xref: /openbmc/linux/net/9p/trans_fd.c (revision 6c33a6f4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/fs/9p/trans_fd.c
4  *
5  * Fd transport layer.  Includes deprecated socket layer.
6  *
7  *  Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
8  *  Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
9  *  Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
10  *  Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
11  */
12 
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 
15 #include <linux/in.h>
16 #include <linux/module.h>
17 #include <linux/net.h>
18 #include <linux/ipv6.h>
19 #include <linux/kthread.h>
20 #include <linux/errno.h>
21 #include <linux/kernel.h>
22 #include <linux/un.h>
23 #include <linux/uaccess.h>
24 #include <linux/inet.h>
25 #include <linux/idr.h>
26 #include <linux/file.h>
27 #include <linux/parser.h>
28 #include <linux/slab.h>
29 #include <linux/seq_file.h>
30 #include <net/9p/9p.h>
31 #include <net/9p/client.h>
32 #include <net/9p/transport.h>
33 
34 #include <linux/syscalls.h> /* killme */
35 
36 #define P9_PORT 564
37 #define MAX_SOCK_BUF (64*1024)
38 #define MAXPOLLWADDR	2
39 
40 static struct p9_trans_module p9_tcp_trans;
41 static struct p9_trans_module p9_fd_trans;
42 
43 /**
44  * struct p9_fd_opts - per-transport options
45  * @rfd: file descriptor for reading (trans=fd)
46  * @wfd: file descriptor for writing (trans=fd)
47  * @port: port to connect to (trans=tcp)
48  *
49  */
50 
51 struct p9_fd_opts {
52 	int rfd;
53 	int wfd;
54 	u16 port;
55 	bool privport;
56 };
57 
58 /*
59   * Option Parsing (code inspired by NFS code)
60   *  - a little lazy - parse all fd-transport options
61   */
62 
63 enum {
64 	/* Options that take integer arguments */
65 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
66 	/* Options that take no arguments */
67 	Opt_privport,
68 };
69 
70 static const match_table_t tokens = {
71 	{Opt_port, "port=%u"},
72 	{Opt_rfdno, "rfdno=%u"},
73 	{Opt_wfdno, "wfdno=%u"},
74 	{Opt_privport, "privport"},
75 	{Opt_err, NULL},
76 };
77 
78 enum {
79 	Rworksched = 1,		/* read work scheduled or running */
80 	Rpending = 2,		/* can read */
81 	Wworksched = 4,		/* write work scheduled or running */
82 	Wpending = 8,		/* can write */
83 };
84 
85 struct p9_poll_wait {
86 	struct p9_conn *conn;
87 	wait_queue_entry_t wait;
88 	wait_queue_head_t *wait_addr;
89 };
90 
91 /**
92  * struct p9_conn - fd mux connection state information
93  * @mux_list: list link for mux to manage multiple connections (?)
94  * @client: reference to client instance for this connection
95  * @err: error state
96  * @req_list: accounting for requests which have been sent
97  * @unsent_req_list: accounting for requests that haven't been sent
98  * @req: current request being processed (if any)
99  * @tmp_buf: temporary buffer to read in header
100  * @rc: temporary fcall for reading current frame
101  * @wpos: write position for current frame
102  * @wsize: amount of data to write for current frame
103  * @wbuf: current write buffer
104  * @poll_pending_link: pending links to be polled per conn
105  * @poll_wait: array of wait_q's for various worker threads
106  * @pt: poll state
107  * @rq: current read work
108  * @wq: current write work
109  * @wsched: ????
110  *
111  */
112 
113 struct p9_conn {
114 	struct list_head mux_list;
115 	struct p9_client *client;
116 	int err;
117 	struct list_head req_list;
118 	struct list_head unsent_req_list;
119 	struct p9_req_t *rreq;
120 	struct p9_req_t *wreq;
121 	char tmp_buf[7];
122 	struct p9_fcall rc;
123 	int wpos;
124 	int wsize;
125 	char *wbuf;
126 	struct list_head poll_pending_link;
127 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
128 	poll_table pt;
129 	struct work_struct rq;
130 	struct work_struct wq;
131 	unsigned long wsched;
132 };
133 
134 /**
135  * struct p9_trans_fd - transport state
136  * @rd: reference to file to read from
137  * @wr: reference of file to write to
138  * @conn: connection state reference
139  *
140  */
141 
142 struct p9_trans_fd {
143 	struct file *rd;
144 	struct file *wr;
145 	struct p9_conn conn;
146 };
147 
148 static void p9_poll_workfn(struct work_struct *work);
149 
150 static DEFINE_SPINLOCK(p9_poll_lock);
151 static LIST_HEAD(p9_poll_pending_list);
152 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
153 
154 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
155 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
156 
157 static void p9_mux_poll_stop(struct p9_conn *m)
158 {
159 	unsigned long flags;
160 	int i;
161 
162 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
163 		struct p9_poll_wait *pwait = &m->poll_wait[i];
164 
165 		if (pwait->wait_addr) {
166 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
167 			pwait->wait_addr = NULL;
168 		}
169 	}
170 
171 	spin_lock_irqsave(&p9_poll_lock, flags);
172 	list_del_init(&m->poll_pending_link);
173 	spin_unlock_irqrestore(&p9_poll_lock, flags);
174 
175 	flush_work(&p9_poll_work);
176 }
177 
178 /**
179  * p9_conn_cancel - cancel all pending requests with error
180  * @m: mux data
181  * @err: error code
182  *
183  */
184 
185 static void p9_conn_cancel(struct p9_conn *m, int err)
186 {
187 	struct p9_req_t *req, *rtmp;
188 	LIST_HEAD(cancel_list);
189 
190 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
191 
192 	spin_lock(&m->client->lock);
193 
194 	if (m->err) {
195 		spin_unlock(&m->client->lock);
196 		return;
197 	}
198 
199 	m->err = err;
200 
201 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
202 		list_move(&req->req_list, &cancel_list);
203 	}
204 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
205 		list_move(&req->req_list, &cancel_list);
206 	}
207 
208 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
209 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
210 		list_del(&req->req_list);
211 		if (!req->t_err)
212 			req->t_err = err;
213 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
214 	}
215 	spin_unlock(&m->client->lock);
216 }
217 
218 static __poll_t
219 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
220 {
221 	__poll_t ret;
222 	struct p9_trans_fd *ts = NULL;
223 
224 	if (client && client->status == Connected)
225 		ts = client->trans;
226 
227 	if (!ts) {
228 		if (err)
229 			*err = -EREMOTEIO;
230 		return EPOLLERR;
231 	}
232 
233 	ret = vfs_poll(ts->rd, pt);
234 	if (ts->rd != ts->wr)
235 		ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN);
236 	return ret;
237 }
238 
239 /**
240  * p9_fd_read- read from a fd
241  * @client: client instance
242  * @v: buffer to receive data into
243  * @len: size of receive buffer
244  *
245  */
246 
247 static int p9_fd_read(struct p9_client *client, void *v, int len)
248 {
249 	int ret;
250 	struct p9_trans_fd *ts = NULL;
251 	loff_t pos;
252 
253 	if (client && client->status != Disconnected)
254 		ts = client->trans;
255 
256 	if (!ts)
257 		return -EREMOTEIO;
258 
259 	if (!(ts->rd->f_flags & O_NONBLOCK))
260 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
261 
262 	pos = ts->rd->f_pos;
263 	ret = kernel_read(ts->rd, v, len, &pos);
264 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
265 		client->status = Disconnected;
266 	return ret;
267 }
268 
269 /**
270  * p9_read_work - called when there is some data to be read from a transport
271  * @work: container of work to be done
272  *
273  */
274 
275 static void p9_read_work(struct work_struct *work)
276 {
277 	__poll_t n;
278 	int err;
279 	struct p9_conn *m;
280 
281 	m = container_of(work, struct p9_conn, rq);
282 
283 	if (m->err < 0)
284 		return;
285 
286 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
287 
288 	if (!m->rc.sdata) {
289 		m->rc.sdata = m->tmp_buf;
290 		m->rc.offset = 0;
291 		m->rc.capacity = 7; /* start by reading header */
292 	}
293 
294 	clear_bit(Rpending, &m->wsched);
295 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
296 		 m, m->rc.offset, m->rc.capacity,
297 		 m->rc.capacity - m->rc.offset);
298 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
299 			 m->rc.capacity - m->rc.offset);
300 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
301 	if (err == -EAGAIN)
302 		goto end_clear;
303 
304 	if (err <= 0)
305 		goto error;
306 
307 	m->rc.offset += err;
308 
309 	/* header read in */
310 	if ((!m->rreq) && (m->rc.offset == m->rc.capacity)) {
311 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
312 
313 		/* Header size */
314 		m->rc.size = 7;
315 		err = p9_parse_header(&m->rc, &m->rc.size, NULL, NULL, 0);
316 		if (err) {
317 			p9_debug(P9_DEBUG_ERROR,
318 				 "error parsing header: %d\n", err);
319 			goto error;
320 		}
321 
322 		if (m->rc.size >= m->client->msize) {
323 			p9_debug(P9_DEBUG_ERROR,
324 				 "requested packet size too big: %d\n",
325 				 m->rc.size);
326 			err = -EIO;
327 			goto error;
328 		}
329 
330 		p9_debug(P9_DEBUG_TRANS,
331 			 "mux %p pkt: size: %d bytes tag: %d\n",
332 			 m, m->rc.size, m->rc.tag);
333 
334 		m->rreq = p9_tag_lookup(m->client, m->rc.tag);
335 		if (!m->rreq || (m->rreq->status != REQ_STATUS_SENT)) {
336 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
337 				 m->rc.tag);
338 			err = -EIO;
339 			goto error;
340 		}
341 
342 		if (!m->rreq->rc.sdata) {
343 			p9_debug(P9_DEBUG_ERROR,
344 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
345 				 m->rc.tag, m->rreq);
346 			m->rreq = NULL;
347 			err = -EIO;
348 			goto error;
349 		}
350 		m->rc.sdata = m->rreq->rc.sdata;
351 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
352 		m->rc.capacity = m->rc.size;
353 	}
354 
355 	/* packet is read in
356 	 * not an else because some packets (like clunk) have no payload
357 	 */
358 	if ((m->rreq) && (m->rc.offset == m->rc.capacity)) {
359 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
360 		m->rreq->rc.size = m->rc.offset;
361 		spin_lock(&m->client->lock);
362 		if (m->rreq->status == REQ_STATUS_SENT) {
363 			list_del(&m->rreq->req_list);
364 			p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD);
365 		} else {
366 			spin_unlock(&m->client->lock);
367 			p9_debug(P9_DEBUG_ERROR,
368 				 "Request tag %d errored out while we were reading the reply\n",
369 				 m->rc.tag);
370 			err = -EIO;
371 			goto error;
372 		}
373 		spin_unlock(&m->client->lock);
374 		m->rc.sdata = NULL;
375 		m->rc.offset = 0;
376 		m->rc.capacity = 0;
377 		p9_req_put(m->rreq);
378 		m->rreq = NULL;
379 	}
380 
381 end_clear:
382 	clear_bit(Rworksched, &m->wsched);
383 
384 	if (!list_empty(&m->req_list)) {
385 		if (test_and_clear_bit(Rpending, &m->wsched))
386 			n = EPOLLIN;
387 		else
388 			n = p9_fd_poll(m->client, NULL, NULL);
389 
390 		if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
391 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
392 			schedule_work(&m->rq);
393 		}
394 	}
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 	ssize_t ret;
413 	struct p9_trans_fd *ts = NULL;
414 
415 	if (client && client->status != Disconnected)
416 		ts = client->trans;
417 
418 	if (!ts)
419 		return -EREMOTEIO;
420 
421 	if (!(ts->wr->f_flags & O_NONBLOCK))
422 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
423 
424 	ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
425 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
426 		client->status = Disconnected;
427 	return ret;
428 }
429 
430 /**
431  * p9_write_work - called when a transport can send some data
432  * @work: container for work to be done
433  *
434  */
435 
436 static void p9_write_work(struct work_struct *work)
437 {
438 	__poll_t n;
439 	int err;
440 	struct p9_conn *m;
441 	struct p9_req_t *req;
442 
443 	m = container_of(work, struct p9_conn, wq);
444 
445 	if (m->err < 0) {
446 		clear_bit(Wworksched, &m->wsched);
447 		return;
448 	}
449 
450 	if (!m->wsize) {
451 		spin_lock(&m->client->lock);
452 		if (list_empty(&m->unsent_req_list)) {
453 			clear_bit(Wworksched, &m->wsched);
454 			spin_unlock(&m->client->lock);
455 			return;
456 		}
457 
458 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
459 			       req_list);
460 		req->status = REQ_STATUS_SENT;
461 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
462 		list_move_tail(&req->req_list, &m->req_list);
463 
464 		m->wbuf = req->tc.sdata;
465 		m->wsize = req->tc.size;
466 		m->wpos = 0;
467 		p9_req_get(req);
468 		m->wreq = req;
469 		spin_unlock(&m->client->lock);
470 	}
471 
472 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
473 		 m, m->wpos, 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_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
477 	if (err == -EAGAIN)
478 		goto end_clear;
479 
480 
481 	if (err < 0)
482 		goto error;
483 	else if (err == 0) {
484 		err = -EREMOTEIO;
485 		goto error;
486 	}
487 
488 	m->wpos += err;
489 	if (m->wpos == m->wsize) {
490 		m->wpos = m->wsize = 0;
491 		p9_req_put(m->wreq);
492 		m->wreq = NULL;
493 	}
494 
495 end_clear:
496 	clear_bit(Wworksched, &m->wsched);
497 
498 	if (m->wsize || !list_empty(&m->unsent_req_list)) {
499 		if (test_and_clear_bit(Wpending, &m->wsched))
500 			n = EPOLLOUT;
501 		else
502 			n = p9_fd_poll(m->client, NULL, NULL);
503 
504 		if ((n & EPOLLOUT) &&
505 		   !test_and_set_bit(Wworksched, &m->wsched)) {
506 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
507 			schedule_work(&m->wq);
508 		}
509 	}
510 
511 	return;
512 
513 error:
514 	p9_conn_cancel(m, err);
515 	clear_bit(Wworksched, &m->wsched);
516 }
517 
518 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
519 {
520 	struct p9_poll_wait *pwait =
521 		container_of(wait, struct p9_poll_wait, wait);
522 	struct p9_conn *m = pwait->conn;
523 	unsigned long flags;
524 
525 	spin_lock_irqsave(&p9_poll_lock, flags);
526 	if (list_empty(&m->poll_pending_link))
527 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
528 	spin_unlock_irqrestore(&p9_poll_lock, flags);
529 
530 	schedule_work(&p9_poll_work);
531 	return 1;
532 }
533 
534 /**
535  * p9_pollwait - add poll task to the wait queue
536  * @filp: file pointer being polled
537  * @wait_address: wait_q to block on
538  * @p: poll state
539  *
540  * called by files poll operation to add v9fs-poll task to files wait queue
541  */
542 
543 static void
544 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
545 {
546 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
547 	struct p9_poll_wait *pwait = NULL;
548 	int i;
549 
550 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
551 		if (m->poll_wait[i].wait_addr == NULL) {
552 			pwait = &m->poll_wait[i];
553 			break;
554 		}
555 	}
556 
557 	if (!pwait) {
558 		p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
559 		return;
560 	}
561 
562 	pwait->conn = m;
563 	pwait->wait_addr = wait_address;
564 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
565 	add_wait_queue(wait_address, &pwait->wait);
566 }
567 
568 /**
569  * p9_conn_create - initialize the per-session mux data
570  * @client: client instance
571  *
572  * Note: Creates the polling task if this is the first session.
573  */
574 
575 static void p9_conn_create(struct p9_client *client)
576 {
577 	__poll_t n;
578 	struct p9_trans_fd *ts = client->trans;
579 	struct p9_conn *m = &ts->conn;
580 
581 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
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, NULL);
594 	if (n & EPOLLIN) {
595 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
596 		set_bit(Rpending, &m->wsched);
597 	}
598 
599 	if (n & EPOLLOUT) {
600 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
601 		set_bit(Wpending, &m->wsched);
602 	}
603 }
604 
605 /**
606  * p9_poll_mux - polls a mux and schedules read or write works if necessary
607  * @m: connection to poll
608  *
609  */
610 
611 static void p9_poll_mux(struct p9_conn *m)
612 {
613 	__poll_t n;
614 	int err = -ECONNRESET;
615 
616 	if (m->err < 0)
617 		return;
618 
619 	n = p9_fd_poll(m->client, NULL, &err);
620 	if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
621 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
622 		p9_conn_cancel(m, err);
623 	}
624 
625 	if (n & EPOLLIN) {
626 		set_bit(Rpending, &m->wsched);
627 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
628 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
629 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
630 			schedule_work(&m->rq);
631 		}
632 	}
633 
634 	if (n & EPOLLOUT) {
635 		set_bit(Wpending, &m->wsched);
636 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
637 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
638 		    !test_and_set_bit(Wworksched, &m->wsched)) {
639 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
640 			schedule_work(&m->wq);
641 		}
642 	}
643 }
644 
645 /**
646  * p9_fd_request - send 9P request
647  * The function can sleep until the request is scheduled for sending.
648  * The function can be interrupted. Return from the function is not
649  * a guarantee that the request is sent successfully.
650  *
651  * @client: client instance
652  * @req: request to be sent
653  *
654  */
655 
656 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
657 {
658 	__poll_t n;
659 	struct p9_trans_fd *ts = client->trans;
660 	struct p9_conn *m = &ts->conn;
661 
662 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
663 		 m, current, &req->tc, req->tc.id);
664 	if (m->err < 0)
665 		return m->err;
666 
667 	spin_lock(&client->lock);
668 	req->status = REQ_STATUS_UNSENT;
669 	list_add_tail(&req->req_list, &m->unsent_req_list);
670 	spin_unlock(&client->lock);
671 
672 	if (test_and_clear_bit(Wpending, &m->wsched))
673 		n = EPOLLOUT;
674 	else
675 		n = p9_fd_poll(m->client, NULL, NULL);
676 
677 	if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
678 		schedule_work(&m->wq);
679 
680 	return 0;
681 }
682 
683 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
684 {
685 	int ret = 1;
686 
687 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
688 
689 	spin_lock(&client->lock);
690 
691 	if (req->status == REQ_STATUS_UNSENT) {
692 		list_del(&req->req_list);
693 		req->status = REQ_STATUS_FLSHD;
694 		p9_req_put(req);
695 		ret = 0;
696 	}
697 	spin_unlock(&client->lock);
698 
699 	return ret;
700 }
701 
702 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
703 {
704 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
705 
706 	/* we haven't received a response for oldreq,
707 	 * remove it from the list.
708 	 */
709 	spin_lock(&client->lock);
710 	list_del(&req->req_list);
711 	spin_unlock(&client->lock);
712 	p9_req_put(req);
713 
714 	return 0;
715 }
716 
717 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
718 {
719 	if (clnt->trans_mod == &p9_tcp_trans) {
720 		if (clnt->trans_opts.tcp.port != P9_PORT)
721 			seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
722 	} else if (clnt->trans_mod == &p9_fd_trans) {
723 		if (clnt->trans_opts.fd.rfd != ~0)
724 			seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
725 		if (clnt->trans_opts.fd.wfd != ~0)
726 			seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
727 	}
728 	return 0;
729 }
730 
731 /**
732  * parse_opts - parse mount options into p9_fd_opts structure
733  * @params: options string passed from mount
734  * @opts: fd transport-specific structure to parse options into
735  *
736  * Returns 0 upon success, -ERRNO upon failure
737  */
738 
739 static int parse_opts(char *params, struct p9_fd_opts *opts)
740 {
741 	char *p;
742 	substring_t args[MAX_OPT_ARGS];
743 	int option;
744 	char *options, *tmp_options;
745 
746 	opts->port = P9_PORT;
747 	opts->rfd = ~0;
748 	opts->wfd = ~0;
749 	opts->privport = false;
750 
751 	if (!params)
752 		return 0;
753 
754 	tmp_options = kstrdup(params, GFP_KERNEL);
755 	if (!tmp_options) {
756 		p9_debug(P9_DEBUG_ERROR,
757 			 "failed to allocate copy of option string\n");
758 		return -ENOMEM;
759 	}
760 	options = tmp_options;
761 
762 	while ((p = strsep(&options, ",")) != NULL) {
763 		int token;
764 		int r;
765 		if (!*p)
766 			continue;
767 		token = match_token(p, tokens, args);
768 		if ((token != Opt_err) && (token != Opt_privport)) {
769 			r = match_int(&args[0], &option);
770 			if (r < 0) {
771 				p9_debug(P9_DEBUG_ERROR,
772 					 "integer field, but no integer?\n");
773 				continue;
774 			}
775 		}
776 		switch (token) {
777 		case Opt_port:
778 			opts->port = option;
779 			break;
780 		case Opt_rfdno:
781 			opts->rfd = option;
782 			break;
783 		case Opt_wfdno:
784 			opts->wfd = option;
785 			break;
786 		case Opt_privport:
787 			opts->privport = true;
788 			break;
789 		default:
790 			continue;
791 		}
792 	}
793 
794 	kfree(tmp_options);
795 	return 0;
796 }
797 
798 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
799 {
800 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
801 					   GFP_KERNEL);
802 	if (!ts)
803 		return -ENOMEM;
804 
805 	ts->rd = fget(rfd);
806 	ts->wr = fget(wfd);
807 	if (!ts->rd || !ts->wr) {
808 		if (ts->rd)
809 			fput(ts->rd);
810 		if (ts->wr)
811 			fput(ts->wr);
812 		kfree(ts);
813 		return -EIO;
814 	}
815 
816 	client->trans = ts;
817 	client->status = Connected;
818 
819 	return 0;
820 }
821 
822 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
823 {
824 	struct p9_trans_fd *p;
825 	struct file *file;
826 
827 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
828 	if (!p)
829 		return -ENOMEM;
830 
831 	csocket->sk->sk_allocation = GFP_NOIO;
832 	file = sock_alloc_file(csocket, 0, NULL);
833 	if (IS_ERR(file)) {
834 		pr_err("%s (%d): failed to map fd\n",
835 		       __func__, task_pid_nr(current));
836 		kfree(p);
837 		return PTR_ERR(file);
838 	}
839 
840 	get_file(file);
841 	p->wr = p->rd = file;
842 	client->trans = p;
843 	client->status = Connected;
844 
845 	p->rd->f_flags |= O_NONBLOCK;
846 
847 	p9_conn_create(client);
848 	return 0;
849 }
850 
851 /**
852  * p9_mux_destroy - cancels all pending requests of mux
853  * @m: mux to destroy
854  *
855  */
856 
857 static void p9_conn_destroy(struct p9_conn *m)
858 {
859 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
860 		 m, m->mux_list.prev, m->mux_list.next);
861 
862 	p9_mux_poll_stop(m);
863 	cancel_work_sync(&m->rq);
864 	if (m->rreq) {
865 		p9_req_put(m->rreq);
866 		m->rreq = NULL;
867 	}
868 	cancel_work_sync(&m->wq);
869 	if (m->wreq) {
870 		p9_req_put(m->wreq);
871 		m->wreq = NULL;
872 	}
873 
874 	p9_conn_cancel(m, -ECONNRESET);
875 
876 	m->client = NULL;
877 }
878 
879 /**
880  * p9_fd_close - shutdown file descriptor transport
881  * @client: client instance
882  *
883  */
884 
885 static void p9_fd_close(struct p9_client *client)
886 {
887 	struct p9_trans_fd *ts;
888 
889 	if (!client)
890 		return;
891 
892 	ts = client->trans;
893 	if (!ts)
894 		return;
895 
896 	client->status = Disconnected;
897 
898 	p9_conn_destroy(&ts->conn);
899 
900 	if (ts->rd)
901 		fput(ts->rd);
902 	if (ts->wr)
903 		fput(ts->wr);
904 
905 	kfree(ts);
906 }
907 
908 /*
909  * stolen from NFS - maybe should be made a generic function?
910  */
911 static inline int valid_ipaddr4(const char *buf)
912 {
913 	int rc, count, in[4];
914 
915 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
916 	if (rc != 4)
917 		return -EINVAL;
918 	for (count = 0; count < 4; count++) {
919 		if (in[count] > 255)
920 			return -EINVAL;
921 	}
922 	return 0;
923 }
924 
925 static int p9_bind_privport(struct socket *sock)
926 {
927 	struct sockaddr_in cl;
928 	int port, err = -EINVAL;
929 
930 	memset(&cl, 0, sizeof(cl));
931 	cl.sin_family = AF_INET;
932 	cl.sin_addr.s_addr = INADDR_ANY;
933 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
934 		cl.sin_port = htons((ushort)port);
935 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
936 		if (err != -EADDRINUSE)
937 			break;
938 	}
939 	return err;
940 }
941 
942 
943 static int
944 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
945 {
946 	int err;
947 	struct socket *csocket;
948 	struct sockaddr_in sin_server;
949 	struct p9_fd_opts opts;
950 
951 	err = parse_opts(args, &opts);
952 	if (err < 0)
953 		return err;
954 
955 	if (addr == NULL || valid_ipaddr4(addr) < 0)
956 		return -EINVAL;
957 
958 	csocket = NULL;
959 
960 	client->trans_opts.tcp.port = opts.port;
961 	client->trans_opts.tcp.privport = opts.privport;
962 	sin_server.sin_family = AF_INET;
963 	sin_server.sin_addr.s_addr = in_aton(addr);
964 	sin_server.sin_port = htons(opts.port);
965 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
966 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
967 	if (err) {
968 		pr_err("%s (%d): problem creating socket\n",
969 		       __func__, task_pid_nr(current));
970 		return err;
971 	}
972 
973 	if (opts.privport) {
974 		err = p9_bind_privport(csocket);
975 		if (err < 0) {
976 			pr_err("%s (%d): problem binding to privport\n",
977 			       __func__, task_pid_nr(current));
978 			sock_release(csocket);
979 			return err;
980 		}
981 	}
982 
983 	err = csocket->ops->connect(csocket,
984 				    (struct sockaddr *)&sin_server,
985 				    sizeof(struct sockaddr_in), 0);
986 	if (err < 0) {
987 		pr_err("%s (%d): problem connecting socket to %s\n",
988 		       __func__, task_pid_nr(current), addr);
989 		sock_release(csocket);
990 		return err;
991 	}
992 
993 	return p9_socket_open(client, csocket);
994 }
995 
996 static int
997 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
998 {
999 	int err;
1000 	struct socket *csocket;
1001 	struct sockaddr_un sun_server;
1002 
1003 	csocket = NULL;
1004 
1005 	if (addr == NULL)
1006 		return -EINVAL;
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_workfn - poll worker thread
1101  * @work: work queue
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