xref: /linux/net/9p/trans_fd.c (revision bd628c1bed7902ec1f24ba0fe70758949146abbe)
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 *rreq;
135 	struct p9_req_t *wreq;
136 	char tmp_buf[7];
137 	struct p9_fcall rc;
138 	int wpos;
139 	int wsize;
140 	char *wbuf;
141 	struct list_head poll_pending_link;
142 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
143 	poll_table pt;
144 	struct work_struct rq;
145 	struct work_struct wq;
146 	unsigned long wsched;
147 };
148 
149 /**
150  * struct p9_trans_fd - transport state
151  * @rd: reference to file to read from
152  * @wr: reference of file to write to
153  * @conn: connection state reference
154  *
155  */
156 
157 struct p9_trans_fd {
158 	struct file *rd;
159 	struct file *wr;
160 	struct p9_conn conn;
161 };
162 
163 static void p9_poll_workfn(struct work_struct *work);
164 
165 static DEFINE_SPINLOCK(p9_poll_lock);
166 static LIST_HEAD(p9_poll_pending_list);
167 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
168 
169 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
170 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
171 
172 static void p9_mux_poll_stop(struct p9_conn *m)
173 {
174 	unsigned long flags;
175 	int i;
176 
177 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
178 		struct p9_poll_wait *pwait = &m->poll_wait[i];
179 
180 		if (pwait->wait_addr) {
181 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
182 			pwait->wait_addr = NULL;
183 		}
184 	}
185 
186 	spin_lock_irqsave(&p9_poll_lock, flags);
187 	list_del_init(&m->poll_pending_link);
188 	spin_unlock_irqrestore(&p9_poll_lock, flags);
189 
190 	flush_work(&p9_poll_work);
191 }
192 
193 /**
194  * p9_conn_cancel - cancel all pending requests with error
195  * @m: mux data
196  * @err: error code
197  *
198  */
199 
200 static void p9_conn_cancel(struct p9_conn *m, int err)
201 {
202 	struct p9_req_t *req, *rtmp;
203 	LIST_HEAD(cancel_list);
204 
205 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
206 
207 	spin_lock(&m->client->lock);
208 
209 	if (m->err) {
210 		spin_unlock(&m->client->lock);
211 		return;
212 	}
213 
214 	m->err = err;
215 
216 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
217 		list_move(&req->req_list, &cancel_list);
218 	}
219 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
220 		list_move(&req->req_list, &cancel_list);
221 	}
222 
223 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
224 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
225 		list_del(&req->req_list);
226 		if (!req->t_err)
227 			req->t_err = err;
228 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
229 	}
230 	spin_unlock(&m->client->lock);
231 }
232 
233 static __poll_t
234 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
235 {
236 	__poll_t ret;
237 	struct p9_trans_fd *ts = NULL;
238 
239 	if (client && client->status == Connected)
240 		ts = client->trans;
241 
242 	if (!ts) {
243 		if (err)
244 			*err = -EREMOTEIO;
245 		return EPOLLERR;
246 	}
247 
248 	ret = vfs_poll(ts->rd, pt);
249 	if (ts->rd != ts->wr)
250 		ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN);
251 	return ret;
252 }
253 
254 /**
255  * p9_fd_read- read from a fd
256  * @client: client instance
257  * @v: buffer to receive data into
258  * @len: size of receive buffer
259  *
260  */
261 
262 static int p9_fd_read(struct p9_client *client, void *v, int len)
263 {
264 	int ret;
265 	struct p9_trans_fd *ts = NULL;
266 	loff_t pos;
267 
268 	if (client && client->status != Disconnected)
269 		ts = client->trans;
270 
271 	if (!ts)
272 		return -EREMOTEIO;
273 
274 	if (!(ts->rd->f_flags & O_NONBLOCK))
275 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
276 
277 	pos = ts->rd->f_pos;
278 	ret = kernel_read(ts->rd, v, len, &pos);
279 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
280 		client->status = Disconnected;
281 	return ret;
282 }
283 
284 /**
285  * p9_read_work - called when there is some data to be read from a transport
286  * @work: container of work to be done
287  *
288  */
289 
290 static void p9_read_work(struct work_struct *work)
291 {
292 	__poll_t n;
293 	int err;
294 	struct p9_conn *m;
295 
296 	m = container_of(work, struct p9_conn, rq);
297 
298 	if (m->err < 0)
299 		return;
300 
301 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
302 
303 	if (!m->rc.sdata) {
304 		m->rc.sdata = m->tmp_buf;
305 		m->rc.offset = 0;
306 		m->rc.capacity = 7; /* start by reading header */
307 	}
308 
309 	clear_bit(Rpending, &m->wsched);
310 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
311 		 m, m->rc.offset, m->rc.capacity,
312 		 m->rc.capacity - m->rc.offset);
313 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
314 			 m->rc.capacity - m->rc.offset);
315 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
316 	if (err == -EAGAIN)
317 		goto end_clear;
318 
319 	if (err <= 0)
320 		goto error;
321 
322 	m->rc.offset += err;
323 
324 	/* header read in */
325 	if ((!m->rreq) && (m->rc.offset == m->rc.capacity)) {
326 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
327 
328 		/* Header size */
329 		m->rc.size = 7;
330 		err = p9_parse_header(&m->rc, &m->rc.size, NULL, NULL, 0);
331 		if (err) {
332 			p9_debug(P9_DEBUG_ERROR,
333 				 "error parsing header: %d\n", err);
334 			goto error;
335 		}
336 
337 		if (m->rc.size >= m->client->msize) {
338 			p9_debug(P9_DEBUG_ERROR,
339 				 "requested packet size too big: %d\n",
340 				 m->rc.size);
341 			err = -EIO;
342 			goto error;
343 		}
344 
345 		p9_debug(P9_DEBUG_TRANS,
346 			 "mux %p pkt: size: %d bytes tag: %d\n",
347 			 m, m->rc.size, m->rc.tag);
348 
349 		m->rreq = p9_tag_lookup(m->client, m->rc.tag);
350 		if (!m->rreq || (m->rreq->status != REQ_STATUS_SENT)) {
351 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
352 				 m->rc.tag);
353 			err = -EIO;
354 			goto error;
355 		}
356 
357 		if (!m->rreq->rc.sdata) {
358 			p9_debug(P9_DEBUG_ERROR,
359 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
360 				 m->rc.tag, m->rreq);
361 			m->rreq = NULL;
362 			err = -EIO;
363 			goto error;
364 		}
365 		m->rc.sdata = m->rreq->rc.sdata;
366 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
367 		m->rc.capacity = m->rc.size;
368 	}
369 
370 	/* packet is read in
371 	 * not an else because some packets (like clunk) have no payload
372 	 */
373 	if ((m->rreq) && (m->rc.offset == m->rc.capacity)) {
374 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
375 		m->rreq->rc.size = m->rc.offset;
376 		spin_lock(&m->client->lock);
377 		if (m->rreq->status == REQ_STATUS_SENT) {
378 			list_del(&m->rreq->req_list);
379 			p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD);
380 		} else {
381 			spin_unlock(&m->client->lock);
382 			p9_debug(P9_DEBUG_ERROR,
383 				 "Request tag %d errored out while we were reading the reply\n",
384 				 m->rc.tag);
385 			err = -EIO;
386 			goto error;
387 		}
388 		spin_unlock(&m->client->lock);
389 		m->rc.sdata = NULL;
390 		m->rc.offset = 0;
391 		m->rc.capacity = 0;
392 		p9_req_put(m->rreq);
393 		m->rreq = NULL;
394 	}
395 
396 end_clear:
397 	clear_bit(Rworksched, &m->wsched);
398 
399 	if (!list_empty(&m->req_list)) {
400 		if (test_and_clear_bit(Rpending, &m->wsched))
401 			n = EPOLLIN;
402 		else
403 			n = p9_fd_poll(m->client, NULL, NULL);
404 
405 		if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
406 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
407 			schedule_work(&m->rq);
408 		}
409 	}
410 
411 	return;
412 error:
413 	p9_conn_cancel(m, err);
414 	clear_bit(Rworksched, &m->wsched);
415 }
416 
417 /**
418  * p9_fd_write - write to a socket
419  * @client: client instance
420  * @v: buffer to send data from
421  * @len: size of send buffer
422  *
423  */
424 
425 static int p9_fd_write(struct p9_client *client, void *v, int len)
426 {
427 	ssize_t ret;
428 	struct p9_trans_fd *ts = NULL;
429 
430 	if (client && client->status != Disconnected)
431 		ts = client->trans;
432 
433 	if (!ts)
434 		return -EREMOTEIO;
435 
436 	if (!(ts->wr->f_flags & O_NONBLOCK))
437 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
438 
439 	ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
440 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
441 		client->status = Disconnected;
442 	return ret;
443 }
444 
445 /**
446  * p9_write_work - called when a transport can send some data
447  * @work: container for work to be done
448  *
449  */
450 
451 static void p9_write_work(struct work_struct *work)
452 {
453 	__poll_t n;
454 	int err;
455 	struct p9_conn *m;
456 	struct p9_req_t *req;
457 
458 	m = container_of(work, struct p9_conn, wq);
459 
460 	if (m->err < 0) {
461 		clear_bit(Wworksched, &m->wsched);
462 		return;
463 	}
464 
465 	if (!m->wsize) {
466 		spin_lock(&m->client->lock);
467 		if (list_empty(&m->unsent_req_list)) {
468 			clear_bit(Wworksched, &m->wsched);
469 			spin_unlock(&m->client->lock);
470 			return;
471 		}
472 
473 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
474 			       req_list);
475 		req->status = REQ_STATUS_SENT;
476 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
477 		list_move_tail(&req->req_list, &m->req_list);
478 
479 		m->wbuf = req->tc.sdata;
480 		m->wsize = req->tc.size;
481 		m->wpos = 0;
482 		p9_req_get(req);
483 		m->wreq = req;
484 		spin_unlock(&m->client->lock);
485 	}
486 
487 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
488 		 m, m->wpos, m->wsize);
489 	clear_bit(Wpending, &m->wsched);
490 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
491 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
492 	if (err == -EAGAIN)
493 		goto end_clear;
494 
495 
496 	if (err < 0)
497 		goto error;
498 	else if (err == 0) {
499 		err = -EREMOTEIO;
500 		goto error;
501 	}
502 
503 	m->wpos += err;
504 	if (m->wpos == m->wsize) {
505 		m->wpos = m->wsize = 0;
506 		p9_req_put(m->wreq);
507 		m->wreq = NULL;
508 	}
509 
510 end_clear:
511 	clear_bit(Wworksched, &m->wsched);
512 
513 	if (m->wsize || !list_empty(&m->unsent_req_list)) {
514 		if (test_and_clear_bit(Wpending, &m->wsched))
515 			n = EPOLLOUT;
516 		else
517 			n = p9_fd_poll(m->client, NULL, NULL);
518 
519 		if ((n & EPOLLOUT) &&
520 		   !test_and_set_bit(Wworksched, &m->wsched)) {
521 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
522 			schedule_work(&m->wq);
523 		}
524 	}
525 
526 	return;
527 
528 error:
529 	p9_conn_cancel(m, err);
530 	clear_bit(Wworksched, &m->wsched);
531 }
532 
533 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
534 {
535 	struct p9_poll_wait *pwait =
536 		container_of(wait, struct p9_poll_wait, wait);
537 	struct p9_conn *m = pwait->conn;
538 	unsigned long flags;
539 
540 	spin_lock_irqsave(&p9_poll_lock, flags);
541 	if (list_empty(&m->poll_pending_link))
542 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
543 	spin_unlock_irqrestore(&p9_poll_lock, flags);
544 
545 	schedule_work(&p9_poll_work);
546 	return 1;
547 }
548 
549 /**
550  * p9_pollwait - add poll task to the wait queue
551  * @filp: file pointer being polled
552  * @wait_address: wait_q to block on
553  * @p: poll state
554  *
555  * called by files poll operation to add v9fs-poll task to files wait queue
556  */
557 
558 static void
559 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
560 {
561 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
562 	struct p9_poll_wait *pwait = NULL;
563 	int i;
564 
565 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
566 		if (m->poll_wait[i].wait_addr == NULL) {
567 			pwait = &m->poll_wait[i];
568 			break;
569 		}
570 	}
571 
572 	if (!pwait) {
573 		p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
574 		return;
575 	}
576 
577 	pwait->conn = m;
578 	pwait->wait_addr = wait_address;
579 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
580 	add_wait_queue(wait_address, &pwait->wait);
581 }
582 
583 /**
584  * p9_conn_create - initialize the per-session mux data
585  * @client: client instance
586  *
587  * Note: Creates the polling task if this is the first session.
588  */
589 
590 static void p9_conn_create(struct p9_client *client)
591 {
592 	__poll_t n;
593 	struct p9_trans_fd *ts = client->trans;
594 	struct p9_conn *m = &ts->conn;
595 
596 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
597 
598 	INIT_LIST_HEAD(&m->mux_list);
599 	m->client = client;
600 
601 	INIT_LIST_HEAD(&m->req_list);
602 	INIT_LIST_HEAD(&m->unsent_req_list);
603 	INIT_WORK(&m->rq, p9_read_work);
604 	INIT_WORK(&m->wq, p9_write_work);
605 	INIT_LIST_HEAD(&m->poll_pending_link);
606 	init_poll_funcptr(&m->pt, p9_pollwait);
607 
608 	n = p9_fd_poll(client, &m->pt, NULL);
609 	if (n & EPOLLIN) {
610 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
611 		set_bit(Rpending, &m->wsched);
612 	}
613 
614 	if (n & EPOLLOUT) {
615 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
616 		set_bit(Wpending, &m->wsched);
617 	}
618 }
619 
620 /**
621  * p9_poll_mux - polls a mux and schedules read or write works if necessary
622  * @m: connection to poll
623  *
624  */
625 
626 static void p9_poll_mux(struct p9_conn *m)
627 {
628 	__poll_t n;
629 	int err = -ECONNRESET;
630 
631 	if (m->err < 0)
632 		return;
633 
634 	n = p9_fd_poll(m->client, NULL, &err);
635 	if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
636 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
637 		p9_conn_cancel(m, err);
638 	}
639 
640 	if (n & EPOLLIN) {
641 		set_bit(Rpending, &m->wsched);
642 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
643 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
644 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
645 			schedule_work(&m->rq);
646 		}
647 	}
648 
649 	if (n & EPOLLOUT) {
650 		set_bit(Wpending, &m->wsched);
651 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
652 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
653 		    !test_and_set_bit(Wworksched, &m->wsched)) {
654 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
655 			schedule_work(&m->wq);
656 		}
657 	}
658 }
659 
660 /**
661  * p9_fd_request - send 9P request
662  * The function can sleep until the request is scheduled for sending.
663  * The function can be interrupted. Return from the function is not
664  * a guarantee that the request is sent successfully.
665  *
666  * @client: client instance
667  * @req: request to be sent
668  *
669  */
670 
671 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
672 {
673 	__poll_t n;
674 	struct p9_trans_fd *ts = client->trans;
675 	struct p9_conn *m = &ts->conn;
676 
677 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
678 		 m, current, &req->tc, req->tc.id);
679 	if (m->err < 0)
680 		return m->err;
681 
682 	spin_lock(&client->lock);
683 	req->status = REQ_STATUS_UNSENT;
684 	list_add_tail(&req->req_list, &m->unsent_req_list);
685 	spin_unlock(&client->lock);
686 
687 	if (test_and_clear_bit(Wpending, &m->wsched))
688 		n = EPOLLOUT;
689 	else
690 		n = p9_fd_poll(m->client, NULL, NULL);
691 
692 	if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
693 		schedule_work(&m->wq);
694 
695 	return 0;
696 }
697 
698 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
699 {
700 	int ret = 1;
701 
702 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
703 
704 	spin_lock(&client->lock);
705 
706 	if (req->status == REQ_STATUS_UNSENT) {
707 		list_del(&req->req_list);
708 		req->status = REQ_STATUS_FLSHD;
709 		p9_req_put(req);
710 		ret = 0;
711 	}
712 	spin_unlock(&client->lock);
713 
714 	return ret;
715 }
716 
717 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
718 {
719 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
720 
721 	/* we haven't received a response for oldreq,
722 	 * remove it from the list.
723 	 */
724 	spin_lock(&client->lock);
725 	list_del(&req->req_list);
726 	spin_unlock(&client->lock);
727 	p9_req_put(req);
728 
729 	return 0;
730 }
731 
732 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
733 {
734 	if (clnt->trans_mod == &p9_tcp_trans) {
735 		if (clnt->trans_opts.tcp.port != P9_PORT)
736 			seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
737 	} else if (clnt->trans_mod == &p9_fd_trans) {
738 		if (clnt->trans_opts.fd.rfd != ~0)
739 			seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
740 		if (clnt->trans_opts.fd.wfd != ~0)
741 			seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
742 	}
743 	return 0;
744 }
745 
746 /**
747  * parse_opts - parse mount options into p9_fd_opts structure
748  * @params: options string passed from mount
749  * @opts: fd transport-specific structure to parse options into
750  *
751  * Returns 0 upon success, -ERRNO upon failure
752  */
753 
754 static int parse_opts(char *params, struct p9_fd_opts *opts)
755 {
756 	char *p;
757 	substring_t args[MAX_OPT_ARGS];
758 	int option;
759 	char *options, *tmp_options;
760 
761 	opts->port = P9_PORT;
762 	opts->rfd = ~0;
763 	opts->wfd = ~0;
764 	opts->privport = false;
765 
766 	if (!params)
767 		return 0;
768 
769 	tmp_options = kstrdup(params, GFP_KERNEL);
770 	if (!tmp_options) {
771 		p9_debug(P9_DEBUG_ERROR,
772 			 "failed to allocate copy of option string\n");
773 		return -ENOMEM;
774 	}
775 	options = tmp_options;
776 
777 	while ((p = strsep(&options, ",")) != NULL) {
778 		int token;
779 		int r;
780 		if (!*p)
781 			continue;
782 		token = match_token(p, tokens, args);
783 		if ((token != Opt_err) && (token != Opt_privport)) {
784 			r = match_int(&args[0], &option);
785 			if (r < 0) {
786 				p9_debug(P9_DEBUG_ERROR,
787 					 "integer field, but no integer?\n");
788 				continue;
789 			}
790 		}
791 		switch (token) {
792 		case Opt_port:
793 			opts->port = option;
794 			break;
795 		case Opt_rfdno:
796 			opts->rfd = option;
797 			break;
798 		case Opt_wfdno:
799 			opts->wfd = option;
800 			break;
801 		case Opt_privport:
802 			opts->privport = true;
803 			break;
804 		default:
805 			continue;
806 		}
807 	}
808 
809 	kfree(tmp_options);
810 	return 0;
811 }
812 
813 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
814 {
815 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
816 					   GFP_KERNEL);
817 	if (!ts)
818 		return -ENOMEM;
819 
820 	ts->rd = fget(rfd);
821 	ts->wr = fget(wfd);
822 	if (!ts->rd || !ts->wr) {
823 		if (ts->rd)
824 			fput(ts->rd);
825 		if (ts->wr)
826 			fput(ts->wr);
827 		kfree(ts);
828 		return -EIO;
829 	}
830 
831 	client->trans = ts;
832 	client->status = Connected;
833 
834 	return 0;
835 }
836 
837 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
838 {
839 	struct p9_trans_fd *p;
840 	struct file *file;
841 
842 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
843 	if (!p)
844 		return -ENOMEM;
845 
846 	csocket->sk->sk_allocation = GFP_NOIO;
847 	file = sock_alloc_file(csocket, 0, NULL);
848 	if (IS_ERR(file)) {
849 		pr_err("%s (%d): failed to map fd\n",
850 		       __func__, task_pid_nr(current));
851 		kfree(p);
852 		return PTR_ERR(file);
853 	}
854 
855 	get_file(file);
856 	p->wr = p->rd = file;
857 	client->trans = p;
858 	client->status = Connected;
859 
860 	p->rd->f_flags |= O_NONBLOCK;
861 
862 	p9_conn_create(client);
863 	return 0;
864 }
865 
866 /**
867  * p9_mux_destroy - cancels all pending requests of mux
868  * @m: mux to destroy
869  *
870  */
871 
872 static void p9_conn_destroy(struct p9_conn *m)
873 {
874 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
875 		 m, m->mux_list.prev, m->mux_list.next);
876 
877 	p9_mux_poll_stop(m);
878 	cancel_work_sync(&m->rq);
879 	if (m->rreq) {
880 		p9_req_put(m->rreq);
881 		m->rreq = NULL;
882 	}
883 	cancel_work_sync(&m->wq);
884 	if (m->wreq) {
885 		p9_req_put(m->wreq);
886 		m->wreq = NULL;
887 	}
888 
889 	p9_conn_cancel(m, -ECONNRESET);
890 
891 	m->client = NULL;
892 }
893 
894 /**
895  * p9_fd_close - shutdown file descriptor transport
896  * @client: client instance
897  *
898  */
899 
900 static void p9_fd_close(struct p9_client *client)
901 {
902 	struct p9_trans_fd *ts;
903 
904 	if (!client)
905 		return;
906 
907 	ts = client->trans;
908 	if (!ts)
909 		return;
910 
911 	client->status = Disconnected;
912 
913 	p9_conn_destroy(&ts->conn);
914 
915 	if (ts->rd)
916 		fput(ts->rd);
917 	if (ts->wr)
918 		fput(ts->wr);
919 
920 	kfree(ts);
921 }
922 
923 /*
924  * stolen from NFS - maybe should be made a generic function?
925  */
926 static inline int valid_ipaddr4(const char *buf)
927 {
928 	int rc, count, in[4];
929 
930 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
931 	if (rc != 4)
932 		return -EINVAL;
933 	for (count = 0; count < 4; count++) {
934 		if (in[count] > 255)
935 			return -EINVAL;
936 	}
937 	return 0;
938 }
939 
940 static int p9_bind_privport(struct socket *sock)
941 {
942 	struct sockaddr_in cl;
943 	int port, err = -EINVAL;
944 
945 	memset(&cl, 0, sizeof(cl));
946 	cl.sin_family = AF_INET;
947 	cl.sin_addr.s_addr = INADDR_ANY;
948 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
949 		cl.sin_port = htons((ushort)port);
950 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
951 		if (err != -EADDRINUSE)
952 			break;
953 	}
954 	return err;
955 }
956 
957 
958 static int
959 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
960 {
961 	int err;
962 	struct socket *csocket;
963 	struct sockaddr_in sin_server;
964 	struct p9_fd_opts opts;
965 
966 	err = parse_opts(args, &opts);
967 	if (err < 0)
968 		return err;
969 
970 	if (addr == NULL || valid_ipaddr4(addr) < 0)
971 		return -EINVAL;
972 
973 	csocket = NULL;
974 
975 	client->trans_opts.tcp.port = opts.port;
976 	client->trans_opts.tcp.privport = opts.privport;
977 	sin_server.sin_family = AF_INET;
978 	sin_server.sin_addr.s_addr = in_aton(addr);
979 	sin_server.sin_port = htons(opts.port);
980 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
981 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
982 	if (err) {
983 		pr_err("%s (%d): problem creating socket\n",
984 		       __func__, task_pid_nr(current));
985 		return err;
986 	}
987 
988 	if (opts.privport) {
989 		err = p9_bind_privport(csocket);
990 		if (err < 0) {
991 			pr_err("%s (%d): problem binding to privport\n",
992 			       __func__, task_pid_nr(current));
993 			sock_release(csocket);
994 			return err;
995 		}
996 	}
997 
998 	err = csocket->ops->connect(csocket,
999 				    (struct sockaddr *)&sin_server,
1000 				    sizeof(struct sockaddr_in), 0);
1001 	if (err < 0) {
1002 		pr_err("%s (%d): problem connecting socket to %s\n",
1003 		       __func__, task_pid_nr(current), addr);
1004 		sock_release(csocket);
1005 		return err;
1006 	}
1007 
1008 	return p9_socket_open(client, csocket);
1009 }
1010 
1011 static int
1012 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
1013 {
1014 	int err;
1015 	struct socket *csocket;
1016 	struct sockaddr_un sun_server;
1017 
1018 	csocket = NULL;
1019 
1020 	if (addr == NULL)
1021 		return -EINVAL;
1022 
1023 	if (strlen(addr) >= UNIX_PATH_MAX) {
1024 		pr_err("%s (%d): address too long: %s\n",
1025 		       __func__, task_pid_nr(current), addr);
1026 		return -ENAMETOOLONG;
1027 	}
1028 
1029 	sun_server.sun_family = PF_UNIX;
1030 	strcpy(sun_server.sun_path, addr);
1031 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1032 			    SOCK_STREAM, 0, &csocket, 1);
1033 	if (err < 0) {
1034 		pr_err("%s (%d): problem creating socket\n",
1035 		       __func__, task_pid_nr(current));
1036 
1037 		return err;
1038 	}
1039 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1040 			sizeof(struct sockaddr_un) - 1, 0);
1041 	if (err < 0) {
1042 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1043 		       __func__, task_pid_nr(current), addr, err);
1044 		sock_release(csocket);
1045 		return err;
1046 	}
1047 
1048 	return p9_socket_open(client, csocket);
1049 }
1050 
1051 static int
1052 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1053 {
1054 	int err;
1055 	struct p9_fd_opts opts;
1056 
1057 	parse_opts(args, &opts);
1058 	client->trans_opts.fd.rfd = opts.rfd;
1059 	client->trans_opts.fd.wfd = opts.wfd;
1060 
1061 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1062 		pr_err("Insufficient options for proto=fd\n");
1063 		return -ENOPROTOOPT;
1064 	}
1065 
1066 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1067 	if (err < 0)
1068 		return err;
1069 
1070 	p9_conn_create(client);
1071 
1072 	return 0;
1073 }
1074 
1075 static struct p9_trans_module p9_tcp_trans = {
1076 	.name = "tcp",
1077 	.maxsize = MAX_SOCK_BUF,
1078 	.def = 0,
1079 	.create = p9_fd_create_tcp,
1080 	.close = p9_fd_close,
1081 	.request = p9_fd_request,
1082 	.cancel = p9_fd_cancel,
1083 	.cancelled = p9_fd_cancelled,
1084 	.show_options = p9_fd_show_options,
1085 	.owner = THIS_MODULE,
1086 };
1087 
1088 static struct p9_trans_module p9_unix_trans = {
1089 	.name = "unix",
1090 	.maxsize = MAX_SOCK_BUF,
1091 	.def = 0,
1092 	.create = p9_fd_create_unix,
1093 	.close = p9_fd_close,
1094 	.request = p9_fd_request,
1095 	.cancel = p9_fd_cancel,
1096 	.cancelled = p9_fd_cancelled,
1097 	.show_options = p9_fd_show_options,
1098 	.owner = THIS_MODULE,
1099 };
1100 
1101 static struct p9_trans_module p9_fd_trans = {
1102 	.name = "fd",
1103 	.maxsize = MAX_SOCK_BUF,
1104 	.def = 0,
1105 	.create = p9_fd_create,
1106 	.close = p9_fd_close,
1107 	.request = p9_fd_request,
1108 	.cancel = p9_fd_cancel,
1109 	.cancelled = p9_fd_cancelled,
1110 	.show_options = p9_fd_show_options,
1111 	.owner = THIS_MODULE,
1112 };
1113 
1114 /**
1115  * p9_poll_workfn - poll worker thread
1116  * @work: work queue
1117  *
1118  * polls all v9fs transports for new events and queues the appropriate
1119  * work to the work queue
1120  *
1121  */
1122 
1123 static void p9_poll_workfn(struct work_struct *work)
1124 {
1125 	unsigned long flags;
1126 
1127 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1128 
1129 	spin_lock_irqsave(&p9_poll_lock, flags);
1130 	while (!list_empty(&p9_poll_pending_list)) {
1131 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1132 							struct p9_conn,
1133 							poll_pending_link);
1134 		list_del_init(&conn->poll_pending_link);
1135 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1136 
1137 		p9_poll_mux(conn);
1138 
1139 		spin_lock_irqsave(&p9_poll_lock, flags);
1140 	}
1141 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1142 
1143 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1144 }
1145 
1146 int p9_trans_fd_init(void)
1147 {
1148 	v9fs_register_trans(&p9_tcp_trans);
1149 	v9fs_register_trans(&p9_unix_trans);
1150 	v9fs_register_trans(&p9_fd_trans);
1151 
1152 	return 0;
1153 }
1154 
1155 void p9_trans_fd_exit(void)
1156 {
1157 	flush_work(&p9_poll_work);
1158 	v9fs_unregister_trans(&p9_tcp_trans);
1159 	v9fs_unregister_trans(&p9_unix_trans);
1160 	v9fs_unregister_trans(&p9_fd_trans);
1161 }
1162