xref: /openbmc/linux/net/9p/trans_fd.c (revision 1849f9f0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Fd transport layer.  Includes deprecated socket layer.
4  *
5  *  Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
6  *  Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
7  *  Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
8  *  Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
9  */
10 
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 
13 #include <linux/in.h>
14 #include <linux/module.h>
15 #include <linux/net.h>
16 #include <linux/ipv6.h>
17 #include <linux/kthread.h>
18 #include <linux/errno.h>
19 #include <linux/kernel.h>
20 #include <linux/un.h>
21 #include <linux/uaccess.h>
22 #include <linux/inet.h>
23 #include <linux/idr.h>
24 #include <linux/file.h>
25 #include <linux/parser.h>
26 #include <linux/slab.h>
27 #include <linux/seq_file.h>
28 #include <net/9p/9p.h>
29 #include <net/9p/client.h>
30 #include <net/9p/transport.h>
31 
32 #include <linux/syscalls.h> /* killme */
33 
34 #define P9_PORT 564
35 #define MAX_SOCK_BUF (1024*1024)
36 #define MAXPOLLWADDR	2
37 
38 static struct p9_trans_module p9_tcp_trans;
39 static struct p9_trans_module p9_fd_trans;
40 
41 /**
42  * struct p9_fd_opts - per-transport options
43  * @rfd: file descriptor for reading (trans=fd)
44  * @wfd: file descriptor for writing (trans=fd)
45  * @port: port to connect to (trans=tcp)
46  * @privport: port is privileged
47  */
48 
49 struct p9_fd_opts {
50 	int rfd;
51 	int wfd;
52 	u16 port;
53 	bool privport;
54 };
55 
56 /*
57   * Option Parsing (code inspired by NFS code)
58   *  - a little lazy - parse all fd-transport options
59   */
60 
61 enum {
62 	/* Options that take integer arguments */
63 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
64 	/* Options that take no arguments */
65 	Opt_privport,
66 };
67 
68 static const match_table_t tokens = {
69 	{Opt_port, "port=%u"},
70 	{Opt_rfdno, "rfdno=%u"},
71 	{Opt_wfdno, "wfdno=%u"},
72 	{Opt_privport, "privport"},
73 	{Opt_err, NULL},
74 };
75 
76 enum {
77 	Rworksched = 1,		/* read work scheduled or running */
78 	Rpending = 2,		/* can read */
79 	Wworksched = 4,		/* write work scheduled or running */
80 	Wpending = 8,		/* can write */
81 };
82 
83 struct p9_poll_wait {
84 	struct p9_conn *conn;
85 	wait_queue_entry_t wait;
86 	wait_queue_head_t *wait_addr;
87 };
88 
89 /**
90  * struct p9_conn - fd mux connection state information
91  * @mux_list: list link for mux to manage multiple connections (?)
92  * @client: reference to client instance for this connection
93  * @err: error state
94  * @req_lock: lock protecting req_list and requests statuses
95  * @req_list: accounting for requests which have been sent
96  * @unsent_req_list: accounting for requests that haven't been sent
97  * @rreq: read request
98  * @wreq: write request
99  * @req: current request being processed (if any)
100  * @tmp_buf: temporary buffer to read in header
101  * @rc: temporary fcall for reading current frame
102  * @wpos: write position for current frame
103  * @wsize: amount of data to write for current frame
104  * @wbuf: current write buffer
105  * @poll_pending_link: pending links to be polled per conn
106  * @poll_wait: array of wait_q's for various worker threads
107  * @pt: poll state
108  * @rq: current read work
109  * @wq: current write work
110  * @wsched: ????
111  *
112  */
113 
114 struct p9_conn {
115 	struct list_head mux_list;
116 	struct p9_client *client;
117 	int err;
118 	spinlock_t req_lock;
119 	struct list_head req_list;
120 	struct list_head unsent_req_list;
121 	struct p9_req_t *rreq;
122 	struct p9_req_t *wreq;
123 	char tmp_buf[P9_HDRSZ];
124 	struct p9_fcall rc;
125 	int wpos;
126 	int wsize;
127 	char *wbuf;
128 	struct list_head poll_pending_link;
129 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
130 	poll_table pt;
131 	struct work_struct rq;
132 	struct work_struct wq;
133 	unsigned long wsched;
134 };
135 
136 /**
137  * struct p9_trans_fd - transport state
138  * @rd: reference to file to read from
139  * @wr: reference of file to write to
140  * @conn: connection state reference
141  *
142  */
143 
144 struct p9_trans_fd {
145 	struct file *rd;
146 	struct file *wr;
147 	struct p9_conn conn;
148 };
149 
150 static void p9_poll_workfn(struct work_struct *work);
151 
152 static DEFINE_SPINLOCK(p9_poll_lock);
153 static LIST_HEAD(p9_poll_pending_list);
154 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
155 
156 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
157 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
158 
159 static void p9_mux_poll_stop(struct p9_conn *m)
160 {
161 	unsigned long flags;
162 	int i;
163 
164 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
165 		struct p9_poll_wait *pwait = &m->poll_wait[i];
166 
167 		if (pwait->wait_addr) {
168 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
169 			pwait->wait_addr = NULL;
170 		}
171 	}
172 
173 	spin_lock_irqsave(&p9_poll_lock, flags);
174 	list_del_init(&m->poll_pending_link);
175 	spin_unlock_irqrestore(&p9_poll_lock, flags);
176 
177 	flush_work(&p9_poll_work);
178 }
179 
180 /**
181  * p9_conn_cancel - cancel all pending requests with error
182  * @m: mux data
183  * @err: error code
184  *
185  */
186 
187 static void p9_conn_cancel(struct p9_conn *m, int err)
188 {
189 	struct p9_req_t *req, *rtmp;
190 	LIST_HEAD(cancel_list);
191 
192 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
193 
194 	spin_lock(&m->req_lock);
195 
196 	if (m->err) {
197 		spin_unlock(&m->req_lock);
198 		return;
199 	}
200 
201 	m->err = err;
202 
203 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
204 		list_move(&req->req_list, &cancel_list);
205 		req->status = REQ_STATUS_ERROR;
206 	}
207 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
208 		list_move(&req->req_list, &cancel_list);
209 		req->status = REQ_STATUS_ERROR;
210 	}
211 
212 	spin_unlock(&m->req_lock);
213 
214 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
215 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
216 		list_del(&req->req_list);
217 		if (!req->t_err)
218 			req->t_err = err;
219 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
220 	}
221 }
222 
223 static __poll_t
224 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
225 {
226 	__poll_t ret;
227 	struct p9_trans_fd *ts = NULL;
228 
229 	if (client && client->status == Connected)
230 		ts = client->trans;
231 
232 	if (!ts) {
233 		if (err)
234 			*err = -EREMOTEIO;
235 		return EPOLLERR;
236 	}
237 
238 	ret = vfs_poll(ts->rd, pt);
239 	if (ts->rd != ts->wr)
240 		ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN);
241 	return ret;
242 }
243 
244 /**
245  * p9_fd_read- read from a fd
246  * @client: client instance
247  * @v: buffer to receive data into
248  * @len: size of receive buffer
249  *
250  */
251 
252 static int p9_fd_read(struct p9_client *client, void *v, int len)
253 {
254 	int ret;
255 	struct p9_trans_fd *ts = NULL;
256 	loff_t pos;
257 
258 	if (client && client->status != Disconnected)
259 		ts = client->trans;
260 
261 	if (!ts)
262 		return -EREMOTEIO;
263 
264 	if (!(ts->rd->f_flags & O_NONBLOCK))
265 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
266 
267 	pos = ts->rd->f_pos;
268 	ret = kernel_read(ts->rd, v, len, &pos);
269 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
270 		client->status = Disconnected;
271 	return ret;
272 }
273 
274 /**
275  * p9_read_work - called when there is some data to be read from a transport
276  * @work: container of work to be done
277  *
278  */
279 
280 static void p9_read_work(struct work_struct *work)
281 {
282 	__poll_t n;
283 	int err;
284 	struct p9_conn *m;
285 
286 	m = container_of(work, struct p9_conn, rq);
287 
288 	if (m->err < 0)
289 		return;
290 
291 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
292 
293 	if (!m->rc.sdata) {
294 		m->rc.sdata = m->tmp_buf;
295 		m->rc.offset = 0;
296 		m->rc.capacity = P9_HDRSZ; /* start by reading header */
297 	}
298 
299 	clear_bit(Rpending, &m->wsched);
300 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
301 		 m, m->rc.offset, m->rc.capacity,
302 		 m->rc.capacity - m->rc.offset);
303 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
304 			 m->rc.capacity - m->rc.offset);
305 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
306 	if (err == -EAGAIN)
307 		goto end_clear;
308 
309 	if (err <= 0)
310 		goto error;
311 
312 	m->rc.offset += err;
313 
314 	/* header read in */
315 	if ((!m->rreq) && (m->rc.offset == m->rc.capacity)) {
316 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
317 
318 		/* Header size */
319 		m->rc.size = P9_HDRSZ;
320 		err = p9_parse_header(&m->rc, &m->rc.size, NULL, NULL, 0);
321 		if (err) {
322 			p9_debug(P9_DEBUG_ERROR,
323 				 "error parsing header: %d\n", err);
324 			goto error;
325 		}
326 
327 		p9_debug(P9_DEBUG_TRANS,
328 			 "mux %p pkt: size: %d bytes tag: %d\n",
329 			 m, m->rc.size, m->rc.tag);
330 
331 		m->rreq = p9_tag_lookup(m->client, m->rc.tag);
332 		if (!m->rreq || (m->rreq->status != REQ_STATUS_SENT)) {
333 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
334 				 m->rc.tag);
335 			err = -EIO;
336 			goto error;
337 		}
338 
339 		if (m->rc.size > m->rreq->rc.capacity) {
340 			p9_debug(P9_DEBUG_ERROR,
341 				 "requested packet size too big: %d for tag %d with capacity %zd\n",
342 				 m->rc.size, m->rc.tag, m->rreq->rc.capacity);
343 			err = -EIO;
344 			goto error;
345 		}
346 
347 		if (!m->rreq->rc.sdata) {
348 			p9_debug(P9_DEBUG_ERROR,
349 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
350 				 m->rc.tag, m->rreq);
351 			p9_req_put(m->client, m->rreq);
352 			m->rreq = NULL;
353 			err = -EIO;
354 			goto error;
355 		}
356 		m->rc.sdata = m->rreq->rc.sdata;
357 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
358 		m->rc.capacity = m->rc.size;
359 	}
360 
361 	/* packet is read in
362 	 * not an else because some packets (like clunk) have no payload
363 	 */
364 	if ((m->rreq) && (m->rc.offset == m->rc.capacity)) {
365 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
366 		m->rreq->rc.size = m->rc.offset;
367 		spin_lock(&m->req_lock);
368 		if (m->rreq->status == REQ_STATUS_SENT) {
369 			list_del(&m->rreq->req_list);
370 			p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD);
371 		} else if (m->rreq->status == REQ_STATUS_FLSHD) {
372 			/* Ignore replies associated with a cancelled request. */
373 			p9_debug(P9_DEBUG_TRANS,
374 				 "Ignore replies associated with a cancelled request\n");
375 		} else {
376 			spin_unlock(&m->req_lock);
377 			p9_debug(P9_DEBUG_ERROR,
378 				 "Request tag %d errored out while we were reading the reply\n",
379 				 m->rc.tag);
380 			err = -EIO;
381 			goto error;
382 		}
383 		spin_unlock(&m->req_lock);
384 		m->rc.sdata = NULL;
385 		m->rc.offset = 0;
386 		m->rc.capacity = 0;
387 		p9_req_put(m->client, m->rreq);
388 		m->rreq = NULL;
389 	}
390 
391 end_clear:
392 	clear_bit(Rworksched, &m->wsched);
393 
394 	if (!list_empty(&m->req_list)) {
395 		if (test_and_clear_bit(Rpending, &m->wsched))
396 			n = EPOLLIN;
397 		else
398 			n = p9_fd_poll(m->client, NULL, NULL);
399 
400 		if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
401 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
402 			schedule_work(&m->rq);
403 		}
404 	}
405 
406 	return;
407 error:
408 	p9_conn_cancel(m, err);
409 	clear_bit(Rworksched, &m->wsched);
410 }
411 
412 /**
413  * p9_fd_write - write to a socket
414  * @client: client instance
415  * @v: buffer to send data from
416  * @len: size of send buffer
417  *
418  */
419 
420 static int p9_fd_write(struct p9_client *client, void *v, int len)
421 {
422 	ssize_t ret;
423 	struct p9_trans_fd *ts = NULL;
424 
425 	if (client && client->status != Disconnected)
426 		ts = client->trans;
427 
428 	if (!ts)
429 		return -EREMOTEIO;
430 
431 	if (!(ts->wr->f_flags & O_NONBLOCK))
432 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
433 
434 	ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
435 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
436 		client->status = Disconnected;
437 	return ret;
438 }
439 
440 /**
441  * p9_write_work - called when a transport can send some data
442  * @work: container for work to be done
443  *
444  */
445 
446 static void p9_write_work(struct work_struct *work)
447 {
448 	__poll_t n;
449 	int err;
450 	struct p9_conn *m;
451 	struct p9_req_t *req;
452 
453 	m = container_of(work, struct p9_conn, wq);
454 
455 	if (m->err < 0) {
456 		clear_bit(Wworksched, &m->wsched);
457 		return;
458 	}
459 
460 	if (!m->wsize) {
461 		spin_lock(&m->req_lock);
462 		if (list_empty(&m->unsent_req_list)) {
463 			clear_bit(Wworksched, &m->wsched);
464 			spin_unlock(&m->req_lock);
465 			return;
466 		}
467 
468 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
469 			       req_list);
470 		req->status = REQ_STATUS_SENT;
471 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
472 		list_move_tail(&req->req_list, &m->req_list);
473 
474 		m->wbuf = req->tc.sdata;
475 		m->wsize = req->tc.size;
476 		m->wpos = 0;
477 		p9_req_get(req);
478 		m->wreq = req;
479 		spin_unlock(&m->req_lock);
480 	}
481 
482 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
483 		 m, m->wpos, m->wsize);
484 	clear_bit(Wpending, &m->wsched);
485 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
486 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
487 	if (err == -EAGAIN)
488 		goto end_clear;
489 
490 
491 	if (err < 0)
492 		goto error;
493 	else if (err == 0) {
494 		err = -EREMOTEIO;
495 		goto error;
496 	}
497 
498 	m->wpos += err;
499 	if (m->wpos == m->wsize) {
500 		m->wpos = m->wsize = 0;
501 		p9_req_put(m->client, m->wreq);
502 		m->wreq = NULL;
503 	}
504 
505 end_clear:
506 	clear_bit(Wworksched, &m->wsched);
507 
508 	if (m->wsize || !list_empty(&m->unsent_req_list)) {
509 		if (test_and_clear_bit(Wpending, &m->wsched))
510 			n = EPOLLOUT;
511 		else
512 			n = p9_fd_poll(m->client, NULL, NULL);
513 
514 		if ((n & EPOLLOUT) &&
515 		   !test_and_set_bit(Wworksched, &m->wsched)) {
516 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
517 			schedule_work(&m->wq);
518 		}
519 	}
520 
521 	return;
522 
523 error:
524 	p9_conn_cancel(m, err);
525 	clear_bit(Wworksched, &m->wsched);
526 }
527 
528 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
529 {
530 	struct p9_poll_wait *pwait =
531 		container_of(wait, struct p9_poll_wait, wait);
532 	struct p9_conn *m = pwait->conn;
533 	unsigned long flags;
534 
535 	spin_lock_irqsave(&p9_poll_lock, flags);
536 	if (list_empty(&m->poll_pending_link))
537 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
538 	spin_unlock_irqrestore(&p9_poll_lock, flags);
539 
540 	schedule_work(&p9_poll_work);
541 	return 1;
542 }
543 
544 /**
545  * p9_pollwait - add poll task to the wait queue
546  * @filp: file pointer being polled
547  * @wait_address: wait_q to block on
548  * @p: poll state
549  *
550  * called by files poll operation to add v9fs-poll task to files wait queue
551  */
552 
553 static void
554 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
555 {
556 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
557 	struct p9_poll_wait *pwait = NULL;
558 	int i;
559 
560 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
561 		if (m->poll_wait[i].wait_addr == NULL) {
562 			pwait = &m->poll_wait[i];
563 			break;
564 		}
565 	}
566 
567 	if (!pwait) {
568 		p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
569 		return;
570 	}
571 
572 	pwait->conn = m;
573 	pwait->wait_addr = wait_address;
574 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
575 	add_wait_queue(wait_address, &pwait->wait);
576 }
577 
578 /**
579  * p9_conn_create - initialize the per-session mux data
580  * @client: client instance
581  *
582  * Note: Creates the polling task if this is the first session.
583  */
584 
585 static void p9_conn_create(struct p9_client *client)
586 {
587 	__poll_t n;
588 	struct p9_trans_fd *ts = client->trans;
589 	struct p9_conn *m = &ts->conn;
590 
591 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
592 
593 	INIT_LIST_HEAD(&m->mux_list);
594 	m->client = client;
595 
596 	spin_lock_init(&m->req_lock);
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, NULL);
605 	if (n & EPOLLIN) {
606 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
607 		set_bit(Rpending, &m->wsched);
608 	}
609 
610 	if (n & EPOLLOUT) {
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 	__poll_t n;
625 	int err = -ECONNRESET;
626 
627 	if (m->err < 0)
628 		return;
629 
630 	n = p9_fd_poll(m->client, NULL, &err);
631 	if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
632 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
633 		p9_conn_cancel(m, err);
634 	}
635 
636 	if (n & EPOLLIN) {
637 		set_bit(Rpending, &m->wsched);
638 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
639 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
640 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
641 			schedule_work(&m->rq);
642 		}
643 	}
644 
645 	if (n & EPOLLOUT) {
646 		set_bit(Wpending, &m->wsched);
647 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
648 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
649 		    !test_and_set_bit(Wworksched, &m->wsched)) {
650 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
651 			schedule_work(&m->wq);
652 		}
653 	}
654 }
655 
656 /**
657  * p9_fd_request - send 9P request
658  * The function can sleep until the request is scheduled for sending.
659  * The function can be interrupted. Return from the function is not
660  * a guarantee that the request is sent successfully.
661  *
662  * @client: client instance
663  * @req: request to be sent
664  *
665  */
666 
667 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
668 {
669 	__poll_t n;
670 	struct p9_trans_fd *ts = client->trans;
671 	struct p9_conn *m = &ts->conn;
672 
673 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
674 		 m, current, &req->tc, req->tc.id);
675 	if (m->err < 0)
676 		return m->err;
677 
678 	spin_lock(&m->req_lock);
679 	req->status = REQ_STATUS_UNSENT;
680 	list_add_tail(&req->req_list, &m->unsent_req_list);
681 	spin_unlock(&m->req_lock);
682 
683 	if (test_and_clear_bit(Wpending, &m->wsched))
684 		n = EPOLLOUT;
685 	else
686 		n = p9_fd_poll(m->client, NULL, NULL);
687 
688 	if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
689 		schedule_work(&m->wq);
690 
691 	return 0;
692 }
693 
694 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
695 {
696 	struct p9_trans_fd *ts = client->trans;
697 	struct p9_conn *m = &ts->conn;
698 	int ret = 1;
699 
700 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
701 
702 	spin_lock(&m->req_lock);
703 
704 	if (req->status == REQ_STATUS_UNSENT) {
705 		list_del(&req->req_list);
706 		req->status = REQ_STATUS_FLSHD;
707 		p9_req_put(client, req);
708 		ret = 0;
709 	}
710 	spin_unlock(&m->req_lock);
711 
712 	return ret;
713 }
714 
715 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
716 {
717 	struct p9_trans_fd *ts = client->trans;
718 	struct p9_conn *m = &ts->conn;
719 
720 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
721 
722 	spin_lock(&m->req_lock);
723 	/* Ignore cancelled request if message has been received
724 	 * before lock.
725 	 */
726 	if (req->status == REQ_STATUS_RCVD) {
727 		spin_unlock(&m->req_lock);
728 		return 0;
729 	}
730 
731 	/* we haven't received a response for oldreq,
732 	 * remove it from the list.
733 	 */
734 	list_del(&req->req_list);
735 	req->status = REQ_STATUS_FLSHD;
736 	spin_unlock(&m->req_lock);
737 
738 	p9_req_put(client, req);
739 
740 	return 0;
741 }
742 
743 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
744 {
745 	if (clnt->trans_mod == &p9_tcp_trans) {
746 		if (clnt->trans_opts.tcp.port != P9_PORT)
747 			seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
748 	} else if (clnt->trans_mod == &p9_fd_trans) {
749 		if (clnt->trans_opts.fd.rfd != ~0)
750 			seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
751 		if (clnt->trans_opts.fd.wfd != ~0)
752 			seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
753 	}
754 	return 0;
755 }
756 
757 /**
758  * parse_opts - parse mount options into p9_fd_opts structure
759  * @params: options string passed from mount
760  * @opts: fd transport-specific structure to parse options into
761  *
762  * Returns 0 upon success, -ERRNO upon failure
763  */
764 
765 static int parse_opts(char *params, struct p9_fd_opts *opts)
766 {
767 	char *p;
768 	substring_t args[MAX_OPT_ARGS];
769 	int option;
770 	char *options, *tmp_options;
771 
772 	opts->port = P9_PORT;
773 	opts->rfd = ~0;
774 	opts->wfd = ~0;
775 	opts->privport = false;
776 
777 	if (!params)
778 		return 0;
779 
780 	tmp_options = kstrdup(params, GFP_KERNEL);
781 	if (!tmp_options) {
782 		p9_debug(P9_DEBUG_ERROR,
783 			 "failed to allocate copy of option string\n");
784 		return -ENOMEM;
785 	}
786 	options = tmp_options;
787 
788 	while ((p = strsep(&options, ",")) != NULL) {
789 		int token;
790 		int r;
791 		if (!*p)
792 			continue;
793 		token = match_token(p, tokens, args);
794 		if ((token != Opt_err) && (token != Opt_privport)) {
795 			r = match_int(&args[0], &option);
796 			if (r < 0) {
797 				p9_debug(P9_DEBUG_ERROR,
798 					 "integer field, but no integer?\n");
799 				continue;
800 			}
801 		}
802 		switch (token) {
803 		case Opt_port:
804 			opts->port = option;
805 			break;
806 		case Opt_rfdno:
807 			opts->rfd = option;
808 			break;
809 		case Opt_wfdno:
810 			opts->wfd = option;
811 			break;
812 		case Opt_privport:
813 			opts->privport = true;
814 			break;
815 		default:
816 			continue;
817 		}
818 	}
819 
820 	kfree(tmp_options);
821 	return 0;
822 }
823 
824 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
825 {
826 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
827 					   GFP_KERNEL);
828 	if (!ts)
829 		return -ENOMEM;
830 
831 	ts->rd = fget(rfd);
832 	if (!ts->rd)
833 		goto out_free_ts;
834 	if (!(ts->rd->f_mode & FMODE_READ))
835 		goto out_put_rd;
836 	/* prevent workers from hanging on IO when fd is a pipe */
837 	ts->rd->f_flags |= O_NONBLOCK;
838 	ts->wr = fget(wfd);
839 	if (!ts->wr)
840 		goto out_put_rd;
841 	if (!(ts->wr->f_mode & FMODE_WRITE))
842 		goto out_put_wr;
843 	ts->wr->f_flags |= O_NONBLOCK;
844 
845 	client->trans = ts;
846 	client->status = Connected;
847 
848 	return 0;
849 
850 out_put_wr:
851 	fput(ts->wr);
852 out_put_rd:
853 	fput(ts->rd);
854 out_free_ts:
855 	kfree(ts);
856 	return -EIO;
857 }
858 
859 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
860 {
861 	struct p9_trans_fd *p;
862 	struct file *file;
863 
864 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
865 	if (!p)
866 		return -ENOMEM;
867 
868 	csocket->sk->sk_allocation = GFP_NOIO;
869 	file = sock_alloc_file(csocket, 0, NULL);
870 	if (IS_ERR(file)) {
871 		pr_err("%s (%d): failed to map fd\n",
872 		       __func__, task_pid_nr(current));
873 		kfree(p);
874 		return PTR_ERR(file);
875 	}
876 
877 	get_file(file);
878 	p->wr = p->rd = file;
879 	client->trans = p;
880 	client->status = Connected;
881 
882 	p->rd->f_flags |= O_NONBLOCK;
883 
884 	p9_conn_create(client);
885 	return 0;
886 }
887 
888 /**
889  * p9_conn_destroy - cancels all pending requests of mux
890  * @m: mux to destroy
891  *
892  */
893 
894 static void p9_conn_destroy(struct p9_conn *m)
895 {
896 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
897 		 m, m->mux_list.prev, m->mux_list.next);
898 
899 	p9_mux_poll_stop(m);
900 	cancel_work_sync(&m->rq);
901 	if (m->rreq) {
902 		p9_req_put(m->client, m->rreq);
903 		m->rreq = NULL;
904 	}
905 	cancel_work_sync(&m->wq);
906 	if (m->wreq) {
907 		p9_req_put(m->client, m->wreq);
908 		m->wreq = NULL;
909 	}
910 
911 	p9_conn_cancel(m, -ECONNRESET);
912 
913 	m->client = NULL;
914 }
915 
916 /**
917  * p9_fd_close - shutdown file descriptor transport
918  * @client: client instance
919  *
920  */
921 
922 static void p9_fd_close(struct p9_client *client)
923 {
924 	struct p9_trans_fd *ts;
925 
926 	if (!client)
927 		return;
928 
929 	ts = client->trans;
930 	if (!ts)
931 		return;
932 
933 	client->status = Disconnected;
934 
935 	p9_conn_destroy(&ts->conn);
936 
937 	if (ts->rd)
938 		fput(ts->rd);
939 	if (ts->wr)
940 		fput(ts->wr);
941 
942 	kfree(ts);
943 }
944 
945 /*
946  * stolen from NFS - maybe should be made a generic function?
947  */
948 static inline int valid_ipaddr4(const char *buf)
949 {
950 	int rc, count, in[4];
951 
952 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
953 	if (rc != 4)
954 		return -EINVAL;
955 	for (count = 0; count < 4; count++) {
956 		if (in[count] > 255)
957 			return -EINVAL;
958 	}
959 	return 0;
960 }
961 
962 static int p9_bind_privport(struct socket *sock)
963 {
964 	struct sockaddr_in cl;
965 	int port, err = -EINVAL;
966 
967 	memset(&cl, 0, sizeof(cl));
968 	cl.sin_family = AF_INET;
969 	cl.sin_addr.s_addr = htonl(INADDR_ANY);
970 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
971 		cl.sin_port = htons((ushort)port);
972 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
973 		if (err != -EADDRINUSE)
974 			break;
975 	}
976 	return err;
977 }
978 
979 
980 static int
981 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
982 {
983 	int err;
984 	struct socket *csocket;
985 	struct sockaddr_in sin_server;
986 	struct p9_fd_opts opts;
987 
988 	err = parse_opts(args, &opts);
989 	if (err < 0)
990 		return err;
991 
992 	if (addr == NULL || valid_ipaddr4(addr) < 0)
993 		return -EINVAL;
994 
995 	csocket = NULL;
996 
997 	client->trans_opts.tcp.port = opts.port;
998 	client->trans_opts.tcp.privport = opts.privport;
999 	sin_server.sin_family = AF_INET;
1000 	sin_server.sin_addr.s_addr = in_aton(addr);
1001 	sin_server.sin_port = htons(opts.port);
1002 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
1003 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
1004 	if (err) {
1005 		pr_err("%s (%d): problem creating socket\n",
1006 		       __func__, task_pid_nr(current));
1007 		return err;
1008 	}
1009 
1010 	if (opts.privport) {
1011 		err = p9_bind_privport(csocket);
1012 		if (err < 0) {
1013 			pr_err("%s (%d): problem binding to privport\n",
1014 			       __func__, task_pid_nr(current));
1015 			sock_release(csocket);
1016 			return err;
1017 		}
1018 	}
1019 
1020 	err = csocket->ops->connect(csocket,
1021 				    (struct sockaddr *)&sin_server,
1022 				    sizeof(struct sockaddr_in), 0);
1023 	if (err < 0) {
1024 		pr_err("%s (%d): problem connecting socket to %s\n",
1025 		       __func__, task_pid_nr(current), addr);
1026 		sock_release(csocket);
1027 		return err;
1028 	}
1029 
1030 	return p9_socket_open(client, csocket);
1031 }
1032 
1033 static int
1034 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
1035 {
1036 	int err;
1037 	struct socket *csocket;
1038 	struct sockaddr_un sun_server;
1039 
1040 	csocket = NULL;
1041 
1042 	if (!addr || !strlen(addr))
1043 		return -EINVAL;
1044 
1045 	if (strlen(addr) >= UNIX_PATH_MAX) {
1046 		pr_err("%s (%d): address too long: %s\n",
1047 		       __func__, task_pid_nr(current), addr);
1048 		return -ENAMETOOLONG;
1049 	}
1050 
1051 	sun_server.sun_family = PF_UNIX;
1052 	strcpy(sun_server.sun_path, addr);
1053 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1054 			    SOCK_STREAM, 0, &csocket, 1);
1055 	if (err < 0) {
1056 		pr_err("%s (%d): problem creating socket\n",
1057 		       __func__, task_pid_nr(current));
1058 
1059 		return err;
1060 	}
1061 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1062 			sizeof(struct sockaddr_un) - 1, 0);
1063 	if (err < 0) {
1064 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1065 		       __func__, task_pid_nr(current), addr, err);
1066 		sock_release(csocket);
1067 		return err;
1068 	}
1069 
1070 	return p9_socket_open(client, csocket);
1071 }
1072 
1073 static int
1074 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1075 {
1076 	int err;
1077 	struct p9_fd_opts opts;
1078 
1079 	err = parse_opts(args, &opts);
1080 	if (err < 0)
1081 		return err;
1082 	client->trans_opts.fd.rfd = opts.rfd;
1083 	client->trans_opts.fd.wfd = opts.wfd;
1084 
1085 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1086 		pr_err("Insufficient options for proto=fd\n");
1087 		return -ENOPROTOOPT;
1088 	}
1089 
1090 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1091 	if (err < 0)
1092 		return err;
1093 
1094 	p9_conn_create(client);
1095 
1096 	return 0;
1097 }
1098 
1099 static struct p9_trans_module p9_tcp_trans = {
1100 	.name = "tcp",
1101 	.maxsize = MAX_SOCK_BUF,
1102 	.pooled_rbuffers = false,
1103 	.def = 0,
1104 	.create = p9_fd_create_tcp,
1105 	.close = p9_fd_close,
1106 	.request = p9_fd_request,
1107 	.cancel = p9_fd_cancel,
1108 	.cancelled = p9_fd_cancelled,
1109 	.show_options = p9_fd_show_options,
1110 	.owner = THIS_MODULE,
1111 };
1112 MODULE_ALIAS_9P("tcp");
1113 
1114 static struct p9_trans_module p9_unix_trans = {
1115 	.name = "unix",
1116 	.maxsize = MAX_SOCK_BUF,
1117 	.def = 0,
1118 	.create = p9_fd_create_unix,
1119 	.close = p9_fd_close,
1120 	.request = p9_fd_request,
1121 	.cancel = p9_fd_cancel,
1122 	.cancelled = p9_fd_cancelled,
1123 	.show_options = p9_fd_show_options,
1124 	.owner = THIS_MODULE,
1125 };
1126 MODULE_ALIAS_9P("unix");
1127 
1128 static struct p9_trans_module p9_fd_trans = {
1129 	.name = "fd",
1130 	.maxsize = MAX_SOCK_BUF,
1131 	.def = 0,
1132 	.create = p9_fd_create,
1133 	.close = p9_fd_close,
1134 	.request = p9_fd_request,
1135 	.cancel = p9_fd_cancel,
1136 	.cancelled = p9_fd_cancelled,
1137 	.show_options = p9_fd_show_options,
1138 	.owner = THIS_MODULE,
1139 };
1140 MODULE_ALIAS_9P("fd");
1141 
1142 /**
1143  * p9_poll_workfn - poll worker thread
1144  * @work: work queue
1145  *
1146  * polls all v9fs transports for new events and queues the appropriate
1147  * work to the work queue
1148  *
1149  */
1150 
1151 static void p9_poll_workfn(struct work_struct *work)
1152 {
1153 	unsigned long flags;
1154 
1155 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1156 
1157 	spin_lock_irqsave(&p9_poll_lock, flags);
1158 	while (!list_empty(&p9_poll_pending_list)) {
1159 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1160 							struct p9_conn,
1161 							poll_pending_link);
1162 		list_del_init(&conn->poll_pending_link);
1163 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1164 
1165 		p9_poll_mux(conn);
1166 
1167 		spin_lock_irqsave(&p9_poll_lock, flags);
1168 	}
1169 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1170 
1171 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1172 }
1173 
1174 static int __init p9_trans_fd_init(void)
1175 {
1176 	v9fs_register_trans(&p9_tcp_trans);
1177 	v9fs_register_trans(&p9_unix_trans);
1178 	v9fs_register_trans(&p9_fd_trans);
1179 
1180 	return 0;
1181 }
1182 
1183 static void __exit p9_trans_fd_exit(void)
1184 {
1185 	flush_work(&p9_poll_work);
1186 	v9fs_unregister_trans(&p9_tcp_trans);
1187 	v9fs_unregister_trans(&p9_unix_trans);
1188 	v9fs_unregister_trans(&p9_fd_trans);
1189 }
1190 
1191 module_init(p9_trans_fd_init);
1192 module_exit(p9_trans_fd_exit);
1193 
1194 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
1195 MODULE_DESCRIPTION("Filedescriptor Transport for 9P");
1196 MODULE_LICENSE("GPL");
1197