rpending, &m->wsched);
604 	}
605 
606 	if (n & POLLOUT) {
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 
618 static void p9_poll_mux(struct p9_conn *m)
619 {
620 	int n;
621 
622 	if (m->err < 0)
623 		return;
624 
625 	n = p9_fd_poll(m->client, NULL);
626 	if (n < 0 || n & (POLLERR | POLLHUP | POLLNVAL)) {
627 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
628 		if (n >= 0)
629 			n = -ECONNRESET;
630 		p9_conn_cancel(m, n);
631 	}
632 
633 	if (n & POLLIN) {
634 		set_bit(Rpending, &m->wsched);
635 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
636 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
637 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
638 			schedule_work(&m->rq);
639 		}
640 	}
641 
642 	if (n & POLLOUT) {
643 		set_bit(Wpending, &m->wsched);
644 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
645 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
646 		    !test_and_set_bit(Wworksched, &m->wsched)) {
647 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
648 			schedule_work(&m->wq);
649 		}
650 	}
651 }
652 
653 /**
654  * p9_fd_request - send 9P request
655  * The function can sleep until the request is scheduled for sending.
656  * The function can be interrupted. Return from the function is not
657  * a guarantee that the request is sent successfully.
658  *
659  * @client: client instance
660  * @req: request to be sent
661  *
662  */
663 
664 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
665 {
666 	int n;
667 	struct p9_trans_fd *ts = client->trans;
668 	struct p9_conn *m = &ts->conn;
669 
670 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
671 		 m, current, req->tc, req->tc->id);
672 	if (m->err < 0)
673 		return m->err;
674 
675 	spin_lock(&client->lock);
676 	req->status = REQ_STATUS_UNSENT;
677 	list_add_tail(&req->req_list, &m->unsent_req_list);
678 	spin_unlock(&client->lock);
679 
680 	if (test_and_clear_bit(Wpending, &m->wsched))
681 		n = POLLOUT;
682 	else
683 		n = p9_fd_poll(m->client, NULL);
684 
685 	if (n & POLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
686 		schedule_work(&m->wq);
687 
688 	return 0;
689 }
690 
691 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
692 {
693 	int ret = 1;
694 
695 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
696 
697 	spin_lock(&client->lock);
698 
699 	if (req->status == REQ_STATUS_UNSENT) {
700 		list_del(&req->req_list);
701 		req->status = REQ_STATUS_FLSHD;
702 		ret = 0;
703 	}
704 	spin_unlock(&client->lock);
705 
706 	return ret;
707 }
708 
709 static 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 	/* we haven't received a response for oldreq,
714 	 * remove it from the list.
715 	 */
716 	spin_lock(&client->lock);
717 	list_del(&req->req_list);
718 	spin_unlock(&client->lock);
719 
720 	return 0;
721 }
722 
723 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
724 {
725 	if (clnt->trans_mod == &p9_tcp_trans) {
726 		if (clnt->trans_opts.tcp.port != P9_PORT)
727 			seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
728 	} else if (clnt->trans_mod == &p9_fd_trans) {
729 		if (clnt->trans_opts.fd.rfd != ~0)
730 			seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
731 		if (clnt->trans_opts.fd.wfd != ~0)
732 			seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
733 	}
734 	return 0;
735 }
736 
737 /**
738  * parse_opts - parse mount options into p9_fd_opts structure
739  * @params: options string passed from mount
740  * @opts: fd transport-specific structure to parse options into
741  *
742  * Returns 0 upon success, -ERRNO upon failure
743  */
744 
745 static int parse_opts(char *params, struct p9_fd_opts *opts)
746 {
747 	char *p;
748 	substring_t args[MAX_OPT_ARGS];
749 	int option;
750 	char *options, *tmp_options;
751 
752 	opts->port = P9_PORT;
753 	opts->rfd = ~0;
754 	opts->wfd = ~0;
755 	opts->privport = false;
756 
757 	if (!params)
758 		return 0;
759 
760 	tmp_options = kstrdup(params, GFP_KERNEL);
761 	if (!tmp_options) {
762 		p9_debug(P9_DEBUG_ERROR,
763 			 "failed to allocate copy of option string\n");
764 		return -ENOMEM;
765 	}
766 	options = tmp_options;
767 
768 	while ((p = strsep(&options, ",")) != NULL) {
769 		int token;
770 		int r;
771 		if (!*p)
772 			continue;
773 		token = match_token(p, tokens, args);
774 		if ((token != Opt_err) && (token != Opt_privport)) {
775 			r = match_int(&args[0], &option);
776 			if (r < 0) {
777 				p9_debug(P9_DEBUG_ERROR,
778 					 "integer field, but no integer?\n");
779 				continue;
780 			}
781 		}
782 		switch (token) {
783 		case Opt_port:
784 			opts->port = option;
785 			break;
786 		case Opt_rfdno:
787 			opts->rfd = option;
788 			break;
789 		case Opt_wfdno:
790 			opts->wfd = option;
791 			break;
792 		case Opt_privport:
793 			opts->privport = true;
794 			break;
795 		default:
796 			continue;
797 		}
798 	}
799 
800 	kfree(tmp_options);
801 	return 0;
802 }
803 
804 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
805 {
806 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
807 					   GFP_KERNEL);
808 	if (!ts)
809 		return -ENOMEM;
810 
811 	ts->rd = fget(rfd);
812 	ts->wr = fget(wfd);
813 	if (!ts->rd || !ts->wr) {
814 		if (ts->rd)
815 			fput(ts->rd);
816 		if (ts->wr)
817 			fput(ts->wr);
818 		kfree(ts);
819 		return -EIO;
820 	}
821 
822 	client->trans = ts;
823 	client->status = Connected;
824 
825 	return 0;
826 }
827 
828 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
829 {
830 	struct p9_trans_fd *p;
831 	struct file *file;
832 
833 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
834 	if (!p)
835 		return -ENOMEM;
836 
837 	csocket->sk->sk_allocation = GFP_NOIO;
838 	file = sock_alloc_file(csocket, 0, NULL);
839 	if (IS_ERR(file)) {
840 		pr_err("%s (%d): failed to map fd\n",
841 		       __func__, task_pid_nr(current));
842 		kfree(p);
843 		return PTR_ERR(file);
844 	}
845 
846 	get_file(file);
847 	p->wr = p->rd = file;
848 	client->trans = p;
849 	client->status = Connected;
850 
851 	p->rd->f_flags |= O_NONBLOCK;
852 
853 	p9_conn_create(client);
854 	return 0;
855 }
856 
857 /**
858  * p9_mux_destroy - cancels all pending requests of mux
859  * @m: mux to destroy
860  *
861  */
862 
863 static void p9_conn_destroy(struct p9_conn *m)
864 {
865 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
866 		 m, m->mux_list.prev, m->mux_list.next);
867 
868 	p9_mux_poll_stop(m);
869 	cancel_work_sync(&m->rq);
870 	cancel_work_sync(&m->wq);
871 
872 	p9_conn_cancel(m, -ECONNRESET);
873 
874 	m->client = NULL;
875 }
876 
877 /**
878  * p9_fd_close - shutdown file descriptor transport
879  * @client: client instance
880  *
881  */
882 
883 static void p9_fd_close(struct p9_client *client)
884 {
885 	struct p9_trans_fd *ts;
886 
887 	if (!client)
888 		return;
889 
890 	ts = client->trans;
891 	if (!ts)
892 		return;
893 
894 	client->status = Disconnected;
895 
896 	p9_conn_destroy(&ts->conn);
897 
898 	if (ts->rd)
899 		fput(ts->rd);
900 	if (ts->wr)
901 		fput(ts->wr);
902 
903 	kfree(ts);
904 }
905 
906 /*
907  * stolen from NFS - maybe should be made a generic function?
908  */
909 static inline int valid_ipaddr4(const char *buf)
910 {
911 	int rc, count, in[4];
912 
913 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
914 	if (rc != 4)
915 		return -EINVAL;
916 	for (count = 0; count < 4; count++) {
917 		if (in[count] > 255)
918 			return -EINVAL;
919 	}
920 	return 0;
921 }
922 
923 static int p9_bind_privport(struct socket *sock)
924 {
925 	struct sockaddr_in cl;
926 	int port, err = -EINVAL;
927 
928 	memset(&cl, 0, sizeof(cl));
929 	cl.sin_family = AF_INET;
930 	cl.sin_addr.s_addr = INADDR_ANY;
931 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
932 		cl.sin_port = htons((ushort)port);
933 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
934 		if (err != -EADDRINUSE)
935 			break;
936 	}
937 	return err;
938 }
939 
940 
941 static int
942 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
943 {
944 	int err;
945 	struct socket *csocket;
946 	struct sockaddr_in sin_server;
947 	struct p9_fd_opts opts;
948 
949 	err = parse_opts(args, &opts);
950 	if (err < 0)
951 		return err;
952 
953 	if (valid_ipaddr4(addr) < 0)
954 		return -EINVAL;
955 
956 	csocket = NULL;
957 
958 	client->trans_opts.tcp.port = opts.port;
959 	client->trans_opts.tcp.privport = opts.privport;
960 	sin_server.sin_family = AF_INET;
961 	sin_server.sin_addr.s_addr = in_aton(addr);
962 	sin_server.sin_port = htons(opts.port);
963 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
964 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
965 	if (err) {
966 		pr_err("%s (%d): problem creating socket\n",
967 		       __func__, task_pid_nr(current));
968 		return err;
969 	}
970 
971 	if (opts.privport) {
972 		err = p9_bind_privport(csocket);
973 		if (err < 0) {
974 			pr_err("%s (%d): problem binding to privport\n",
975 			       __func__, task_pid_nr(current));
976 			sock_release(csocket);
977 			return err;
978 		}
979 	}
980 
981 	err = csocket->ops->connect(csocket,
982 				    (struct sockaddr *)&sin_server,
983 				    sizeof(struct sockaddr_in), 0);
984 	if (err < 0) {
985 		pr_err("%s (%d): problem connecting socket to %s\n",
986 		       __func__, task_pid_nr(current), addr);
987 		sock_release(csocket);
988 		return err;
989 	}
990 
991 	return p9_socket_open(client, csocket);
992 }
993 
994 static int
995 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
996 {
997 	int err;
998 	struct socket *csocket;
999 	struct sockaddr_un sun_server;
1000 
1001 	csocket = NULL;
1002 
1003 	if (strlen(addr) >= UNIX_PATH_MAX) {
1004 		pr_err("%s (%d): address too long: %s\n",
1005 		       __func__, task_pid_nr(current), addr);
1006 		return -ENAMETOOLONG;
1007 	}
1008 
1009 	sun_server.sun_family = PF_UNIX;
1010 	strcpy(sun_server.sun_path, addr);
1011 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1012 			    SOCK_STREAM, 0, &csocket, 1);
1013 	if (err < 0) {
1014 		pr_err("%s (%d): problem creating socket\n",
1015 		       __func__, task_pid_nr(current));
1016 
1017 		return err;
1018 	}
1019 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1020 			sizeof(struct sockaddr_un) - 1, 0);
1021 	if (err < 0) {
1022 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1023 		       __func__, task_pid_nr(current), addr, err);
1024 		sock_release(csocket);
1025 		return err;
1026 	}
1027 
1028 	return p9_socket_open(client, csocket);
1029 }
1030 
1031 static int
1032 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1033 {
1034 	int err;
1035 	struct p9_fd_opts opts;
1036 
1037 	parse_opts(args, &opts);
1038 	client->trans_opts.fd.rfd = opts.rfd;
1039 	client->trans_opts.fd.wfd = opts.wfd;
1040 
1041 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1042 		pr_err("Insufficient options for proto=fd\n");
1043 		return -ENOPROTOOPT;
1044 	}
1045 
1046 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1047 	if (err < 0)
1048 		return err;
1049 
1050 	p9_conn_create(client);
1051 
1052 	return 0;
1053 }
1054 
1055 static struct p9_trans_module p9_tcp_trans = {
1056 	.name = "tcp",
1057 	.maxsize = MAX_SOCK_BUF,
1058 	.def = 0,
1059 	.create = p9_fd_create_tcp,
1060 	.close = p9_fd_close,
1061 	.request = p9_fd_request,
1062 	.cancel = p9_fd_cancel,
1063 	.cancelled = p9_fd_cancelled,
1064 	.show_options = p9_fd_show_options,
1065 	.owner = THIS_MODULE,
1066 };
1067 
1068 static struct p9_trans_module p9_unix_trans = {
1069 	.name = "unix",
1070 	.maxsize = MAX_SOCK_BUF,
1071 	.def = 0,
1072 	.create = p9_fd_create_unix,
1073 	.close = p9_fd_close,
1074 	.request = p9_fd_request,
1075 	.cancel = p9_fd_cancel,
1076 	.cancelled = p9_fd_cancelled,
1077 	.show_options = p9_fd_show_options,
1078 	.owner = THIS_MODULE,
1079 };
1080 
1081 static struct p9_trans_module p9_fd_trans = {
1082 	.name = "fd",
1083 	.maxsize = MAX_SOCK_BUF,
1084 	.def = 0,
1085 	.create = p9_fd_create,
1086 	.close = p9_fd_close,
1087 	.request = p9_fd_request,
1088 	.cancel = p9_fd_cancel,
1089 	.cancelled = p9_fd_cancelled,
1090 	.show_options = p9_fd_show_options,
1091 	.owner = THIS_MODULE,
1092 };
1093 
1094 /**
1095  * p9_poll_proc - poll worker thread
1096  * @a: thread state and arguments
1097  *
1098  * polls all v9fs transports for new events and queues the appropriate
1099  * work to the work queue
1100  *
1101  */
1102 
1103 static void p9_poll_workfn(struct work_struct *work)
1104 {
1105 	unsigned long flags;
1106 
1107 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1108 
1109 	spin_lock_irqsave(&p9_poll_lock, flags);
1110 	while (!list_empty(&p9_poll_pending_list)) {
1111 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1112 							struct p9_conn,
1113 							poll_pending_link);
1114 		list_del_init(&conn->poll_pending_link);
1115 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1116 
1117 		p9_poll_mux(conn);
1118 
1119 		spin_lock_irqsave(&p9_poll_lock, flags);
1120 	}
1121 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1122 
1123 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1124 }
1125 
1126 int p9_trans_fd_init(void)
1127 {
1128 	v9fs_register_trans(&p9_tcp_trans);
1129 	v9fs_register_trans(&p9_unix_trans);
1130 	v9fs_register_trans(&p9_fd_trans);
1131 
1132 	return 0;
1133 }
1134 
1135 void p9_trans_fd_exit(void)
1136 {
1137 	flush_work(&p9_poll_work);
1138 	v9fs_unregister_trans(&p9_tcp_trans);
1139 	v9fs_unregister_trans(&p9_unix_trans);
1140 	v9fs_unregister_trans(&p9_fd_trans);
1141 }
1142