xref: /openbmc/linux/net/9p/trans_fd.c (revision 565d76cb)
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 #include <linux/in.h>
29 #include <linux/module.h>
30 #include <linux/net.h>
31 #include <linux/ipv6.h>
32 #include <linux/kthread.h>
33 #include <linux/errno.h>
34 #include <linux/kernel.h>
35 #include <linux/un.h>
36 #include <linux/uaccess.h>
37 #include <linux/inet.h>
38 #include <linux/idr.h>
39 #include <linux/file.h>
40 #include <linux/parser.h>
41 #include <linux/slab.h>
42 #include <net/9p/9p.h>
43 #include <net/9p/client.h>
44 #include <net/9p/transport.h>
45 
46 #include <linux/syscalls.h> /* killme */
47 
48 #define P9_PORT 564
49 #define MAX_SOCK_BUF (64*1024)
50 #define MAXPOLLWADDR	2
51 
52 /**
53  * struct p9_fd_opts - per-transport options
54  * @rfd: file descriptor for reading (trans=fd)
55  * @wfd: file descriptor for writing (trans=fd)
56  * @port: port to connect to (trans=tcp)
57  *
58  */
59 
60 struct p9_fd_opts {
61 	int rfd;
62 	int wfd;
63 	u16 port;
64 };
65 
66 /**
67  * struct p9_trans_fd - transport state
68  * @rd: reference to file to read from
69  * @wr: reference of file to write to
70  * @conn: connection state reference
71  *
72  */
73 
74 struct p9_trans_fd {
75 	struct file *rd;
76 	struct file *wr;
77 	struct p9_conn *conn;
78 };
79 
80 /*
81   * Option Parsing (code inspired by NFS code)
82   *  - a little lazy - parse all fd-transport options
83   */
84 
85 enum {
86 	/* Options that take integer arguments */
87 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
88 };
89 
90 static const match_table_t tokens = {
91 	{Opt_port, "port=%u"},
92 	{Opt_rfdno, "rfdno=%u"},
93 	{Opt_wfdno, "wfdno=%u"},
94 	{Opt_err, NULL},
95 };
96 
97 enum {
98 	Rworksched = 1,		/* read work scheduled or running */
99 	Rpending = 2,		/* can read */
100 	Wworksched = 4,		/* write work scheduled or running */
101 	Wpending = 8,		/* can write */
102 };
103 
104 struct p9_poll_wait {
105 	struct p9_conn *conn;
106 	wait_queue_t wait;
107 	wait_queue_head_t *wait_addr;
108 };
109 
110 /**
111  * struct p9_conn - fd mux connection state information
112  * @mux_list: list link for mux to manage multiple connections (?)
113  * @client: reference to client instance for this connection
114  * @err: error state
115  * @req_list: accounting for requests which have been sent
116  * @unsent_req_list: accounting for requests that haven't been sent
117  * @req: current request being processed (if any)
118  * @tmp_buf: temporary buffer to read in header
119  * @rsize: amount to read for current frame
120  * @rpos: read position in current frame
121  * @rbuf: current read buffer
122  * @wpos: write position for current frame
123  * @wsize: amount of data to write for current frame
124  * @wbuf: current write buffer
125  * @poll_pending_link: pending links to be polled per conn
126  * @poll_wait: array of wait_q's for various worker threads
127  * @pt: poll state
128  * @rq: current read work
129  * @wq: current write work
130  * @wsched: ????
131  *
132  */
133 
134 struct p9_conn {
135 	struct list_head mux_list;
136 	struct p9_client *client;
137 	int err;
138 	struct list_head req_list;
139 	struct list_head unsent_req_list;
140 	struct p9_req_t *req;
141 	char tmp_buf[7];
142 	int rsize;
143 	int rpos;
144 	char *rbuf;
145 	int wpos;
146 	int wsize;
147 	char *wbuf;
148 	struct list_head poll_pending_link;
149 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
150 	poll_table pt;
151 	struct work_struct rq;
152 	struct work_struct wq;
153 	unsigned long wsched;
154 };
155 
156 static void p9_poll_workfn(struct work_struct *work);
157 
158 static DEFINE_SPINLOCK(p9_poll_lock);
159 static LIST_HEAD(p9_poll_pending_list);
160 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
161 
162 static void p9_mux_poll_stop(struct p9_conn *m)
163 {
164 	unsigned long flags;
165 	int i;
166 
167 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
168 		struct p9_poll_wait *pwait = &m->poll_wait[i];
169 
170 		if (pwait->wait_addr) {
171 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
172 			pwait->wait_addr = NULL;
173 		}
174 	}
175 
176 	spin_lock_irqsave(&p9_poll_lock, flags);
177 	list_del_init(&m->poll_pending_link);
178 	spin_unlock_irqrestore(&p9_poll_lock, flags);
179 }
180 
181 /**
182  * p9_conn_cancel - cancel all pending requests with error
183  * @m: mux data
184  * @err: error code
185  *
186  */
187 
188 static void p9_conn_cancel(struct p9_conn *m, int err)
189 {
190 	struct p9_req_t *req, *rtmp;
191 	unsigned long flags;
192 	LIST_HEAD(cancel_list);
193 
194 	P9_DPRINTK(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
195 
196 	spin_lock_irqsave(&m->client->lock, flags);
197 
198 	if (m->err) {
199 		spin_unlock_irqrestore(&m->client->lock, flags);
200 		return;
201 	}
202 
203 	m->err = err;
204 
205 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
206 		req->status = REQ_STATUS_ERROR;
207 		if (!req->t_err)
208 			req->t_err = err;
209 		list_move(&req->req_list, &cancel_list);
210 	}
211 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
212 		req->status = REQ_STATUS_ERROR;
213 		if (!req->t_err)
214 			req->t_err = err;
215 		list_move(&req->req_list, &cancel_list);
216 	}
217 	spin_unlock_irqrestore(&m->client->lock, flags);
218 
219 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
220 		P9_DPRINTK(P9_DEBUG_ERROR, "call back req %p\n", req);
221 		list_del(&req->req_list);
222 		p9_client_cb(m->client, req);
223 	}
224 }
225 
226 static int
227 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt)
228 {
229 	int ret, n;
230 	struct p9_trans_fd *ts = NULL;
231 
232 	if (client && client->status == Connected)
233 		ts = client->trans;
234 
235 	if (!ts)
236 		return -EREMOTEIO;
237 
238 	if (!ts->rd->f_op || !ts->rd->f_op->poll)
239 		return -EIO;
240 
241 	if (!ts->wr->f_op || !ts->wr->f_op->poll)
242 		return -EIO;
243 
244 	ret = ts->rd->f_op->poll(ts->rd, pt);
245 	if (ret < 0)
246 		return ret;
247 
248 	if (ts->rd != ts->wr) {
249 		n = ts->wr->f_op->poll(ts->wr, pt);
250 		if (n < 0)
251 			return n;
252 		ret = (ret & ~POLLOUT) | (n & ~POLLIN);
253 	}
254 
255 	return ret;
256 }
257 
258 /**
259  * p9_fd_read- read from a fd
260  * @client: client instance
261  * @v: buffer to receive data into
262  * @len: size of receive buffer
263  *
264  */
265 
266 static int p9_fd_read(struct p9_client *client, void *v, int len)
267 {
268 	int ret;
269 	struct p9_trans_fd *ts = NULL;
270 
271 	if (client && client->status != Disconnected)
272 		ts = client->trans;
273 
274 	if (!ts)
275 		return -EREMOTEIO;
276 
277 	if (!(ts->rd->f_flags & O_NONBLOCK))
278 		P9_DPRINTK(P9_DEBUG_ERROR, "blocking read ...\n");
279 
280 	ret = kernel_read(ts->rd, ts->rd->f_pos, v, len);
281 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
282 		client->status = Disconnected;
283 	return ret;
284 }
285 
286 /**
287  * p9_read_work - called when there is some data to be read from a transport
288  * @work: container of work to be done
289  *
290  */
291 
292 static void p9_read_work(struct work_struct *work)
293 {
294 	int n, err;
295 	struct p9_conn *m;
296 
297 	m = container_of(work, struct p9_conn, rq);
298 
299 	if (m->err < 0)
300 		return;
301 
302 	P9_DPRINTK(P9_DEBUG_TRANS, "start mux %p pos %d\n", m, m->rpos);
303 
304 	if (!m->rbuf) {
305 		m->rbuf = m->tmp_buf;
306 		m->rpos = 0;
307 		m->rsize = 7; /* start by reading header */
308 	}
309 
310 	clear_bit(Rpending, &m->wsched);
311 	P9_DPRINTK(P9_DEBUG_TRANS, "read mux %p pos %d size: %d = %d\n", m,
312 					m->rpos, m->rsize, m->rsize-m->rpos);
313 	err = p9_fd_read(m->client, m->rbuf + m->rpos,
314 						m->rsize - m->rpos);
315 	P9_DPRINTK(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
316 	if (err == -EAGAIN) {
317 		clear_bit(Rworksched, &m->wsched);
318 		return;
319 	}
320 
321 	if (err <= 0)
322 		goto error;
323 
324 	m->rpos += err;
325 
326 	if ((!m->req) && (m->rpos == m->rsize)) { /* header read in */
327 		u16 tag;
328 		P9_DPRINTK(P9_DEBUG_TRANS, "got new header\n");
329 
330 		n = le32_to_cpu(*(__le32 *) m->rbuf); /* read packet size */
331 		if (n >= m->client->msize) {
332 			P9_DPRINTK(P9_DEBUG_ERROR,
333 				"requested packet size too big: %d\n", n);
334 			err = -EIO;
335 			goto error;
336 		}
337 
338 		tag = le16_to_cpu(*(__le16 *) (m->rbuf+5)); /* read tag */
339 		P9_DPRINTK(P9_DEBUG_TRANS,
340 			"mux %p pkt: size: %d bytes tag: %d\n", m, n, tag);
341 
342 		m->req = p9_tag_lookup(m->client, tag);
343 		if (!m->req || (m->req->status != REQ_STATUS_SENT &&
344 					m->req->status != REQ_STATUS_FLSH)) {
345 			P9_DPRINTK(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
346 								 tag);
347 			err = -EIO;
348 			goto error;
349 		}
350 
351 		if (m->req->rc == NULL) {
352 			m->req->rc = kmalloc(sizeof(struct p9_fcall) +
353 						m->client->msize, GFP_NOFS);
354 			if (!m->req->rc) {
355 				m->req = NULL;
356 				err = -ENOMEM;
357 				goto error;
358 			}
359 		}
360 		m->rbuf = (char *)m->req->rc + sizeof(struct p9_fcall);
361 		memcpy(m->rbuf, m->tmp_buf, m->rsize);
362 		m->rsize = n;
363 	}
364 
365 	/* not an else because some packets (like clunk) have no payload */
366 	if ((m->req) && (m->rpos == m->rsize)) { /* packet is read in */
367 		P9_DPRINTK(P9_DEBUG_TRANS, "got new packet\n");
368 		spin_lock(&m->client->lock);
369 		if (m->req->status != REQ_STATUS_ERROR)
370 			m->req->status = REQ_STATUS_RCVD;
371 		list_del(&m->req->req_list);
372 		spin_unlock(&m->client->lock);
373 		p9_client_cb(m->client, m->req);
374 		m->rbuf = NULL;
375 		m->rpos = 0;
376 		m->rsize = 0;
377 		m->req = NULL;
378 	}
379 
380 	if (!list_empty(&m->req_list)) {
381 		if (test_and_clear_bit(Rpending, &m->wsched))
382 			n = POLLIN;
383 		else
384 			n = p9_fd_poll(m->client, NULL);
385 
386 		if (n & POLLIN) {
387 			P9_DPRINTK(P9_DEBUG_TRANS, "sched read work %p\n", m);
388 			schedule_work(&m->rq);
389 		} else
390 			clear_bit(Rworksched, &m->wsched);
391 	} else
392 		clear_bit(Rworksched, &m->wsched);
393 
394 	return;
395 error:
396 	p9_conn_cancel(m, err);
397 	clear_bit(Rworksched, &m->wsched);
398 }
399 
400 /**
401  * p9_fd_write - write to a socket
402  * @client: client instance
403  * @v: buffer to send data from
404  * @len: size of send buffer
405  *
406  */
407 
408 static int p9_fd_write(struct p9_client *client, void *v, int len)
409 {
410 	int ret;
411 	mm_segment_t oldfs;
412 	struct p9_trans_fd *ts = NULL;
413 
414 	if (client && client->status != Disconnected)
415 		ts = client->trans;
416 
417 	if (!ts)
418 		return -EREMOTEIO;
419 
420 	if (!(ts->wr->f_flags & O_NONBLOCK))
421 		P9_DPRINTK(P9_DEBUG_ERROR, "blocking write ...\n");
422 
423 	oldfs = get_fs();
424 	set_fs(get_ds());
425 	/* The cast to a user pointer is valid due to the set_fs() */
426 	ret = vfs_write(ts->wr, (__force void __user *)v, len, &ts->wr->f_pos);
427 	set_fs(oldfs);
428 
429 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
430 		client->status = Disconnected;
431 	return ret;
432 }
433 
434 /**
435  * p9_write_work - called when a transport can send some data
436  * @work: container for work to be done
437  *
438  */
439 
440 static void p9_write_work(struct work_struct *work)
441 {
442 	int n, err;
443 	struct p9_conn *m;
444 	struct p9_req_t *req;
445 
446 	m = container_of(work, struct p9_conn, wq);
447 
448 	if (m->err < 0) {
449 		clear_bit(Wworksched, &m->wsched);
450 		return;
451 	}
452 
453 	if (!m->wsize) {
454 		if (list_empty(&m->unsent_req_list)) {
455 			clear_bit(Wworksched, &m->wsched);
456 			return;
457 		}
458 
459 		spin_lock(&m->client->lock);
460 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
461 			       req_list);
462 		req->status = REQ_STATUS_SENT;
463 		P9_DPRINTK(P9_DEBUG_TRANS, "move req %p\n", req);
464 		list_move_tail(&req->req_list, &m->req_list);
465 
466 		m->wbuf = req->tc->sdata;
467 		m->wsize = req->tc->size;
468 		m->wpos = 0;
469 		spin_unlock(&m->client->lock);
470 	}
471 
472 	P9_DPRINTK(P9_DEBUG_TRANS, "mux %p pos %d size %d\n", m, m->wpos,
473 								m->wsize);
474 	clear_bit(Wpending, &m->wsched);
475 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
476 	P9_DPRINTK(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
477 	if (err == -EAGAIN) {
478 		clear_bit(Wworksched, &m->wsched);
479 		return;
480 	}
481 
482 	if (err < 0)
483 		goto error;
484 	else if (err == 0) {
485 		err = -EREMOTEIO;
486 		goto error;
487 	}
488 
489 	m->wpos += err;
490 	if (m->wpos == m->wsize)
491 		m->wpos = m->wsize = 0;
492 
493 	if (m->wsize == 0 && !list_empty(&m->unsent_req_list)) {
494 		if (test_and_clear_bit(Wpending, &m->wsched))
495 			n = POLLOUT;
496 		else
497 			n = p9_fd_poll(m->client, NULL);
498 
499 		if (n & POLLOUT) {
500 			P9_DPRINTK(P9_DEBUG_TRANS, "sched write work %p\n", m);
501 			schedule_work(&m->wq);
502 		} else
503 			clear_bit(Wworksched, &m->wsched);
504 	} else
505 		clear_bit(Wworksched, &m->wsched);
506 
507 	return;
508 
509 error:
510 	p9_conn_cancel(m, err);
511 	clear_bit(Wworksched, &m->wsched);
512 }
513 
514 static int p9_pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
515 {
516 	struct p9_poll_wait *pwait =
517 		container_of(wait, struct p9_poll_wait, wait);
518 	struct p9_conn *m = pwait->conn;
519 	unsigned long flags;
520 
521 	spin_lock_irqsave(&p9_poll_lock, flags);
522 	if (list_empty(&m->poll_pending_link))
523 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
524 	spin_unlock_irqrestore(&p9_poll_lock, flags);
525 
526 	schedule_work(&p9_poll_work);
527 	return 1;
528 }
529 
530 /**
531  * p9_pollwait - add poll task to the wait queue
532  * @filp: file pointer being polled
533  * @wait_address: wait_q to block on
534  * @p: poll state
535  *
536  * called by files poll operation to add v9fs-poll task to files wait queue
537  */
538 
539 static void
540 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
541 {
542 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
543 	struct p9_poll_wait *pwait = NULL;
544 	int i;
545 
546 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
547 		if (m->poll_wait[i].wait_addr == NULL) {
548 			pwait = &m->poll_wait[i];
549 			break;
550 		}
551 	}
552 
553 	if (!pwait) {
554 		P9_DPRINTK(P9_DEBUG_ERROR, "not enough wait_address slots\n");
555 		return;
556 	}
557 
558 	pwait->conn = m;
559 	pwait->wait_addr = wait_address;
560 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
561 	add_wait_queue(wait_address, &pwait->wait);
562 }
563 
564 /**
565  * p9_conn_create - allocate and initialize the per-session mux data
566  * @client: client instance
567  *
568  * Note: Creates the polling task if this is the first session.
569  */
570 
571 static struct p9_conn *p9_conn_create(struct p9_client *client)
572 {
573 	int n;
574 	struct p9_conn *m;
575 
576 	P9_DPRINTK(P9_DEBUG_TRANS, "client %p msize %d\n", client,
577 								client->msize);
578 	m = kzalloc(sizeof(struct p9_conn), GFP_KERNEL);
579 	if (!m)
580 		return ERR_PTR(-ENOMEM);
581 
582 	INIT_LIST_HEAD(&m->mux_list);
583 	m->client = client;
584 
585 	INIT_LIST_HEAD(&m->req_list);
586 	INIT_LIST_HEAD(&m->unsent_req_list);
587 	INIT_WORK(&m->rq, p9_read_work);
588 	INIT_WORK(&m->wq, p9_write_work);
589 	INIT_LIST_HEAD(&m->poll_pending_link);
590 	init_poll_funcptr(&m->pt, p9_pollwait);
591 
592 	n = p9_fd_poll(client, &m->pt);
593 	if (n & POLLIN) {
594 		P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can read\n", m);
595 		set_bit(Rpending, &m->wsched);
596 	}
597 
598 	if (n & POLLOUT) {
599 		P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can write\n", m);
600 		set_bit(Wpending, &m->wsched);
601 	}
602 
603 	return m;
604 }
605 
606 /**
607  * p9_poll_mux - polls a mux and schedules read or write works if necessary
608  * @m: connection to poll
609  *
610  */
611 
612 static void p9_poll_mux(struct p9_conn *m)
613 {
614 	int n;
615 
616 	if (m->err < 0)
617 		return;
618 
619 	n = p9_fd_poll(m->client, NULL);
620 	if (n < 0 || n & (POLLERR | POLLHUP | POLLNVAL)) {
621 		P9_DPRINTK(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
622 		if (n >= 0)
623 			n = -ECONNRESET;
624 		p9_conn_cancel(m, n);
625 	}
626 
627 	if (n & POLLIN) {
628 		set_bit(Rpending, &m->wsched);
629 		P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can read\n", m);
630 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
631 			P9_DPRINTK(P9_DEBUG_TRANS, "sched read work %p\n", m);
632 			schedule_work(&m->rq);
633 		}
634 	}
635 
636 	if (n & POLLOUT) {
637 		set_bit(Wpending, &m->wsched);
638 		P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can write\n", m);
639 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
640 		    !test_and_set_bit(Wworksched, &m->wsched)) {
641 			P9_DPRINTK(P9_DEBUG_TRANS, "sched write work %p\n", m);
642 			schedule_work(&m->wq);
643 		}
644 	}
645 }
646 
647 /**
648  * p9_fd_request - send 9P request
649  * The function can sleep until the request is scheduled for sending.
650  * The function can be interrupted. Return from the function is not
651  * a guarantee that the request is sent successfully.
652  *
653  * @client: client instance
654  * @req: request to be sent
655  *
656  */
657 
658 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
659 {
660 	int n;
661 	struct p9_trans_fd *ts = client->trans;
662 	struct p9_conn *m = ts->conn;
663 
664 	P9_DPRINTK(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n", m,
665 						current, req->tc, req->tc->id);
666 	if (m->err < 0)
667 		return m->err;
668 
669 	spin_lock(&client->lock);
670 	req->status = REQ_STATUS_UNSENT;
671 	list_add_tail(&req->req_list, &m->unsent_req_list);
672 	spin_unlock(&client->lock);
673 
674 	if (test_and_clear_bit(Wpending, &m->wsched))
675 		n = POLLOUT;
676 	else
677 		n = p9_fd_poll(m->client, NULL);
678 
679 	if (n & POLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
680 		schedule_work(&m->wq);
681 
682 	return 0;
683 }
684 
685 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
686 {
687 	int ret = 1;
688 
689 	P9_DPRINTK(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
690 
691 	spin_lock(&client->lock);
692 
693 	if (req->status == REQ_STATUS_UNSENT) {
694 		list_del(&req->req_list);
695 		req->status = REQ_STATUS_FLSHD;
696 		ret = 0;
697 	} else if (req->status == REQ_STATUS_SENT)
698 		req->status = REQ_STATUS_FLSH;
699 
700 	spin_unlock(&client->lock);
701 
702 	return ret;
703 }
704 
705 /**
706  * parse_opts - parse mount options into p9_fd_opts structure
707  * @params: options string passed from mount
708  * @opts: fd transport-specific structure to parse options into
709  *
710  * Returns 0 upon success, -ERRNO upon failure
711  */
712 
713 static int parse_opts(char *params, struct p9_fd_opts *opts)
714 {
715 	char *p;
716 	substring_t args[MAX_OPT_ARGS];
717 	int option;
718 	char *options, *tmp_options;
719 	int ret;
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_DPRINTK(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_DPRINTK(P9_DEBUG_ERROR,
746 				"integer field, but no integer?\n");
747 				ret = r;
748 				continue;
749 			}
750 		}
751 		switch (token) {
752 		case Opt_port:
753 			opts->port = option;
754 			break;
755 		case Opt_rfdno:
756 			opts->rfd = option;
757 			break;
758 		case Opt_wfdno:
759 			opts->wfd = option;
760 			break;
761 		default:
762 			continue;
763 		}
764 	}
765 
766 	kfree(tmp_options);
767 	return 0;
768 }
769 
770 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
771 {
772 	struct p9_trans_fd *ts = kmalloc(sizeof(struct p9_trans_fd),
773 					   GFP_KERNEL);
774 	if (!ts)
775 		return -ENOMEM;
776 
777 	ts->rd = fget(rfd);
778 	ts->wr = fget(wfd);
779 	if (!ts->rd || !ts->wr) {
780 		if (ts->rd)
781 			fput(ts->rd);
782 		if (ts->wr)
783 			fput(ts->wr);
784 		kfree(ts);
785 		return -EIO;
786 	}
787 
788 	client->trans = ts;
789 	client->status = Connected;
790 
791 	return 0;
792 }
793 
794 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
795 {
796 	struct p9_trans_fd *p;
797 	int ret, fd;
798 
799 	p = kmalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
800 	if (!p)
801 		return -ENOMEM;
802 
803 	csocket->sk->sk_allocation = GFP_NOIO;
804 	fd = sock_map_fd(csocket, 0);
805 	if (fd < 0) {
806 		P9_EPRINTK(KERN_ERR, "p9_socket_open: failed to map fd\n");
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_DPRINTK(P9_DEBUG_TRANS, "mux %p prev %p next %p\n", m,
843 		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_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &csocket);
922 
923 	if (err) {
924 		P9_EPRINTK(KERN_ERR, "p9_trans_tcp: problem creating socket\n");
925 		return err;
926 	}
927 
928 	err = csocket->ops->connect(csocket,
929 				    (struct sockaddr *)&sin_server,
930 				    sizeof(struct sockaddr_in), 0);
931 	if (err < 0) {
932 		P9_EPRINTK(KERN_ERR,
933 			"p9_trans_tcp: problem connecting socket to %s\n",
934 			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 		P9_EPRINTK(KERN_ERR, "p9_trans_unix: address too long: %s\n",
953 			addr);
954 		return -ENAMETOOLONG;
955 	}
956 
957 	sun_server.sun_family = PF_UNIX;
958 	strcpy(sun_server.sun_path, addr);
959 	err = sock_create_kern(PF_UNIX, SOCK_STREAM, 0, &csocket);
960 	if (err < 0) {
961 		P9_EPRINTK(KERN_ERR, "p9_trans_unix: problem creating socket\n");
962 		return err;
963 	}
964 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
965 			sizeof(struct sockaddr_un) - 1, 0);
966 	if (err < 0) {
967 		P9_EPRINTK(KERN_ERR,
968 			"p9_trans_unix: problem connecting socket: %s: %d\n",
969 			addr, err);
970 		sock_release(csocket);
971 		return err;
972 	}
973 
974 	return p9_socket_open(client, csocket);
975 }
976 
977 static int
978 p9_fd_create(struct p9_client *client, const char *addr, char *args)
979 {
980 	int err;
981 	struct p9_fd_opts opts;
982 	struct p9_trans_fd *p;
983 
984 	parse_opts(args, &opts);
985 
986 	if (opts.rfd == ~0 || opts.wfd == ~0) {
987 		printk(KERN_ERR "v9fs: Insufficient options for proto=fd\n");
988 		return -ENOPROTOOPT;
989 	}
990 
991 	err = p9_fd_open(client, opts.rfd, opts.wfd);
992 	if (err < 0)
993 		return err;
994 
995 	p = (struct p9_trans_fd *) client->trans;
996 	p->conn = p9_conn_create(client);
997 	if (IS_ERR(p->conn)) {
998 		err = PTR_ERR(p->conn);
999 		p->conn = NULL;
1000 		fput(p->rd);
1001 		fput(p->wr);
1002 		return err;
1003 	}
1004 
1005 	return 0;
1006 }
1007 
1008 static struct p9_trans_module p9_tcp_trans = {
1009 	.name = "tcp",
1010 	.maxsize = MAX_SOCK_BUF,
1011 	.def = 1,
1012 	.create = p9_fd_create_tcp,
1013 	.close = p9_fd_close,
1014 	.request = p9_fd_request,
1015 	.cancel = p9_fd_cancel,
1016 	.owner = THIS_MODULE,
1017 };
1018 
1019 static struct p9_trans_module p9_unix_trans = {
1020 	.name = "unix",
1021 	.maxsize = MAX_SOCK_BUF,
1022 	.def = 0,
1023 	.create = p9_fd_create_unix,
1024 	.close = p9_fd_close,
1025 	.request = p9_fd_request,
1026 	.cancel = p9_fd_cancel,
1027 	.owner = THIS_MODULE,
1028 };
1029 
1030 static struct p9_trans_module p9_fd_trans = {
1031 	.name = "fd",
1032 	.maxsize = MAX_SOCK_BUF,
1033 	.def = 0,
1034 	.create = p9_fd_create,
1035 	.close = p9_fd_close,
1036 	.request = p9_fd_request,
1037 	.cancel = p9_fd_cancel,
1038 	.owner = THIS_MODULE,
1039 };
1040 
1041 /**
1042  * p9_poll_proc - poll worker thread
1043  * @a: thread state and arguments
1044  *
1045  * polls all v9fs transports for new events and queues the appropriate
1046  * work to the work queue
1047  *
1048  */
1049 
1050 static void p9_poll_workfn(struct work_struct *work)
1051 {
1052 	unsigned long flags;
1053 
1054 	P9_DPRINTK(P9_DEBUG_TRANS, "start %p\n", current);
1055 
1056 	spin_lock_irqsave(&p9_poll_lock, flags);
1057 	while (!list_empty(&p9_poll_pending_list)) {
1058 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1059 							struct p9_conn,
1060 							poll_pending_link);
1061 		list_del_init(&conn->poll_pending_link);
1062 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1063 
1064 		p9_poll_mux(conn);
1065 
1066 		spin_lock_irqsave(&p9_poll_lock, flags);
1067 	}
1068 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1069 
1070 	P9_DPRINTK(P9_DEBUG_TRANS, "finish\n");
1071 }
1072 
1073 int p9_trans_fd_init(void)
1074 {
1075 	v9fs_register_trans(&p9_tcp_trans);
1076 	v9fs_register_trans(&p9_unix_trans);
1077 	v9fs_register_trans(&p9_fd_trans);
1078 
1079 	return 0;
1080 }
1081 
1082 void p9_trans_fd_exit(void)
1083 {
1084 	flush_work_sync(&p9_poll_work);
1085 	v9fs_unregister_trans(&p9_tcp_trans);
1086 	v9fs_unregister_trans(&p9_unix_trans);
1087 	v9fs_unregister_trans(&p9_fd_trans);
1088 }
1089