xref: /openbmc/linux/net/9p/trans_fd.c (revision dea54fba)
1 /*
2  * linux/fs/9p/trans_fd.c
3  *
4  * Fd transport layer.  Includes deprecated socket layer.
5  *
6  *  Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
7  *  Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
8  *  Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
9  *  Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
10  *
11  *  This program is free software; you can redistribute it and/or modify
12  *  it under the terms of the GNU General Public License version 2
13  *  as published by the Free Software Foundation.
14  *
15  *  This program is distributed in the hope that it will be useful,
16  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
17  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  *  GNU General Public License for more details.
19  *
20  *  You should have received a copy of the GNU General Public License
21  *  along with this program; if not, write to:
22  *  Free Software Foundation
23  *  51 Franklin Street, Fifth Floor
24  *  Boston, MA  02111-1301  USA
25  *
26  */
27 
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 
30 #include <linux/in.h>
31 #include <linux/module.h>
32 #include <linux/net.h>
33 #include <linux/ipv6.h>
34 #include <linux/kthread.h>
35 #include <linux/errno.h>
36 #include <linux/kernel.h>
37 #include <linux/un.h>
38 #include <linux/uaccess.h>
39 #include <linux/inet.h>
40 #include <linux/idr.h>
41 #include <linux/file.h>
42 #include <linux/parser.h>
43 #include <linux/slab.h>
44 #include <linux/seq_file.h>
45 #include <net/9p/9p.h>
46 #include <net/9p/client.h>
47 #include <net/9p/transport.h>
48 
49 #include <linux/syscalls.h> /* killme */
50 
51 #define P9_PORT 564
52 #define MAX_SOCK_BUF (64*1024)
53 #define MAXPOLLWADDR	2
54 
55 static struct p9_trans_module p9_tcp_trans;
56 static struct p9_trans_module p9_fd_trans;
57 
58 /**
59  * struct p9_fd_opts - per-transport options
60  * @rfd: file descriptor for reading (trans=fd)
61  * @wfd: file descriptor for writing (trans=fd)
62  * @port: port to connect to (trans=tcp)
63  *
64  */
65 
66 struct p9_fd_opts {
67 	int rfd;
68 	int wfd;
69 	u16 port;
70 	bool privport;
71 };
72 
73 /*
74   * Option Parsing (code inspired by NFS code)
75   *  - a little lazy - parse all fd-transport options
76   */
77 
78 enum {
79 	/* Options that take integer arguments */
80 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
81 	/* Options that take no arguments */
82 	Opt_privport,
83 };
84 
85 static const match_table_t tokens = {
86 	{Opt_port, "port=%u"},
87 	{Opt_rfdno, "rfdno=%u"},
88 	{Opt_wfdno, "wfdno=%u"},
89 	{Opt_privport, "privport"},
90 	{Opt_err, NULL},
91 };
92 
93 enum {
94 	Rworksched = 1,		/* read work scheduled or running */
95 	Rpending = 2,		/* can read */
96 	Wworksched = 4,		/* write work scheduled or running */
97 	Wpending = 8,		/* can write */
98 };
99 
100 struct p9_poll_wait {
101 	struct p9_conn *conn;
102 	wait_queue_entry_t wait;
103 	wait_queue_head_t *wait_addr;
104 };
105 
106 /**
107  * struct p9_conn - fd mux connection state information
108  * @mux_list: list link for mux to manage multiple connections (?)
109  * @client: reference to client instance for this connection
110  * @err: error state
111  * @req_list: accounting for requests which have been sent
112  * @unsent_req_list: accounting for requests that haven't been sent
113  * @req: current request being processed (if any)
114  * @tmp_buf: temporary buffer to read in header
115  * @rc: temporary fcall for reading current frame
116  * @wpos: write position for current frame
117  * @wsize: amount of data to write for current frame
118  * @wbuf: current write buffer
119  * @poll_pending_link: pending links to be polled per conn
120  * @poll_wait: array of wait_q's for various worker threads
121  * @pt: poll state
122  * @rq: current read work
123  * @wq: current write work
124  * @wsched: ????
125  *
126  */
127 
128 struct p9_conn {
129 	struct list_head mux_list;
130 	struct p9_client *client;
131 	int err;
132 	struct list_head req_list;
133 	struct list_head unsent_req_list;
134 	struct p9_req_t *req;
135 	char tmp_buf[7];
136 	struct p9_fcall rc;
137 	int wpos;
138 	int wsize;
139 	char *wbuf;
140 	struct list_head poll_pending_link;
141 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
142 	poll_table pt;
143 	struct work_struct rq;
144 	struct work_struct wq;
145 	unsigned long wsched;
146 };
147 
148 /**
149  * struct p9_trans_fd - transport state
150  * @rd: reference to file to read from
151  * @wr: reference of file to write to
152  * @conn: connection state reference
153  *
154  */
155 
156 struct p9_trans_fd {
157 	struct file *rd;
158 	struct file *wr;
159 	struct p9_conn conn;
160 };
161 
162 static void p9_poll_workfn(struct work_struct *work);
163 
164 static DEFINE_SPINLOCK(p9_poll_lock);
165 static LIST_HEAD(p9_poll_pending_list);
166 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
167 
168 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
169 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
170 
171 static void p9_mux_poll_stop(struct p9_conn *m)
172 {
173 	unsigned long flags;
174 	int i;
175 
176 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
177 		struct p9_poll_wait *pwait = &m->poll_wait[i];
178 
179 		if (pwait->wait_addr) {
180 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
181 			pwait->wait_addr = NULL;
182 		}
183 	}
184 
185 	spin_lock_irqsave(&p9_poll_lock, flags);
186 	list_del_init(&m->poll_pending_link);
187 	spin_unlock_irqrestore(&p9_poll_lock, flags);
188 }
189 
190 /**
191  * p9_conn_cancel - cancel all pending requests with error
192  * @m: mux data
193  * @err: error code
194  *
195  */
196 
197 static void p9_conn_cancel(struct p9_conn *m, int err)
198 {
199 	struct p9_req_t *req, *rtmp;
200 	unsigned long flags;
201 	LIST_HEAD(cancel_list);
202 
203 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
204 
205 	spin_lock_irqsave(&m->client->lock, flags);
206 
207 	if (m->err) {
208 		spin_unlock_irqrestore(&m->client->lock, flags);
209 		return;
210 	}
211 
212 	m->err = err;
213 
214 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
215 		list_move(&req->req_list, &cancel_list);
216 	}
217 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
218 		list_move(&req->req_list, &cancel_list);
219 	}
220 	spin_unlock_irqrestore(&m->client->lock, flags);
221 
222 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
223 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
224 		list_del(&req->req_list);
225 		if (!req->t_err)
226 			req->t_err = err;
227 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
228 	}
229 }
230 
231 static int
232 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt)
233 {
234 	int ret, n;
235 	struct p9_trans_fd *ts = NULL;
236 
237 	if (client && client->status == Connected)
238 		ts = client->trans;
239 
240 	if (!ts)
241 		return -EREMOTEIO;
242 
243 	if (!ts->rd->f_op->poll)
244 		return -EIO;
245 
246 	if (!ts->wr->f_op->poll)
247 		return -EIO;
248 
249 	ret = ts->rd->f_op->poll(ts->rd, pt);
250 	if (ret < 0)
251 		return ret;
252 
253 	if (ts->rd != ts->wr) {
254 		n = ts->wr->f_op->poll(ts->wr, pt);
255 		if (n < 0)
256 			return n;
257 		ret = (ret & ~POLLOUT) | (n & ~POLLIN);
258 	}
259 
260 	return ret;
261 }
262 
263 /**
264  * p9_fd_read- read from a fd
265  * @client: client instance
266  * @v: buffer to receive data into
267  * @len: size of receive buffer
268  *
269  */
270 
271 static int p9_fd_read(struct p9_client *client, void *v, int len)
272 {
273 	int ret;
274 	struct p9_trans_fd *ts = NULL;
275 
276 	if (client && client->status != Disconnected)
277 		ts = client->trans;
278 
279 	if (!ts)
280 		return -EREMOTEIO;
281 
282 	if (!(ts->rd->f_flags & O_NONBLOCK))
283 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
284 
285 	ret = kernel_read(ts->rd, ts->rd->f_pos, v, len);
286 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
287 		client->status = Disconnected;
288 	return ret;
289 }
290 
291 /**
292  * p9_read_work - called when there is some data to be read from a transport
293  * @work: container of work to be done
294  *
295  */
296 
297 static void p9_read_work(struct work_struct *work)
298 {
299 	int n, err;
300 	struct p9_conn *m;
301 	int status = REQ_STATUS_ERROR;
302 
303 	m = container_of(work, struct p9_conn, rq);
304 
305 	if (m->err < 0)
306 		return;
307 
308 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
309 
310 	if (!m->rc.sdata) {
311 		m->rc.sdata = m->tmp_buf;
312 		m->rc.offset = 0;
313 		m->rc.capacity = 7; /* start by reading header */
314 	}
315 
316 	clear_bit(Rpending, &m->wsched);
317 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
318 		 m, m->rc.offset, m->rc.capacity,
319 		 m->rc.capacity - m->rc.offset);
320 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
321 			 m->rc.capacity - m->rc.offset);
322 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
323 	if (err == -EAGAIN)
324 		goto end_clear;
325 
326 	if (err <= 0)
327 		goto error;
328 
329 	m->rc.offset += err;
330 
331 	/* header read in */
332 	if ((!m->req) && (m->rc.offset == m->rc.capacity)) {
333 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
334 
335 		err = p9_parse_header(&m->rc, NULL, NULL, NULL, 0);
336 		if (err) {
337 			p9_debug(P9_DEBUG_ERROR,
338 				 "error parsing header: %d\n", err);
339 			goto error;
340 		}
341 
342 		if (m->rc.size >= m->client->msize) {
343 			p9_debug(P9_DEBUG_ERROR,
344 				 "requested packet size too big: %d\n",
345 				 m->rc.size);
346 			err = -EIO;
347 			goto error;
348 		}
349 
350 		p9_debug(P9_DEBUG_TRANS,
351 			 "mux %p pkt: size: %d bytes tag: %d\n",
352 			 m, m->rc.size, m->rc.tag);
353 
354 		m->req = p9_tag_lookup(m->client, m->rc.tag);
355 		if (!m->req || (m->req->status != REQ_STATUS_SENT)) {
356 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
357 				 m->rc.tag);
358 			err = -EIO;
359 			goto error;
360 		}
361 
362 		if (m->req->rc == NULL) {
363 			p9_debug(P9_DEBUG_ERROR,
364 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
365 				 m->rc.tag, m->req);
366 			m->req = NULL;
367 			err = -EIO;
368 			goto error;
369 		}
370 		m->rc.sdata = (char *)m->req->rc + sizeof(struct p9_fcall);
371 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
372 		m->rc.capacity = m->rc.size;
373 	}
374 
375 	/* packet is read in
376 	 * not an else because some packets (like clunk) have no payload
377 	 */
378 	if ((m->req) && (m->rc.offset == m->rc.capacity)) {
379 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
380 		spin_lock(&m->client->lock);
381 		if (m->req->status != REQ_STATUS_ERROR)
382 			status = REQ_STATUS_RCVD;
383 		list_del(&m->req->req_list);
384 		spin_unlock(&m->client->lock);
385 		p9_client_cb(m->client, m->req, status);
386 		m->rc.sdata = NULL;
387 		m->rc.offset = 0;
388 		m->rc.capacity = 0;
389 		m->req = NULL;
390 	}
391 
392 end_clear:
393 	clear_bit(Rworksched, &m->wsched);
394 
395 	if (!list_empty(&m->req_list)) {
396 		if (test_and_clear_bit(Rpending, &m->wsched))
397 			n = POLLIN;
398 		else
399 			n = p9_fd_poll(m->client, NULL);
400 
401 		if ((n & POLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
402 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
403 			schedule_work(&m->rq);
404 		}
405 	}
406 
407 	return;
408 error:
409 	p9_conn_cancel(m, err);
410 	clear_bit(Rworksched, &m->wsched);
411 }
412 
413 /**
414  * p9_fd_write - write to a socket
415  * @client: client instance
416  * @v: buffer to send data from
417  * @len: size of send buffer
418  *
419  */
420 
421 static int p9_fd_write(struct p9_client *client, void *v, int len)
422 {
423 	int ret;
424 	mm_segment_t oldfs;
425 	struct p9_trans_fd *ts = NULL;
426 
427 	if (client && client->status != Disconnected)
428 		ts = client->trans;
429 
430 	if (!ts)
431 		return -EREMOTEIO;
432 
433 	if (!(ts->wr->f_flags & O_NONBLOCK))
434 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
435 
436 	oldfs = get_fs();
437 	set_fs(get_ds());
438 	/* The cast to a user pointer is valid due to the set_fs() */
439 	ret = vfs_write(ts->wr, (__force void __user *)v, len, &ts->wr->f_pos);
440 	set_fs(oldfs);
441 
442 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
443 		client->status = Disconnected;
444 	return ret;
445 }
446 
447 /**
448  * p9_write_work - called when a transport can send some data
449  * @work: container for work to be done
450  *
451  */
452 
453 static void p9_write_work(struct work_struct *work)
454 {
455 	int n, err;
456 	struct p9_conn *m;
457 	struct p9_req_t *req;
458 
459 	m = container_of(work, struct p9_conn, wq);
460 
461 	if (m->err < 0) {
462 		clear_bit(Wworksched, &m->wsched);
463 		return;
464 	}
465 
466 	if (!m->wsize) {
467 		spin_lock(&m->client->lock);
468 		if (list_empty(&m->unsent_req_list)) {
469 			clear_bit(Wworksched, &m->wsched);
470 			spin_unlock(&m->client->lock);
471 			return;
472 		}
473 
474 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
475 			       req_list);
476 		req->status = REQ_STATUS_SENT;
477 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
478 		list_move_tail(&req->req_list, &m->req_list);
479 
480 		m->wbuf = req->tc->sdata;
481 		m->wsize = req->tc->size;
482 		m->wpos = 0;
483 		spin_unlock(&m->client->lock);
484 	}
485 
486 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
487 		 m, m->wpos, m->wsize);
488 	clear_bit(Wpending, &m->wsched);
489 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
490 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
491 	if (err == -EAGAIN)
492 		goto end_clear;
493 
494 
495 	if (err < 0)
496 		goto error;
497 	else if (err == 0) {
498 		err = -EREMOTEIO;
499 		goto error;
500 	}
501 
502 	m->wpos += err;
503 	if (m->wpos == m->wsize)
504 		m->wpos = m->wsize = 0;
505 
506 end_clear:
507 	clear_bit(Wworksched, &m->wsched);
508 
509 	if (m->wsize || !list_empty(&m->unsent_req_list)) {
510 		if (test_and_clear_bit(Wpending, &m->wsched))
511 			n = POLLOUT;
512 		else
513 			n = p9_fd_poll(m->client, NULL);
514 
515 		if ((n & POLLOUT) &&
516 		   !test_and_set_bit(Wworksched, &m->wsched)) {
517 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
518 			schedule_work(&m->wq);
519 		}
520 	}
521 
522 	return;
523 
524 error:
525 	p9_conn_cancel(m, err);
526 	clear_bit(Wworksched, &m->wsched);
527 }
528 
529 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
530 {
531 	struct p9_poll_wait *pwait =
532 		container_of(wait, struct p9_poll_wait, wait);
533 	struct p9_conn *m = pwait->conn;
534 	unsigned long flags;
535 
536 	spin_lock_irqsave(&p9_poll_lock, flags);
537 	if (list_empty(&m->poll_pending_link))
538 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
539 	spin_unlock_irqrestore(&p9_poll_lock, flags);
540 
541 	schedule_work(&p9_poll_work);
542 	return 1;
543 }
544 
545 /**
546  * p9_pollwait - add poll task to the wait queue
547  * @filp: file pointer being polled
548  * @wait_address: wait_q to block on
549  * @p: poll state
550  *
551  * called by files poll operation to add v9fs-poll task to files wait queue
552  */
553 
554 static void
555 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
556 {
557 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
558 	struct p9_poll_wait *pwait = NULL;
559 	int i;
560 
561 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
562 		if (m->poll_wait[i].wait_addr == NULL) {
563 			pwait = &m->poll_wait[i];
564 			break;
565 		}
566 	}
567 
568 	if (!pwait) {
569 		p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
570 		return;
571 	}
572 
573 	pwait->conn = m;
574 	pwait->wait_addr = wait_address;
575 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
576 	add_wait_queue(wait_address, &pwait->wait);
577 }
578 
579 /**
580  * p9_conn_create - initialize the per-session mux data
581  * @client: client instance
582  *
583  * Note: Creates the polling task if this is the first session.
584  */
585 
586 static void p9_conn_create(struct p9_client *client)
587 {
588 	int n;
589 	struct p9_trans_fd *ts = client->trans;
590 	struct p9_conn *m = &ts->conn;
591 
592 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
593 
594 	INIT_LIST_HEAD(&m->mux_list);
595 	m->client = client;
596 
597 	INIT_LIST_HEAD(&m->req_list);
598 	INIT_LIST_HEAD(&m->unsent_req_list);
599 	INIT_WORK(&m->rq, p9_read_work);
600 	INIT_WORK(&m->wq, p9_write_work);
601 	INIT_LIST_HEAD(&m->poll_pending_link);
602 	init_poll_funcptr(&m->pt, p9_pollwait);
603 
604 	n = p9_fd_poll(client, &m->pt);
605 	if (n & POLLIN) {
606 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
607 		set_bit(Rpending, &m->wsched);
608 	}
609 
610 	if (n & POLLOUT) {
611 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
612 		set_bit(Wpending, &m->wsched);
613 	}
614 }
615 
616 /**
617  * p9_poll_mux - polls a mux and schedules read or write works if necessary
618  * @m: connection to poll
619  *
620  */
621 
622 static void p9_poll_mux(struct p9_conn *m)
623 {
624 	int n;
625 
626 	if (m->err < 0)
627 		return;
628 
629 	n = p9_fd_poll(m->client, NULL);
630 	if (n < 0 || n & (POLLERR | POLLHUP | POLLNVAL)) {
631 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
632 		if (n >= 0)
633 			n = -ECONNRESET;
634 		p9_conn_cancel(m, n);
635 	}
636 
637 	if (n & POLLIN) {
638 		set_bit(Rpending, &m->wsched);
639 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
640 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
641 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
642 			schedule_work(&m->rq);
643 		}
644 	}
645 
646 	if (n & POLLOUT) {
647 		set_bit(Wpending, &m->wsched);
648 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
649 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
650 		    !test_and_set_bit(Wworksched, &m->wsched)) {
651 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
652 			schedule_work(&m->wq);
653 		}
654 	}
655 }
656 
657 /**
658  * p9_fd_request - send 9P request
659  * The function can sleep until the request is scheduled for sending.
660  * The function can be interrupted. Return from the function is not
661  * a guarantee that the request is sent successfully.
662  *
663  * @client: client instance
664  * @req: request to be sent
665  *
666  */
667 
668 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
669 {
670 	int n;
671 	struct p9_trans_fd *ts = client->trans;
672 	struct p9_conn *m = &ts->conn;
673 
674 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
675 		 m, current, req->tc, req->tc->id);
676 	if (m->err < 0)
677 		return m->err;
678 
679 	spin_lock(&client->lock);
680 	req->status = REQ_STATUS_UNSENT;
681 	list_add_tail(&req->req_list, &m->unsent_req_list);
682 	spin_unlock(&client->lock);
683 
684 	if (test_and_clear_bit(Wpending, &m->wsched))
685 		n = POLLOUT;
686 	else
687 		n = p9_fd_poll(m->client, NULL);
688 
689 	if (n & POLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
690 		schedule_work(&m->wq);
691 
692 	return 0;
693 }
694 
695 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
696 {
697 	int ret = 1;
698 
699 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
700 
701 	spin_lock(&client->lock);
702 
703 	if (req->status == REQ_STATUS_UNSENT) {
704 		list_del(&req->req_list);
705 		req->status = REQ_STATUS_FLSHD;
706 		ret = 0;
707 	}
708 	spin_unlock(&client->lock);
709 
710 	return ret;
711 }
712 
713 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
714 {
715 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
716 
717 	/* we haven't received a response for oldreq,
718 	 * remove it from the list.
719 	 */
720 	spin_lock(&client->lock);
721 	list_del(&req->req_list);
722 	spin_unlock(&client->lock);
723 
724 	return 0;
725 }
726 
727 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
728 {
729 	if (clnt->trans_mod == &p9_tcp_trans) {
730 		if (clnt->trans_opts.tcp.port != P9_PORT)
731 			seq_printf(m, "port=%u", clnt->trans_opts.tcp.port);
732 	} else if (clnt->trans_mod == &p9_fd_trans) {
733 		if (clnt->trans_opts.fd.rfd != ~0)
734 			seq_printf(m, "rfd=%u", clnt->trans_opts.fd.rfd);
735 		if (clnt->trans_opts.fd.wfd != ~0)
736 			seq_printf(m, "wfd=%u", clnt->trans_opts.fd.wfd);
737 	}
738 	return 0;
739 }
740 
741 /**
742  * parse_opts - parse mount options into p9_fd_opts structure
743  * @params: options string passed from mount
744  * @opts: fd transport-specific structure to parse options into
745  *
746  * Returns 0 upon success, -ERRNO upon failure
747  */
748 
749 static int parse_opts(char *params, struct p9_fd_opts *opts)
750 {
751 	char *p;
752 	substring_t args[MAX_OPT_ARGS];
753 	int option;
754 	char *options, *tmp_options;
755 
756 	opts->port = P9_PORT;
757 	opts->rfd = ~0;
758 	opts->wfd = ~0;
759 	opts->privport = false;
760 
761 	if (!params)
762 		return 0;
763 
764 	tmp_options = kstrdup(params, GFP_KERNEL);
765 	if (!tmp_options) {
766 		p9_debug(P9_DEBUG_ERROR,
767 			 "failed to allocate copy of option string\n");
768 		return -ENOMEM;
769 	}
770 	options = tmp_options;
771 
772 	while ((p = strsep(&options, ",")) != NULL) {
773 		int token;
774 		int r;
775 		if (!*p)
776 			continue;
777 		token = match_token(p, tokens, args);
778 		if ((token != Opt_err) && (token != Opt_privport)) {
779 			r = match_int(&args[0], &option);
780 			if (r < 0) {
781 				p9_debug(P9_DEBUG_ERROR,
782 					 "integer field, but no integer?\n");
783 				continue;
784 			}
785 		}
786 		switch (token) {
787 		case Opt_port:
788 			opts->port = option;
789 			break;
790 		case Opt_rfdno:
791 			opts->rfd = option;
792 			break;
793 		case Opt_wfdno:
794 			opts->wfd = option;
795 			break;
796 		case Opt_privport:
797 			opts->privport = true;
798 			break;
799 		default:
800 			continue;
801 		}
802 	}
803 
804 	kfree(tmp_options);
805 	return 0;
806 }
807 
808 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
809 {
810 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
811 					   GFP_KERNEL);
812 	if (!ts)
813 		return -ENOMEM;
814 
815 	ts->rd = fget(rfd);
816 	ts->wr = fget(wfd);
817 	if (!ts->rd || !ts->wr) {
818 		if (ts->rd)
819 			fput(ts->rd);
820 		if (ts->wr)
821 			fput(ts->wr);
822 		kfree(ts);
823 		return -EIO;
824 	}
825 
826 	client->trans = ts;
827 	client->status = Connected;
828 
829 	return 0;
830 }
831 
832 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
833 {
834 	struct p9_trans_fd *p;
835 	struct file *file;
836 
837 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
838 	if (!p)
839 		return -ENOMEM;
840 
841 	csocket->sk->sk_allocation = GFP_NOIO;
842 	file = sock_alloc_file(csocket, 0, NULL);
843 	if (IS_ERR(file)) {
844 		pr_err("%s (%d): failed to map fd\n",
845 		       __func__, task_pid_nr(current));
846 		sock_release(csocket);
847 		kfree(p);
848 		return PTR_ERR(file);
849 	}
850 
851 	get_file(file);
852 	p->wr = p->rd = file;
853 	client->trans = p;
854 	client->status = Connected;
855 
856 	p->rd->f_flags |= O_NONBLOCK;
857 
858 	p9_conn_create(client);
859 	return 0;
860 }
861 
862 /**
863  * p9_mux_destroy - cancels all pending requests of mux
864  * @m: mux to destroy
865  *
866  */
867 
868 static void p9_conn_destroy(struct p9_conn *m)
869 {
870 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
871 		 m, m->mux_list.prev, m->mux_list.next);
872 
873 	p9_mux_poll_stop(m);
874 	cancel_work_sync(&m->rq);
875 	cancel_work_sync(&m->wq);
876 
877 	p9_conn_cancel(m, -ECONNRESET);
878 
879 	m->client = NULL;
880 }
881 
882 /**
883  * p9_fd_close - shutdown file descriptor transport
884  * @client: client instance
885  *
886  */
887 
888 static void p9_fd_close(struct p9_client *client)
889 {
890 	struct p9_trans_fd *ts;
891 
892 	if (!client)
893 		return;
894 
895 	ts = client->trans;
896 	if (!ts)
897 		return;
898 
899 	client->status = Disconnected;
900 
901 	p9_conn_destroy(&ts->conn);
902 
903 	if (ts->rd)
904 		fput(ts->rd);
905 	if (ts->wr)
906 		fput(ts->wr);
907 
908 	kfree(ts);
909 }
910 
911 /*
912  * stolen from NFS - maybe should be made a generic function?
913  */
914 static inline int valid_ipaddr4(const char *buf)
915 {
916 	int rc, count, in[4];
917 
918 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
919 	if (rc != 4)
920 		return -EINVAL;
921 	for (count = 0; count < 4; count++) {
922 		if (in[count] > 255)
923 			return -EINVAL;
924 	}
925 	return 0;
926 }
927 
928 static int p9_bind_privport(struct socket *sock)
929 {
930 	struct sockaddr_in cl;
931 	int port, err = -EINVAL;
932 
933 	memset(&cl, 0, sizeof(cl));
934 	cl.sin_family = AF_INET;
935 	cl.sin_addr.s_addr = INADDR_ANY;
936 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
937 		cl.sin_port = htons((ushort)port);
938 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
939 		if (err != -EADDRINUSE)
940 			break;
941 	}
942 	return err;
943 }
944 
945 
946 static int
947 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
948 {
949 	int err;
950 	struct socket *csocket;
951 	struct sockaddr_in sin_server;
952 	struct p9_fd_opts opts;
953 
954 	err = parse_opts(args, &opts);
955 	if (err < 0)
956 		return err;
957 
958 	if (valid_ipaddr4(addr) < 0)
959 		return -EINVAL;
960 
961 	csocket = NULL;
962 
963 	client->trans_opts.tcp.port = opts.port;
964 	client->trans_opts.tcp.privport = opts.privport;
965 	sin_server.sin_family = AF_INET;
966 	sin_server.sin_addr.s_addr = in_aton(addr);
967 	sin_server.sin_port = htons(opts.port);
968 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
969 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
970 	if (err) {
971 		pr_err("%s (%d): problem creating socket\n",
972 		       __func__, task_pid_nr(current));
973 		return err;
974 	}
975 
976 	if (opts.privport) {
977 		err = p9_bind_privport(csocket);
978 		if (err < 0) {
979 			pr_err("%s (%d): problem binding to privport\n",
980 			       __func__, task_pid_nr(current));
981 			sock_release(csocket);
982 			return err;
983 		}
984 	}
985 
986 	err = csocket->ops->connect(csocket,
987 				    (struct sockaddr *)&sin_server,
988 				    sizeof(struct sockaddr_in), 0);
989 	if (err < 0) {
990 		pr_err("%s (%d): problem connecting socket to %s\n",
991 		       __func__, task_pid_nr(current), addr);
992 		sock_release(csocket);
993 		return err;
994 	}
995 
996 	return p9_socket_open(client, csocket);
997 }
998 
999 static int
1000 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
1001 {
1002 	int err;
1003 	struct socket *csocket;
1004 	struct sockaddr_un sun_server;
1005 
1006 	csocket = NULL;
1007 
1008 	if (strlen(addr) >= UNIX_PATH_MAX) {
1009 		pr_err("%s (%d): address too long: %s\n",
1010 		       __func__, task_pid_nr(current), addr);
1011 		return -ENAMETOOLONG;
1012 	}
1013 
1014 	sun_server.sun_family = PF_UNIX;
1015 	strcpy(sun_server.sun_path, addr);
1016 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1017 			    SOCK_STREAM, 0, &csocket, 1);
1018 	if (err < 0) {
1019 		pr_err("%s (%d): problem creating socket\n",
1020 		       __func__, task_pid_nr(current));
1021 
1022 		return err;
1023 	}
1024 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1025 			sizeof(struct sockaddr_un) - 1, 0);
1026 	if (err < 0) {
1027 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1028 		       __func__, task_pid_nr(current), addr, err);
1029 		sock_release(csocket);
1030 		return err;
1031 	}
1032 
1033 	return p9_socket_open(client, csocket);
1034 }
1035 
1036 static int
1037 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1038 {
1039 	int err;
1040 	struct p9_fd_opts opts;
1041 
1042 	parse_opts(args, &opts);
1043 	client->trans_opts.fd.rfd = opts.rfd;
1044 	client->trans_opts.fd.wfd = opts.wfd;
1045 
1046 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1047 		pr_err("Insufficient options for proto=fd\n");
1048 		return -ENOPROTOOPT;
1049 	}
1050 
1051 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1052 	if (err < 0)
1053 		return err;
1054 
1055 	p9_conn_create(client);
1056 
1057 	return 0;
1058 }
1059 
1060 static struct p9_trans_module p9_tcp_trans = {
1061 	.name = "tcp",
1062 	.maxsize = MAX_SOCK_BUF,
1063 	.def = 0,
1064 	.create = p9_fd_create_tcp,
1065 	.close = p9_fd_close,
1066 	.request = p9_fd_request,
1067 	.cancel = p9_fd_cancel,
1068 	.cancelled = p9_fd_cancelled,
1069 	.show_options = p9_fd_show_options,
1070 	.owner = THIS_MODULE,
1071 };
1072 
1073 static struct p9_trans_module p9_unix_trans = {
1074 	.name = "unix",
1075 	.maxsize = MAX_SOCK_BUF,
1076 	.def = 0,
1077 	.create = p9_fd_create_unix,
1078 	.close = p9_fd_close,
1079 	.request = p9_fd_request,
1080 	.cancel = p9_fd_cancel,
1081 	.cancelled = p9_fd_cancelled,
1082 	.show_options = p9_fd_show_options,
1083 	.owner = THIS_MODULE,
1084 };
1085 
1086 static struct p9_trans_module p9_fd_trans = {
1087 	.name = "fd",
1088 	.maxsize = MAX_SOCK_BUF,
1089 	.def = 0,
1090 	.create = p9_fd_create,
1091 	.close = p9_fd_close,
1092 	.request = p9_fd_request,
1093 	.cancel = p9_fd_cancel,
1094 	.cancelled = p9_fd_cancelled,
1095 	.show_options = p9_fd_show_options,
1096 	.owner = THIS_MODULE,
1097 };
1098 
1099 /**
1100  * p9_poll_proc - poll worker thread
1101  * @a: thread state and arguments
1102  *
1103  * polls all v9fs transports for new events and queues the appropriate
1104  * work to the work queue
1105  *
1106  */
1107 
1108 static void p9_poll_workfn(struct work_struct *work)
1109 {
1110 	unsigned long flags;
1111 
1112 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1113 
1114 	spin_lock_irqsave(&p9_poll_lock, flags);
1115 	while (!list_empty(&p9_poll_pending_list)) {
1116 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1117 							struct p9_conn,
1118 							poll_pending_link);
1119 		list_del_init(&conn->poll_pending_link);
1120 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1121 
1122 		p9_poll_mux(conn);
1123 
1124 		spin_lock_irqsave(&p9_poll_lock, flags);
1125 	}
1126 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1127 
1128 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1129 }
1130 
1131 int p9_trans_fd_init(void)
1132 {
1133 	v9fs_register_trans(&p9_tcp_trans);
1134 	v9fs_register_trans(&p9_unix_trans);
1135 	v9fs_register_trans(&p9_fd_trans);
1136 
1137 	return 0;
1138 }
1139 
1140 void p9_trans_fd_exit(void)
1141 {
1142 	flush_work(&p9_poll_work);
1143 	v9fs_unregister_trans(&p9_tcp_trans);
1144 	v9fs_unregister_trans(&p9_unix_trans);
1145 	v9fs_unregister_trans(&p9_fd_trans);
1146 }
1147