Loading...
1/*
2 * linux/fs/9p/trans_fd.c
3 *
4 * Fd transport layer. Includes deprecated socket layer.
5 *
6 * Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
7 * Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
8 * Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
9 * Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License version 2
13 * as published by the Free Software Foundation.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to:
22 * Free Software Foundation
23 * 51 Franklin Street, Fifth Floor
24 * Boston, MA 02111-1301 USA
25 *
26 */
27
28#include <linux/in.h>
29#include <linux/module.h>
30#include <linux/net.h>
31#include <linux/ipv6.h>
32#include <linux/kthread.h>
33#include <linux/errno.h>
34#include <linux/kernel.h>
35#include <linux/un.h>
36#include <linux/uaccess.h>
37#include <linux/inet.h>
38#include <linux/idr.h>
39#include <linux/file.h>
40#include <linux/parser.h>
41#include <linux/slab.h>
42#include <net/9p/9p.h>
43#include <net/9p/client.h>
44#include <net/9p/transport.h>
45
46#include <linux/syscalls.h> /* killme */
47
48#define P9_PORT 564
49#define MAX_SOCK_BUF (64*1024)
50#define MAXPOLLWADDR 2
51
52/**
53 * struct p9_fd_opts - per-transport options
54 * @rfd: file descriptor for reading (trans=fd)
55 * @wfd: file descriptor for writing (trans=fd)
56 * @port: port to connect to (trans=tcp)
57 *
58 */
59
60struct p9_fd_opts {
61 int rfd;
62 int wfd;
63 u16 port;
64};
65
66/**
67 * struct p9_trans_fd - transport state
68 * @rd: reference to file to read from
69 * @wr: reference of file to write to
70 * @conn: connection state reference
71 *
72 */
73
74struct p9_trans_fd {
75 struct file *rd;
76 struct file *wr;
77 struct p9_conn *conn;
78};
79
80/*
81 * Option Parsing (code inspired by NFS code)
82 * - a little lazy - parse all fd-transport options
83 */
84
85enum {
86 /* Options that take integer arguments */
87 Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
88};
89
90static const match_table_t tokens = {
91 {Opt_port, "port=%u"},
92 {Opt_rfdno, "rfdno=%u"},
93 {Opt_wfdno, "wfdno=%u"},
94 {Opt_err, NULL},
95};
96
97enum {
98 Rworksched = 1, /* read work scheduled or running */
99 Rpending = 2, /* can read */
100 Wworksched = 4, /* write work scheduled or running */
101 Wpending = 8, /* can write */
102};
103
104struct p9_poll_wait {
105 struct p9_conn *conn;
106 wait_queue_t wait;
107 wait_queue_head_t *wait_addr;
108};
109
110/**
111 * struct p9_conn - fd mux connection state information
112 * @mux_list: list link for mux to manage multiple connections (?)
113 * @client: reference to client instance for this connection
114 * @err: error state
115 * @req_list: accounting for requests which have been sent
116 * @unsent_req_list: accounting for requests that haven't been sent
117 * @req: current request being processed (if any)
118 * @tmp_buf: temporary buffer to read in header
119 * @rsize: amount to read for current frame
120 * @rpos: read position in current frame
121 * @rbuf: current read buffer
122 * @wpos: write position for current frame
123 * @wsize: amount of data to write for current frame
124 * @wbuf: current write buffer
125 * @poll_pending_link: pending links to be polled per conn
126 * @poll_wait: array of wait_q's for various worker threads
127 * @pt: poll state
128 * @rq: current read work
129 * @wq: current write work
130 * @wsched: ????
131 *
132 */
133
134struct p9_conn {
135 struct list_head mux_list;
136 struct p9_client *client;
137 int err;
138 struct list_head req_list;
139 struct list_head unsent_req_list;
140 struct p9_req_t *req;
141 char tmp_buf[7];
142 int rsize;
143 int rpos;
144 char *rbuf;
145 int wpos;
146 int wsize;
147 char *wbuf;
148 struct list_head poll_pending_link;
149 struct p9_poll_wait poll_wait[MAXPOLLWADDR];
150 poll_table pt;
151 struct work_struct rq;
152 struct work_struct wq;
153 unsigned long wsched;
154};
155
156static void p9_poll_workfn(struct work_struct *work);
157
158static DEFINE_SPINLOCK(p9_poll_lock);
159static LIST_HEAD(p9_poll_pending_list);
160static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
161
162static void p9_mux_poll_stop(struct p9_conn *m)
163{
164 unsigned long flags;
165 int i;
166
167 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
168 struct p9_poll_wait *pwait = &m->poll_wait[i];
169
170 if (pwait->wait_addr) {
171 remove_wait_queue(pwait->wait_addr, &pwait->wait);
172 pwait->wait_addr = NULL;
173 }
174 }
175
176 spin_lock_irqsave(&p9_poll_lock, flags);
177 list_del_init(&m->poll_pending_link);
178 spin_unlock_irqrestore(&p9_poll_lock, flags);
179}
180
181/**
182 * p9_conn_cancel - cancel all pending requests with error
183 * @m: mux data
184 * @err: error code
185 *
186 */
187
188static void p9_conn_cancel(struct p9_conn *m, int err)
189{
190 struct p9_req_t *req, *rtmp;
191 unsigned long flags;
192 LIST_HEAD(cancel_list);
193
194 P9_DPRINTK(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
195
196 spin_lock_irqsave(&m->client->lock, flags);
197
198 if (m->err) {
199 spin_unlock_irqrestore(&m->client->lock, flags);
200 return;
201 }
202
203 m->err = err;
204
205 list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
206 req->status = REQ_STATUS_ERROR;
207 if (!req->t_err)
208 req->t_err = err;
209 list_move(&req->req_list, &cancel_list);
210 }
211 list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
212 req->status = REQ_STATUS_ERROR;
213 if (!req->t_err)
214 req->t_err = err;
215 list_move(&req->req_list, &cancel_list);
216 }
217 spin_unlock_irqrestore(&m->client->lock, flags);
218
219 list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
220 P9_DPRINTK(P9_DEBUG_ERROR, "call back req %p\n", req);
221 list_del(&req->req_list);
222 p9_client_cb(m->client, req);
223 }
224}
225
226static int
227p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt)
228{
229 int ret, n;
230 struct p9_trans_fd *ts = NULL;
231
232 if (client && client->status == Connected)
233 ts = client->trans;
234
235 if (!ts)
236 return -EREMOTEIO;
237
238 if (!ts->rd->f_op || !ts->rd->f_op->poll)
239 return -EIO;
240
241 if (!ts->wr->f_op || !ts->wr->f_op->poll)
242 return -EIO;
243
244 ret = ts->rd->f_op->poll(ts->rd, pt);
245 if (ret < 0)
246 return ret;
247
248 if (ts->rd != ts->wr) {
249 n = ts->wr->f_op->poll(ts->wr, pt);
250 if (n < 0)
251 return n;
252 ret = (ret & ~POLLOUT) | (n & ~POLLIN);
253 }
254
255 return ret;
256}
257
258/**
259 * p9_fd_read- read from a fd
260 * @client: client instance
261 * @v: buffer to receive data into
262 * @len: size of receive buffer
263 *
264 */
265
266static int p9_fd_read(struct p9_client *client, void *v, int len)
267{
268 int ret;
269 struct p9_trans_fd *ts = NULL;
270
271 if (client && client->status != Disconnected)
272 ts = client->trans;
273
274 if (!ts)
275 return -EREMOTEIO;
276
277 if (!(ts->rd->f_flags & O_NONBLOCK))
278 P9_DPRINTK(P9_DEBUG_ERROR, "blocking read ...\n");
279
280 ret = kernel_read(ts->rd, ts->rd->f_pos, v, len);
281 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
282 client->status = Disconnected;
283 return ret;
284}
285
286/**
287 * p9_read_work - called when there is some data to be read from a transport
288 * @work: container of work to be done
289 *
290 */
291
292static void p9_read_work(struct work_struct *work)
293{
294 int n, err;
295 struct p9_conn *m;
296
297 m = container_of(work, struct p9_conn, rq);
298
299 if (m->err < 0)
300 return;
301
302 P9_DPRINTK(P9_DEBUG_TRANS, "start mux %p pos %d\n", m, m->rpos);
303
304 if (!m->rbuf) {
305 m->rbuf = m->tmp_buf;
306 m->rpos = 0;
307 m->rsize = 7; /* start by reading header */
308 }
309
310 clear_bit(Rpending, &m->wsched);
311 P9_DPRINTK(P9_DEBUG_TRANS, "read mux %p pos %d size: %d = %d\n", m,
312 m->rpos, m->rsize, m->rsize-m->rpos);
313 err = p9_fd_read(m->client, m->rbuf + m->rpos,
314 m->rsize - m->rpos);
315 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
316 if (err == -EAGAIN) {
317 clear_bit(Rworksched, &m->wsched);
318 return;
319 }
320
321 if (err <= 0)
322 goto error;
323
324 m->rpos += err;
325
326 if ((!m->req) && (m->rpos == m->rsize)) { /* header read in */
327 u16 tag;
328 P9_DPRINTK(P9_DEBUG_TRANS, "got new header\n");
329
330 n = le32_to_cpu(*(__le32 *) m->rbuf); /* read packet size */
331 if (n >= m->client->msize) {
332 P9_DPRINTK(P9_DEBUG_ERROR,
333 "requested packet size too big: %d\n", n);
334 err = -EIO;
335 goto error;
336 }
337
338 tag = le16_to_cpu(*(__le16 *) (m->rbuf+5)); /* read tag */
339 P9_DPRINTK(P9_DEBUG_TRANS,
340 "mux %p pkt: size: %d bytes tag: %d\n", m, n, tag);
341
342 m->req = p9_tag_lookup(m->client, tag);
343 if (!m->req || (m->req->status != REQ_STATUS_SENT &&
344 m->req->status != REQ_STATUS_FLSH)) {
345 P9_DPRINTK(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
346 tag);
347 err = -EIO;
348 goto error;
349 }
350
351 if (m->req->rc == NULL) {
352 m->req->rc = kmalloc(sizeof(struct p9_fcall) +
353 m->client->msize, GFP_NOFS);
354 if (!m->req->rc) {
355 m->req = NULL;
356 err = -ENOMEM;
357 goto error;
358 }
359 }
360 m->rbuf = (char *)m->req->rc + sizeof(struct p9_fcall);
361 memcpy(m->rbuf, m->tmp_buf, m->rsize);
362 m->rsize = n;
363 }
364
365 /* not an else because some packets (like clunk) have no payload */
366 if ((m->req) && (m->rpos == m->rsize)) { /* packet is read in */
367 P9_DPRINTK(P9_DEBUG_TRANS, "got new packet\n");
368 spin_lock(&m->client->lock);
369 if (m->req->status != REQ_STATUS_ERROR)
370 m->req->status = REQ_STATUS_RCVD;
371 list_del(&m->req->req_list);
372 spin_unlock(&m->client->lock);
373 p9_client_cb(m->client, m->req);
374 m->rbuf = NULL;
375 m->rpos = 0;
376 m->rsize = 0;
377 m->req = NULL;
378 }
379
380 if (!list_empty(&m->req_list)) {
381 if (test_and_clear_bit(Rpending, &m->wsched))
382 n = POLLIN;
383 else
384 n = p9_fd_poll(m->client, NULL);
385
386 if (n & POLLIN) {
387 P9_DPRINTK(P9_DEBUG_TRANS, "sched read work %p\n", m);
388 schedule_work(&m->rq);
389 } else
390 clear_bit(Rworksched, &m->wsched);
391 } else
392 clear_bit(Rworksched, &m->wsched);
393
394 return;
395error:
396 p9_conn_cancel(m, err);
397 clear_bit(Rworksched, &m->wsched);
398}
399
400/**
401 * p9_fd_write - write to a socket
402 * @client: client instance
403 * @v: buffer to send data from
404 * @len: size of send buffer
405 *
406 */
407
408static int p9_fd_write(struct p9_client *client, void *v, int len)
409{
410 int ret;
411 mm_segment_t oldfs;
412 struct p9_trans_fd *ts = NULL;
413
414 if (client && client->status != Disconnected)
415 ts = client->trans;
416
417 if (!ts)
418 return -EREMOTEIO;
419
420 if (!(ts->wr->f_flags & O_NONBLOCK))
421 P9_DPRINTK(P9_DEBUG_ERROR, "blocking write ...\n");
422
423 oldfs = get_fs();
424 set_fs(get_ds());
425 /* The cast to a user pointer is valid due to the set_fs() */
426 ret = vfs_write(ts->wr, (__force void __user *)v, len, &ts->wr->f_pos);
427 set_fs(oldfs);
428
429 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
430 client->status = Disconnected;
431 return ret;
432}
433
434/**
435 * p9_write_work - called when a transport can send some data
436 * @work: container for work to be done
437 *
438 */
439
440static void p9_write_work(struct work_struct *work)
441{
442 int n, err;
443 struct p9_conn *m;
444 struct p9_req_t *req;
445
446 m = container_of(work, struct p9_conn, wq);
447
448 if (m->err < 0) {
449 clear_bit(Wworksched, &m->wsched);
450 return;
451 }
452
453 if (!m->wsize) {
454 if (list_empty(&m->unsent_req_list)) {
455 clear_bit(Wworksched, &m->wsched);
456 return;
457 }
458
459 spin_lock(&m->client->lock);
460 req = list_entry(m->unsent_req_list.next, struct p9_req_t,
461 req_list);
462 req->status = REQ_STATUS_SENT;
463 P9_DPRINTK(P9_DEBUG_TRANS, "move req %p\n", req);
464 list_move_tail(&req->req_list, &m->req_list);
465
466 m->wbuf = req->tc->sdata;
467 m->wsize = req->tc->size;
468 m->wpos = 0;
469 spin_unlock(&m->client->lock);
470 }
471
472 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p pos %d size %d\n", m, m->wpos,
473 m->wsize);
474 clear_bit(Wpending, &m->wsched);
475 err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
476 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
477 if (err == -EAGAIN) {
478 clear_bit(Wworksched, &m->wsched);
479 return;
480 }
481
482 if (err < 0)
483 goto error;
484 else if (err == 0) {
485 err = -EREMOTEIO;
486 goto error;
487 }
488
489 m->wpos += err;
490 if (m->wpos == m->wsize)
491 m->wpos = m->wsize = 0;
492
493 if (m->wsize == 0 && !list_empty(&m->unsent_req_list)) {
494 if (test_and_clear_bit(Wpending, &m->wsched))
495 n = POLLOUT;
496 else
497 n = p9_fd_poll(m->client, NULL);
498
499 if (n & POLLOUT) {
500 P9_DPRINTK(P9_DEBUG_TRANS, "sched write work %p\n", m);
501 schedule_work(&m->wq);
502 } else
503 clear_bit(Wworksched, &m->wsched);
504 } else
505 clear_bit(Wworksched, &m->wsched);
506
507 return;
508
509error:
510 p9_conn_cancel(m, err);
511 clear_bit(Wworksched, &m->wsched);
512}
513
514static int p9_pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
515{
516 struct p9_poll_wait *pwait =
517 container_of(wait, struct p9_poll_wait, wait);
518 struct p9_conn *m = pwait->conn;
519 unsigned long flags;
520
521 spin_lock_irqsave(&p9_poll_lock, flags);
522 if (list_empty(&m->poll_pending_link))
523 list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
524 spin_unlock_irqrestore(&p9_poll_lock, flags);
525
526 schedule_work(&p9_poll_work);
527 return 1;
528}
529
530/**
531 * p9_pollwait - add poll task to the wait queue
532 * @filp: file pointer being polled
533 * @wait_address: wait_q to block on
534 * @p: poll state
535 *
536 * called by files poll operation to add v9fs-poll task to files wait queue
537 */
538
539static void
540p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
541{
542 struct p9_conn *m = container_of(p, struct p9_conn, pt);
543 struct p9_poll_wait *pwait = NULL;
544 int i;
545
546 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
547 if (m->poll_wait[i].wait_addr == NULL) {
548 pwait = &m->poll_wait[i];
549 break;
550 }
551 }
552
553 if (!pwait) {
554 P9_DPRINTK(P9_DEBUG_ERROR, "not enough wait_address slots\n");
555 return;
556 }
557
558 pwait->conn = m;
559 pwait->wait_addr = wait_address;
560 init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
561 add_wait_queue(wait_address, &pwait->wait);
562}
563
564/**
565 * p9_conn_create - allocate and initialize the per-session mux data
566 * @client: client instance
567 *
568 * Note: Creates the polling task if this is the first session.
569 */
570
571static struct p9_conn *p9_conn_create(struct p9_client *client)
572{
573 int n;
574 struct p9_conn *m;
575
576 P9_DPRINTK(P9_DEBUG_TRANS, "client %p msize %d\n", client,
577 client->msize);
578 m = kzalloc(sizeof(struct p9_conn), GFP_KERNEL);
579 if (!m)
580 return ERR_PTR(-ENOMEM);
581
582 INIT_LIST_HEAD(&m->mux_list);
583 m->client = client;
584
585 INIT_LIST_HEAD(&m->req_list);
586 INIT_LIST_HEAD(&m->unsent_req_list);
587 INIT_WORK(&m->rq, p9_read_work);
588 INIT_WORK(&m->wq, p9_write_work);
589 INIT_LIST_HEAD(&m->poll_pending_link);
590 init_poll_funcptr(&m->pt, p9_pollwait);
591
592 n = p9_fd_poll(client, &m->pt);
593 if (n & POLLIN) {
594 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can read\n", m);
595 set_bit(Rpending, &m->wsched);
596 }
597
598 if (n & POLLOUT) {
599 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can write\n", m);
600 set_bit(Wpending, &m->wsched);
601 }
602
603 return m;
604}
605
606/**
607 * p9_poll_mux - polls a mux and schedules read or write works if necessary
608 * @m: connection to poll
609 *
610 */
611
612static void p9_poll_mux(struct p9_conn *m)
613{
614 int n;
615
616 if (m->err < 0)
617 return;
618
619 n = p9_fd_poll(m->client, NULL);
620 if (n < 0 || n & (POLLERR | POLLHUP | POLLNVAL)) {
621 P9_DPRINTK(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
622 if (n >= 0)
623 n = -ECONNRESET;
624 p9_conn_cancel(m, n);
625 }
626
627 if (n & POLLIN) {
628 set_bit(Rpending, &m->wsched);
629 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can read\n", m);
630 if (!test_and_set_bit(Rworksched, &m->wsched)) {
631 P9_DPRINTK(P9_DEBUG_TRANS, "sched read work %p\n", m);
632 schedule_work(&m->rq);
633 }
634 }
635
636 if (n & POLLOUT) {
637 set_bit(Wpending, &m->wsched);
638 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p can write\n", m);
639 if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
640 !test_and_set_bit(Wworksched, &m->wsched)) {
641 P9_DPRINTK(P9_DEBUG_TRANS, "sched write work %p\n", m);
642 schedule_work(&m->wq);
643 }
644 }
645}
646
647/**
648 * p9_fd_request - send 9P request
649 * The function can sleep until the request is scheduled for sending.
650 * The function can be interrupted. Return from the function is not
651 * a guarantee that the request is sent successfully.
652 *
653 * @client: client instance
654 * @req: request to be sent
655 *
656 */
657
658static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
659{
660 int n;
661 struct p9_trans_fd *ts = client->trans;
662 struct p9_conn *m = ts->conn;
663
664 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n", m,
665 current, req->tc, req->tc->id);
666 if (m->err < 0)
667 return m->err;
668
669 spin_lock(&client->lock);
670 req->status = REQ_STATUS_UNSENT;
671 list_add_tail(&req->req_list, &m->unsent_req_list);
672 spin_unlock(&client->lock);
673
674 if (test_and_clear_bit(Wpending, &m->wsched))
675 n = POLLOUT;
676 else
677 n = p9_fd_poll(m->client, NULL);
678
679 if (n & POLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
680 schedule_work(&m->wq);
681
682 return 0;
683}
684
685static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
686{
687 int ret = 1;
688
689 P9_DPRINTK(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
690
691 spin_lock(&client->lock);
692
693 if (req->status == REQ_STATUS_UNSENT) {
694 list_del(&req->req_list);
695 req->status = REQ_STATUS_FLSHD;
696 ret = 0;
697 } else if (req->status == REQ_STATUS_SENT)
698 req->status = REQ_STATUS_FLSH;
699
700 spin_unlock(&client->lock);
701
702 return ret;
703}
704
705/**
706 * parse_opts - parse mount options into p9_fd_opts structure
707 * @params: options string passed from mount
708 * @opts: fd transport-specific structure to parse options into
709 *
710 * Returns 0 upon success, -ERRNO upon failure
711 */
712
713static int parse_opts(char *params, struct p9_fd_opts *opts)
714{
715 char *p;
716 substring_t args[MAX_OPT_ARGS];
717 int option;
718 char *options, *tmp_options;
719
720 opts->port = P9_PORT;
721 opts->rfd = ~0;
722 opts->wfd = ~0;
723
724 if (!params)
725 return 0;
726
727 tmp_options = kstrdup(params, GFP_KERNEL);
728 if (!tmp_options) {
729 P9_DPRINTK(P9_DEBUG_ERROR,
730 "failed to allocate copy of option string\n");
731 return -ENOMEM;
732 }
733 options = tmp_options;
734
735 while ((p = strsep(&options, ",")) != NULL) {
736 int token;
737 int r;
738 if (!*p)
739 continue;
740 token = match_token(p, tokens, args);
741 if (token != Opt_err) {
742 r = match_int(&args[0], &option);
743 if (r < 0) {
744 P9_DPRINTK(P9_DEBUG_ERROR,
745 "integer field, but no integer?\n");
746 continue;
747 }
748 }
749 switch (token) {
750 case Opt_port:
751 opts->port = option;
752 break;
753 case Opt_rfdno:
754 opts->rfd = option;
755 break;
756 case Opt_wfdno:
757 opts->wfd = option;
758 break;
759 default:
760 continue;
761 }
762 }
763
764 kfree(tmp_options);
765 return 0;
766}
767
768static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
769{
770 struct p9_trans_fd *ts = kmalloc(sizeof(struct p9_trans_fd),
771 GFP_KERNEL);
772 if (!ts)
773 return -ENOMEM;
774
775 ts->rd = fget(rfd);
776 ts->wr = fget(wfd);
777 if (!ts->rd || !ts->wr) {
778 if (ts->rd)
779 fput(ts->rd);
780 if (ts->wr)
781 fput(ts->wr);
782 kfree(ts);
783 return -EIO;
784 }
785
786 client->trans = ts;
787 client->status = Connected;
788
789 return 0;
790}
791
792static int p9_socket_open(struct p9_client *client, struct socket *csocket)
793{
794 struct p9_trans_fd *p;
795 int ret, fd;
796
797 p = kmalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
798 if (!p)
799 return -ENOMEM;
800
801 csocket->sk->sk_allocation = GFP_NOIO;
802 fd = sock_map_fd(csocket, 0);
803 if (fd < 0) {
804 P9_EPRINTK(KERN_ERR, "p9_socket_open: failed to map fd\n");
805 sock_release(csocket);
806 kfree(p);
807 return fd;
808 }
809
810 get_file(csocket->file);
811 get_file(csocket->file);
812 p->wr = p->rd = csocket->file;
813 client->trans = p;
814 client->status = Connected;
815
816 sys_close(fd); /* still racy */
817
818 p->rd->f_flags |= O_NONBLOCK;
819
820 p->conn = p9_conn_create(client);
821 if (IS_ERR(p->conn)) {
822 ret = PTR_ERR(p->conn);
823 p->conn = NULL;
824 kfree(p);
825 sockfd_put(csocket);
826 sockfd_put(csocket);
827 return ret;
828 }
829 return 0;
830}
831
832/**
833 * p9_mux_destroy - cancels all pending requests and frees mux resources
834 * @m: mux to destroy
835 *
836 */
837
838static void p9_conn_destroy(struct p9_conn *m)
839{
840 P9_DPRINTK(P9_DEBUG_TRANS, "mux %p prev %p next %p\n", m,
841 m->mux_list.prev, m->mux_list.next);
842
843 p9_mux_poll_stop(m);
844 cancel_work_sync(&m->rq);
845 cancel_work_sync(&m->wq);
846
847 p9_conn_cancel(m, -ECONNRESET);
848
849 m->client = NULL;
850 kfree(m);
851}
852
853/**
854 * p9_fd_close - shutdown file descriptor transport
855 * @client: client instance
856 *
857 */
858
859static void p9_fd_close(struct p9_client *client)
860{
861 struct p9_trans_fd *ts;
862
863 if (!client)
864 return;
865
866 ts = client->trans;
867 if (!ts)
868 return;
869
870 client->status = Disconnected;
871
872 p9_conn_destroy(ts->conn);
873
874 if (ts->rd)
875 fput(ts->rd);
876 if (ts->wr)
877 fput(ts->wr);
878
879 kfree(ts);
880}
881
882/*
883 * stolen from NFS - maybe should be made a generic function?
884 */
885static inline int valid_ipaddr4(const char *buf)
886{
887 int rc, count, in[4];
888
889 rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
890 if (rc != 4)
891 return -EINVAL;
892 for (count = 0; count < 4; count++) {
893 if (in[count] > 255)
894 return -EINVAL;
895 }
896 return 0;
897}
898
899static int
900p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
901{
902 int err;
903 struct socket *csocket;
904 struct sockaddr_in sin_server;
905 struct p9_fd_opts opts;
906
907 err = parse_opts(args, &opts);
908 if (err < 0)
909 return err;
910
911 if (valid_ipaddr4(addr) < 0)
912 return -EINVAL;
913
914 csocket = NULL;
915
916 sin_server.sin_family = AF_INET;
917 sin_server.sin_addr.s_addr = in_aton(addr);
918 sin_server.sin_port = htons(opts.port);
919 err = __sock_create(read_pnet(¤t->nsproxy->net_ns), PF_INET,
920 SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
921 if (err) {
922 P9_EPRINTK(KERN_ERR, "p9_trans_tcp: problem creating socket\n");
923 return err;
924 }
925
926 err = csocket->ops->connect(csocket,
927 (struct sockaddr *)&sin_server,
928 sizeof(struct sockaddr_in), 0);
929 if (err < 0) {
930 P9_EPRINTK(KERN_ERR,
931 "p9_trans_tcp: problem connecting socket to %s\n",
932 addr);
933 sock_release(csocket);
934 return err;
935 }
936
937 return p9_socket_open(client, csocket);
938}
939
940static int
941p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
942{
943 int err;
944 struct socket *csocket;
945 struct sockaddr_un sun_server;
946
947 csocket = NULL;
948
949 if (strlen(addr) >= UNIX_PATH_MAX) {
950 P9_EPRINTK(KERN_ERR, "p9_trans_unix: address too long: %s\n",
951 addr);
952 return -ENAMETOOLONG;
953 }
954
955 sun_server.sun_family = PF_UNIX;
956 strcpy(sun_server.sun_path, addr);
957 err = __sock_create(read_pnet(¤t->nsproxy->net_ns), PF_UNIX,
958 SOCK_STREAM, 0, &csocket, 1);
959 if (err < 0) {
960 P9_EPRINTK(KERN_ERR, "p9_trans_unix: problem creating socket\n");
961 return err;
962 }
963 err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
964 sizeof(struct sockaddr_un) - 1, 0);
965 if (err < 0) {
966 P9_EPRINTK(KERN_ERR,
967 "p9_trans_unix: problem connecting socket: %s: %d\n",
968 addr, err);
969 sock_release(csocket);
970 return err;
971 }
972
973 return p9_socket_open(client, csocket);
974}
975
976static int
977p9_fd_create(struct p9_client *client, const char *addr, char *args)
978{
979 int err;
980 struct p9_fd_opts opts;
981 struct p9_trans_fd *p;
982
983 parse_opts(args, &opts);
984
985 if (opts.rfd == ~0 || opts.wfd == ~0) {
986 printk(KERN_ERR "v9fs: Insufficient options for proto=fd\n");
987 return -ENOPROTOOPT;
988 }
989
990 err = p9_fd_open(client, opts.rfd, opts.wfd);
991 if (err < 0)
992 return err;
993
994 p = (struct p9_trans_fd *) client->trans;
995 p->conn = p9_conn_create(client);
996 if (IS_ERR(p->conn)) {
997 err = PTR_ERR(p->conn);
998 p->conn = NULL;
999 fput(p->rd);
1000 fput(p->wr);
1001 return err;
1002 }
1003
1004 return 0;
1005}
1006
1007static struct p9_trans_module p9_tcp_trans = {
1008 .name = "tcp",
1009 .maxsize = MAX_SOCK_BUF,
1010 .def = 1,
1011 .create = p9_fd_create_tcp,
1012 .close = p9_fd_close,
1013 .request = p9_fd_request,
1014 .cancel = p9_fd_cancel,
1015 .owner = THIS_MODULE,
1016};
1017
1018static struct p9_trans_module p9_unix_trans = {
1019 .name = "unix",
1020 .maxsize = MAX_SOCK_BUF,
1021 .def = 0,
1022 .create = p9_fd_create_unix,
1023 .close = p9_fd_close,
1024 .request = p9_fd_request,
1025 .cancel = p9_fd_cancel,
1026 .owner = THIS_MODULE,
1027};
1028
1029static struct p9_trans_module p9_fd_trans = {
1030 .name = "fd",
1031 .maxsize = MAX_SOCK_BUF,
1032 .def = 0,
1033 .create = p9_fd_create,
1034 .close = p9_fd_close,
1035 .request = p9_fd_request,
1036 .cancel = p9_fd_cancel,
1037 .owner = THIS_MODULE,
1038};
1039
1040/**
1041 * p9_poll_proc - poll worker thread
1042 * @a: thread state and arguments
1043 *
1044 * polls all v9fs transports for new events and queues the appropriate
1045 * work to the work queue
1046 *
1047 */
1048
1049static void p9_poll_workfn(struct work_struct *work)
1050{
1051 unsigned long flags;
1052
1053 P9_DPRINTK(P9_DEBUG_TRANS, "start %p\n", current);
1054
1055 spin_lock_irqsave(&p9_poll_lock, flags);
1056 while (!list_empty(&p9_poll_pending_list)) {
1057 struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1058 struct p9_conn,
1059 poll_pending_link);
1060 list_del_init(&conn->poll_pending_link);
1061 spin_unlock_irqrestore(&p9_poll_lock, flags);
1062
1063 p9_poll_mux(conn);
1064
1065 spin_lock_irqsave(&p9_poll_lock, flags);
1066 }
1067 spin_unlock_irqrestore(&p9_poll_lock, flags);
1068
1069 P9_DPRINTK(P9_DEBUG_TRANS, "finish\n");
1070}
1071
1072int p9_trans_fd_init(void)
1073{
1074 v9fs_register_trans(&p9_tcp_trans);
1075 v9fs_register_trans(&p9_unix_trans);
1076 v9fs_register_trans(&p9_fd_trans);
1077
1078 return 0;
1079}
1080
1081void p9_trans_fd_exit(void)
1082{
1083 flush_work_sync(&p9_poll_work);
1084 v9fs_unregister_trans(&p9_tcp_trans);
1085 v9fs_unregister_trans(&p9_unix_trans);
1086 v9fs_unregister_trans(&p9_fd_trans);
1087}
1/*
2 * linux/fs/9p/trans_fd.c
3 *
4 * Fd transport layer. Includes deprecated socket layer.
5 *
6 * Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
7 * Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
8 * Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
9 * Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License version 2
13 * as published by the Free Software Foundation.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to:
22 * Free Software Foundation
23 * 51 Franklin Street, Fifth Floor
24 * Boston, MA 02111-1301 USA
25 *
26 */
27
28#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29
30#include <linux/in.h>
31#include <linux/module.h>
32#include <linux/net.h>
33#include <linux/ipv6.h>
34#include <linux/kthread.h>
35#include <linux/errno.h>
36#include <linux/kernel.h>
37#include <linux/un.h>
38#include <linux/uaccess.h>
39#include <linux/inet.h>
40#include <linux/idr.h>
41#include <linux/file.h>
42#include <linux/parser.h>
43#include <linux/slab.h>
44#include <linux/seq_file.h>
45#include <net/9p/9p.h>
46#include <net/9p/client.h>
47#include <net/9p/transport.h>
48
49#include <linux/syscalls.h> /* killme */
50
51#define P9_PORT 564
52#define MAX_SOCK_BUF (64*1024)
53#define MAXPOLLWADDR 2
54
55static struct p9_trans_module p9_tcp_trans;
56static struct p9_trans_module p9_fd_trans;
57
58/**
59 * struct p9_fd_opts - per-transport options
60 * @rfd: file descriptor for reading (trans=fd)
61 * @wfd: file descriptor for writing (trans=fd)
62 * @port: port to connect to (trans=tcp)
63 *
64 */
65
66struct p9_fd_opts {
67 int rfd;
68 int wfd;
69 u16 port;
70 bool privport;
71};
72
73/*
74 * Option Parsing (code inspired by NFS code)
75 * - a little lazy - parse all fd-transport options
76 */
77
78enum {
79 /* Options that take integer arguments */
80 Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
81 /* Options that take no arguments */
82 Opt_privport,
83};
84
85static const match_table_t tokens = {
86 {Opt_port, "port=%u"},
87 {Opt_rfdno, "rfdno=%u"},
88 {Opt_wfdno, "wfdno=%u"},
89 {Opt_privport, "privport"},
90 {Opt_err, NULL},
91};
92
93enum {
94 Rworksched = 1, /* read work scheduled or running */
95 Rpending = 2, /* can read */
96 Wworksched = 4, /* write work scheduled or running */
97 Wpending = 8, /* can write */
98};
99
100struct p9_poll_wait {
101 struct p9_conn *conn;
102 wait_queue_entry_t wait;
103 wait_queue_head_t *wait_addr;
104};
105
106/**
107 * struct p9_conn - fd mux connection state information
108 * @mux_list: list link for mux to manage multiple connections (?)
109 * @client: reference to client instance for this connection
110 * @err: error state
111 * @req_list: accounting for requests which have been sent
112 * @unsent_req_list: accounting for requests that haven't been sent
113 * @req: current request being processed (if any)
114 * @tmp_buf: temporary buffer to read in header
115 * @rc: temporary fcall for reading current frame
116 * @wpos: write position for current frame
117 * @wsize: amount of data to write for current frame
118 * @wbuf: current write buffer
119 * @poll_pending_link: pending links to be polled per conn
120 * @poll_wait: array of wait_q's for various worker threads
121 * @pt: poll state
122 * @rq: current read work
123 * @wq: current write work
124 * @wsched: ????
125 *
126 */
127
128struct p9_conn {
129 struct list_head mux_list;
130 struct p9_client *client;
131 int err;
132 struct list_head req_list;
133 struct list_head unsent_req_list;
134 struct p9_req_t *req;
135 char tmp_buf[7];
136 struct p9_fcall rc;
137 int wpos;
138 int wsize;
139 char *wbuf;
140 struct list_head poll_pending_link;
141 struct p9_poll_wait poll_wait[MAXPOLLWADDR];
142 poll_table pt;
143 struct work_struct rq;
144 struct work_struct wq;
145 unsigned long wsched;
146};
147
148/**
149 * struct p9_trans_fd - transport state
150 * @rd: reference to file to read from
151 * @wr: reference of file to write to
152 * @conn: connection state reference
153 *
154 */
155
156struct p9_trans_fd {
157 struct file *rd;
158 struct file *wr;
159 struct p9_conn conn;
160};
161
162static void p9_poll_workfn(struct work_struct *work);
163
164static DEFINE_SPINLOCK(p9_poll_lock);
165static LIST_HEAD(p9_poll_pending_list);
166static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
167
168static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
169static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
170
171static void p9_mux_poll_stop(struct p9_conn *m)
172{
173 unsigned long flags;
174 int i;
175
176 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
177 struct p9_poll_wait *pwait = &m->poll_wait[i];
178
179 if (pwait->wait_addr) {
180 remove_wait_queue(pwait->wait_addr, &pwait->wait);
181 pwait->wait_addr = NULL;
182 }
183 }
184
185 spin_lock_irqsave(&p9_poll_lock, flags);
186 list_del_init(&m->poll_pending_link);
187 spin_unlock_irqrestore(&p9_poll_lock, flags);
188}
189
190/**
191 * p9_conn_cancel - cancel all pending requests with error
192 * @m: mux data
193 * @err: error code
194 *
195 */
196
197static void p9_conn_cancel(struct p9_conn *m, int err)
198{
199 struct p9_req_t *req, *rtmp;
200 unsigned long flags;
201 LIST_HEAD(cancel_list);
202
203 p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
204
205 spin_lock_irqsave(&m->client->lock, flags);
206
207 if (m->err) {
208 spin_unlock_irqrestore(&m->client->lock, flags);
209 return;
210 }
211
212 m->err = err;
213
214 list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
215 list_move(&req->req_list, &cancel_list);
216 }
217 list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
218 list_move(&req->req_list, &cancel_list);
219 }
220 spin_unlock_irqrestore(&m->client->lock, flags);
221
222 list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
223 p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
224 list_del(&req->req_list);
225 if (!req->t_err)
226 req->t_err = err;
227 p9_client_cb(m->client, req, REQ_STATUS_ERROR);
228 }
229}
230
231static __poll_t
232p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
233{
234 __poll_t ret, n;
235 struct p9_trans_fd *ts = NULL;
236
237 if (client && client->status == Connected)
238 ts = client->trans;
239
240 if (!ts) {
241 if (err)
242 *err = -EREMOTEIO;
243 return EPOLLERR;
244 }
245
246 if (!ts->rd->f_op->poll)
247 ret = DEFAULT_POLLMASK;
248 else
249 ret = ts->rd->f_op->poll(ts->rd, pt);
250
251 if (ts->rd != ts->wr) {
252 if (!ts->wr->f_op->poll)
253 n = DEFAULT_POLLMASK;
254 else
255 n = ts->wr->f_op->poll(ts->wr, pt);
256 ret = (ret & ~EPOLLOUT) | (n & ~EPOLLIN);
257 }
258
259 return ret;
260}
261
262/**
263 * p9_fd_read- read from a fd
264 * @client: client instance
265 * @v: buffer to receive data into
266 * @len: size of receive buffer
267 *
268 */
269
270static int p9_fd_read(struct p9_client *client, void *v, int len)
271{
272 int ret;
273 struct p9_trans_fd *ts = NULL;
274 loff_t pos;
275
276 if (client && client->status != Disconnected)
277 ts = client->trans;
278
279 if (!ts)
280 return -EREMOTEIO;
281
282 if (!(ts->rd->f_flags & O_NONBLOCK))
283 p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
284
285 pos = ts->rd->f_pos;
286 ret = kernel_read(ts->rd, v, len, &pos);
287 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
288 client->status = Disconnected;
289 return ret;
290}
291
292/**
293 * p9_read_work - called when there is some data to be read from a transport
294 * @work: container of work to be done
295 *
296 */
297
298static void p9_read_work(struct work_struct *work)
299{
300 __poll_t n;
301 int err;
302 struct p9_conn *m;
303 int status = REQ_STATUS_ERROR;
304
305 m = container_of(work, struct p9_conn, rq);
306
307 if (m->err < 0)
308 return;
309
310 p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
311
312 if (!m->rc.sdata) {
313 m->rc.sdata = m->tmp_buf;
314 m->rc.offset = 0;
315 m->rc.capacity = 7; /* start by reading header */
316 }
317
318 clear_bit(Rpending, &m->wsched);
319 p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
320 m, m->rc.offset, m->rc.capacity,
321 m->rc.capacity - m->rc.offset);
322 err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
323 m->rc.capacity - m->rc.offset);
324 p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
325 if (err == -EAGAIN)
326 goto end_clear;
327
328 if (err <= 0)
329 goto error;
330
331 m->rc.offset += err;
332
333 /* header read in */
334 if ((!m->req) && (m->rc.offset == m->rc.capacity)) {
335 p9_debug(P9_DEBUG_TRANS, "got new header\n");
336
337 err = p9_parse_header(&m->rc, NULL, NULL, NULL, 0);
338 if (err) {
339 p9_debug(P9_DEBUG_ERROR,
340 "error parsing header: %d\n", err);
341 goto error;
342 }
343
344 if (m->rc.size >= m->client->msize) {
345 p9_debug(P9_DEBUG_ERROR,
346 "requested packet size too big: %d\n",
347 m->rc.size);
348 err = -EIO;
349 goto error;
350 }
351
352 p9_debug(P9_DEBUG_TRANS,
353 "mux %p pkt: size: %d bytes tag: %d\n",
354 m, m->rc.size, m->rc.tag);
355
356 m->req = p9_tag_lookup(m->client, m->rc.tag);
357 if (!m->req || (m->req->status != REQ_STATUS_SENT)) {
358 p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
359 m->rc.tag);
360 err = -EIO;
361 goto error;
362 }
363
364 if (m->req->rc == NULL) {
365 p9_debug(P9_DEBUG_ERROR,
366 "No recv fcall for tag %d (req %p), disconnecting!\n",
367 m->rc.tag, m->req);
368 m->req = NULL;
369 err = -EIO;
370 goto error;
371 }
372 m->rc.sdata = (char *)m->req->rc + sizeof(struct p9_fcall);
373 memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
374 m->rc.capacity = m->rc.size;
375 }
376
377 /* packet is read in
378 * not an else because some packets (like clunk) have no payload
379 */
380 if ((m->req) && (m->rc.offset == m->rc.capacity)) {
381 p9_debug(P9_DEBUG_TRANS, "got new packet\n");
382 spin_lock(&m->client->lock);
383 if (m->req->status != REQ_STATUS_ERROR)
384 status = REQ_STATUS_RCVD;
385 list_del(&m->req->req_list);
386 spin_unlock(&m->client->lock);
387 p9_client_cb(m->client, m->req, status);
388 m->rc.sdata = NULL;
389 m->rc.offset = 0;
390 m->rc.capacity = 0;
391 m->req = NULL;
392 }
393
394end_clear:
395 clear_bit(Rworksched, &m->wsched);
396
397 if (!list_empty(&m->req_list)) {
398 if (test_and_clear_bit(Rpending, &m->wsched))
399 n = EPOLLIN;
400 else
401 n = p9_fd_poll(m->client, NULL, NULL);
402
403 if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
404 p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
405 schedule_work(&m->rq);
406 }
407 }
408
409 return;
410error:
411 p9_conn_cancel(m, err);
412 clear_bit(Rworksched, &m->wsched);
413}
414
415/**
416 * p9_fd_write - write to a socket
417 * @client: client instance
418 * @v: buffer to send data from
419 * @len: size of send buffer
420 *
421 */
422
423static int p9_fd_write(struct p9_client *client, void *v, int len)
424{
425 ssize_t ret;
426 struct p9_trans_fd *ts = NULL;
427
428 if (client && client->status != Disconnected)
429 ts = client->trans;
430
431 if (!ts)
432 return -EREMOTEIO;
433
434 if (!(ts->wr->f_flags & O_NONBLOCK))
435 p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
436
437 ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
438 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
439 client->status = Disconnected;
440 return ret;
441}
442
443/**
444 * p9_write_work - called when a transport can send some data
445 * @work: container for work to be done
446 *
447 */
448
449static void p9_write_work(struct work_struct *work)
450{
451 __poll_t n;
452 int err;
453 struct p9_conn *m;
454 struct p9_req_t *req;
455
456 m = container_of(work, struct p9_conn, wq);
457
458 if (m->err < 0) {
459 clear_bit(Wworksched, &m->wsched);
460 return;
461 }
462
463 if (!m->wsize) {
464 spin_lock(&m->client->lock);
465 if (list_empty(&m->unsent_req_list)) {
466 clear_bit(Wworksched, &m->wsched);
467 spin_unlock(&m->client->lock);
468 return;
469 }
470
471 req = list_entry(m->unsent_req_list.next, struct p9_req_t,
472 req_list);
473 req->status = REQ_STATUS_SENT;
474 p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
475 list_move_tail(&req->req_list, &m->req_list);
476
477 m->wbuf = req->tc->sdata;
478 m->wsize = req->tc->size;
479 m->wpos = 0;
480 spin_unlock(&m->client->lock);
481 }
482
483 p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
484 m, m->wpos, m->wsize);
485 clear_bit(Wpending, &m->wsched);
486 err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
487 p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
488 if (err == -EAGAIN)
489 goto end_clear;
490
491
492 if (err < 0)
493 goto error;
494 else if (err == 0) {
495 err = -EREMOTEIO;
496 goto error;
497 }
498
499 m->wpos += err;
500 if (m->wpos == m->wsize)
501 m->wpos = m->wsize = 0;
502
503end_clear:
504 clear_bit(Wworksched, &m->wsched);
505
506 if (m->wsize || !list_empty(&m->unsent_req_list)) {
507 if (test_and_clear_bit(Wpending, &m->wsched))
508 n = EPOLLOUT;
509 else
510 n = p9_fd_poll(m->client, NULL, NULL);
511
512 if ((n & EPOLLOUT) &&
513 !test_and_set_bit(Wworksched, &m->wsched)) {
514 p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
515 schedule_work(&m->wq);
516 }
517 }
518
519 return;
520
521error:
522 p9_conn_cancel(m, err);
523 clear_bit(Wworksched, &m->wsched);
524}
525
526static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
527{
528 struct p9_poll_wait *pwait =
529 container_of(wait, struct p9_poll_wait, wait);
530 struct p9_conn *m = pwait->conn;
531 unsigned long flags;
532
533 spin_lock_irqsave(&p9_poll_lock, flags);
534 if (list_empty(&m->poll_pending_link))
535 list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
536 spin_unlock_irqrestore(&p9_poll_lock, flags);
537
538 schedule_work(&p9_poll_work);
539 return 1;
540}
541
542/**
543 * p9_pollwait - add poll task to the wait queue
544 * @filp: file pointer being polled
545 * @wait_address: wait_q to block on
546 * @p: poll state
547 *
548 * called by files poll operation to add v9fs-poll task to files wait queue
549 */
550
551static void
552p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
553{
554 struct p9_conn *m = container_of(p, struct p9_conn, pt);
555 struct p9_poll_wait *pwait = NULL;
556 int i;
557
558 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
559 if (m->poll_wait[i].wait_addr == NULL) {
560 pwait = &m->poll_wait[i];
561 break;
562 }
563 }
564
565 if (!pwait) {
566 p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
567 return;
568 }
569
570 pwait->conn = m;
571 pwait->wait_addr = wait_address;
572 init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
573 add_wait_queue(wait_address, &pwait->wait);
574}
575
576/**
577 * p9_conn_create - initialize the per-session mux data
578 * @client: client instance
579 *
580 * Note: Creates the polling task if this is the first session.
581 */
582
583static void p9_conn_create(struct p9_client *client)
584{
585 __poll_t n;
586 struct p9_trans_fd *ts = client->trans;
587 struct p9_conn *m = &ts->conn;
588
589 p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
590
591 INIT_LIST_HEAD(&m->mux_list);
592 m->client = client;
593
594 INIT_LIST_HEAD(&m->req_list);
595 INIT_LIST_HEAD(&m->unsent_req_list);
596 INIT_WORK(&m->rq, p9_read_work);
597 INIT_WORK(&m->wq, p9_write_work);
598 INIT_LIST_HEAD(&m->poll_pending_link);
599 init_poll_funcptr(&m->pt, p9_pollwait);
600
601 n = p9_fd_poll(client, &m->pt, NULL);
602 if (n & EPOLLIN) {
603 p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
604 set_bit(Rpending, &m->wsched);
605 }
606
607 if (n & EPOLLOUT) {
608 p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
609 set_bit(Wpending, &m->wsched);
610 }
611}
612
613/**
614 * p9_poll_mux - polls a mux and schedules read or write works if necessary
615 * @m: connection to poll
616 *
617 */
618
619static void p9_poll_mux(struct p9_conn *m)
620{
621 __poll_t n;
622 int err = -ECONNRESET;
623
624 if (m->err < 0)
625 return;
626
627 n = p9_fd_poll(m->client, NULL, &err);
628 if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
629 p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
630 p9_conn_cancel(m, err);
631 }
632
633 if (n & EPOLLIN) {
634 set_bit(Rpending, &m->wsched);
635 p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
636 if (!test_and_set_bit(Rworksched, &m->wsched)) {
637 p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
638 schedule_work(&m->rq);
639 }
640 }
641
642 if (n & EPOLLOUT) {
643 set_bit(Wpending, &m->wsched);
644 p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
645 if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
646 !test_and_set_bit(Wworksched, &m->wsched)) {
647 p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
648 schedule_work(&m->wq);
649 }
650 }
651}
652
653/**
654 * p9_fd_request - send 9P request
655 * The function can sleep until the request is scheduled for sending.
656 * The function can be interrupted. Return from the function is not
657 * a guarantee that the request is sent successfully.
658 *
659 * @client: client instance
660 * @req: request to be sent
661 *
662 */
663
664static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
665{
666 __poll_t n;
667 struct p9_trans_fd *ts = client->trans;
668 struct p9_conn *m = &ts->conn;
669
670 p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
671 m, current, req->tc, req->tc->id);
672 if (m->err < 0)
673 return m->err;
674
675 spin_lock(&client->lock);
676 req->status = REQ_STATUS_UNSENT;
677 list_add_tail(&req->req_list, &m->unsent_req_list);
678 spin_unlock(&client->lock);
679
680 if (test_and_clear_bit(Wpending, &m->wsched))
681 n = EPOLLOUT;
682 else
683 n = p9_fd_poll(m->client, NULL, NULL);
684
685 if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
686 schedule_work(&m->wq);
687
688 return 0;
689}
690
691static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
692{
693 int ret = 1;
694
695 p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
696
697 spin_lock(&client->lock);
698
699 if (req->status == REQ_STATUS_UNSENT) {
700 list_del(&req->req_list);
701 req->status = REQ_STATUS_FLSHD;
702 ret = 0;
703 }
704 spin_unlock(&client->lock);
705
706 return ret;
707}
708
709static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
710{
711 p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
712
713 /* we haven't received a response for oldreq,
714 * remove it from the list.
715 */
716 spin_lock(&client->lock);
717 list_del(&req->req_list);
718 spin_unlock(&client->lock);
719
720 return 0;
721}
722
723static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
724{
725 if (clnt->trans_mod == &p9_tcp_trans) {
726 if (clnt->trans_opts.tcp.port != P9_PORT)
727 seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
728 } else if (clnt->trans_mod == &p9_fd_trans) {
729 if (clnt->trans_opts.fd.rfd != ~0)
730 seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
731 if (clnt->trans_opts.fd.wfd != ~0)
732 seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
733 }
734 return 0;
735}
736
737/**
738 * parse_opts - parse mount options into p9_fd_opts structure
739 * @params: options string passed from mount
740 * @opts: fd transport-specific structure to parse options into
741 *
742 * Returns 0 upon success, -ERRNO upon failure
743 */
744
745static int parse_opts(char *params, struct p9_fd_opts *opts)
746{
747 char *p;
748 substring_t args[MAX_OPT_ARGS];
749 int option;
750 char *options, *tmp_options;
751
752 opts->port = P9_PORT;
753 opts->rfd = ~0;
754 opts->wfd = ~0;
755 opts->privport = false;
756
757 if (!params)
758 return 0;
759
760 tmp_options = kstrdup(params, GFP_KERNEL);
761 if (!tmp_options) {
762 p9_debug(P9_DEBUG_ERROR,
763 "failed to allocate copy of option string\n");
764 return -ENOMEM;
765 }
766 options = tmp_options;
767
768 while ((p = strsep(&options, ",")) != NULL) {
769 int token;
770 int r;
771 if (!*p)
772 continue;
773 token = match_token(p, tokens, args);
774 if ((token != Opt_err) && (token != Opt_privport)) {
775 r = match_int(&args[0], &option);
776 if (r < 0) {
777 p9_debug(P9_DEBUG_ERROR,
778 "integer field, but no integer?\n");
779 continue;
780 }
781 }
782 switch (token) {
783 case Opt_port:
784 opts->port = option;
785 break;
786 case Opt_rfdno:
787 opts->rfd = option;
788 break;
789 case Opt_wfdno:
790 opts->wfd = option;
791 break;
792 case Opt_privport:
793 opts->privport = true;
794 break;
795 default:
796 continue;
797 }
798 }
799
800 kfree(tmp_options);
801 return 0;
802}
803
804static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
805{
806 struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
807 GFP_KERNEL);
808 if (!ts)
809 return -ENOMEM;
810
811 ts->rd = fget(rfd);
812 ts->wr = fget(wfd);
813 if (!ts->rd || !ts->wr) {
814 if (ts->rd)
815 fput(ts->rd);
816 if (ts->wr)
817 fput(ts->wr);
818 kfree(ts);
819 return -EIO;
820 }
821
822 client->trans = ts;
823 client->status = Connected;
824
825 return 0;
826}
827
828static int p9_socket_open(struct p9_client *client, struct socket *csocket)
829{
830 struct p9_trans_fd *p;
831 struct file *file;
832
833 p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
834 if (!p)
835 return -ENOMEM;
836
837 csocket->sk->sk_allocation = GFP_NOIO;
838 file = sock_alloc_file(csocket, 0, NULL);
839 if (IS_ERR(file)) {
840 pr_err("%s (%d): failed to map fd\n",
841 __func__, task_pid_nr(current));
842 kfree(p);
843 return PTR_ERR(file);
844 }
845
846 get_file(file);
847 p->wr = p->rd = file;
848 client->trans = p;
849 client->status = Connected;
850
851 p->rd->f_flags |= O_NONBLOCK;
852
853 p9_conn_create(client);
854 return 0;
855}
856
857/**
858 * p9_mux_destroy - cancels all pending requests of mux
859 * @m: mux to destroy
860 *
861 */
862
863static void p9_conn_destroy(struct p9_conn *m)
864{
865 p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
866 m, m->mux_list.prev, m->mux_list.next);
867
868 p9_mux_poll_stop(m);
869 cancel_work_sync(&m->rq);
870 cancel_work_sync(&m->wq);
871
872 p9_conn_cancel(m, -ECONNRESET);
873
874 m->client = NULL;
875}
876
877/**
878 * p9_fd_close - shutdown file descriptor transport
879 * @client: client instance
880 *
881 */
882
883static void p9_fd_close(struct p9_client *client)
884{
885 struct p9_trans_fd *ts;
886
887 if (!client)
888 return;
889
890 ts = client->trans;
891 if (!ts)
892 return;
893
894 client->status = Disconnected;
895
896 p9_conn_destroy(&ts->conn);
897
898 if (ts->rd)
899 fput(ts->rd);
900 if (ts->wr)
901 fput(ts->wr);
902
903 kfree(ts);
904}
905
906/*
907 * stolen from NFS - maybe should be made a generic function?
908 */
909static inline int valid_ipaddr4(const char *buf)
910{
911 int rc, count, in[4];
912
913 rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
914 if (rc != 4)
915 return -EINVAL;
916 for (count = 0; count < 4; count++) {
917 if (in[count] > 255)
918 return -EINVAL;
919 }
920 return 0;
921}
922
923static int p9_bind_privport(struct socket *sock)
924{
925 struct sockaddr_in cl;
926 int port, err = -EINVAL;
927
928 memset(&cl, 0, sizeof(cl));
929 cl.sin_family = AF_INET;
930 cl.sin_addr.s_addr = INADDR_ANY;
931 for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
932 cl.sin_port = htons((ushort)port);
933 err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
934 if (err != -EADDRINUSE)
935 break;
936 }
937 return err;
938}
939
940
941static int
942p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
943{
944 int err;
945 struct socket *csocket;
946 struct sockaddr_in sin_server;
947 struct p9_fd_opts opts;
948
949 err = parse_opts(args, &opts);
950 if (err < 0)
951 return err;
952
953 if (valid_ipaddr4(addr) < 0)
954 return -EINVAL;
955
956 csocket = NULL;
957
958 client->trans_opts.tcp.port = opts.port;
959 client->trans_opts.tcp.privport = opts.privport;
960 sin_server.sin_family = AF_INET;
961 sin_server.sin_addr.s_addr = in_aton(addr);
962 sin_server.sin_port = htons(opts.port);
963 err = __sock_create(current->nsproxy->net_ns, PF_INET,
964 SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
965 if (err) {
966 pr_err("%s (%d): problem creating socket\n",
967 __func__, task_pid_nr(current));
968 return err;
969 }
970
971 if (opts.privport) {
972 err = p9_bind_privport(csocket);
973 if (err < 0) {
974 pr_err("%s (%d): problem binding to privport\n",
975 __func__, task_pid_nr(current));
976 sock_release(csocket);
977 return err;
978 }
979 }
980
981 err = csocket->ops->connect(csocket,
982 (struct sockaddr *)&sin_server,
983 sizeof(struct sockaddr_in), 0);
984 if (err < 0) {
985 pr_err("%s (%d): problem connecting socket to %s\n",
986 __func__, task_pid_nr(current), addr);
987 sock_release(csocket);
988 return err;
989 }
990
991 return p9_socket_open(client, csocket);
992}
993
994static int
995p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
996{
997 int err;
998 struct socket *csocket;
999 struct sockaddr_un sun_server;
1000
1001 csocket = NULL;
1002
1003 if (strlen(addr) >= UNIX_PATH_MAX) {
1004 pr_err("%s (%d): address too long: %s\n",
1005 __func__, task_pid_nr(current), addr);
1006 return -ENAMETOOLONG;
1007 }
1008
1009 sun_server.sun_family = PF_UNIX;
1010 strcpy(sun_server.sun_path, addr);
1011 err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1012 SOCK_STREAM, 0, &csocket, 1);
1013 if (err < 0) {
1014 pr_err("%s (%d): problem creating socket\n",
1015 __func__, task_pid_nr(current));
1016
1017 return err;
1018 }
1019 err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1020 sizeof(struct sockaddr_un) - 1, 0);
1021 if (err < 0) {
1022 pr_err("%s (%d): problem connecting socket: %s: %d\n",
1023 __func__, task_pid_nr(current), addr, err);
1024 sock_release(csocket);
1025 return err;
1026 }
1027
1028 return p9_socket_open(client, csocket);
1029}
1030
1031static int
1032p9_fd_create(struct p9_client *client, const char *addr, char *args)
1033{
1034 int err;
1035 struct p9_fd_opts opts;
1036
1037 parse_opts(args, &opts);
1038 client->trans_opts.fd.rfd = opts.rfd;
1039 client->trans_opts.fd.wfd = opts.wfd;
1040
1041 if (opts.rfd == ~0 || opts.wfd == ~0) {
1042 pr_err("Insufficient options for proto=fd\n");
1043 return -ENOPROTOOPT;
1044 }
1045
1046 err = p9_fd_open(client, opts.rfd, opts.wfd);
1047 if (err < 0)
1048 return err;
1049
1050 p9_conn_create(client);
1051
1052 return 0;
1053}
1054
1055static struct p9_trans_module p9_tcp_trans = {
1056 .name = "tcp",
1057 .maxsize = MAX_SOCK_BUF,
1058 .def = 0,
1059 .create = p9_fd_create_tcp,
1060 .close = p9_fd_close,
1061 .request = p9_fd_request,
1062 .cancel = p9_fd_cancel,
1063 .cancelled = p9_fd_cancelled,
1064 .show_options = p9_fd_show_options,
1065 .owner = THIS_MODULE,
1066};
1067
1068static struct p9_trans_module p9_unix_trans = {
1069 .name = "unix",
1070 .maxsize = MAX_SOCK_BUF,
1071 .def = 0,
1072 .create = p9_fd_create_unix,
1073 .close = p9_fd_close,
1074 .request = p9_fd_request,
1075 .cancel = p9_fd_cancel,
1076 .cancelled = p9_fd_cancelled,
1077 .show_options = p9_fd_show_options,
1078 .owner = THIS_MODULE,
1079};
1080
1081static struct p9_trans_module p9_fd_trans = {
1082 .name = "fd",
1083 .maxsize = MAX_SOCK_BUF,
1084 .def = 0,
1085 .create = p9_fd_create,
1086 .close = p9_fd_close,
1087 .request = p9_fd_request,
1088 .cancel = p9_fd_cancel,
1089 .cancelled = p9_fd_cancelled,
1090 .show_options = p9_fd_show_options,
1091 .owner = THIS_MODULE,
1092};
1093
1094/**
1095 * p9_poll_workfn - poll worker thread
1096 * @work: work queue
1097 *
1098 * polls all v9fs transports for new events and queues the appropriate
1099 * work to the work queue
1100 *
1101 */
1102
1103static void p9_poll_workfn(struct work_struct *work)
1104{
1105 unsigned long flags;
1106
1107 p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1108
1109 spin_lock_irqsave(&p9_poll_lock, flags);
1110 while (!list_empty(&p9_poll_pending_list)) {
1111 struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1112 struct p9_conn,
1113 poll_pending_link);
1114 list_del_init(&conn->poll_pending_link);
1115 spin_unlock_irqrestore(&p9_poll_lock, flags);
1116
1117 p9_poll_mux(conn);
1118
1119 spin_lock_irqsave(&p9_poll_lock, flags);
1120 }
1121 spin_unlock_irqrestore(&p9_poll_lock, flags);
1122
1123 p9_debug(P9_DEBUG_TRANS, "finish\n");
1124}
1125
1126int p9_trans_fd_init(void)
1127{
1128 v9fs_register_trans(&p9_tcp_trans);
1129 v9fs_register_trans(&p9_unix_trans);
1130 v9fs_register_trans(&p9_fd_trans);
1131
1132 return 0;
1133}
1134
1135void p9_trans_fd_exit(void)
1136{
1137 flush_work(&p9_poll_work);
1138 v9fs_unregister_trans(&p9_tcp_trans);
1139 v9fs_unregister_trans(&p9_unix_trans);
1140 v9fs_unregister_trans(&p9_fd_trans);
1141}