xref: /linux/net/9p/trans_fd.c (revision 79997eda0d31bc68203c95ecb978773ee6ce7a1f)
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/file.h>
24 #include <linux/parser.h>
25 #include <linux/slab.h>
26 #include <linux/seq_file.h>
27 #include <net/9p/9p.h>
28 #include <net/9p/client.h>
29 #include <net/9p/transport.h>
30 
31 #include <linux/syscalls.h> /* killme */
32 
33 #define P9_PORT 564
34 #define MAX_SOCK_BUF (1024*1024)
35 #define MAXPOLLWADDR	2
36 
37 static struct p9_trans_module p9_tcp_trans;
38 static struct p9_trans_module p9_fd_trans;
39 
40 /**
41  * struct p9_fd_opts - per-transport options
42  * @rfd: file descriptor for reading (trans=fd)
43  * @wfd: file descriptor for writing (trans=fd)
44  * @port: port to connect to (trans=tcp)
45  * @privport: port is privileged
46  */
47 
48 struct p9_fd_opts {
49 	int rfd;
50 	int wfd;
51 	u16 port;
52 	bool privport;
53 };
54 
55 /*
56   * Option Parsing (code inspired by NFS code)
57   *  - a little lazy - parse all fd-transport options
58   */
59 
60 enum {
61 	/* Options that take integer arguments */
62 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
63 	/* Options that take no arguments */
64 	Opt_privport,
65 };
66 
67 static const match_table_t tokens = {
68 	{Opt_port, "port=%u"},
69 	{Opt_rfdno, "rfdno=%u"},
70 	{Opt_wfdno, "wfdno=%u"},
71 	{Opt_privport, "privport"},
72 	{Opt_err, NULL},
73 };
74 
75 enum {
76 	Rworksched = 1,		/* read work scheduled or running */
77 	Rpending = 2,		/* can read */
78 	Wworksched = 4,		/* write work scheduled or running */
79 	Wpending = 8,		/* can write */
80 };
81 
82 struct p9_poll_wait {
83 	struct p9_conn *conn;
84 	wait_queue_entry_t wait;
85 	wait_queue_head_t *wait_addr;
86 };
87 
88 /**
89  * struct p9_conn - fd mux connection state information
90  * @mux_list: list link for mux to manage multiple connections (?)
91  * @client: reference to client instance for this connection
92  * @err: error state
93  * @req_lock: lock protecting req_list and requests statuses
94  * @req_list: accounting for requests which have been sent
95  * @unsent_req_list: accounting for requests that haven't been sent
96  * @rreq: read request
97  * @wreq: write request
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 	spinlock_t req_lock;
118 	struct list_head req_list;
119 	struct list_head unsent_req_list;
120 	struct p9_req_t *rreq;
121 	struct p9_req_t *wreq;
122 	char tmp_buf[P9_HDRSZ];
123 	struct p9_fcall rc;
124 	int wpos;
125 	int wsize;
126 	char *wbuf;
127 	struct list_head poll_pending_link;
128 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
129 	poll_table pt;
130 	struct work_struct rq;
131 	struct work_struct wq;
132 	unsigned long wsched;
133 };
134 
135 /**
136  * struct p9_trans_fd - transport state
137  * @rd: reference to file to read from
138  * @wr: reference of file to write to
139  * @conn: connection state reference
140  *
141  */
142 
143 struct p9_trans_fd {
144 	struct file *rd;
145 	struct file *wr;
146 	struct p9_conn conn;
147 };
148 
149 static void p9_poll_workfn(struct work_struct *work);
150 
151 static DEFINE_SPINLOCK(p9_poll_lock);
152 static LIST_HEAD(p9_poll_pending_list);
153 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
154 
155 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
156 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
157 
158 static void p9_mux_poll_stop(struct p9_conn *m)
159 {
160 	unsigned long flags;
161 	int i;
162 
163 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
164 		struct p9_poll_wait *pwait = &m->poll_wait[i];
165 
166 		if (pwait->wait_addr) {
167 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
168 			pwait->wait_addr = NULL;
169 		}
170 	}
171 
172 	spin_lock_irqsave(&p9_poll_lock, flags);
173 	list_del_init(&m->poll_pending_link);
174 	spin_unlock_irqrestore(&p9_poll_lock, flags);
175 
176 	flush_work(&p9_poll_work);
177 }
178 
179 /**
180  * p9_conn_cancel - cancel all pending requests with error
181  * @m: mux data
182  * @err: error code
183  *
184  */
185 
186 static void p9_conn_cancel(struct p9_conn *m, int err)
187 {
188 	struct p9_req_t *req, *rtmp;
189 	LIST_HEAD(cancel_list);
190 
191 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
192 
193 	spin_lock(&m->req_lock);
194 
195 	if (m->err) {
196 		spin_unlock(&m->req_lock);
197 		return;
198 	}
199 
200 	m->err = err;
201 
202 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
203 		list_move(&req->req_list, &cancel_list);
204 		WRITE_ONCE(req->status, REQ_STATUS_ERROR);
205 	}
206 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
207 		list_move(&req->req_list, &cancel_list);
208 		WRITE_ONCE(req->status, REQ_STATUS_ERROR);
209 	}
210 
211 	spin_unlock(&m->req_lock);
212 
213 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
214 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
215 		list_del(&req->req_list);
216 		if (!req->t_err)
217 			req->t_err = err;
218 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
219 	}
220 }
221 
222 static __poll_t
223 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
224 {
225 	__poll_t ret;
226 	struct p9_trans_fd *ts = NULL;
227 
228 	if (client && client->status == Connected)
229 		ts = client->trans;
230 
231 	if (!ts) {
232 		if (err)
233 			*err = -EREMOTEIO;
234 		return EPOLLERR;
235 	}
236 
237 	ret = vfs_poll(ts->rd, pt);
238 	if (ts->rd != ts->wr)
239 		ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN);
240 	return ret;
241 }
242 
243 /**
244  * p9_fd_read- read from a fd
245  * @client: client instance
246  * @v: buffer to receive data into
247  * @len: size of receive buffer
248  *
249  */
250 
251 static int p9_fd_read(struct p9_client *client, void *v, int len)
252 {
253 	int ret;
254 	struct p9_trans_fd *ts = NULL;
255 	loff_t pos;
256 
257 	if (client && client->status != Disconnected)
258 		ts = client->trans;
259 
260 	if (!ts)
261 		return -EREMOTEIO;
262 
263 	if (!(ts->rd->f_flags & O_NONBLOCK))
264 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
265 
266 	pos = ts->rd->f_pos;
267 	ret = kernel_read(ts->rd, v, len, &pos);
268 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
269 		client->status = Disconnected;
270 	return ret;
271 }
272 
273 /**
274  * p9_read_work - called when there is some data to be read from a transport
275  * @work: container of work to be done
276  *
277  */
278 
279 static void p9_read_work(struct work_struct *work)
280 {
281 	__poll_t n;
282 	int err;
283 	struct p9_conn *m;
284 
285 	m = container_of(work, struct p9_conn, rq);
286 
287 	if (m->err < 0)
288 		return;
289 
290 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
291 
292 	if (!m->rc.sdata) {
293 		m->rc.sdata = m->tmp_buf;
294 		m->rc.offset = 0;
295 		m->rc.capacity = P9_HDRSZ; /* start by reading header */
296 	}
297 
298 	clear_bit(Rpending, &m->wsched);
299 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
300 		 m, m->rc.offset, m->rc.capacity,
301 		 m->rc.capacity - m->rc.offset);
302 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
303 			 m->rc.capacity - m->rc.offset);
304 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
305 	if (err == -EAGAIN)
306 		goto end_clear;
307 
308 	if (err <= 0)
309 		goto error;
310 
311 	m->rc.offset += err;
312 
313 	/* header read in */
314 	if ((!m->rreq) && (m->rc.offset == m->rc.capacity)) {
315 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
316 
317 		/* Header size */
318 		m->rc.size = P9_HDRSZ;
319 		err = p9_parse_header(&m->rc, &m->rc.size, NULL, NULL, 0);
320 		if (err) {
321 			p9_debug(P9_DEBUG_ERROR,
322 				 "error parsing header: %d\n", err);
323 			goto error;
324 		}
325 
326 		p9_debug(P9_DEBUG_TRANS,
327 			 "mux %p pkt: size: %d bytes tag: %d\n",
328 			 m, m->rc.size, m->rc.tag);
329 
330 		m->rreq = p9_tag_lookup(m->client, m->rc.tag);
331 		if (!m->rreq || (m->rreq->status != REQ_STATUS_SENT)) {
332 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
333 				 m->rc.tag);
334 			err = -EIO;
335 			goto error;
336 		}
337 
338 		if (m->rc.size > m->rreq->rc.capacity) {
339 			p9_debug(P9_DEBUG_ERROR,
340 				 "requested packet size too big: %d for tag %d with capacity %zd\n",
341 				 m->rc.size, m->rc.tag, m->rreq->rc.capacity);
342 			err = -EIO;
343 			goto error;
344 		}
345 
346 		if (!m->rreq->rc.sdata) {
347 			p9_debug(P9_DEBUG_ERROR,
348 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
349 				 m->rc.tag, m->rreq);
350 			p9_req_put(m->client, m->rreq);
351 			m->rreq = NULL;
352 			err = -EIO;
353 			goto error;
354 		}
355 		m->rc.sdata = m->rreq->rc.sdata;
356 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
357 		m->rc.capacity = m->rc.size;
358 	}
359 
360 	/* packet is read in
361 	 * not an else because some packets (like clunk) have no payload
362 	 */
363 	if ((m->rreq) && (m->rc.offset == m->rc.capacity)) {
364 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
365 		m->rreq->rc.size = m->rc.offset;
366 		spin_lock(&m->req_lock);
367 		if (m->rreq->status == REQ_STATUS_SENT) {
368 			list_del(&m->rreq->req_list);
369 			p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD);
370 		} else if (m->rreq->status == REQ_STATUS_FLSHD) {
371 			/* Ignore replies associated with a cancelled request. */
372 			p9_debug(P9_DEBUG_TRANS,
373 				 "Ignore replies associated with a cancelled request\n");
374 		} else {
375 			spin_unlock(&m->req_lock);
376 			p9_debug(P9_DEBUG_ERROR,
377 				 "Request tag %d errored out while we were reading the reply\n",
378 				 m->rc.tag);
379 			err = -EIO;
380 			goto error;
381 		}
382 		spin_unlock(&m->req_lock);
383 		m->rc.sdata = NULL;
384 		m->rc.offset = 0;
385 		m->rc.capacity = 0;
386 		p9_req_put(m->client, m->rreq);
387 		m->rreq = NULL;
388 	}
389 
390 end_clear:
391 	clear_bit(Rworksched, &m->wsched);
392 
393 	if (!list_empty(&m->req_list)) {
394 		if (test_and_clear_bit(Rpending, &m->wsched))
395 			n = EPOLLIN;
396 		else
397 			n = p9_fd_poll(m->client, NULL, NULL);
398 
399 		if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
400 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
401 			schedule_work(&m->rq);
402 		}
403 	}
404 
405 	return;
406 error:
407 	p9_conn_cancel(m, err);
408 	clear_bit(Rworksched, &m->wsched);
409 }
410 
411 /**
412  * p9_fd_write - write to a socket
413  * @client: client instance
414  * @v: buffer to send data from
415  * @len: size of send buffer
416  *
417  */
418 
419 static int p9_fd_write(struct p9_client *client, void *v, int len)
420 {
421 	ssize_t ret;
422 	struct p9_trans_fd *ts = NULL;
423 
424 	if (client && client->status != Disconnected)
425 		ts = client->trans;
426 
427 	if (!ts)
428 		return -EREMOTEIO;
429 
430 	if (!(ts->wr->f_flags & O_NONBLOCK))
431 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
432 
433 	ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
434 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
435 		client->status = Disconnected;
436 	return ret;
437 }
438 
439 /**
440  * p9_write_work - called when a transport can send some data
441  * @work: container for work to be done
442  *
443  */
444 
445 static void p9_write_work(struct work_struct *work)
446 {
447 	__poll_t n;
448 	int err;
449 	struct p9_conn *m;
450 	struct p9_req_t *req;
451 
452 	m = container_of(work, struct p9_conn, wq);
453 
454 	if (m->err < 0) {
455 		clear_bit(Wworksched, &m->wsched);
456 		return;
457 	}
458 
459 	if (!m->wsize) {
460 		spin_lock(&m->req_lock);
461 		if (list_empty(&m->unsent_req_list)) {
462 			clear_bit(Wworksched, &m->wsched);
463 			spin_unlock(&m->req_lock);
464 			return;
465 		}
466 
467 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
468 			       req_list);
469 		WRITE_ONCE(req->status, REQ_STATUS_SENT);
470 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
471 		list_move_tail(&req->req_list, &m->req_list);
472 
473 		m->wbuf = req->tc.sdata;
474 		m->wsize = req->tc.size;
475 		m->wpos = 0;
476 		p9_req_get(req);
477 		m->wreq = req;
478 		spin_unlock(&m->req_lock);
479 	}
480 
481 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
482 		 m, m->wpos, m->wsize);
483 	clear_bit(Wpending, &m->wsched);
484 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
485 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
486 	if (err == -EAGAIN)
487 		goto end_clear;
488 
489 
490 	if (err < 0)
491 		goto error;
492 	else if (err == 0) {
493 		err = -EREMOTEIO;
494 		goto error;
495 	}
496 
497 	m->wpos += err;
498 	if (m->wpos == m->wsize) {
499 		m->wpos = m->wsize = 0;
500 		p9_req_put(m->client, m->wreq);
501 		m->wreq = NULL;
502 	}
503 
504 end_clear:
505 	clear_bit(Wworksched, &m->wsched);
506 
507 	if (m->wsize || !list_empty(&m->unsent_req_list)) {
508 		if (test_and_clear_bit(Wpending, &m->wsched))
509 			n = EPOLLOUT;
510 		else
511 			n = p9_fd_poll(m->client, NULL, NULL);
512 
513 		if ((n & EPOLLOUT) &&
514 		   !test_and_set_bit(Wworksched, &m->wsched)) {
515 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
516 			schedule_work(&m->wq);
517 		}
518 	}
519 
520 	return;
521 
522 error:
523 	p9_conn_cancel(m, err);
524 	clear_bit(Wworksched, &m->wsched);
525 }
526 
527 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
528 {
529 	struct p9_poll_wait *pwait =
530 		container_of(wait, struct p9_poll_wait, wait);
531 	struct p9_conn *m = pwait->conn;
532 	unsigned long flags;
533 
534 	spin_lock_irqsave(&p9_poll_lock, flags);
535 	if (list_empty(&m->poll_pending_link))
536 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
537 	spin_unlock_irqrestore(&p9_poll_lock, flags);
538 
539 	schedule_work(&p9_poll_work);
540 	return 1;
541 }
542 
543 /**
544  * p9_pollwait - add poll task to the wait queue
545  * @filp: file pointer being polled
546  * @wait_address: wait_q to block on
547  * @p: poll state
548  *
549  * called by files poll operation to add v9fs-poll task to files wait queue
550  */
551 
552 static void
553 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
554 {
555 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
556 	struct p9_poll_wait *pwait = NULL;
557 	int i;
558 
559 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
560 		if (m->poll_wait[i].wait_addr == NULL) {
561 			pwait = &m->poll_wait[i];
562 			break;
563 		}
564 	}
565 
566 	if (!pwait) {
567 		p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
568 		return;
569 	}
570 
571 	pwait->conn = m;
572 	pwait->wait_addr = wait_address;
573 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
574 	add_wait_queue(wait_address, &pwait->wait);
575 }
576 
577 /**
578  * p9_conn_create - initialize the per-session mux data
579  * @client: client instance
580  *
581  * Note: Creates the polling task if this is the first session.
582  */
583 
584 static void p9_conn_create(struct p9_client *client)
585 {
586 	__poll_t n;
587 	struct p9_trans_fd *ts = client->trans;
588 	struct p9_conn *m = &ts->conn;
589 
590 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
591 
592 	INIT_LIST_HEAD(&m->mux_list);
593 	m->client = client;
594 
595 	spin_lock_init(&m->req_lock);
596 	INIT_LIST_HEAD(&m->req_list);
597 	INIT_LIST_HEAD(&m->unsent_req_list);
598 	INIT_WORK(&m->rq, p9_read_work);
599 	INIT_WORK(&m->wq, p9_write_work);
600 	INIT_LIST_HEAD(&m->poll_pending_link);
601 	init_poll_funcptr(&m->pt, p9_pollwait);
602 
603 	n = p9_fd_poll(client, &m->pt, NULL);
604 	if (n & EPOLLIN) {
605 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
606 		set_bit(Rpending, &m->wsched);
607 	}
608 
609 	if (n & EPOLLOUT) {
610 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
611 		set_bit(Wpending, &m->wsched);
612 	}
613 }
614 
615 /**
616  * p9_poll_mux - polls a mux and schedules read or write works if necessary
617  * @m: connection to poll
618  *
619  */
620 
621 static void p9_poll_mux(struct p9_conn *m)
622 {
623 	__poll_t n;
624 	int err = -ECONNRESET;
625 
626 	if (m->err < 0)
627 		return;
628 
629 	n = p9_fd_poll(m->client, NULL, &err);
630 	if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
631 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
632 		p9_conn_cancel(m, err);
633 	}
634 
635 	if (n & EPOLLIN) {
636 		set_bit(Rpending, &m->wsched);
637 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
638 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
639 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
640 			schedule_work(&m->rq);
641 		}
642 	}
643 
644 	if (n & EPOLLOUT) {
645 		set_bit(Wpending, &m->wsched);
646 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
647 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
648 		    !test_and_set_bit(Wworksched, &m->wsched)) {
649 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
650 			schedule_work(&m->wq);
651 		}
652 	}
653 }
654 
655 /**
656  * p9_fd_request - send 9P request
657  * The function can sleep until the request is scheduled for sending.
658  * The function can be interrupted. Return from the function is not
659  * a guarantee that the request is sent successfully.
660  *
661  * @client: client instance
662  * @req: request to be sent
663  *
664  */
665 
666 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
667 {
668 	__poll_t n;
669 	struct p9_trans_fd *ts = client->trans;
670 	struct p9_conn *m = &ts->conn;
671 
672 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
673 		 m, current, &req->tc, req->tc.id);
674 
675 	spin_lock(&m->req_lock);
676 
677 	if (m->err < 0) {
678 		spin_unlock(&m->req_lock);
679 		return m->err;
680 	}
681 
682 	WRITE_ONCE(req->status, REQ_STATUS_UNSENT);
683 	list_add_tail(&req->req_list, &m->unsent_req_list);
684 	spin_unlock(&m->req_lock);
685 
686 	if (test_and_clear_bit(Wpending, &m->wsched))
687 		n = EPOLLOUT;
688 	else
689 		n = p9_fd_poll(m->client, NULL, NULL);
690 
691 	if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
692 		schedule_work(&m->wq);
693 
694 	return 0;
695 }
696 
697 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
698 {
699 	struct p9_trans_fd *ts = client->trans;
700 	struct p9_conn *m = &ts->conn;
701 	int ret = 1;
702 
703 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
704 
705 	spin_lock(&m->req_lock);
706 
707 	if (req->status == REQ_STATUS_UNSENT) {
708 		list_del(&req->req_list);
709 		WRITE_ONCE(req->status, REQ_STATUS_FLSHD);
710 		p9_req_put(client, req);
711 		ret = 0;
712 	}
713 	spin_unlock(&m->req_lock);
714 
715 	return ret;
716 }
717 
718 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
719 {
720 	struct p9_trans_fd *ts = client->trans;
721 	struct p9_conn *m = &ts->conn;
722 
723 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
724 
725 	spin_lock(&m->req_lock);
726 	/* Ignore cancelled request if message has been received
727 	 * before lock.
728 	 */
729 	if (req->status == REQ_STATUS_RCVD) {
730 		spin_unlock(&m->req_lock);
731 		return 0;
732 	}
733 
734 	/* we haven't received a response for oldreq,
735 	 * remove it from the list.
736 	 */
737 	list_del(&req->req_list);
738 	WRITE_ONCE(req->status, REQ_STATUS_FLSHD);
739 	spin_unlock(&m->req_lock);
740 
741 	p9_req_put(client, req);
742 
743 	return 0;
744 }
745 
746 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
747 {
748 	if (clnt->trans_mod == &p9_tcp_trans) {
749 		if (clnt->trans_opts.tcp.port != P9_PORT)
750 			seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
751 	} else if (clnt->trans_mod == &p9_fd_trans) {
752 		if (clnt->trans_opts.fd.rfd != ~0)
753 			seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
754 		if (clnt->trans_opts.fd.wfd != ~0)
755 			seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
756 	}
757 	return 0;
758 }
759 
760 /**
761  * parse_opts - parse mount options into p9_fd_opts structure
762  * @params: options string passed from mount
763  * @opts: fd transport-specific structure to parse options into
764  *
765  * Returns 0 upon success, -ERRNO upon failure
766  */
767 
768 static int parse_opts(char *params, struct p9_fd_opts *opts)
769 {
770 	char *p;
771 	substring_t args[MAX_OPT_ARGS];
772 	int option;
773 	char *options, *tmp_options;
774 
775 	opts->port = P9_PORT;
776 	opts->rfd = ~0;
777 	opts->wfd = ~0;
778 	opts->privport = false;
779 
780 	if (!params)
781 		return 0;
782 
783 	tmp_options = kstrdup(params, GFP_KERNEL);
784 	if (!tmp_options) {
785 		p9_debug(P9_DEBUG_ERROR,
786 			 "failed to allocate copy of option string\n");
787 		return -ENOMEM;
788 	}
789 	options = tmp_options;
790 
791 	while ((p = strsep(&options, ",")) != NULL) {
792 		int token;
793 		int r;
794 		if (!*p)
795 			continue;
796 		token = match_token(p, tokens, args);
797 		if ((token != Opt_err) && (token != Opt_privport)) {
798 			r = match_int(&args[0], &option);
799 			if (r < 0) {
800 				p9_debug(P9_DEBUG_ERROR,
801 					 "integer field, but no integer?\n");
802 				continue;
803 			}
804 		}
805 		switch (token) {
806 		case Opt_port:
807 			opts->port = option;
808 			break;
809 		case Opt_rfdno:
810 			opts->rfd = option;
811 			break;
812 		case Opt_wfdno:
813 			opts->wfd = option;
814 			break;
815 		case Opt_privport:
816 			opts->privport = true;
817 			break;
818 		default:
819 			continue;
820 		}
821 	}
822 
823 	kfree(tmp_options);
824 	return 0;
825 }
826 
827 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
828 {
829 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
830 					   GFP_KERNEL);
831 	if (!ts)
832 		return -ENOMEM;
833 
834 	ts->rd = fget(rfd);
835 	if (!ts->rd)
836 		goto out_free_ts;
837 	if (!(ts->rd->f_mode & FMODE_READ))
838 		goto out_put_rd;
839 	/* Prevent workers from hanging on IO when fd is a pipe.
840 	 * It's technically possible for userspace or concurrent mounts to
841 	 * modify this flag concurrently, which will likely result in a
842 	 * broken filesystem. However, just having bad flags here should
843 	 * not crash the kernel or cause any other sort of bug, so mark this
844 	 * particular data race as intentional so that tooling (like KCSAN)
845 	 * can allow it and detect further problems.
846 	 */
847 	data_race(ts->rd->f_flags |= O_NONBLOCK);
848 	ts->wr = fget(wfd);
849 	if (!ts->wr)
850 		goto out_put_rd;
851 	if (!(ts->wr->f_mode & FMODE_WRITE))
852 		goto out_put_wr;
853 	data_race(ts->wr->f_flags |= O_NONBLOCK);
854 
855 	client->trans = ts;
856 	client->status = Connected;
857 
858 	return 0;
859 
860 out_put_wr:
861 	fput(ts->wr);
862 out_put_rd:
863 	fput(ts->rd);
864 out_free_ts:
865 	kfree(ts);
866 	return -EIO;
867 }
868 
869 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
870 {
871 	struct p9_trans_fd *p;
872 	struct file *file;
873 
874 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
875 	if (!p) {
876 		sock_release(csocket);
877 		return -ENOMEM;
878 	}
879 
880 	csocket->sk->sk_allocation = GFP_NOIO;
881 	csocket->sk->sk_use_task_frag = false;
882 	file = sock_alloc_file(csocket, 0, NULL);
883 	if (IS_ERR(file)) {
884 		pr_err("%s (%d): failed to map fd\n",
885 		       __func__, task_pid_nr(current));
886 		kfree(p);
887 		return PTR_ERR(file);
888 	}
889 
890 	get_file(file);
891 	p->wr = p->rd = file;
892 	client->trans = p;
893 	client->status = Connected;
894 
895 	p->rd->f_flags |= O_NONBLOCK;
896 
897 	p9_conn_create(client);
898 	return 0;
899 }
900 
901 /**
902  * p9_conn_destroy - cancels all pending requests of mux
903  * @m: mux to destroy
904  *
905  */
906 
907 static void p9_conn_destroy(struct p9_conn *m)
908 {
909 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
910 		 m, m->mux_list.prev, m->mux_list.next);
911 
912 	p9_mux_poll_stop(m);
913 	cancel_work_sync(&m->rq);
914 	if (m->rreq) {
915 		p9_req_put(m->client, m->rreq);
916 		m->rreq = NULL;
917 	}
918 	cancel_work_sync(&m->wq);
919 	if (m->wreq) {
920 		p9_req_put(m->client, m->wreq);
921 		m->wreq = NULL;
922 	}
923 
924 	p9_conn_cancel(m, -ECONNRESET);
925 
926 	m->client = NULL;
927 }
928 
929 /**
930  * p9_fd_close - shutdown file descriptor transport
931  * @client: client instance
932  *
933  */
934 
935 static void p9_fd_close(struct p9_client *client)
936 {
937 	struct p9_trans_fd *ts;
938 
939 	if (!client)
940 		return;
941 
942 	ts = client->trans;
943 	if (!ts)
944 		return;
945 
946 	client->status = Disconnected;
947 
948 	p9_conn_destroy(&ts->conn);
949 
950 	if (ts->rd)
951 		fput(ts->rd);
952 	if (ts->wr)
953 		fput(ts->wr);
954 
955 	kfree(ts);
956 }
957 
958 /*
959  * stolen from NFS - maybe should be made a generic function?
960  */
961 static inline int valid_ipaddr4(const char *buf)
962 {
963 	int rc, count, in[4];
964 
965 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
966 	if (rc != 4)
967 		return -EINVAL;
968 	for (count = 0; count < 4; count++) {
969 		if (in[count] > 255)
970 			return -EINVAL;
971 	}
972 	return 0;
973 }
974 
975 static int p9_bind_privport(struct socket *sock)
976 {
977 	struct sockaddr_in cl;
978 	int port, err = -EINVAL;
979 
980 	memset(&cl, 0, sizeof(cl));
981 	cl.sin_family = AF_INET;
982 	cl.sin_addr.s_addr = htonl(INADDR_ANY);
983 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
984 		cl.sin_port = htons((ushort)port);
985 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
986 		if (err != -EADDRINUSE)
987 			break;
988 	}
989 	return err;
990 }
991 
992 
993 static int
994 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
995 {
996 	int err;
997 	struct socket *csocket;
998 	struct sockaddr_in sin_server;
999 	struct p9_fd_opts opts;
1000 
1001 	err = parse_opts(args, &opts);
1002 	if (err < 0)
1003 		return err;
1004 
1005 	if (addr == NULL || valid_ipaddr4(addr) < 0)
1006 		return -EINVAL;
1007 
1008 	csocket = NULL;
1009 
1010 	client->trans_opts.tcp.port = opts.port;
1011 	client->trans_opts.tcp.privport = opts.privport;
1012 	sin_server.sin_family = AF_INET;
1013 	sin_server.sin_addr.s_addr = in_aton(addr);
1014 	sin_server.sin_port = htons(opts.port);
1015 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
1016 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
1017 	if (err) {
1018 		pr_err("%s (%d): problem creating socket\n",
1019 		       __func__, task_pid_nr(current));
1020 		return err;
1021 	}
1022 
1023 	if (opts.privport) {
1024 		err = p9_bind_privport(csocket);
1025 		if (err < 0) {
1026 			pr_err("%s (%d): problem binding to privport\n",
1027 			       __func__, task_pid_nr(current));
1028 			sock_release(csocket);
1029 			return err;
1030 		}
1031 	}
1032 
1033 	err = READ_ONCE(csocket->ops)->connect(csocket,
1034 				    (struct sockaddr *)&sin_server,
1035 				    sizeof(struct sockaddr_in), 0);
1036 	if (err < 0) {
1037 		pr_err("%s (%d): problem connecting socket to %s\n",
1038 		       __func__, task_pid_nr(current), addr);
1039 		sock_release(csocket);
1040 		return err;
1041 	}
1042 
1043 	return p9_socket_open(client, csocket);
1044 }
1045 
1046 static int
1047 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
1048 {
1049 	int err;
1050 	struct socket *csocket;
1051 	struct sockaddr_un sun_server;
1052 
1053 	csocket = NULL;
1054 
1055 	if (!addr || !strlen(addr))
1056 		return -EINVAL;
1057 
1058 	if (strlen(addr) >= UNIX_PATH_MAX) {
1059 		pr_err("%s (%d): address too long: %s\n",
1060 		       __func__, task_pid_nr(current), addr);
1061 		return -ENAMETOOLONG;
1062 	}
1063 
1064 	sun_server.sun_family = PF_UNIX;
1065 	strcpy(sun_server.sun_path, addr);
1066 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1067 			    SOCK_STREAM, 0, &csocket, 1);
1068 	if (err < 0) {
1069 		pr_err("%s (%d): problem creating socket\n",
1070 		       __func__, task_pid_nr(current));
1071 
1072 		return err;
1073 	}
1074 	err = READ_ONCE(csocket->ops)->connect(csocket, (struct sockaddr *)&sun_server,
1075 			sizeof(struct sockaddr_un) - 1, 0);
1076 	if (err < 0) {
1077 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1078 		       __func__, task_pid_nr(current), addr, err);
1079 		sock_release(csocket);
1080 		return err;
1081 	}
1082 
1083 	return p9_socket_open(client, csocket);
1084 }
1085 
1086 static int
1087 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1088 {
1089 	int err;
1090 	struct p9_fd_opts opts;
1091 
1092 	err = parse_opts(args, &opts);
1093 	if (err < 0)
1094 		return err;
1095 	client->trans_opts.fd.rfd = opts.rfd;
1096 	client->trans_opts.fd.wfd = opts.wfd;
1097 
1098 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1099 		pr_err("Insufficient options for proto=fd\n");
1100 		return -ENOPROTOOPT;
1101 	}
1102 
1103 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1104 	if (err < 0)
1105 		return err;
1106 
1107 	p9_conn_create(client);
1108 
1109 	return 0;
1110 }
1111 
1112 static struct p9_trans_module p9_tcp_trans = {
1113 	.name = "tcp",
1114 	.maxsize = MAX_SOCK_BUF,
1115 	.pooled_rbuffers = false,
1116 	.def = 0,
1117 	.create = p9_fd_create_tcp,
1118 	.close = p9_fd_close,
1119 	.request = p9_fd_request,
1120 	.cancel = p9_fd_cancel,
1121 	.cancelled = p9_fd_cancelled,
1122 	.show_options = p9_fd_show_options,
1123 	.owner = THIS_MODULE,
1124 };
1125 MODULE_ALIAS_9P("tcp");
1126 
1127 static struct p9_trans_module p9_unix_trans = {
1128 	.name = "unix",
1129 	.maxsize = MAX_SOCK_BUF,
1130 	.def = 0,
1131 	.create = p9_fd_create_unix,
1132 	.close = p9_fd_close,
1133 	.request = p9_fd_request,
1134 	.cancel = p9_fd_cancel,
1135 	.cancelled = p9_fd_cancelled,
1136 	.show_options = p9_fd_show_options,
1137 	.owner = THIS_MODULE,
1138 };
1139 MODULE_ALIAS_9P("unix");
1140 
1141 static struct p9_trans_module p9_fd_trans = {
1142 	.name = "fd",
1143 	.maxsize = MAX_SOCK_BUF,
1144 	.def = 0,
1145 	.create = p9_fd_create,
1146 	.close = p9_fd_close,
1147 	.request = p9_fd_request,
1148 	.cancel = p9_fd_cancel,
1149 	.cancelled = p9_fd_cancelled,
1150 	.show_options = p9_fd_show_options,
1151 	.owner = THIS_MODULE,
1152 };
1153 MODULE_ALIAS_9P("fd");
1154 
1155 /**
1156  * p9_poll_workfn - poll worker thread
1157  * @work: work queue
1158  *
1159  * polls all v9fs transports for new events and queues the appropriate
1160  * work to the work queue
1161  *
1162  */
1163 
1164 static void p9_poll_workfn(struct work_struct *work)
1165 {
1166 	unsigned long flags;
1167 
1168 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1169 
1170 	spin_lock_irqsave(&p9_poll_lock, flags);
1171 	while (!list_empty(&p9_poll_pending_list)) {
1172 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1173 							struct p9_conn,
1174 							poll_pending_link);
1175 		list_del_init(&conn->poll_pending_link);
1176 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1177 
1178 		p9_poll_mux(conn);
1179 
1180 		spin_lock_irqsave(&p9_poll_lock, flags);
1181 	}
1182 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1183 
1184 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1185 }
1186 
1187 static int __init p9_trans_fd_init(void)
1188 {
1189 	v9fs_register_trans(&p9_tcp_trans);
1190 	v9fs_register_trans(&p9_unix_trans);
1191 	v9fs_register_trans(&p9_fd_trans);
1192 
1193 	return 0;
1194 }
1195 
1196 static void __exit p9_trans_fd_exit(void)
1197 {
1198 	flush_work(&p9_poll_work);
1199 	v9fs_unregister_trans(&p9_tcp_trans);
1200 	v9fs_unregister_trans(&p9_unix_trans);
1201 	v9fs_unregister_trans(&p9_fd_trans);
1202 }
1203 
1204 module_init(p9_trans_fd_init);
1205 module_exit(p9_trans_fd_exit);
1206 
1207 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
1208 MODULE_DESCRIPTION("Filedescriptor Transport for 9P");
1209 MODULE_LICENSE("GPL");
1210