xref: /linux/net/9p/trans_fd.c (revision 4413e16d9d21673bb5048a2e542f1aaa00015c2e)
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 };
67 
68 /**
69  * struct p9_trans_fd - transport state
70  * @rd: reference to file to read from
71  * @wr: reference of file to write to
72  * @conn: connection state reference
73  *
74  */
75 
76 struct p9_trans_fd {
77 	struct file *rd;
78 	struct file *wr;
79 	struct p9_conn *conn;
80 };
81 
82 /*
83   * Option Parsing (code inspired by NFS code)
84   *  - a little lazy - parse all fd-transport options
85   */
86 
87 enum {
88 	/* Options that take integer arguments */
89 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
90 };
91 
92 static const match_table_t tokens = {
93 	{Opt_port, "port=%u"},
94 	{Opt_rfdno, "rfdno=%u"},
95 	{Opt_wfdno, "wfdno=%u"},
96 	{Opt_err, NULL},
97 };
98 
99 enum {
100 	Rworksched = 1,		/* read work scheduled or running */
101 	Rpending = 2,		/* can read */
102 	Wworksched = 4,		/* write work scheduled or running */
103 	Wpending = 8,		/* can write */
104 };
105 
106 struct p9_poll_wait {
107 	struct p9_conn *conn;
108 	wait_queue_t wait;
109 	wait_queue_head_t *wait_addr;
110 };
111 
112 /**
113  * struct p9_conn - fd mux connection state information
114  * @mux_list: list link for mux to manage multiple connections (?)
115  * @client: reference to client instance for this connection
116  * @err: error state
117  * @req_list: accounting for requests which have been sent
118  * @unsent_req_list: accounting for requests that haven't been sent
119  * @req: current request being processed (if any)
120  * @tmp_buf: temporary buffer to read in header
121  * @rsize: amount to read for current frame
122  * @rpos: read position in current frame
123  * @rbuf: current read buffer
124  * @wpos: write position for current frame
125  * @wsize: amount of data to write for current frame
126  * @wbuf: current write buffer
127  * @poll_pending_link: pending links to be polled per conn
128  * @poll_wait: array of wait_q's for various worker threads
129  * @pt: poll state
130  * @rq: current read work
131  * @wq: current write work
132  * @wsched: ????
133  *
134  */
135 
136 struct p9_conn {
137 	struct list_head mux_list;
138 	struct p9_client *client;
139 	int err;
140 	struct list_head req_list;
141 	struct list_head unsent_req_list;
142 	struct p9_req_t *req;
143 	char tmp_buf[7];
144 	int rsize;
145 	int rpos;
146 	char *rbuf;
147 	int wpos;
148 	int wsize;
149 	char *wbuf;
150 	struct list_head poll_pending_link;
151 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
152 	poll_table pt;
153 	struct work_struct rq;
154 	struct work_struct wq;
155 	unsigned long wsched;
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 void p9_mux_poll_stop(struct p9_conn *m)
165 {
166 	unsigned long flags;
167 	int i;
168 
169 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
170 		struct p9_poll_wait *pwait = &m->poll_wait[i];
171 
172 		if (pwait->wait_addr) {
173 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
174 			pwait->wait_addr = NULL;
175 		}
176 	}
177 
178 	spin_lock_irqsave(&p9_poll_lock, flags);
179 	list_del_init(&m->poll_pending_link);
180 	spin_unlock_irqrestore(&p9_poll_lock, flags);
181 }
182 
183 /**
184  * p9_conn_cancel - cancel all pending requests with error
185  * @m: mux data
186  * @err: error code
187  *
188  */
189 
190 static void p9_conn_cancel(struct p9_conn *m, int err)
191 {
192 	struct p9_req_t *req, *rtmp;
193 	unsigned long flags;
194 	LIST_HEAD(cancel_list);
195 
196 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
197 
198 	spin_lock_irqsave(&m->client->lock, flags);
199 
200 	if (m->err) {
201 		spin_unlock_irqrestore(&m->client->lock, flags);
202 		return;
203 	}
204 
205 	m->err = err;
206 
207 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
208 		req->status = REQ_STATUS_ERROR;
209 		if (!req->t_err)
210 			req->t_err = err;
211 		list_move(&req->req_list, &cancel_list);
212 	}
213 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
214 		req->status = REQ_STATUS_ERROR;
215 		if (!req->t_err)
216 			req->t_err = err;
217 		list_move(&req->req_list, &cancel_list);
218 	}
219 	spin_unlock_irqrestore(&m->client->lock, flags);
220 
221 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
222 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
223 		list_del(&req->req_list);
224 		p9_client_cb(m->client, req);
225 	}
226 }
227 
228 static int
229 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt)
230 {
231 	int ret, n;
232 	struct p9_trans_fd *ts = NULL;
233 
234 	if (client && client->status == Connected)
235 		ts = client->trans;
236 
237 	if (!ts)
238 		return -EREMOTEIO;
239 
240 	if (!ts->rd->f_op || !ts->rd->f_op->poll)
241 		return -EIO;
242 
243 	if (!ts->wr->f_op || !ts->wr->f_op->poll)
244 		return -EIO;
245 
246 	ret = ts->rd->f_op->poll(ts->rd, pt);
247 	if (ret < 0)
248 		return ret;
249 
250 	if (ts->rd != ts->wr) {
251 		n = ts->wr->f_op->poll(ts->wr, pt);
252 		if (n < 0)
253 			return n;
254 		ret = (ret & ~POLLOUT) | (n & ~POLLIN);
255 	}
256 
257 	return ret;
258 }
259 
260 /**
261  * p9_fd_read- read from a fd
262  * @client: client instance
263  * @v: buffer to receive data into
264  * @len: size of receive buffer
265  *
266  */
267 
268 static int p9_fd_read(struct p9_client *client, void *v, int len)
269 {
270 	int ret;
271 	struct p9_trans_fd *ts = NULL;
272 
273 	if (client && client->status != Disconnected)
274 		ts = client->trans;
275 
276 	if (!ts)
277 		return -EREMOTEIO;
278 
279 	if (!(ts->rd->f_flags & O_NONBLOCK))
280 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
281 
282 	ret = kernel_read(ts->rd, ts->rd->f_pos, v, len);
283 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
284 		client->status = Disconnected;
285 	return ret;
286 }
287 
288 /**
289  * p9_read_work - called when there is some data to be read from a transport
290  * @work: container of work to be done
291  *
292  */
293 
294 static void p9_read_work(struct work_struct *work)
295 {
296 	int n, err;
297 	struct p9_conn *m;
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 %d\n", m, m->rpos);
305 
306 	if (!m->rbuf) {
307 		m->rbuf = m->tmp_buf;
308 		m->rpos = 0;
309 		m->rsize = 7; /* start by reading header */
310 	}
311 
312 	clear_bit(Rpending, &m->wsched);
313 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %d size: %d = %d\n",
314 		 m, m->rpos, m->rsize, m->rsize-m->rpos);
315 	err = p9_fd_read(m->client, m->rbuf + m->rpos,
316 						m->rsize - m->rpos);
317 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
318 	if (err == -EAGAIN) {
319 		clear_bit(Rworksched, &m->wsched);
320 		return;
321 	}
322 
323 	if (err <= 0)
324 		goto error;
325 
326 	m->rpos += err;
327 
328 	if ((!m->req) && (m->rpos == m->rsize)) { /* header read in */
329 		u16 tag;
330 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
331 
332 		n = le32_to_cpu(*(__le32 *) m->rbuf); /* read packet size */
333 		if (n >= m->client->msize) {
334 			p9_debug(P9_DEBUG_ERROR,
335 				 "requested packet size too big: %d\n", n);
336 			err = -EIO;
337 			goto error;
338 		}
339 
340 		tag = le16_to_cpu(*(__le16 *) (m->rbuf+5)); /* read tag */
341 		p9_debug(P9_DEBUG_TRANS,
342 			 "mux %p pkt: size: %d bytes tag: %d\n", m, n, tag);
343 
344 		m->req = p9_tag_lookup(m->client, tag);
345 		if (!m->req || (m->req->status != REQ_STATUS_SENT &&
346 					m->req->status != REQ_STATUS_FLSH)) {
347 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
348 				 tag);
349 			err = -EIO;
350 			goto error;
351 		}
352 
353 		if (m->req->rc == NULL) {
354 			m->req->rc = kmalloc(sizeof(struct p9_fcall) +
355 						m->client->msize, GFP_NOFS);
356 			if (!m->req->rc) {
357 				m->req = NULL;
358 				err = -ENOMEM;
359 				goto error;
360 			}
361 		}
362 		m->rbuf = (char *)m->req->rc + sizeof(struct p9_fcall);
363 		memcpy(m->rbuf, m->tmp_buf, m->rsize);
364 		m->rsize = n;
365 	}
366 
367 	/* not an else because some packets (like clunk) have no payload */
368 	if ((m->req) && (m->rpos == m->rsize)) { /* packet is read in */
369 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
370 		spin_lock(&m->client->lock);
371 		if (m->req->status != REQ_STATUS_ERROR)
372 			m->req->status = REQ_STATUS_RCVD;
373 		list_del(&m->req->req_list);
374 		spin_unlock(&m->client->lock);
375 		p9_client_cb(m->client, m->req);
376 		m->rbuf = NULL;
377 		m->rpos = 0;
378 		m->rsize = 0;
379 		m->req = NULL;
380 	}
381 
382 	if (!list_empty(&m->req_list)) {
383 		if (test_and_clear_bit(Rpending, &m->wsched))
384 			n = POLLIN;
385 		else
386 			n = p9_fd_poll(m->client, NULL);
387 
388 		if (n & POLLIN) {
389 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
390 			schedule_work(&m->rq);
391 		} else
392 			clear_bit(Rworksched, &m->wsched);
393 	} else
394 		clear_bit(Rworksched, &m->wsched);
395 
396 	return;
397 error:
398 	p9_conn_cancel(m, err);
399 	clear_bit(Rworksched, &m->wsched);
400 }
401 
402 /**
403  * p9_fd_write - write to a socket
404  * @client: client instance
405  * @v: buffer to send data from
406  * @len: size of send buffer
407  *
408  */
409 
410 static int p9_fd_write(struct p9_client *client, void *v, int len)
411 {
412 	int ret;
413 	mm_segment_t oldfs;
414 	struct p9_trans_fd *ts = NULL;
415 
416 	if (client && client->status != Disconnected)
417 		ts = client->trans;
418 
419 	if (!ts)
420 		return -EREMOTEIO;
421 
422 	if (!(ts->wr->f_flags & O_NONBLOCK))
423 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
424 
425 	oldfs = get_fs();
426 	set_fs(get_ds());
427 	/* The cast to a user pointer is valid due to the set_fs() */
428 	ret = vfs_write(ts->wr, (__force void __user *)v, len, &ts->wr->f_pos);
429 	set_fs(oldfs);
430 
431 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
432 		client->status = Disconnected;
433 	return ret;
434 }
435 
436 /**
437  * p9_write_work - called when a transport can send some data
438  * @work: container for work to be done
439  *
440  */
441 
442 static void p9_write_work(struct work_struct *work)
443 {
444 	int n, err;
445 	struct p9_conn *m;
446 	struct p9_req_t *req;
447 
448 	m = container_of(work, struct p9_conn, wq);
449 
450 	if (m->err < 0) {
451 		clear_bit(Wworksched, &m->wsched);
452 		return;
453 	}
454 
455 	if (!m->wsize) {
456 		if (list_empty(&m->unsent_req_list)) {
457 			clear_bit(Wworksched, &m->wsched);
458 			return;
459 		}
460 
461 		spin_lock(&m->client->lock);
462 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
463 			       req_list);
464 		req->status = REQ_STATUS_SENT;
465 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
466 		list_move_tail(&req->req_list, &m->req_list);
467 
468 		m->wbuf = req->tc->sdata;
469 		m->wsize = req->tc->size;
470 		m->wpos = 0;
471 		spin_unlock(&m->client->lock);
472 	}
473 
474 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
475 		 m, m->wpos, m->wsize);
476 	clear_bit(Wpending, &m->wsched);
477 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
478 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
479 	if (err == -EAGAIN) {
480 		clear_bit(Wworksched, &m->wsched);
481 		return;
482 	}
483 
484 	if (err < 0)
485 		goto error;
486 	else if (err == 0) {
487 		err = -EREMOTEIO;
488 		goto error;
489 	}
490 
491 	m->wpos += err;
492 	if (m->wpos == m->wsize)
493 		m->wpos = m->wsize = 0;
494 
495 	if (m->wsize == 0 && !list_empty(&m->unsent_req_list)) {
496 		if (test_and_clear_bit(Wpending, &m->wsched))
497 			n = POLLOUT;
498 		else
499 			n = p9_fd_poll(m->client, NULL);
500 
501 		if (n & POLLOUT) {
502 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
503 			schedule_work(&m->wq);
504 		} else
505 			clear_bit(Wworksched, &m->wsched);
506 	} else
507 		clear_bit(Wworksched, &m->wsched);
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_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 - allocate and 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 struct p9_conn *p9_conn_create(struct p9_client *client)
574 {
575 	int n;
576 	struct p9_conn *m;
577 
578 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
579 	m = kzalloc(sizeof(struct p9_conn), GFP_KERNEL);
580 	if (!m)
581 		return ERR_PTR(-ENOMEM);
582 
583 	INIT_LIST_HEAD(&m->mux_list);
584 	m->client = client;
585 
586 	INIT_LIST_HEAD(&m->req_list);
587 	INIT_LIST_HEAD(&m->unsent_req_list);
588 	INIT_WORK(&m->rq, p9_read_work);
589 	INIT_WORK(&m->wq, p9_write_work);
590 	INIT_LIST_HEAD(&m->poll_pending_link);
591 	init_poll_funcptr(&m->pt, p9_pollwait);
592 
593 	n = p9_fd_poll(client, &m->pt);
594 	if (n & POLLIN) {
595 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
596 		set_bit(Rpending, &m->wsched);
597 	}
598 
599 	if (n & POLLOUT) {
600 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
601 		set_bit(Wpending, &m->wsched);
602 	}
603 
604 	return m;
605 }
606 
607 /**
608  * p9_poll_mux - polls a mux and schedules read or write works if necessary
609  * @m: connection to poll
610  *
611  */
612 
613 static void p9_poll_mux(struct p9_conn *m)
614 {
615 	int n;
616 
617 	if (m->err < 0)
618 		return;
619 
620 	n = p9_fd_poll(m->client, NULL);
621 	if (n < 0 || n & (POLLERR | POLLHUP | POLLNVAL)) {
622 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
623 		if (n >= 0)
624 			n = -ECONNRESET;
625 		p9_conn_cancel(m, n);
626 	}
627 
628 	if (n & POLLIN) {
629 		set_bit(Rpending, &m->wsched);
630 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
631 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
632 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
633 			schedule_work(&m->rq);
634 		}
635 	}
636 
637 	if (n & POLLOUT) {
638 		set_bit(Wpending, &m->wsched);
639 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
640 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
641 		    !test_and_set_bit(Wworksched, &m->wsched)) {
642 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
643 			schedule_work(&m->wq);
644 		}
645 	}
646 }
647 
648 /**
649  * p9_fd_request - send 9P request
650  * The function can sleep until the request is scheduled for sending.
651  * The function can be interrupted. Return from the function is not
652  * a guarantee that the request is sent successfully.
653  *
654  * @client: client instance
655  * @req: request to be sent
656  *
657  */
658 
659 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
660 {
661 	int n;
662 	struct p9_trans_fd *ts = client->trans;
663 	struct p9_conn *m = ts->conn;
664 
665 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
666 		 m, current, req->tc, req->tc->id);
667 	if (m->err < 0)
668 		return m->err;
669 
670 	spin_lock(&client->lock);
671 	req->status = REQ_STATUS_UNSENT;
672 	list_add_tail(&req->req_list, &m->unsent_req_list);
673 	spin_unlock(&client->lock);
674 
675 	if (test_and_clear_bit(Wpending, &m->wsched))
676 		n = POLLOUT;
677 	else
678 		n = p9_fd_poll(m->client, NULL);
679 
680 	if (n & POLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
681 		schedule_work(&m->wq);
682 
683 	return 0;
684 }
685 
686 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
687 {
688 	int ret = 1;
689 
690 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
691 
692 	spin_lock(&client->lock);
693 
694 	if (req->status == REQ_STATUS_UNSENT) {
695 		list_del(&req->req_list);
696 		req->status = REQ_STATUS_FLSHD;
697 		ret = 0;
698 	} else if (req->status == REQ_STATUS_SENT)
699 		req->status = REQ_STATUS_FLSH;
700 
701 	spin_unlock(&client->lock);
702 
703 	return ret;
704 }
705 
706 /**
707  * parse_opts - parse mount options into p9_fd_opts structure
708  * @params: options string passed from mount
709  * @opts: fd transport-specific structure to parse options into
710  *
711  * Returns 0 upon success, -ERRNO upon failure
712  */
713 
714 static int parse_opts(char *params, struct p9_fd_opts *opts)
715 {
716 	char *p;
717 	substring_t args[MAX_OPT_ARGS];
718 	int option;
719 	char *options, *tmp_options;
720 
721 	opts->port = P9_PORT;
722 	opts->rfd = ~0;
723 	opts->wfd = ~0;
724 
725 	if (!params)
726 		return 0;
727 
728 	tmp_options = kstrdup(params, GFP_KERNEL);
729 	if (!tmp_options) {
730 		p9_debug(P9_DEBUG_ERROR,
731 			 "failed to allocate copy of option string\n");
732 		return -ENOMEM;
733 	}
734 	options = tmp_options;
735 
736 	while ((p = strsep(&options, ",")) != NULL) {
737 		int token;
738 		int r;
739 		if (!*p)
740 			continue;
741 		token = match_token(p, tokens, args);
742 		if (token != Opt_err) {
743 			r = match_int(&args[0], &option);
744 			if (r < 0) {
745 				p9_debug(P9_DEBUG_ERROR,
746 					 "integer field, but no integer?\n");
747 				continue;
748 			}
749 		}
750 		switch (token) {
751 		case Opt_port:
752 			opts->port = option;
753 			break;
754 		case Opt_rfdno:
755 			opts->rfd = option;
756 			break;
757 		case Opt_wfdno:
758 			opts->wfd = option;
759 			break;
760 		default:
761 			continue;
762 		}
763 	}
764 
765 	kfree(tmp_options);
766 	return 0;
767 }
768 
769 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
770 {
771 	struct p9_trans_fd *ts = kmalloc(sizeof(struct p9_trans_fd),
772 					   GFP_KERNEL);
773 	if (!ts)
774 		return -ENOMEM;
775 
776 	ts->rd = fget(rfd);
777 	ts->wr = fget(wfd);
778 	if (!ts->rd || !ts->wr) {
779 		if (ts->rd)
780 			fput(ts->rd);
781 		if (ts->wr)
782 			fput(ts->wr);
783 		kfree(ts);
784 		return -EIO;
785 	}
786 
787 	client->trans = ts;
788 	client->status = Connected;
789 
790 	return 0;
791 }
792 
793 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
794 {
795 	struct p9_trans_fd *p;
796 	int ret, fd;
797 
798 	p = kmalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
799 	if (!p)
800 		return -ENOMEM;
801 
802 	csocket->sk->sk_allocation = GFP_NOIO;
803 	fd = sock_map_fd(csocket, 0);
804 	if (fd < 0) {
805 		pr_err("%s (%d): failed to map fd\n",
806 		       __func__, task_pid_nr(current));
807 		sock_release(csocket);
808 		kfree(p);
809 		return fd;
810 	}
811 
812 	get_file(csocket->file);
813 	get_file(csocket->file);
814 	p->wr = p->rd = csocket->file;
815 	client->trans = p;
816 	client->status = Connected;
817 
818 	sys_close(fd);	/* still racy */
819 
820 	p->rd->f_flags |= O_NONBLOCK;
821 
822 	p->conn = p9_conn_create(client);
823 	if (IS_ERR(p->conn)) {
824 		ret = PTR_ERR(p->conn);
825 		p->conn = NULL;
826 		kfree(p);
827 		sockfd_put(csocket);
828 		sockfd_put(csocket);
829 		return ret;
830 	}
831 	return 0;
832 }
833 
834 /**
835  * p9_mux_destroy - cancels all pending requests and frees mux resources
836  * @m: mux to destroy
837  *
838  */
839 
840 static void p9_conn_destroy(struct p9_conn *m)
841 {
842 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
843 		 m, m->mux_list.prev, m->mux_list.next);
844 
845 	p9_mux_poll_stop(m);
846 	cancel_work_sync(&m->rq);
847 	cancel_work_sync(&m->wq);
848 
849 	p9_conn_cancel(m, -ECONNRESET);
850 
851 	m->client = NULL;
852 	kfree(m);
853 }
854 
855 /**
856  * p9_fd_close - shutdown file descriptor transport
857  * @client: client instance
858  *
859  */
860 
861 static void p9_fd_close(struct p9_client *client)
862 {
863 	struct p9_trans_fd *ts;
864 
865 	if (!client)
866 		return;
867 
868 	ts = client->trans;
869 	if (!ts)
870 		return;
871 
872 	client->status = Disconnected;
873 
874 	p9_conn_destroy(ts->conn);
875 
876 	if (ts->rd)
877 		fput(ts->rd);
878 	if (ts->wr)
879 		fput(ts->wr);
880 
881 	kfree(ts);
882 }
883 
884 /*
885  * stolen from NFS - maybe should be made a generic function?
886  */
887 static inline int valid_ipaddr4(const char *buf)
888 {
889 	int rc, count, in[4];
890 
891 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
892 	if (rc != 4)
893 		return -EINVAL;
894 	for (count = 0; count < 4; count++) {
895 		if (in[count] > 255)
896 			return -EINVAL;
897 	}
898 	return 0;
899 }
900 
901 static int
902 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
903 {
904 	int err;
905 	struct socket *csocket;
906 	struct sockaddr_in sin_server;
907 	struct p9_fd_opts opts;
908 
909 	err = parse_opts(args, &opts);
910 	if (err < 0)
911 		return err;
912 
913 	if (valid_ipaddr4(addr) < 0)
914 		return -EINVAL;
915 
916 	csocket = NULL;
917 
918 	sin_server.sin_family = AF_INET;
919 	sin_server.sin_addr.s_addr = in_aton(addr);
920 	sin_server.sin_port = htons(opts.port);
921 	err = __sock_create(read_pnet(&current->nsproxy->net_ns), PF_INET,
922 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
923 	if (err) {
924 		pr_err("%s (%d): problem creating socket\n",
925 		       __func__, task_pid_nr(current));
926 		return err;
927 	}
928 
929 	err = csocket->ops->connect(csocket,
930 				    (struct sockaddr *)&sin_server,
931 				    sizeof(struct sockaddr_in), 0);
932 	if (err < 0) {
933 		pr_err("%s (%d): problem connecting socket to %s\n",
934 		       __func__, task_pid_nr(current), addr);
935 		sock_release(csocket);
936 		return err;
937 	}
938 
939 	return p9_socket_open(client, csocket);
940 }
941 
942 static int
943 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
944 {
945 	int err;
946 	struct socket *csocket;
947 	struct sockaddr_un sun_server;
948 
949 	csocket = NULL;
950 
951 	if (strlen(addr) >= UNIX_PATH_MAX) {
952 		pr_err("%s (%d): address too long: %s\n",
953 		       __func__, task_pid_nr(current), addr);
954 		return -ENAMETOOLONG;
955 	}
956 
957 	sun_server.sun_family = PF_UNIX;
958 	strcpy(sun_server.sun_path, addr);
959 	err = __sock_create(read_pnet(&current->nsproxy->net_ns), PF_UNIX,
960 			    SOCK_STREAM, 0, &csocket, 1);
961 	if (err < 0) {
962 		pr_err("%s (%d): problem creating socket\n",
963 		       __func__, task_pid_nr(current));
964 
965 		return err;
966 	}
967 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
968 			sizeof(struct sockaddr_un) - 1, 0);
969 	if (err < 0) {
970 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
971 		       __func__, task_pid_nr(current), addr, err);
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(struct p9_client *client, const char *addr, char *args)
981 {
982 	int err;
983 	struct p9_fd_opts opts;
984 	struct p9_trans_fd *p;
985 
986 	parse_opts(args, &opts);
987 
988 	if (opts.rfd == ~0 || opts.wfd == ~0) {
989 		pr_err("Insufficient options for proto=fd\n");
990 		return -ENOPROTOOPT;
991 	}
992 
993 	err = p9_fd_open(client, opts.rfd, opts.wfd);
994 	if (err < 0)
995 		return err;
996 
997 	p = (struct p9_trans_fd *) client->trans;
998 	p->conn = p9_conn_create(client);
999 	if (IS_ERR(p->conn)) {
1000 		err = PTR_ERR(p->conn);
1001 		p->conn = NULL;
1002 		fput(p->rd);
1003 		fput(p->wr);
1004 		return err;
1005 	}
1006 
1007 	return 0;
1008 }
1009 
1010 static struct p9_trans_module p9_tcp_trans = {
1011 	.name = "tcp",
1012 	.maxsize = MAX_SOCK_BUF,
1013 	.def = 1,
1014 	.create = p9_fd_create_tcp,
1015 	.close = p9_fd_close,
1016 	.request = p9_fd_request,
1017 	.cancel = p9_fd_cancel,
1018 	.owner = THIS_MODULE,
1019 };
1020 
1021 static struct p9_trans_module p9_unix_trans = {
1022 	.name = "unix",
1023 	.maxsize = MAX_SOCK_BUF,
1024 	.def = 0,
1025 	.create = p9_fd_create_unix,
1026 	.close = p9_fd_close,
1027 	.request = p9_fd_request,
1028 	.cancel = p9_fd_cancel,
1029 	.owner = THIS_MODULE,
1030 };
1031 
1032 static struct p9_trans_module p9_fd_trans = {
1033 	.name = "fd",
1034 	.maxsize = MAX_SOCK_BUF,
1035 	.def = 0,
1036 	.create = p9_fd_create,
1037 	.close = p9_fd_close,
1038 	.request = p9_fd_request,
1039 	.cancel = p9_fd_cancel,
1040 	.owner = THIS_MODULE,
1041 };
1042 
1043 /**
1044  * p9_poll_proc - poll worker thread
1045  * @a: thread state and arguments
1046  *
1047  * polls all v9fs transports for new events and queues the appropriate
1048  * work to the work queue
1049  *
1050  */
1051 
1052 static void p9_poll_workfn(struct work_struct *work)
1053 {
1054 	unsigned long flags;
1055 
1056 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1057 
1058 	spin_lock_irqsave(&p9_poll_lock, flags);
1059 	while (!list_empty(&p9_poll_pending_list)) {
1060 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1061 							struct p9_conn,
1062 							poll_pending_link);
1063 		list_del_init(&conn->poll_pending_link);
1064 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1065 
1066 		p9_poll_mux(conn);
1067 
1068 		spin_lock_irqsave(&p9_poll_lock, flags);
1069 	}
1070 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1071 
1072 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1073 }
1074 
1075 int p9_trans_fd_init(void)
1076 {
1077 	v9fs_register_trans(&p9_tcp_trans);
1078 	v9fs_register_trans(&p9_unix_trans);
1079 	v9fs_register_trans(&p9_fd_trans);
1080 
1081 	return 0;
1082 }
1083 
1084 void p9_trans_fd_exit(void)
1085 {
1086 	flush_work_sync(&p9_poll_work);
1087 	v9fs_unregister_trans(&p9_tcp_trans);
1088 	v9fs_unregister_trans(&p9_unix_trans);
1089 	v9fs_unregister_trans(&p9_fd_trans);
1090 }
1091