Commit | Line | Data |
---|---|---|
1f327613 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b530cc79 | 2 | /* |
fea511a6 | 3 | * The Virtio 9p transport driver |
b530cc79 | 4 | * |
e2735b77 EVH |
5 | * This is a block based transport driver based on the lguest block driver |
6 | * code. | |
b530cc79 | 7 | * |
fea511a6 | 8 | * Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation |
b530cc79 EVH |
9 | * |
10 | * Based on virtio console driver | |
11 | * Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation | |
b530cc79 EVH |
12 | */ |
13 | ||
5d385153 JP |
14 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
15 | ||
b530cc79 EVH |
16 | #include <linux/in.h> |
17 | #include <linux/module.h> | |
18 | #include <linux/net.h> | |
19 | #include <linux/ipv6.h> | |
20 | #include <linux/errno.h> | |
21 | #include <linux/kernel.h> | |
22 | #include <linux/un.h> | |
23 | #include <linux/uaccess.h> | |
24 | #include <linux/inet.h> | |
b530cc79 | 25 | #include <linux/file.h> |
b9cdc88d | 26 | #include <linux/highmem.h> |
5a0e3ad6 | 27 | #include <linux/slab.h> |
b530cc79 EVH |
28 | #include <net/9p/9p.h> |
29 | #include <linux/parser.h> | |
8b81ef58 | 30 | #include <net/9p/client.h> |
b530cc79 EVH |
31 | #include <net/9p/transport.h> |
32 | #include <linux/scatterlist.h> | |
68da9ba4 | 33 | #include <linux/swap.h> |
b530cc79 EVH |
34 | #include <linux/virtio.h> |
35 | #include <linux/virtio_9p.h> | |
4038866d | 36 | #include "trans_common.h" |
b530cc79 | 37 | |
e2735b77 EVH |
38 | #define VIRTQUEUE_NUM 128 |
39 | ||
b530cc79 | 40 | /* a single mutex to manage channel initialization and attachment */ |
c1549497 | 41 | static DEFINE_MUTEX(virtio_9p_lock); |
68da9ba4 VJJ |
42 | static DECLARE_WAIT_QUEUE_HEAD(vp_wq); |
43 | static atomic_t vp_pinned = ATOMIC_INIT(0); | |
b530cc79 | 44 | |
ee443996 EVH |
45 | /** |
46 | * struct virtio_chan - per-instance transport information | |
ee443996 EVH |
47 | * @inuse: whether the channel is in use |
48 | * @lock: protects multiple elements within this structure | |
0e15597e | 49 | * @client: client instance |
ee443996 EVH |
50 | * @vdev: virtio dev associated with this channel |
51 | * @vq: virtio queue associated with this channel | |
760b3d61 AL |
52 | * @ring_bufs_avail: flag to indicate there is some available in the ring buf |
53 | * @vc_wq: wait queue for waiting for thing to be added to ring buf | |
54 | * @p9_max_pages: maximum number of pinned pages | |
ee443996 | 55 | * @sg: scatter gather list which is used to pack a request (protected?) |
760b3d61 | 56 | * @chan_list: linked list of channels |
ee443996 EVH |
57 | * |
58 | * We keep all per-channel information in a structure. | |
b530cc79 EVH |
59 | * This structure is allocated within the devices dev->mem space. |
60 | * A pointer to the structure will get put in the transport private. | |
ee443996 | 61 | * |
b530cc79 | 62 | */ |
ee443996 | 63 | |
37c1209d | 64 | struct virtio_chan { |
ee443996 | 65 | bool inuse; |
b530cc79 | 66 | |
e2735b77 EVH |
67 | spinlock_t lock; |
68 | ||
fea511a6 | 69 | struct p9_client *client; |
b530cc79 | 70 | struct virtio_device *vdev; |
e2735b77 | 71 | struct virtqueue *vq; |
52f44e0d VJJ |
72 | int ring_bufs_avail; |
73 | wait_queue_head_t *vc_wq; | |
68da9ba4 VJJ |
74 | /* This is global limit. Since we don't have a global structure, |
75 | * will be placing it in each channel. | |
76 | */ | |
7293bfba | 77 | unsigned long p9_max_pages; |
e2735b77 EVH |
78 | /* Scatterlist: can be too big for stack. */ |
79 | struct scatterlist sg[VIRTQUEUE_NUM]; | |
760b3d61 AL |
80 | /** |
81 | * @tag: name to identify a mount null terminated | |
97ee9b02 AK |
82 | */ |
83 | char *tag; | |
84 | ||
37c1209d AK |
85 | struct list_head chan_list; |
86 | }; | |
87 | ||
88 | static struct list_head virtio_chan_list; | |
b530cc79 EVH |
89 | |
90 | /* How many bytes left in this page. */ | |
91 | static unsigned int rest_of_page(void *data) | |
92 | { | |
222e4ade | 93 | return PAGE_SIZE - offset_in_page(data); |
b530cc79 EVH |
94 | } |
95 | ||
ee443996 EVH |
96 | /** |
97 | * p9_virtio_close - reclaim resources of a channel | |
0e15597e | 98 | * @client: client instance |
ee443996 EVH |
99 | * |
100 | * This reclaims a channel by freeing its resources and | |
8ab1784d | 101 | * resetting its inuse flag. |
ee443996 EVH |
102 | * |
103 | */ | |
104 | ||
8b81ef58 | 105 | static void p9_virtio_close(struct p9_client *client) |
e2735b77 | 106 | { |
8b81ef58 | 107 | struct virtio_chan *chan = client->trans; |
b530cc79 | 108 | |
c1549497 | 109 | mutex_lock(&virtio_9p_lock); |
fb786100 AK |
110 | if (chan) |
111 | chan->inuse = false; | |
c1549497 | 112 | mutex_unlock(&virtio_9p_lock); |
b530cc79 EVH |
113 | } |
114 | ||
ee443996 EVH |
115 | /** |
116 | * req_done - callback which signals activity from the server | |
117 | * @vq: virtio queue activity was received on | |
118 | * | |
119 | * This notifies us that the server has triggered some activity | |
120 | * on the virtio channel - most likely a response to request we | |
121 | * sent. Figure out which requests now have responses and wake up | |
122 | * those threads. | |
123 | * | |
124 | * Bugs: could do with some additional sanity checking, but appears to work. | |
125 | * | |
126 | */ | |
127 | ||
e2735b77 | 128 | static void req_done(struct virtqueue *vq) |
b530cc79 | 129 | { |
e2735b77 | 130 | struct virtio_chan *chan = vq->vdev->priv; |
e2735b77 | 131 | unsigned int len; |
e2735b77 | 132 | struct p9_req_t *req; |
31934da8 | 133 | bool need_wakeup = false; |
419b3956 | 134 | unsigned long flags; |
e2735b77 | 135 | |
5d385153 | 136 | p9_debug(P9_DEBUG_TRANS, ": request done\n"); |
91b8534f | 137 | |
31934da8 | 138 | spin_lock_irqsave(&chan->lock, flags); |
139 | while ((req = virtqueue_get_buf(chan->vq, &len)) != NULL) { | |
140 | if (!chan->ring_bufs_avail) { | |
141 | chan->ring_bufs_avail = 1; | |
142 | need_wakeup = true; | |
a01a9840 | 143 | } |
31934da8 | 144 | |
f984579a | 145 | if (len) { |
523adb6c | 146 | req->rc.size = len; |
26d99834 | 147 | p9_client_cb(chan->client, req, REQ_STATUS_RCVD); |
f984579a | 148 | } |
a01a9840 | 149 | } |
31934da8 | 150 | spin_unlock_irqrestore(&chan->lock, flags); |
151 | /* Wakeup if anyone waiting for VirtIO ring space. */ | |
152 | if (need_wakeup) | |
153 | wake_up(chan->vc_wq); | |
e2735b77 | 154 | } |
b530cc79 | 155 | |
ee443996 EVH |
156 | /** |
157 | * pack_sg_list - pack a scatter gather list from a linear buffer | |
158 | * @sg: scatter/gather list to pack into | |
159 | * @start: which segment of the sg_list to start at | |
160 | * @limit: maximum segment to pack data to | |
161 | * @data: data to pack into scatter/gather list | |
162 | * @count: amount of data to pack into the scatter/gather list | |
163 | * | |
164 | * sg_lists have multiple segments of various sizes. This will pack | |
165 | * arbitrary data into an existing scatter gather list, segmenting the | |
166 | * data as necessary within constraints. | |
167 | * | |
168 | */ | |
169 | ||
abfa034e AK |
170 | static int pack_sg_list(struct scatterlist *sg, int start, |
171 | int limit, char *data, int count) | |
e2735b77 EVH |
172 | { |
173 | int s; | |
174 | int index = start; | |
175 | ||
176 | while (count) { | |
177 | s = rest_of_page(data); | |
178 | if (s > count) | |
179 | s = count; | |
23cba9cb | 180 | BUG_ON(index >= limit); |
0b36f1ad RR |
181 | /* Make sure we don't terminate early. */ |
182 | sg_unmark_end(&sg[index]); | |
e2735b77 EVH |
183 | sg_set_buf(&sg[index++], data, s); |
184 | count -= s; | |
185 | data += s; | |
e2735b77 | 186 | } |
0b36f1ad RR |
187 | if (index-start) |
188 | sg_mark_end(&sg[index - 1]); | |
e2735b77 | 189 | return index-start; |
b530cc79 EVH |
190 | } |
191 | ||
91b8534f EVH |
192 | /* We don't currently allow canceling of virtio requests */ |
193 | static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req) | |
194 | { | |
195 | return 1; | |
196 | } | |
197 | ||
728356de TB |
198 | /* Reply won't come, so drop req ref */ |
199 | static int p9_virtio_cancelled(struct p9_client *client, struct p9_req_t *req) | |
200 | { | |
8b11ff09 | 201 | p9_req_put(client, req); |
728356de TB |
202 | return 0; |
203 | } | |
204 | ||
4038866d VJJ |
205 | /** |
206 | * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer, | |
207 | * this takes a list of pages. | |
208 | * @sg: scatter/gather list to pack into | |
209 | * @start: which segment of the sg_list to start at | |
760b3d61 | 210 | * @limit: maximum number of pages in sg list. |
2c53040f | 211 | * @pdata: a list of pages to add into sg. |
abfa034e | 212 | * @nr_pages: number of pages to pack into the scatter/gather list |
4f3b35c1 | 213 | * @offs: amount of data in the beginning of first page _not_ to pack |
4038866d VJJ |
214 | * @count: amount of data to pack into the scatter/gather list |
215 | */ | |
216 | static int | |
abfa034e | 217 | pack_sg_list_p(struct scatterlist *sg, int start, int limit, |
4f3b35c1 | 218 | struct page **pdata, int nr_pages, size_t offs, int count) |
4038866d | 219 | { |
abfa034e | 220 | int i = 0, s; |
4f3b35c1 | 221 | int data_off = offs; |
4038866d VJJ |
222 | int index = start; |
223 | ||
abfa034e AK |
224 | BUG_ON(nr_pages > (limit - start)); |
225 | /* | |
226 | * if the first page doesn't start at | |
227 | * page boundary find the offset | |
228 | */ | |
abfa034e | 229 | while (nr_pages) { |
4f3b35c1 | 230 | s = PAGE_SIZE - data_off; |
abfa034e AK |
231 | if (s > count) |
232 | s = count; | |
23cba9cb | 233 | BUG_ON(index >= limit); |
0b36f1ad RR |
234 | /* Make sure we don't terminate early. */ |
235 | sg_unmark_end(&sg[index]); | |
abfa034e AK |
236 | sg_set_page(&sg[index++], pdata[i++], s, data_off); |
237 | data_off = 0; | |
4038866d | 238 | count -= s; |
abfa034e | 239 | nr_pages--; |
4038866d | 240 | } |
0b36f1ad RR |
241 | |
242 | if (index-start) | |
243 | sg_mark_end(&sg[index - 1]); | |
abfa034e | 244 | return index - start; |
4038866d VJJ |
245 | } |
246 | ||
ee443996 | 247 | /** |
91b8534f | 248 | * p9_virtio_request - issue a request |
0e15597e AK |
249 | * @client: client instance issuing the request |
250 | * @req: request to be issued | |
ee443996 EVH |
251 | * |
252 | */ | |
253 | ||
e2735b77 | 254 | static int |
91b8534f | 255 | p9_virtio_request(struct p9_client *client, struct p9_req_t *req) |
b530cc79 | 256 | { |
abfa034e | 257 | int err; |
0b36f1ad | 258 | int in, out, out_sgs, in_sgs; |
419b3956 | 259 | unsigned long flags; |
abfa034e | 260 | struct virtio_chan *chan = client->trans; |
0b36f1ad | 261 | struct scatterlist *sgs[2]; |
b530cc79 | 262 | |
5d385153 | 263 | p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n"); |
b530cc79 | 264 | |
1a4f69ef | 265 | WRITE_ONCE(req->status, REQ_STATUS_SENT); |
abfa034e AK |
266 | req_retry: |
267 | spin_lock_irqsave(&chan->lock, flags); | |
268 | ||
0b36f1ad | 269 | out_sgs = in_sgs = 0; |
abfa034e AK |
270 | /* Handle out VirtIO ring buffers */ |
271 | out = pack_sg_list(chan->sg, 0, | |
523adb6c | 272 | VIRTQUEUE_NUM, req->tc.sdata, req->tc.size); |
0b36f1ad RR |
273 | if (out) |
274 | sgs[out_sgs++] = chan->sg; | |
419b3956 | 275 | |
abfa034e | 276 | in = pack_sg_list(chan->sg, out, |
523adb6c | 277 | VIRTQUEUE_NUM, req->rc.sdata, req->rc.capacity); |
0b36f1ad RR |
278 | if (in) |
279 | sgs[out_sgs + in_sgs++] = chan->sg + out; | |
4038866d | 280 | |
474fe9f7 | 281 | err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req, |
f96fde41 | 282 | GFP_ATOMIC); |
abfa034e AK |
283 | if (err < 0) { |
284 | if (err == -ENOSPC) { | |
285 | chan->ring_bufs_avail = 0; | |
286 | spin_unlock_irqrestore(&chan->lock, flags); | |
9523feac TT |
287 | err = wait_event_killable(*chan->vc_wq, |
288 | chan->ring_bufs_avail); | |
68da9ba4 VJJ |
289 | if (err == -ERESTARTSYS) |
290 | return err; | |
68da9ba4 | 291 | |
5d385153 | 292 | p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n"); |
abfa034e | 293 | goto req_retry; |
4038866d | 294 | } else { |
abfa034e | 295 | spin_unlock_irqrestore(&chan->lock, flags); |
5d385153 | 296 | p9_debug(P9_DEBUG_TRANS, |
0b36f1ad | 297 | "virtio rpc add_sgs returned failure\n"); |
abfa034e | 298 | return -EIO; |
4038866d | 299 | } |
abfa034e AK |
300 | } |
301 | virtqueue_kick(chan->vq); | |
302 | spin_unlock_irqrestore(&chan->lock, flags); | |
4038866d | 303 | |
5d385153 | 304 | p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); |
abfa034e AK |
305 | return 0; |
306 | } | |
307 | ||
308 | static int p9_get_mapped_pages(struct virtio_chan *chan, | |
4f3b35c1 AV |
309 | struct page ***pages, |
310 | struct iov_iter *data, | |
311 | int count, | |
312 | size_t *offs, | |
313 | int *need_drop) | |
abfa034e | 314 | { |
4f3b35c1 | 315 | int nr_pages; |
abfa034e | 316 | int err; |
4f3b35c1 AV |
317 | |
318 | if (!iov_iter_count(data)) | |
319 | return 0; | |
320 | ||
2cbfdf4d | 321 | if (!iov_iter_is_kvec(data)) { |
4f3b35c1 | 322 | int n; |
abfa034e AK |
323 | /* |
324 | * We allow only p9_max_pages pinned. We wait for the | |
325 | * Other zc request to finish here | |
326 | */ | |
327 | if (atomic_read(&vp_pinned) >= chan->p9_max_pages) { | |
9523feac | 328 | err = wait_event_killable(vp_wq, |
abfa034e AK |
329 | (atomic_read(&vp_pinned) < chan->p9_max_pages)); |
330 | if (err == -ERESTARTSYS) | |
331 | return err; | |
332 | } | |
7f024647 | 333 | n = iov_iter_get_pages_alloc2(data, pages, count, offs); |
4f3b35c1 AV |
334 | if (n < 0) |
335 | return n; | |
336 | *need_drop = 1; | |
337 | nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE); | |
abfa034e | 338 | atomic_add(nr_pages, &vp_pinned); |
4f3b35c1 | 339 | return n; |
abfa034e AK |
340 | } else { |
341 | /* kernel buffer, no need to pin pages */ | |
4f3b35c1 AV |
342 | int index; |
343 | size_t len; | |
344 | void *p; | |
345 | ||
346 | /* we'd already checked that it's non-empty */ | |
347 | while (1) { | |
348 | len = iov_iter_single_seg_count(data); | |
349 | if (likely(len)) { | |
350 | p = data->kvec->iov_base + data->iov_offset; | |
351 | break; | |
352 | } | |
353 | iov_iter_advance(data, 0); | |
354 | } | |
355 | if (len > count) | |
356 | len = count; | |
357 | ||
358 | nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) - | |
359 | (unsigned long)p / PAGE_SIZE; | |
360 | ||
6da2ec56 KC |
361 | *pages = kmalloc_array(nr_pages, sizeof(struct page *), |
362 | GFP_NOFS); | |
4f3b35c1 AV |
363 | if (!*pages) |
364 | return -ENOMEM; | |
365 | ||
366 | *need_drop = 0; | |
222e4ade | 367 | p -= (*offs = offset_in_page(p)); |
4f3b35c1 AV |
368 | for (index = 0; index < nr_pages; index++) { |
369 | if (is_vmalloc_addr(p)) | |
370 | (*pages)[index] = vmalloc_to_page(p); | |
b6f52ae2 | 371 | else |
4f3b35c1 AV |
372 | (*pages)[index] = kmap_to_page(p); |
373 | p += PAGE_SIZE; | |
4038866d | 374 | } |
7f024647 | 375 | iov_iter_advance(data, len); |
4f3b35c1 | 376 | return len; |
4038866d | 377 | } |
abfa034e | 378 | } |
4038866d | 379 | |
f615625a AV |
380 | static void handle_rerror(struct p9_req_t *req, int in_hdr_len, |
381 | size_t offs, struct page **pages) | |
382 | { | |
383 | unsigned size, n; | |
384 | void *to = req->rc.sdata + in_hdr_len; | |
385 | ||
386 | // Fits entirely into the static data? Nothing to do. | |
387 | if (req->rc.size < in_hdr_len) | |
388 | return; | |
389 | ||
390 | // Really long error message? Tough, truncate the reply. Might get | |
391 | // rejected (we can't be arsed to adjust the size encoded in header, | |
392 | // or string size for that matter), but it wouldn't be anything valid | |
393 | // anyway. | |
394 | if (unlikely(req->rc.size > P9_ZC_HDR_SZ)) | |
395 | req->rc.size = P9_ZC_HDR_SZ; | |
396 | ||
397 | // data won't span more than two pages | |
398 | size = req->rc.size - in_hdr_len; | |
399 | n = PAGE_SIZE - offs; | |
400 | if (size > n) { | |
401 | memcpy_from_page(to, *pages++, offs, n); | |
402 | offs = 0; | |
403 | to += n; | |
404 | size -= n; | |
405 | } | |
406 | memcpy_from_page(to, *pages, offs, size); | |
407 | } | |
408 | ||
abfa034e AK |
409 | /** |
410 | * p9_virtio_zc_request - issue a zero copy request | |
411 | * @client: client instance issuing the request | |
412 | * @req: request to be issued | |
c7ebbae7 | 413 | * @uidata: user buffer that should be used for zero copy read |
414 | * @uodata: user buffer that should be used for zero copy write | |
abfa034e | 415 | * @inlen: read buffer size |
4a026da9 SL |
416 | * @outlen: write buffer size |
417 | * @in_hdr_len: reader header size, This is the size of response protocol data | |
abfa034e AK |
418 | * |
419 | */ | |
420 | static int | |
421 | p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req, | |
4f3b35c1 AV |
422 | struct iov_iter *uidata, struct iov_iter *uodata, |
423 | int inlen, int outlen, int in_hdr_len) | |
abfa034e | 424 | { |
0b36f1ad | 425 | int in, out, err, out_sgs, in_sgs; |
abfa034e AK |
426 | unsigned long flags; |
427 | int in_nr_pages = 0, out_nr_pages = 0; | |
428 | struct page **in_pages = NULL, **out_pages = NULL; | |
429 | struct virtio_chan *chan = client->trans; | |
0b36f1ad | 430 | struct scatterlist *sgs[4]; |
4f3b35c1 AV |
431 | size_t offs; |
432 | int need_drop = 0; | |
728356de | 433 | int kicked = 0; |
4038866d | 434 | |
5d385153 | 435 | p9_debug(P9_DEBUG_TRANS, "virtio request\n"); |
abfa034e AK |
436 | |
437 | if (uodata) { | |
d28c756c | 438 | __le32 sz; |
4f3b35c1 AV |
439 | int n = p9_get_mapped_pages(chan, &out_pages, uodata, |
440 | outlen, &offs, &need_drop); | |
728356de TB |
441 | if (n < 0) { |
442 | err = n; | |
443 | goto err_out; | |
444 | } | |
4f3b35c1 AV |
445 | out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE); |
446 | if (n != outlen) { | |
447 | __le32 v = cpu_to_le32(n); | |
523adb6c | 448 | memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4); |
4f3b35c1 | 449 | outlen = n; |
4038866d | 450 | } |
d28c756c CE |
451 | /* The size field of the message must include the length of the |
452 | * header and the length of the data. We didn't actually know | |
453 | * the length of the data until this point so add it in now. | |
454 | */ | |
523adb6c DM |
455 | sz = cpu_to_le32(req->tc.size + outlen); |
456 | memcpy(&req->tc.sdata[0], &sz, sizeof(sz)); | |
4f3b35c1 AV |
457 | } else if (uidata) { |
458 | int n = p9_get_mapped_pages(chan, &in_pages, uidata, | |
459 | inlen, &offs, &need_drop); | |
728356de TB |
460 | if (n < 0) { |
461 | err = n; | |
462 | goto err_out; | |
463 | } | |
4f3b35c1 AV |
464 | in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE); |
465 | if (n != inlen) { | |
466 | __le32 v = cpu_to_le32(n); | |
523adb6c | 467 | memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4); |
4f3b35c1 | 468 | inlen = n; |
4038866d | 469 | } |
4038866d | 470 | } |
1a4f69ef | 471 | WRITE_ONCE(req->status, REQ_STATUS_SENT); |
abfa034e AK |
472 | req_retry_pinned: |
473 | spin_lock_irqsave(&chan->lock, flags); | |
0b36f1ad RR |
474 | |
475 | out_sgs = in_sgs = 0; | |
476 | ||
abfa034e AK |
477 | /* out data */ |
478 | out = pack_sg_list(chan->sg, 0, | |
523adb6c | 479 | VIRTQUEUE_NUM, req->tc.sdata, req->tc.size); |
abfa034e | 480 | |
0b36f1ad RR |
481 | if (out) |
482 | sgs[out_sgs++] = chan->sg; | |
483 | ||
484 | if (out_pages) { | |
485 | sgs[out_sgs++] = chan->sg + out; | |
abfa034e | 486 | out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM, |
4f3b35c1 | 487 | out_pages, out_nr_pages, offs, outlen); |
0b36f1ad | 488 | } |
c69f297d | 489 | |
abfa034e AK |
490 | /* |
491 | * Take care of in data | |
492 | * For example TREAD have 11. | |
493 | * 11 is the read/write header = PDU Header(7) + IO Size (4). | |
494 | * Arrange in such a way that server places header in the | |
8ab1784d | 495 | * allocated memory and payload onto the user buffer. |
abfa034e AK |
496 | */ |
497 | in = pack_sg_list(chan->sg, out, | |
523adb6c | 498 | VIRTQUEUE_NUM, req->rc.sdata, in_hdr_len); |
0b36f1ad RR |
499 | if (in) |
500 | sgs[out_sgs + in_sgs++] = chan->sg + out; | |
501 | ||
502 | if (in_pages) { | |
503 | sgs[out_sgs + in_sgs++] = chan->sg + out + in; | |
abfa034e | 504 | in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM, |
4f3b35c1 | 505 | in_pages, in_nr_pages, offs, inlen); |
0b36f1ad | 506 | } |
b530cc79 | 507 | |
0b36f1ad | 508 | BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs)); |
474fe9f7 | 509 | err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req, |
f96fde41 | 510 | GFP_ATOMIC); |
419b3956 | 511 | if (err < 0) { |
52f44e0d VJJ |
512 | if (err == -ENOSPC) { |
513 | chan->ring_bufs_avail = 0; | |
514 | spin_unlock_irqrestore(&chan->lock, flags); | |
9523feac TT |
515 | err = wait_event_killable(*chan->vc_wq, |
516 | chan->ring_bufs_avail); | |
52f44e0d | 517 | if (err == -ERESTARTSYS) |
abfa034e | 518 | goto err_out; |
52f44e0d | 519 | |
5d385153 | 520 | p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n"); |
316ad550 | 521 | goto req_retry_pinned; |
52f44e0d VJJ |
522 | } else { |
523 | spin_unlock_irqrestore(&chan->lock, flags); | |
5d385153 | 524 | p9_debug(P9_DEBUG_TRANS, |
0b36f1ad | 525 | "virtio rpc add_sgs returned failure\n"); |
abfa034e AK |
526 | err = -EIO; |
527 | goto err_out; | |
52f44e0d | 528 | } |
e2735b77 | 529 | } |
dc3f5e68 | 530 | virtqueue_kick(chan->vq); |
419b3956 | 531 | spin_unlock_irqrestore(&chan->lock, flags); |
728356de | 532 | kicked = 1; |
5d385153 | 533 | p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); |
1a4f69ef DM |
534 | err = wait_event_killable(req->wq, |
535 | READ_ONCE(req->status) >= REQ_STATUS_RCVD); | |
f615625a | 536 | // RERROR needs reply (== error string) in static data |
1a4f69ef | 537 | if (READ_ONCE(req->status) == REQ_STATUS_RCVD && |
f615625a AV |
538 | unlikely(req->rc.sdata[4] == P9_RERROR)) |
539 | handle_rerror(req, in_hdr_len, offs, in_pages); | |
540 | ||
abfa034e AK |
541 | /* |
542 | * Non kernel buffers are pinned, unpin them | |
543 | */ | |
544 | err_out: | |
4f3b35c1 | 545 | if (need_drop) { |
abfa034e AK |
546 | if (in_pages) { |
547 | p9_release_pages(in_pages, in_nr_pages); | |
548 | atomic_sub(in_nr_pages, &vp_pinned); | |
549 | } | |
550 | if (out_pages) { | |
551 | p9_release_pages(out_pages, out_nr_pages); | |
552 | atomic_sub(out_nr_pages, &vp_pinned); | |
553 | } | |
554 | /* wakeup anybody waiting for slots to pin pages */ | |
555 | wake_up(&vp_wq); | |
556 | } | |
1b8553c0 VN |
557 | kvfree(in_pages); |
558 | kvfree(out_pages); | |
728356de TB |
559 | if (!kicked) { |
560 | /* reply won't come */ | |
8b11ff09 | 561 | p9_req_put(client, req); |
728356de | 562 | } |
abfa034e | 563 | return err; |
b530cc79 EVH |
564 | } |
565 | ||
86c84373 AK |
566 | static ssize_t p9_mount_tag_show(struct device *dev, |
567 | struct device_attribute *attr, char *buf) | |
568 | { | |
569 | struct virtio_chan *chan; | |
570 | struct virtio_device *vdev; | |
edcd9d97 | 571 | int tag_len; |
86c84373 AK |
572 | |
573 | vdev = dev_to_virtio(dev); | |
574 | chan = vdev->priv; | |
edcd9d97 | 575 | tag_len = strlen(chan->tag); |
86c84373 | 576 | |
edcd9d97 | 577 | memcpy(buf, chan->tag, tag_len + 1); |
179a5bc4 | 578 | |
edcd9d97 | 579 | return tag_len + 1; |
86c84373 AK |
580 | } |
581 | ||
582 | static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL); | |
583 | ||
ee443996 EVH |
584 | /** |
585 | * p9_virtio_probe - probe for existence of 9P virtio channels | |
586 | * @vdev: virtio device to probe | |
587 | * | |
37c1209d | 588 | * This probes for existing virtio channels. |
ee443996 EVH |
589 | * |
590 | */ | |
591 | ||
e2735b77 | 592 | static int p9_virtio_probe(struct virtio_device *vdev) |
b530cc79 | 593 | { |
97ee9b02 AK |
594 | __u16 tag_len; |
595 | char *tag; | |
b530cc79 EVH |
596 | int err; |
597 | struct virtio_chan *chan; | |
b530cc79 | 598 | |
7754f53e MT |
599 | if (!vdev->config->get) { |
600 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
601 | __func__); | |
602 | return -EINVAL; | |
603 | } | |
604 | ||
37c1209d AK |
605 | chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL); |
606 | if (!chan) { | |
5d385153 | 607 | pr_err("Failed to allocate virtio 9P channel\n"); |
b530cc79 EVH |
608 | err = -ENOMEM; |
609 | goto fail; | |
610 | } | |
611 | ||
e2735b77 | 612 | chan->vdev = vdev; |
b530cc79 | 613 | |
e2735b77 | 614 | /* We expect one virtqueue, for requests. */ |
d2a7ddda | 615 | chan->vq = virtio_find_single_vq(vdev, req_done, "requests"); |
e2735b77 EVH |
616 | if (IS_ERR(chan->vq)) { |
617 | err = PTR_ERR(chan->vq); | |
92aef467 | 618 | goto out_free_chan; |
b530cc79 | 619 | } |
e2735b77 EVH |
620 | chan->vq->vdev->priv = chan; |
621 | spin_lock_init(&chan->lock); | |
b530cc79 | 622 | |
e2735b77 | 623 | sg_init_table(chan->sg, VIRTQUEUE_NUM); |
b530cc79 | 624 | |
b530cc79 | 625 | chan->inuse = false; |
97ee9b02 | 626 | if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) { |
855e0c52 | 627 | virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len); |
97ee9b02 AK |
628 | } else { |
629 | err = -EINVAL; | |
630 | goto out_free_vq; | |
631 | } | |
edcd9d97 | 632 | tag = kzalloc(tag_len + 1, GFP_KERNEL); |
97ee9b02 AK |
633 | if (!tag) { |
634 | err = -ENOMEM; | |
635 | goto out_free_vq; | |
636 | } | |
855e0c52 RR |
637 | |
638 | virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag), | |
639 | tag, tag_len); | |
97ee9b02 | 640 | chan->tag = tag; |
86c84373 AK |
641 | err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr); |
642 | if (err) { | |
52f44e0d | 643 | goto out_free_tag; |
86c84373 | 644 | } |
52f44e0d VJJ |
645 | chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL); |
646 | if (!chan->vc_wq) { | |
647 | err = -ENOMEM; | |
f997ea3b | 648 | goto out_remove_file; |
52f44e0d VJJ |
649 | } |
650 | init_waitqueue_head(chan->vc_wq); | |
651 | chan->ring_bufs_avail = 1; | |
68da9ba4 VJJ |
652 | /* Ceiling limit to avoid denial of service attacks */ |
653 | chan->p9_max_pages = nr_free_buffer_pages()/4; | |
52f44e0d | 654 | |
64b4cc39 MT |
655 | virtio_device_ready(vdev); |
656 | ||
37c1209d AK |
657 | mutex_lock(&virtio_9p_lock); |
658 | list_add_tail(&chan->chan_list, &virtio_chan_list); | |
659 | mutex_unlock(&virtio_9p_lock); | |
e0d6cb9c MM |
660 | |
661 | /* Let udev rules use the new mount_tag attribute. */ | |
662 | kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE); | |
663 | ||
b530cc79 EVH |
664 | return 0; |
665 | ||
f997ea3b XY |
666 | out_remove_file: |
667 | sysfs_remove_file(&vdev->dev.kobj, &dev_attr_mount_tag.attr); | |
52f44e0d VJJ |
668 | out_free_tag: |
669 | kfree(tag); | |
e2735b77 | 670 | out_free_vq: |
d2a7ddda | 671 | vdev->config->del_vqs(vdev); |
92aef467 | 672 | out_free_chan: |
37c1209d | 673 | kfree(chan); |
b530cc79 | 674 | fail: |
b530cc79 EVH |
675 | return err; |
676 | } | |
677 | ||
ee443996 EVH |
678 | |
679 | /** | |
680 | * p9_virtio_create - allocate a new virtio channel | |
8b81ef58 | 681 | * @client: client instance invoking this transport |
ee443996 EVH |
682 | * @devname: string identifying the channel to connect to (unused) |
683 | * @args: args passed from sys_mount() for per-transport options (unused) | |
ee443996 EVH |
684 | * |
685 | * This sets up a transport channel for 9p communication. Right now | |
f2745669 | 686 | * we only match the first available channel, but eventually we could look up |
b530cc79 EVH |
687 | * alternate channels by matching devname versus a virtio_config entry. |
688 | * We use a simple reference count mechanism to ensure that only a single | |
ee443996 EVH |
689 | * mount has a channel open at a time. |
690 | * | |
ee443996 EVH |
691 | */ |
692 | ||
8b81ef58 EVH |
693 | static int |
694 | p9_virtio_create(struct p9_client *client, const char *devname, char *args) | |
b530cc79 | 695 | { |
37c1209d | 696 | struct virtio_chan *chan; |
c1a7c226 | 697 | int ret = -ENOENT; |
37c1209d | 698 | int found = 0; |
b530cc79 | 699 | |
10aa1452 TB |
700 | if (devname == NULL) |
701 | return -EINVAL; | |
702 | ||
c1549497 | 703 | mutex_lock(&virtio_9p_lock); |
37c1209d | 704 | list_for_each_entry(chan, &virtio_chan_list, chan_list) { |
edcd9d97 | 705 | if (!strcmp(devname, chan->tag)) { |
f75580c4 AK |
706 | if (!chan->inuse) { |
707 | chan->inuse = true; | |
37c1209d | 708 | found = 1; |
f75580c4 AK |
709 | break; |
710 | } | |
c1a7c226 | 711 | ret = -EBUSY; |
b530cc79 EVH |
712 | } |
713 | } | |
c1549497 | 714 | mutex_unlock(&virtio_9p_lock); |
b530cc79 | 715 | |
37c1209d | 716 | if (!found) { |
c7c72c5a | 717 | pr_err("no channels available for device %s\n", devname); |
c1a7c226 | 718 | return ret; |
e2735b77 EVH |
719 | } |
720 | ||
8b81ef58 | 721 | client->trans = (void *)chan; |
562ada61 | 722 | client->status = Connected; |
fea511a6 | 723 | chan->client = client; |
b530cc79 | 724 | |
8b81ef58 | 725 | return 0; |
b530cc79 EVH |
726 | } |
727 | ||
ee443996 EVH |
728 | /** |
729 | * p9_virtio_remove - clean up resources associated with a virtio device | |
730 | * @vdev: virtio device to remove | |
731 | * | |
732 | */ | |
733 | ||
f3933545 EVH |
734 | static void p9_virtio_remove(struct virtio_device *vdev) |
735 | { | |
736 | struct virtio_chan *chan = vdev->priv; | |
8051a2a5 | 737 | unsigned long warning_time; |
37c1209d AK |
738 | |
739 | mutex_lock(&virtio_9p_lock); | |
8051a2a5 MT |
740 | |
741 | /* Remove self from list so we don't get new users. */ | |
37c1209d | 742 | list_del(&chan->chan_list); |
8051a2a5 MT |
743 | warning_time = jiffies; |
744 | ||
745 | /* Wait for existing users to close. */ | |
746 | while (chan->inuse) { | |
747 | mutex_unlock(&virtio_9p_lock); | |
748 | msleep(250); | |
749 | if (time_after(jiffies, warning_time + 10 * HZ)) { | |
750 | dev_emerg(&vdev->dev, | |
751 | "p9_virtio_remove: waiting for device in use.\n"); | |
752 | warning_time = jiffies; | |
753 | } | |
754 | mutex_lock(&virtio_9p_lock); | |
755 | } | |
756 | ||
37c1209d | 757 | mutex_unlock(&virtio_9p_lock); |
8051a2a5 | 758 | |
d9679d00 | 759 | virtio_reset_device(vdev); |
8051a2a5 MT |
760 | vdev->config->del_vqs(vdev); |
761 | ||
86c84373 | 762 | sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr); |
e0d6cb9c | 763 | kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE); |
97ee9b02 | 764 | kfree(chan->tag); |
52f44e0d | 765 | kfree(chan->vc_wq); |
37c1209d | 766 | kfree(chan); |
f3933545 | 767 | |
f3933545 EVH |
768 | } |
769 | ||
b530cc79 EVH |
770 | static struct virtio_device_id id_table[] = { |
771 | { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID }, | |
772 | { 0 }, | |
773 | }; | |
774 | ||
97ee9b02 AK |
775 | static unsigned int features[] = { |
776 | VIRTIO_9P_MOUNT_TAG, | |
777 | }; | |
778 | ||
b530cc79 EVH |
779 | /* The standard "struct lguest_driver": */ |
780 | static struct virtio_driver p9_virtio_drv = { | |
97ee9b02 AK |
781 | .feature_table = features, |
782 | .feature_table_size = ARRAY_SIZE(features), | |
783 | .driver.name = KBUILD_MODNAME, | |
784 | .driver.owner = THIS_MODULE, | |
785 | .id_table = id_table, | |
786 | .probe = p9_virtio_probe, | |
787 | .remove = p9_virtio_remove, | |
b530cc79 EVH |
788 | }; |
789 | ||
790 | static struct p9_trans_module p9_virtio_trans = { | |
791 | .name = "virtio", | |
792 | .create = p9_virtio_create, | |
8b81ef58 | 793 | .close = p9_virtio_close, |
91b8534f | 794 | .request = p9_virtio_request, |
dc893e19 | 795 | .zc_request = p9_virtio_zc_request, |
91b8534f | 796 | .cancel = p9_virtio_cancel, |
728356de | 797 | .cancelled = p9_virtio_cancelled, |
b49d8b5d AK |
798 | /* |
799 | * We leave one entry for input and one entry for response | |
8ab1784d | 800 | * headers. We also skip one more entry to accommodate, address |
b49d8b5d AK |
801 | * that are not at page boundary, that can result in an extra |
802 | * page in zero copy. | |
803 | */ | |
804 | .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3), | |
01d205d9 | 805 | .pooled_rbuffers = false, |
f94741fd | 806 | .def = 1, |
72029fe8 | 807 | .owner = THIS_MODULE, |
b530cc79 EVH |
808 | }; |
809 | ||
810 | /* The standard init function */ | |
811 | static int __init p9_virtio_init(void) | |
812 | { | |
d4548543 Y |
813 | int rc; |
814 | ||
37c1209d | 815 | INIT_LIST_HEAD(&virtio_chan_list); |
b530cc79 EVH |
816 | |
817 | v9fs_register_trans(&p9_virtio_trans); | |
d4548543 Y |
818 | rc = register_virtio_driver(&p9_virtio_drv); |
819 | if (rc) | |
820 | v9fs_unregister_trans(&p9_virtio_trans); | |
821 | ||
822 | return rc; | |
b530cc79 EVH |
823 | } |
824 | ||
f3933545 EVH |
825 | static void __exit p9_virtio_cleanup(void) |
826 | { | |
827 | unregister_virtio_driver(&p9_virtio_drv); | |
72029fe8 | 828 | v9fs_unregister_trans(&p9_virtio_trans); |
f3933545 EVH |
829 | } |
830 | ||
b530cc79 | 831 | module_init(p9_virtio_init); |
f3933545 | 832 | module_exit(p9_virtio_cleanup); |
4cd82a5b | 833 | MODULE_ALIAS_9P("virtio"); |
b530cc79 EVH |
834 | |
835 | MODULE_DEVICE_TABLE(virtio, id_table); | |
836 | MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>"); | |
837 | MODULE_DESCRIPTION("Virtio 9p Transport"); | |
838 | MODULE_LICENSE("GPL"); |