xref: /openbmc/linux/net/9p/trans_fd.c (revision bc5aa3a0)
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 <net/9p/9p.h>
45 #include <net/9p/client.h>
46 #include <net/9p/transport.h>
47 
48 #include <linux/syscalls.h> /* killme */
49 
50 #define P9_PORT 564
51 #define MAX_SOCK_BUF (64*1024)
52 #define MAXPOLLWADDR	2
53 
54 /**
55  * struct p9_fd_opts - per-transport options
56  * @rfd: file descriptor for reading (trans=fd)
57  * @wfd: file descriptor for writing (trans=fd)
58  * @port: port to connect to (trans=tcp)
59  *
60  */
61 
62 struct p9_fd_opts {
63 	int rfd;
64 	int wfd;
65 	u16 port;
66 	int privport;
67 };
68 
69 /*
70   * Option Parsing (code inspired by NFS code)
71   *  - a little lazy - parse all fd-transport options
72   */
73 
74 enum {
75 	/* Options that take integer arguments */
76 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
77 	/* Options that take no arguments */
78 	Opt_privport,
79 };
80 
81 static const match_table_t tokens = {
82 	{Opt_port, "port=%u"},
83 	{Opt_rfdno, "rfdno=%u"},
84 	{Opt_wfdno, "wfdno=%u"},
85 	{Opt_privport, "privport"},
86 	{Opt_err, NULL},
87 };
88 
89 enum {
90 	Rworksched = 1,		/* read work scheduled or running */
91 	Rpending = 2,		/* can read */
92 	Wworksched = 4,		/* write work scheduled or running */
93 	Wpending = 8,		/* can write */
94 };
95 
96 struct p9_poll_wait {
97 	struct p9_conn *conn;
98 	wait_queue_t wait;
99 	wait_queue_head_t *wait_addr;
100 };
101 
102 /**
103  * struct p9_conn - fd mux connection state information
104  * @mux_list: list link for mux to manage multiple connections (?)
105  * @client: reference to client instance for this connection
106  * @err: error state
107  * @req_list: accounting for requests which have been sent
108  * @unsent_req_list: accounting for requests that haven't been sent
109  * @req: current request being processed (if any)
110  * @tmp_buf: temporary buffer to read in header
111  * @rc: temporary fcall for reading current frame
112  * @wpos: write position for current frame
113  * @wsize: amount of data to write for current frame
114  * @wbuf: current write buffer
115  * @poll_pending_link: pending links to be polled per conn
116  * @poll_wait: array of wait_q's for various worker threads
117  * @pt: poll state
118  * @rq: current read work
119  * @wq: current write work
120  * @wsched: ????
121  *
122  */
123 
124 struct p9_conn {
125 	struct list_head mux_list;
126 	struct p9_client *client;
127 	int err;
128 	struct list_head req_list;
129 	struct list_head unsent_req_list;
130 	struct p9_req_t *req;
131 	char tmp_buf[7];
132 	struct p9_fcall rc;
133 	int wpos;
134 	int wsize;
135 	char *wbuf;
136 	struct list_head poll_pending_link;
137 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
138 	poll_table pt;
139 	struct work_struct rq;
140 	struct work_struct wq;
141 	unsigned long wsched;
142 };
143 
144 /**
145  * struct p9_trans_fd - transport state
146  * @rd: reference to file to read from
147  * @wr: reference of file to write to
148  * @conn: connection state reference
149  *
150  */
151 
152 struct p9_trans_fd {
153 	struct file *rd;
154 	struct file *wr;
155 	struct p9_conn conn;
156 };
157 
158 static void p9_poll_workfn(struct work_struct *work);
159 
160 static DEFINE_SPINLOCK(p9_poll_lock);
161 static LIST_HEAD(p9_poll_pending_list);
162 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
163 
164 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
165 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
166 
167 static void p9_mux_poll_stop(struct p9_conn *m)
168 {
169 	unsigned long flags;
170 	int i;
171 
172 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
173 		struct p9_poll_wait *pwait = &m->poll_wait[i];
174 
175 		if (pwait->wait_addr) {
176 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
177 			pwait->wait_addr = NULL;
178 		}
179 	}
180 
181 	spin_lock_irqsave(&p9_poll_lock, flags);
182 	list_del_init(&m->poll_pending_link);
183 	spin_unlock_irqrestore(&p9_poll_lock, flags);
184 }
185 
186 /**
187  * p9_conn_cancel - cancel all pending requests with error
188  * @m: mux data
189  * @err: error code
190  *
191  */
192 
193 static void p9_conn_cancel(struct p9_conn *m, int err)
194 {
195 	struct p9_req_t *req, *rtmp;
196 	unsigned long flags;
197 	LIST_HEAD(cancel_list);
198 
199 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
200 
201 	spin_lock_irqsave(&m->client->lock, flags);
202 
203 	if (m->err) {
204 		spin_unlock_irqrestore(&m->client->lock, flags);
205 		return;
206 	}
207 
208 	m->err = err;
209 
210 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
211 		list_move(&req->req_list, &cancel_list);
212 	}
213 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
214 		list_move(&req->req_list, &cancel_list);
215 	}
216 	spin_unlock_irqrestore(&m->client->lock, flags);
217 
218 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
219 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
220 		list_del(&req->req_list);
221 		if (!req->t_err)
222 			req->t_err = err;
223 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
224 	}
225 }
226 
227 static int
228 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt)
229 {
230 	int ret, n;
231 	struct p9_trans_fd *ts = NULL;
232 
233 	if (client && client->status == Connected)
234 		ts = client->trans;
235 
236 	if (!ts)
237 		return -EREMOTEIO;
238 
239 	if (!ts->rd->f_op->poll)
240 		return -EIO;
241 
242 	if (!ts->wr->f_op->poll)
243 		return -EIO;
244 
245 	ret = ts->rd->f_op->poll(ts->rd, pt);
246 	if (ret < 0)
247 		return ret;
248 
249 	if (ts->rd != ts->wr) {
250 		n = ts->wr->f_op->poll(ts->wr, pt);
251 		if (n < 0)
252 			return n;
253 		ret = (ret & ~POLLOUT) | (n & ~POLLIN);
254 	}
255 
256 	return ret;
257 }
258 
259 /**
260  * p9_fd_read- read from a fd
261  * @client: client instance
262  * @v: buffer to receive data into
263  * @len: size of receive buffer
264  *
265  */
266 
267 static int p9_fd_read(struct p9_client *client, void *v, int len)
268 {
269 	int ret;
270 	struct p9_trans_fd *ts = NULL;
271 
272 	if (client && client->status != Disconnected)
273 		ts = client->trans;
274 
275 	if (!ts)
276 		return -EREMOTEIO;
277 
278 	if (!(ts->rd->f_flags & O_NONBLOCK))
279 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
280 
281 	ret = kernel_read(ts->rd, ts->rd->f_pos, v, len);
282 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
283 		client->status = Disconnected;
284 	return ret;
285 }
286 
287 /**
288  * p9_read_work - called when there is some data to be read from a transport
289  * @work: container of work to be done
290  *
291  */
292 
293 static void p9_read_work(struct work_struct *work)
294 {
295 	int n, err;
296 	struct p9_conn *m;
297 	int status = REQ_STATUS_ERROR;
298 
299 	m = container_of(work, struct p9_conn, rq);
300 
301 	if (m->err < 0)
302 		return;
303 
304 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
305 
306 	if (!m->rc.sdata) {
307 		m->rc.sdata = m->tmp_buf;
308 		m->rc.offset = 0;
309 		m->rc.capacity = 7; /* start by reading header */
310 	}
311 
312 	clear_bit(Rpending, &m->wsched);
313 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
314 		 m, m->rc.offset, m->rc.capacity,
315 		 m->rc.capacity - m->rc.offset);
316 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
317 			 m->rc.capacity - m->rc.offset);
318 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
319 	if (err == -EAGAIN)
320 		goto end_clear;
321 
322 	if (err <= 0)
323 		goto error;
324 
325 	m->rc.offset += err;
326 
327 	/* header read in */
328 	if ((!m->req) && (m->rc.offset == m->rc.capacity)) {
329 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
330 
331 		err = p9_parse_header(&m->rc, NULL, NULL, NULL, 0);
332 		if (err) {
333 			p9_debug(P9_DEBUG_ERROR,
334 				 "error parsing header: %d\n", err);
335 			goto error;
336 		}
337 
338 		if (m->rc.size >= m->client->msize) {
339 			p9_debug(P9_DEBUG_ERROR,
340 				 "requested packet size too big: %d\n",
341 				 m->rc.size);
342 			err = -EIO;
343 			goto error;
344 		}
345 
346 		p9_debug(P9_DEBUG_TRANS,
347 			 "mux %p pkt: size: %d bytes tag: %d\n",
348 			 m, m->rc.size, m->rc.tag);
349 
350 		m->req = p9_tag_lookup(m->client, m->rc.tag);
351 		if (!m->req || (m->req->status != REQ_STATUS_SENT)) {
352 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
353 				 m->rc.tag);
354 			err = -EIO;
355 			goto error;
356 		}
357 
358 		if (m->req->rc == NULL) {
359 			p9_debug(P9_DEBUG_ERROR,
360 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
361 				 m->rc.tag, m->req);
362 			m->req = NULL;
363 			err = -EIO;
364 			goto error;
365 		}
366 		m->rc.sdata = (char *)m->req->rc + sizeof(struct p9_fcall);
367 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
368 		m->rc.capacity = m->rc.size;
369 	}
370 
371 	/* packet is read in
372 	 * not an else because some packets (like clunk) have no payload
373 	 */
374 	if ((m->req) && (m->rc.offset == m->rc.capacity)) {
375 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
376 		spin_lock(&m->client->lock);
377 		if (m->req->status != REQ_STATUS_ERROR)
378 			status = REQ_STATUS_RCVD;
379 		list_del(&m->req->req_list);
380 		spin_unlock(&m->client->lock);
381 		p9_client_cb(m->client, m->req, status);
382 		m->rc.sdata = NULL;
383 		m->rc.offset = 0;
384 		m->rc.capacity = 0;
385 		m->req = NULL;
386 	}
387 
388 end_clear:
389 	clear_bit(Rworksched, &m->wsched);
390 
391 	if (!list_empty(&m->req_list)) {
392 		if (test_and_clear_bit(Rpending, &m->wsched))
393 			n = POLLIN;
394 		else
395 			n = p9_fd_poll(m->client, NULL);
396 
397 		if ((n & POLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
398 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
399 			schedule_work(&m->rq);
400 		}
401 	}
402 
403 	return;
404 error:
405 	p9_conn_cancel(m, err);
406 	clear_bit(Rworksched, &m->wsched);
407 }
408 
409 /**
410  * p9_fd_write - write to a socket
411  * @client: client instance
412  * @v: buffer to send data from
413  * @len: size of send buffer
414  *
415  */
416 
417 static int p9_fd_write(struct p9_client *client, void *v, int len)
418 {
419 	int ret;
420 	mm_segment_t oldfs;
421 	struct p9_trans_fd *ts = NULL;
422 
423 	if (client && client->status != Disconnected)
424 		ts = client->trans;
425 
426 	if (!ts)
427 		return -EREMOTEIO;
428 
429 	if (!(ts->wr->f_flags & O_NONBLOCK))
430 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
431 
432 	oldfs = get_fs();
433 	set_fs(get_ds());
434 	/* The cast to a user pointer is valid due to the set_fs() */
435 	ret = vfs_write(ts->wr, (__force void __user *)v, len, &ts->wr->f_pos);
436 	set_fs(oldfs);
437 
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_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 /**
724  * parse_opts - parse mount options into p9_fd_opts structure
725  * @params: options string passed from mount
726  * @opts: fd transport-specific structure to parse options into
727  *
728  * Returns 0 upon success, -ERRNO upon failure
729  */
730 
731 static int parse_opts(char *params, struct p9_fd_opts *opts)
732 {
733 	char *p;
734 	substring_t args[MAX_OPT_ARGS];
735 	int option;
736 	char *options, *tmp_options;
737 
738 	opts->port = P9_PORT;
739 	opts->rfd = ~0;
740 	opts->wfd = ~0;
741 	opts->privport = 0;
742 
743 	if (!params)
744 		return 0;
745 
746 	tmp_options = kstrdup(params, GFP_KERNEL);
747 	if (!tmp_options) {
748 		p9_debug(P9_DEBUG_ERROR,
749 			 "failed to allocate copy of option string\n");
750 		return -ENOMEM;
751 	}
752 	options = tmp_options;
753 
754 	while ((p = strsep(&options, ",")) != NULL) {
755 		int token;
756 		int r;
757 		if (!*p)
758 			continue;
759 		token = match_token(p, tokens, args);
760 		if ((token != Opt_err) && (token != Opt_privport)) {
761 			r = match_int(&args[0], &option);
762 			if (r < 0) {
763 				p9_debug(P9_DEBUG_ERROR,
764 					 "integer field, but no integer?\n");
765 				continue;
766 			}
767 		}
768 		switch (token) {
769 		case Opt_port:
770 			opts->port = option;
771 			break;
772 		case Opt_rfdno:
773 			opts->rfd = option;
774 			break;
775 		case Opt_wfdno:
776 			opts->wfd = option;
777 			break;
778 		case Opt_privport:
779 			opts->privport = 1;
780 			break;
781 		default:
782 			continue;
783 		}
784 	}
785 
786 	kfree(tmp_options);
787 	return 0;
788 }
789 
790 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
791 {
792 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
793 					   GFP_KERNEL);
794 	if (!ts)
795 		return -ENOMEM;
796 
797 	ts->rd = fget(rfd);
798 	ts->wr = fget(wfd);
799 	if (!ts->rd || !ts->wr) {
800 		if (ts->rd)
801 			fput(ts->rd);
802 		if (ts->wr)
803 			fput(ts->wr);
804 		kfree(ts);
805 		return -EIO;
806 	}
807 
808 	client->trans = ts;
809 	client->status = Connected;
810 
811 	return 0;
812 }
813 
814 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
815 {
816 	struct p9_trans_fd *p;
817 	struct file *file;
818 
819 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
820 	if (!p)
821 		return -ENOMEM;
822 
823 	csocket->sk->sk_allocation = GFP_NOIO;
824 	file = sock_alloc_file(csocket, 0, NULL);
825 	if (IS_ERR(file)) {
826 		pr_err("%s (%d): failed to map fd\n",
827 		       __func__, task_pid_nr(current));
828 		sock_release(csocket);
829 		kfree(p);
830 		return PTR_ERR(file);
831 	}
832 
833 	get_file(file);
834 	p->wr = p->rd = file;
835 	client->trans = p;
836 	client->status = Connected;
837 
838 	p->rd->f_flags |= O_NONBLOCK;
839 
840 	p9_conn_create(client);
841 	return 0;
842 }
843 
844 /**
845  * p9_mux_destroy - cancels all pending requests of mux
846  * @m: mux to destroy
847  *
848  */
849 
850 static void p9_conn_destroy(struct p9_conn *m)
851 {
852 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
853 		 m, m->mux_list.prev, m->mux_list.next);
854 
855 	p9_mux_poll_stop(m);
856 	cancel_work_sync(&m->rq);
857 	cancel_work_sync(&m->wq);
858 
859 	p9_conn_cancel(m, -ECONNRESET);
860 
861 	m->client = NULL;
862 }
863 
864 /**
865  * p9_fd_close - shutdown file descriptor transport
866  * @client: client instance
867  *
868  */
869 
870 static void p9_fd_close(struct p9_client *client)
871 {
872 	struct p9_trans_fd *ts;
873 
874 	if (!client)
875 		return;
876 
877 	ts = client->trans;
878 	if (!ts)
879 		return;
880 
881 	client->status = Disconnected;
882 
883 	p9_conn_destroy(&ts->conn);
884 
885 	if (ts->rd)
886 		fput(ts->rd);
887 	if (ts->wr)
888 		fput(ts->wr);
889 
890 	kfree(ts);
891 }
892 
893 /*
894  * stolen from NFS - maybe should be made a generic function?
895  */
896 static inline int valid_ipaddr4(const char *buf)
897 {
898 	int rc, count, in[4];
899 
900 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
901 	if (rc != 4)
902 		return -EINVAL;
903 	for (count = 0; count < 4; count++) {
904 		if (in[count] > 255)
905 			return -EINVAL;
906 	}
907 	return 0;
908 }
909 
910 static int p9_bind_privport(struct socket *sock)
911 {
912 	struct sockaddr_in cl;
913 	int port, err = -EINVAL;
914 
915 	memset(&cl, 0, sizeof(cl));
916 	cl.sin_family = AF_INET;
917 	cl.sin_addr.s_addr = INADDR_ANY;
918 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
919 		cl.sin_port = htons((ushort)port);
920 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
921 		if (err != -EADDRINUSE)
922 			break;
923 	}
924 	return err;
925 }
926 
927 
928 static int
929 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
930 {
931 	int err;
932 	struct socket *csocket;
933 	struct sockaddr_in sin_server;
934 	struct p9_fd_opts opts;
935 
936 	err = parse_opts(args, &opts);
937 	if (err < 0)
938 		return err;
939 
940 	if (valid_ipaddr4(addr) < 0)
941 		return -EINVAL;
942 
943 	csocket = NULL;
944 
945 	sin_server.sin_family = AF_INET;
946 	sin_server.sin_addr.s_addr = in_aton(addr);
947 	sin_server.sin_port = htons(opts.port);
948 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
949 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
950 	if (err) {
951 		pr_err("%s (%d): problem creating socket\n",
952 		       __func__, task_pid_nr(current));
953 		return err;
954 	}
955 
956 	if (opts.privport) {
957 		err = p9_bind_privport(csocket);
958 		if (err < 0) {
959 			pr_err("%s (%d): problem binding to privport\n",
960 			       __func__, task_pid_nr(current));
961 			sock_release(csocket);
962 			return err;
963 		}
964 	}
965 
966 	err = csocket->ops->connect(csocket,
967 				    (struct sockaddr *)&sin_server,
968 				    sizeof(struct sockaddr_in), 0);
969 	if (err < 0) {
970 		pr_err("%s (%d): problem connecting socket to %s\n",
971 		       __func__, task_pid_nr(current), addr);
972 		sock_release(csocket);
973 		return err;
974 	}
975 
976 	return p9_socket_open(client, csocket);
977 }
978 
979 static int
980 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
981 {
982 	int err;
983 	struct socket *csocket;
984 	struct sockaddr_un sun_server;
985 
986 	csocket = NULL;
987 
988 	if (strlen(addr) >= UNIX_PATH_MAX) {
989 		pr_err("%s (%d): address too long: %s\n",
990 		       __func__, task_pid_nr(current), addr);
991 		return -ENAMETOOLONG;
992 	}
993 
994 	sun_server.sun_family = PF_UNIX;
995 	strcpy(sun_server.sun_path, addr);
996 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
997 			    SOCK_STREAM, 0, &csocket, 1);
998 	if (err < 0) {
999 		pr_err("%s (%d): problem creating socket\n",
1000 		       __func__, task_pid_nr(current));
1001 
1002 		return err;
1003 	}
1004 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1005 			sizeof(struct sockaddr_un) - 1, 0);
1006 	if (err < 0) {
1007 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1008 		       __func__, task_pid_nr(current), addr, err);
1009 		sock_release(csocket);
1010 		return err;
1011 	}
1012 
1013 	return p9_socket_open(client, csocket);
1014 }
1015 
1016 static int
1017 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1018 {
1019 	int err;
1020 	struct p9_fd_opts opts;
1021 
1022 	parse_opts(args, &opts);
1023 
1024 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1025 		pr_err("Insufficient options for proto=fd\n");
1026 		return -ENOPROTOOPT;
1027 	}
1028 
1029 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1030 	if (err < 0)
1031 		return err;
1032 
1033 	p9_conn_create(client);
1034 
1035 	return 0;
1036 }
1037 
1038 static struct p9_trans_module p9_tcp_trans = {
1039 	.name = "tcp",
1040 	.maxsize = MAX_SOCK_BUF,
1041 	.def = 0,
1042 	.create = p9_fd_create_tcp,
1043 	.close = p9_fd_close,
1044 	.request = p9_fd_request,
1045 	.cancel = p9_fd_cancel,
1046 	.cancelled = p9_fd_cancelled,
1047 	.owner = THIS_MODULE,
1048 };
1049 
1050 static struct p9_trans_module p9_unix_trans = {
1051 	.name = "unix",
1052 	.maxsize = MAX_SOCK_BUF,
1053 	.def = 0,
1054 	.create = p9_fd_create_unix,
1055 	.close = p9_fd_close,
1056 	.request = p9_fd_request,
1057 	.cancel = p9_fd_cancel,
1058 	.cancelled = p9_fd_cancelled,
1059 	.owner = THIS_MODULE,
1060 };
1061 
1062 static struct p9_trans_module p9_fd_trans = {
1063 	.name = "fd",
1064 	.maxsize = MAX_SOCK_BUF,
1065 	.def = 0,
1066 	.create = p9_fd_create,
1067 	.close = p9_fd_close,
1068 	.request = p9_fd_request,
1069 	.cancel = p9_fd_cancel,
1070 	.cancelled = p9_fd_cancelled,
1071 	.owner = THIS_MODULE,
1072 };
1073 
1074 /**
1075  * p9_poll_proc - poll worker thread
1076  * @a: thread state and arguments
1077  *
1078  * polls all v9fs transports for new events and queues the appropriate
1079  * work to the work queue
1080  *
1081  */
1082 
1083 static void p9_poll_workfn(struct work_struct *work)
1084 {
1085 	unsigned long flags;
1086 
1087 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1088 
1089 	spin_lock_irqsave(&p9_poll_lock, flags);
1090 	while (!list_empty(&p9_poll_pending_list)) {
1091 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1092 							struct p9_conn,
1093 							poll_pending_link);
1094 		list_del_init(&conn->poll_pending_link);
1095 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1096 
1097 		p9_poll_mux(conn);
1098 
1099 		spin_lock_irqsave(&p9_poll_lock, flags);
1100 	}
1101 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1102 
1103 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1104 }
1105 
1106 int p9_trans_fd_init(void)
1107 {
1108 	v9fs_register_trans(&p9_tcp_trans);
1109 	v9fs_register_trans(&p9_unix_trans);
1110 	v9fs_register_trans(&p9_fd_trans);
1111 
1112 	return 0;
1113 }
1114 
1115 void p9_trans_fd_exit(void)
1116 {
1117 	flush_work(&p9_poll_work);
1118 	v9fs_unregister_trans(&p9_tcp_trans);
1119 	v9fs_unregister_trans(&p9_unix_trans);
1120 	v9fs_unregister_trans(&p9_fd_trans);
1121 }
1122