| 1 | // SPDX-License-Identifier: GPL-2.0-only | 
|---|
| 2 | /* | 
|---|
| 3 | * The Virtio 9p transport driver | 
|---|
| 4 | * | 
|---|
| 5 | * This is a block based transport driver based on the lguest block driver | 
|---|
| 6 | * code. | 
|---|
| 7 | * | 
|---|
| 8 | *  Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation | 
|---|
| 9 | * | 
|---|
| 10 | *  Based on virtio console driver | 
|---|
| 11 | *  Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation | 
|---|
| 12 | */ | 
|---|
| 13 |  | 
|---|
| 14 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | 
|---|
| 15 |  | 
|---|
| 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> | 
|---|
| 25 | #include <linux/file.h> | 
|---|
| 26 | #include <linux/highmem.h> | 
|---|
| 27 | #include <linux/slab.h> | 
|---|
| 28 | #include <net/9p/9p.h> | 
|---|
| 29 | #include <linux/parser.h> | 
|---|
| 30 | #include <net/9p/client.h> | 
|---|
| 31 | #include <net/9p/transport.h> | 
|---|
| 32 | #include <linux/scatterlist.h> | 
|---|
| 33 | #include <linux/swap.h> | 
|---|
| 34 | #include <linux/virtio.h> | 
|---|
| 35 | #include <linux/virtio_9p.h> | 
|---|
| 36 | #include "trans_common.h" | 
|---|
| 37 |  | 
|---|
| 38 | #define VIRTQUEUE_NUM	128 | 
|---|
| 39 |  | 
|---|
| 40 | /* a single mutex to manage channel initialization and attachment */ | 
|---|
| 41 | static DEFINE_MUTEX(virtio_9p_lock); | 
|---|
| 42 | static DECLARE_WAIT_QUEUE_HEAD(vp_wq); | 
|---|
| 43 | static atomic_t vp_pinned = ATOMIC_INIT(0); | 
|---|
| 44 |  | 
|---|
| 45 | /** | 
|---|
| 46 | * struct virtio_chan - per-instance transport information | 
|---|
| 47 | * @inuse: whether the channel is in use | 
|---|
| 48 | * @lock: protects multiple elements within this structure | 
|---|
| 49 | * @client: client instance | 
|---|
| 50 | * @vdev: virtio dev associated with this channel | 
|---|
| 51 | * @vq: virtio queue associated with this channel | 
|---|
| 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 | 
|---|
| 55 | * @sg: scatter gather list which is used to pack a request (protected?) | 
|---|
| 56 | * @chan_list: linked list of channels | 
|---|
| 57 | * | 
|---|
| 58 | * We keep all per-channel information in a structure. | 
|---|
| 59 | * This structure is allocated within the devices dev->mem space. | 
|---|
| 60 | * A pointer to the structure will get put in the transport private. | 
|---|
| 61 | * | 
|---|
| 62 | */ | 
|---|
| 63 |  | 
|---|
| 64 | struct virtio_chan { | 
|---|
| 65 | bool inuse; | 
|---|
| 66 |  | 
|---|
| 67 | spinlock_t lock; | 
|---|
| 68 |  | 
|---|
| 69 | struct p9_client *client; | 
|---|
| 70 | struct virtio_device *vdev; | 
|---|
| 71 | struct virtqueue *vq; | 
|---|
| 72 | int ring_bufs_avail; | 
|---|
| 73 | wait_queue_head_t *vc_wq; | 
|---|
| 74 | /* This is global limit. Since we don't have a global structure, | 
|---|
| 75 | * will be placing it in each channel. | 
|---|
| 76 | */ | 
|---|
| 77 | unsigned long p9_max_pages; | 
|---|
| 78 | /* Scatterlist: can be too big for stack. */ | 
|---|
| 79 | struct scatterlist sg[VIRTQUEUE_NUM]; | 
|---|
| 80 | /** | 
|---|
| 81 | * @tag: name to identify a mount null terminated | 
|---|
| 82 | */ | 
|---|
| 83 | char *tag; | 
|---|
| 84 |  | 
|---|
| 85 | struct list_head chan_list; | 
|---|
| 86 | }; | 
|---|
| 87 |  | 
|---|
| 88 | static struct list_head virtio_chan_list; | 
|---|
| 89 |  | 
|---|
| 90 | /* How many bytes left in this page. */ | 
|---|
| 91 | static unsigned int rest_of_page(void *data) | 
|---|
| 92 | { | 
|---|
| 93 | return PAGE_SIZE - offset_in_page(data); | 
|---|
| 94 | } | 
|---|
| 95 |  | 
|---|
| 96 | /** | 
|---|
| 97 | * p9_virtio_close - reclaim resources of a channel | 
|---|
| 98 | * @client: client instance | 
|---|
| 99 | * | 
|---|
| 100 | * This reclaims a channel by freeing its resources and | 
|---|
| 101 | * resetting its inuse flag. | 
|---|
| 102 | * | 
|---|
| 103 | */ | 
|---|
| 104 |  | 
|---|
| 105 | static void p9_virtio_close(struct p9_client *client) | 
|---|
| 106 | { | 
|---|
| 107 | struct virtio_chan *chan = client->trans; | 
|---|
| 108 |  | 
|---|
| 109 | mutex_lock(lock: &virtio_9p_lock); | 
|---|
| 110 | if (chan) | 
|---|
| 111 | chan->inuse = false; | 
|---|
| 112 | mutex_unlock(lock: &virtio_9p_lock); | 
|---|
| 113 | } | 
|---|
| 114 |  | 
|---|
| 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 |  | 
|---|
| 128 | static void req_done(struct virtqueue *vq) | 
|---|
| 129 | { | 
|---|
| 130 | struct virtio_chan *chan = vq->vdev->priv; | 
|---|
| 131 | unsigned int len; | 
|---|
| 132 | struct p9_req_t *req; | 
|---|
| 133 | bool need_wakeup = false; | 
|---|
| 134 | unsigned long flags; | 
|---|
| 135 |  | 
|---|
| 136 | p9_debug(P9_DEBUG_TRANS, ": request done\n"); | 
|---|
| 137 |  | 
|---|
| 138 | spin_lock_irqsave(&chan->lock, flags); | 
|---|
| 139 | while ((req = virtqueue_get_buf(vq: chan->vq, len: &len)) != NULL) { | 
|---|
| 140 | if (!chan->ring_bufs_avail) { | 
|---|
| 141 | chan->ring_bufs_avail = 1; | 
|---|
| 142 | need_wakeup = true; | 
|---|
| 143 | } | 
|---|
| 144 |  | 
|---|
| 145 | if (len) { | 
|---|
| 146 | req->rc.size = len; | 
|---|
| 147 | p9_client_cb(c: chan->client, req, status: REQ_STATUS_RCVD); | 
|---|
| 148 | } | 
|---|
| 149 | } | 
|---|
| 150 | spin_unlock_irqrestore(lock: &chan->lock, flags); | 
|---|
| 151 | /* Wakeup if anyone waiting for VirtIO ring space. */ | 
|---|
| 152 | if (need_wakeup) | 
|---|
| 153 | wake_up(chan->vc_wq); | 
|---|
| 154 | } | 
|---|
| 155 |  | 
|---|
| 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 |  | 
|---|
| 170 | static int pack_sg_list(struct scatterlist *sg, int start, | 
|---|
| 171 | int limit, char *data, int count) | 
|---|
| 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; | 
|---|
| 180 | BUG_ON(index >= limit); | 
|---|
| 181 | /* Make sure we don't terminate early. */ | 
|---|
| 182 | sg_unmark_end(sg: &sg[index]); | 
|---|
| 183 | sg_set_buf(sg: &sg[index++], buf: data, buflen: s); | 
|---|
| 184 | count -= s; | 
|---|
| 185 | data += s; | 
|---|
| 186 | } | 
|---|
| 187 | if (index-start) | 
|---|
| 188 | sg_mark_end(sg: &sg[index - 1]); | 
|---|
| 189 | return index-start; | 
|---|
| 190 | } | 
|---|
| 191 |  | 
|---|
| 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 |  | 
|---|
| 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 | { | 
|---|
| 201 | p9_req_put(c: client, r: req); | 
|---|
| 202 | return 0; | 
|---|
| 203 | } | 
|---|
| 204 |  | 
|---|
| 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 | 
|---|
| 210 | * @limit: maximum number of pages in sg list. | 
|---|
| 211 | * @pdata: a list of pages to add into sg. | 
|---|
| 212 | * @nr_pages: number of pages to pack into the scatter/gather list | 
|---|
| 213 | * @offs: amount of data in the beginning of first page _not_ to pack | 
|---|
| 214 | * @count: amount of data to pack into the scatter/gather list | 
|---|
| 215 | */ | 
|---|
| 216 | static int | 
|---|
| 217 | pack_sg_list_p(struct scatterlist *sg, int start, int limit, | 
|---|
| 218 | struct page **pdata, int nr_pages, size_t offs, int count) | 
|---|
| 219 | { | 
|---|
| 220 | int i = 0, s; | 
|---|
| 221 | int data_off = offs; | 
|---|
| 222 | int index = start; | 
|---|
| 223 |  | 
|---|
| 224 | BUG_ON(nr_pages > (limit - start)); | 
|---|
| 225 | /* | 
|---|
| 226 | * if the first page doesn't start at | 
|---|
| 227 | * page boundary find the offset | 
|---|
| 228 | */ | 
|---|
| 229 | while (nr_pages) { | 
|---|
| 230 | s = PAGE_SIZE - data_off; | 
|---|
| 231 | if (s > count) | 
|---|
| 232 | s = count; | 
|---|
| 233 | BUG_ON(index >= limit); | 
|---|
| 234 | /* Make sure we don't terminate early. */ | 
|---|
| 235 | sg_unmark_end(sg: &sg[index]); | 
|---|
| 236 | sg_set_page(sg: &sg[index++], page: pdata[i++], len: s, offset: data_off); | 
|---|
| 237 | data_off = 0; | 
|---|
| 238 | count -= s; | 
|---|
| 239 | nr_pages--; | 
|---|
| 240 | } | 
|---|
| 241 |  | 
|---|
| 242 | if (index-start) | 
|---|
| 243 | sg_mark_end(sg: &sg[index - 1]); | 
|---|
| 244 | return index - start; | 
|---|
| 245 | } | 
|---|
| 246 |  | 
|---|
| 247 | /** | 
|---|
| 248 | * p9_virtio_request - issue a request | 
|---|
| 249 | * @client: client instance issuing the request | 
|---|
| 250 | * @req: request to be issued | 
|---|
| 251 | * | 
|---|
| 252 | */ | 
|---|
| 253 |  | 
|---|
| 254 | static int | 
|---|
| 255 | p9_virtio_request(struct p9_client *client, struct p9_req_t *req) | 
|---|
| 256 | { | 
|---|
| 257 | int err; | 
|---|
| 258 | int in, out, out_sgs, in_sgs; | 
|---|
| 259 | unsigned long flags; | 
|---|
| 260 | struct virtio_chan *chan = client->trans; | 
|---|
| 261 | struct scatterlist *sgs[2]; | 
|---|
| 262 |  | 
|---|
| 263 | p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n"); | 
|---|
| 264 |  | 
|---|
| 265 | WRITE_ONCE(req->status, REQ_STATUS_SENT); | 
|---|
| 266 | req_retry: | 
|---|
| 267 | spin_lock_irqsave(&chan->lock, flags); | 
|---|
| 268 |  | 
|---|
| 269 | out_sgs = in_sgs = 0; | 
|---|
| 270 | /* Handle out VirtIO ring buffers */ | 
|---|
| 271 | out = pack_sg_list(sg: chan->sg, start: 0, | 
|---|
| 272 | VIRTQUEUE_NUM, data: req->tc.sdata, count: req->tc.size); | 
|---|
| 273 | if (out) | 
|---|
| 274 | sgs[out_sgs++] = chan->sg; | 
|---|
| 275 |  | 
|---|
| 276 | in = pack_sg_list(sg: chan->sg, start: out, | 
|---|
| 277 | VIRTQUEUE_NUM, data: req->rc.sdata, count: req->rc.capacity); | 
|---|
| 278 | if (in) | 
|---|
| 279 | sgs[out_sgs + in_sgs++] = chan->sg + out; | 
|---|
| 280 |  | 
|---|
| 281 | err = virtqueue_add_sgs(vq: chan->vq, sgs, out_sgs, in_sgs, data: req, | 
|---|
| 282 | GFP_ATOMIC); | 
|---|
| 283 | if (err < 0) { | 
|---|
| 284 | if (err == -ENOSPC) { | 
|---|
| 285 | chan->ring_bufs_avail = 0; | 
|---|
| 286 | spin_unlock_irqrestore(lock: &chan->lock, flags); | 
|---|
| 287 | err = wait_event_killable(*chan->vc_wq, | 
|---|
| 288 | chan->ring_bufs_avail); | 
|---|
| 289 | if (err  == -ERESTARTSYS) | 
|---|
| 290 | return err; | 
|---|
| 291 |  | 
|---|
| 292 | p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n"); | 
|---|
| 293 | goto req_retry; | 
|---|
| 294 | } else { | 
|---|
| 295 | spin_unlock_irqrestore(lock: &chan->lock, flags); | 
|---|
| 296 | p9_debug(P9_DEBUG_TRANS, | 
|---|
| 297 | "virtio rpc add_sgs returned failure\n"); | 
|---|
| 298 | return -EIO; | 
|---|
| 299 | } | 
|---|
| 300 | } | 
|---|
| 301 | virtqueue_kick(vq: chan->vq); | 
|---|
| 302 | spin_unlock_irqrestore(lock: &chan->lock, flags); | 
|---|
| 303 |  | 
|---|
| 304 | p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); | 
|---|
| 305 | return 0; | 
|---|
| 306 | } | 
|---|
| 307 |  | 
|---|
| 308 | static int p9_get_mapped_pages(struct virtio_chan *chan, | 
|---|
| 309 | struct page ***pages, | 
|---|
| 310 | struct iov_iter *data, | 
|---|
| 311 | int count, | 
|---|
| 312 | size_t *offs, | 
|---|
| 313 | int *need_drop) | 
|---|
| 314 | { | 
|---|
| 315 | int nr_pages; | 
|---|
| 316 | int err; | 
|---|
| 317 |  | 
|---|
| 318 | if (!iov_iter_count(i: data)) | 
|---|
| 319 | return 0; | 
|---|
| 320 |  | 
|---|
| 321 | if (!iov_iter_is_kvec(i: data)) { | 
|---|
| 322 | int n; | 
|---|
| 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(v: &vp_pinned) >= chan->p9_max_pages) { | 
|---|
| 328 | err = wait_event_killable(vp_wq, | 
|---|
| 329 | (atomic_read(&vp_pinned) < chan->p9_max_pages)); | 
|---|
| 330 | if (err == -ERESTARTSYS) | 
|---|
| 331 | return err; | 
|---|
| 332 | } | 
|---|
| 333 | n = iov_iter_get_pages_alloc2(i: data, pages, maxsize: count, start: offs); | 
|---|
| 334 | if (n < 0) | 
|---|
| 335 | return n; | 
|---|
| 336 | *need_drop = 1; | 
|---|
| 337 | nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE); | 
|---|
| 338 | atomic_add(i: nr_pages, v: &vp_pinned); | 
|---|
| 339 | return n; | 
|---|
| 340 | } else { | 
|---|
| 341 | /* kernel buffer, no need to pin pages */ | 
|---|
| 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(i: data); | 
|---|
| 349 | if (likely(len)) { | 
|---|
| 350 | p = data->kvec->iov_base + data->iov_offset; | 
|---|
| 351 | break; | 
|---|
| 352 | } | 
|---|
| 353 | iov_iter_advance(i: data, bytes: 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 |  | 
|---|
| 361 | *pages = kmalloc_array(nr_pages, sizeof(struct page *), | 
|---|
| 362 | GFP_NOFS); | 
|---|
| 363 | if (!*pages) | 
|---|
| 364 | return -ENOMEM; | 
|---|
| 365 |  | 
|---|
| 366 | *need_drop = 0; | 
|---|
| 367 | p -= (*offs = offset_in_page(p)); | 
|---|
| 368 | for (index = 0; index < nr_pages; index++) { | 
|---|
| 369 | if (is_vmalloc_addr(x: p)) | 
|---|
| 370 | (*pages)[index] = vmalloc_to_page(addr: p); | 
|---|
| 371 | else | 
|---|
| 372 | (*pages)[index] = kmap_to_page(addr: p); | 
|---|
| 373 | p += PAGE_SIZE; | 
|---|
| 374 | } | 
|---|
| 375 | iov_iter_advance(i: data, bytes: len); | 
|---|
| 376 | return len; | 
|---|
| 377 | } | 
|---|
| 378 | } | 
|---|
| 379 |  | 
|---|
| 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 || !pages) | 
|---|
| 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, page: *pages++, offset: offs, len: n); | 
|---|
| 402 | offs = 0; | 
|---|
| 403 | to += n; | 
|---|
| 404 | size -= n; | 
|---|
| 405 | } | 
|---|
| 406 | memcpy_from_page(to, page: *pages, offset: offs, len: size); | 
|---|
| 407 | } | 
|---|
| 408 |  | 
|---|
| 409 | /** | 
|---|
| 410 | * p9_virtio_zc_request - issue a zero copy request | 
|---|
| 411 | * @client: client instance issuing the request | 
|---|
| 412 | * @req: request to be issued | 
|---|
| 413 | * @uidata: user buffer that should be used for zero copy read | 
|---|
| 414 | * @uodata: user buffer that should be used for zero copy write | 
|---|
| 415 | * @inlen: read buffer size | 
|---|
| 416 | * @outlen: write buffer size | 
|---|
| 417 | * @in_hdr_len: reader header size, This is the size of response protocol data | 
|---|
| 418 | * | 
|---|
| 419 | */ | 
|---|
| 420 | static int | 
|---|
| 421 | p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req, | 
|---|
| 422 | struct iov_iter *uidata, struct iov_iter *uodata, | 
|---|
| 423 | int inlen, int outlen, int in_hdr_len) | 
|---|
| 424 | { | 
|---|
| 425 | int in, out, err, out_sgs, in_sgs; | 
|---|
| 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; | 
|---|
| 430 | struct scatterlist *sgs[4]; | 
|---|
| 431 | size_t offs = 0; | 
|---|
| 432 | int need_drop = 0; | 
|---|
| 433 | int kicked = 0; | 
|---|
| 434 |  | 
|---|
| 435 | p9_debug(P9_DEBUG_TRANS, "virtio request\n"); | 
|---|
| 436 |  | 
|---|
| 437 | if (uodata) { | 
|---|
| 438 | __le32 sz; | 
|---|
| 439 | int n = p9_get_mapped_pages(chan, pages: &out_pages, data: uodata, | 
|---|
| 440 | count: outlen, offs: &offs, need_drop: &need_drop); | 
|---|
| 441 | if (n < 0) { | 
|---|
| 442 | err = n; | 
|---|
| 443 | goto err_out; | 
|---|
| 444 | } | 
|---|
| 445 | out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE); | 
|---|
| 446 | if (n != outlen) { | 
|---|
| 447 | __le32 v = cpu_to_le32(n); | 
|---|
| 448 | memcpy(to: &req->tc.sdata[req->tc.size - 4], from: &v, len: 4); | 
|---|
| 449 | outlen = n; | 
|---|
| 450 | } | 
|---|
| 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 | */ | 
|---|
| 455 | sz = cpu_to_le32(req->tc.size + outlen); | 
|---|
| 456 | memcpy(to: &req->tc.sdata[0], from: &sz, len: sizeof(sz)); | 
|---|
| 457 | } else if (uidata) { | 
|---|
| 458 | int n = p9_get_mapped_pages(chan, pages: &in_pages, data: uidata, | 
|---|
| 459 | count: inlen, offs: &offs, need_drop: &need_drop); | 
|---|
| 460 | if (n < 0) { | 
|---|
| 461 | err = n; | 
|---|
| 462 | goto err_out; | 
|---|
| 463 | } | 
|---|
| 464 | in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE); | 
|---|
| 465 | if (n != inlen) { | 
|---|
| 466 | __le32 v = cpu_to_le32(n); | 
|---|
| 467 | memcpy(to: &req->tc.sdata[req->tc.size - 4], from: &v, len: 4); | 
|---|
| 468 | inlen = n; | 
|---|
| 469 | } | 
|---|
| 470 | } | 
|---|
| 471 | WRITE_ONCE(req->status, REQ_STATUS_SENT); | 
|---|
| 472 | req_retry_pinned: | 
|---|
| 473 | spin_lock_irqsave(&chan->lock, flags); | 
|---|
| 474 |  | 
|---|
| 475 | out_sgs = in_sgs = 0; | 
|---|
| 476 |  | 
|---|
| 477 | /* out data */ | 
|---|
| 478 | out = pack_sg_list(sg: chan->sg, start: 0, | 
|---|
| 479 | VIRTQUEUE_NUM, data: req->tc.sdata, count: req->tc.size); | 
|---|
| 480 |  | 
|---|
| 481 | if (out) | 
|---|
| 482 | sgs[out_sgs++] = chan->sg; | 
|---|
| 483 |  | 
|---|
| 484 | if (out_pages) { | 
|---|
| 485 | sgs[out_sgs++] = chan->sg + out; | 
|---|
| 486 | out += pack_sg_list_p(sg: chan->sg, start: out, VIRTQUEUE_NUM, | 
|---|
| 487 | pdata: out_pages, nr_pages: out_nr_pages, offs, count: outlen); | 
|---|
| 488 | } | 
|---|
| 489 |  | 
|---|
| 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 | 
|---|
| 495 | * allocated memory and payload onto the user buffer. | 
|---|
| 496 | */ | 
|---|
| 497 | in = pack_sg_list(sg: chan->sg, start: out, | 
|---|
| 498 | VIRTQUEUE_NUM, data: req->rc.sdata, count: in_hdr_len); | 
|---|
| 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; | 
|---|
| 504 | pack_sg_list_p(sg: chan->sg, start: out + in, VIRTQUEUE_NUM, | 
|---|
| 505 | pdata: in_pages, nr_pages: in_nr_pages, offs, count: inlen); | 
|---|
| 506 | } | 
|---|
| 507 |  | 
|---|
| 508 | BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs)); | 
|---|
| 509 | err = virtqueue_add_sgs(vq: chan->vq, sgs, out_sgs, in_sgs, data: req, | 
|---|
| 510 | GFP_ATOMIC); | 
|---|
| 511 | if (err < 0) { | 
|---|
| 512 | if (err == -ENOSPC) { | 
|---|
| 513 | chan->ring_bufs_avail = 0; | 
|---|
| 514 | spin_unlock_irqrestore(lock: &chan->lock, flags); | 
|---|
| 515 | err = wait_event_killable(*chan->vc_wq, | 
|---|
| 516 | chan->ring_bufs_avail); | 
|---|
| 517 | if (err  == -ERESTARTSYS) | 
|---|
| 518 | goto err_out; | 
|---|
| 519 |  | 
|---|
| 520 | p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n"); | 
|---|
| 521 | goto req_retry_pinned; | 
|---|
| 522 | } else { | 
|---|
| 523 | spin_unlock_irqrestore(lock: &chan->lock, flags); | 
|---|
| 524 | p9_debug(P9_DEBUG_TRANS, | 
|---|
| 525 | "virtio rpc add_sgs returned failure\n"); | 
|---|
| 526 | err = -EIO; | 
|---|
| 527 | goto err_out; | 
|---|
| 528 | } | 
|---|
| 529 | } | 
|---|
| 530 | virtqueue_kick(vq: chan->vq); | 
|---|
| 531 | spin_unlock_irqrestore(lock: &chan->lock, flags); | 
|---|
| 532 | kicked = 1; | 
|---|
| 533 | p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); | 
|---|
| 534 | err = wait_event_killable(req->wq, | 
|---|
| 535 | READ_ONCE(req->status) >= REQ_STATUS_RCVD); | 
|---|
| 536 | // RERROR needs reply (== error string) in static data | 
|---|
| 537 | if (READ_ONCE(req->status) == REQ_STATUS_RCVD && | 
|---|
| 538 | unlikely(req->rc.sdata[4] == P9_RERROR)) | 
|---|
| 539 | handle_rerror(req, in_hdr_len, offs, pages: in_pages); | 
|---|
| 540 |  | 
|---|
| 541 | /* | 
|---|
| 542 | * Non kernel buffers are pinned, unpin them | 
|---|
| 543 | */ | 
|---|
| 544 | err_out: | 
|---|
| 545 | if (need_drop) { | 
|---|
| 546 | if (in_pages) { | 
|---|
| 547 | p9_release_pages(pages: in_pages, nr_pages: in_nr_pages); | 
|---|
| 548 | atomic_sub(i: in_nr_pages, v: &vp_pinned); | 
|---|
| 549 | } | 
|---|
| 550 | if (out_pages) { | 
|---|
| 551 | p9_release_pages(pages: out_pages, nr_pages: out_nr_pages); | 
|---|
| 552 | atomic_sub(i: out_nr_pages, v: &vp_pinned); | 
|---|
| 553 | } | 
|---|
| 554 | /* wakeup anybody waiting for slots to pin pages */ | 
|---|
| 555 | wake_up(&vp_wq); | 
|---|
| 556 | } | 
|---|
| 557 | kvfree(addr: in_pages); | 
|---|
| 558 | kvfree(addr: out_pages); | 
|---|
| 559 | if (!kicked) { | 
|---|
| 560 | /* reply won't come */ | 
|---|
| 561 | p9_req_put(c: client, r: req); | 
|---|
| 562 | } | 
|---|
| 563 | return err; | 
|---|
| 564 | } | 
|---|
| 565 |  | 
|---|
| 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; | 
|---|
| 571 | int tag_len; | 
|---|
| 572 |  | 
|---|
| 573 | vdev = dev_to_virtio(dev); | 
|---|
| 574 | chan = vdev->priv; | 
|---|
| 575 | tag_len = strlen(chan->tag); | 
|---|
| 576 |  | 
|---|
| 577 | memcpy(to: buf, from: chan->tag, len: tag_len + 1); | 
|---|
| 578 |  | 
|---|
| 579 | return tag_len + 1; | 
|---|
| 580 | } | 
|---|
| 581 |  | 
|---|
| 582 | static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL); | 
|---|
| 583 |  | 
|---|
| 584 | /** | 
|---|
| 585 | * p9_virtio_probe - probe for existence of 9P virtio channels | 
|---|
| 586 | * @vdev: virtio device to probe | 
|---|
| 587 | * | 
|---|
| 588 | * This probes for existing virtio channels. | 
|---|
| 589 | * | 
|---|
| 590 | */ | 
|---|
| 591 |  | 
|---|
| 592 | static int p9_virtio_probe(struct virtio_device *vdev) | 
|---|
| 593 | { | 
|---|
| 594 | __u16 tag_len; | 
|---|
| 595 | char *tag; | 
|---|
| 596 | int err; | 
|---|
| 597 | struct virtio_chan *chan; | 
|---|
| 598 |  | 
|---|
| 599 | if (!vdev->config->get) { | 
|---|
| 600 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | 
|---|
| 601 | __func__); | 
|---|
| 602 | return -EINVAL; | 
|---|
| 603 | } | 
|---|
| 604 |  | 
|---|
| 605 | chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL); | 
|---|
| 606 | if (!chan) { | 
|---|
| 607 | pr_err( "Failed to allocate virtio 9P channel\n"); | 
|---|
| 608 | err = -ENOMEM; | 
|---|
| 609 | goto fail; | 
|---|
| 610 | } | 
|---|
| 611 |  | 
|---|
| 612 | chan->vdev = vdev; | 
|---|
| 613 |  | 
|---|
| 614 | /* We expect one virtqueue, for requests. */ | 
|---|
| 615 | chan->vq = virtio_find_single_vq(vdev, c: req_done, n: "requests"); | 
|---|
| 616 | if (IS_ERR(ptr: chan->vq)) { | 
|---|
| 617 | err = PTR_ERR(ptr: chan->vq); | 
|---|
| 618 | goto out_free_chan; | 
|---|
| 619 | } | 
|---|
| 620 | chan->vq->vdev->priv = chan; | 
|---|
| 621 | spin_lock_init(&chan->lock); | 
|---|
| 622 |  | 
|---|
| 623 | sg_init_table(chan->sg, VIRTQUEUE_NUM); | 
|---|
| 624 |  | 
|---|
| 625 | chan->inuse = false; | 
|---|
| 626 | if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) { | 
|---|
| 627 | virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len); | 
|---|
| 628 | } else { | 
|---|
| 629 | err = -EINVAL; | 
|---|
| 630 | goto out_free_vq; | 
|---|
| 631 | } | 
|---|
| 632 | tag = kzalloc(tag_len + 1, GFP_KERNEL); | 
|---|
| 633 | if (!tag) { | 
|---|
| 634 | err = -ENOMEM; | 
|---|
| 635 | goto out_free_vq; | 
|---|
| 636 | } | 
|---|
| 637 |  | 
|---|
| 638 | virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag), | 
|---|
| 639 | buf: tag, len: tag_len); | 
|---|
| 640 | chan->tag = tag; | 
|---|
| 641 | err = sysfs_create_file(kobj: &(vdev->dev.kobj), attr: &dev_attr_mount_tag.attr); | 
|---|
| 642 | if (err) { | 
|---|
| 643 | goto out_free_tag; | 
|---|
| 644 | } | 
|---|
| 645 | chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL); | 
|---|
| 646 | if (!chan->vc_wq) { | 
|---|
| 647 | err = -ENOMEM; | 
|---|
| 648 | goto out_remove_file; | 
|---|
| 649 | } | 
|---|
| 650 | init_waitqueue_head(chan->vc_wq); | 
|---|
| 651 | chan->ring_bufs_avail = 1; | 
|---|
| 652 | /* Ceiling limit to avoid denial of service attacks */ | 
|---|
| 653 | chan->p9_max_pages = nr_free_buffer_pages()/4; | 
|---|
| 654 |  | 
|---|
| 655 | virtio_device_ready(dev: vdev); | 
|---|
| 656 |  | 
|---|
| 657 | mutex_lock(lock: &virtio_9p_lock); | 
|---|
| 658 | list_add_tail(new: &chan->chan_list, head: &virtio_chan_list); | 
|---|
| 659 | mutex_unlock(lock: &virtio_9p_lock); | 
|---|
| 660 |  | 
|---|
| 661 | /* Let udev rules use the new mount_tag attribute. */ | 
|---|
| 662 | kobject_uevent(kobj: &(vdev->dev.kobj), action: KOBJ_CHANGE); | 
|---|
| 663 |  | 
|---|
| 664 | return 0; | 
|---|
| 665 |  | 
|---|
| 666 | out_remove_file: | 
|---|
| 667 | sysfs_remove_file(kobj: &vdev->dev.kobj, attr: &dev_attr_mount_tag.attr); | 
|---|
| 668 | out_free_tag: | 
|---|
| 669 | kfree(objp: tag); | 
|---|
| 670 | out_free_vq: | 
|---|
| 671 | vdev->config->del_vqs(vdev); | 
|---|
| 672 | out_free_chan: | 
|---|
| 673 | kfree(objp: chan); | 
|---|
| 674 | fail: | 
|---|
| 675 | return err; | 
|---|
| 676 | } | 
|---|
| 677 |  | 
|---|
| 678 |  | 
|---|
| 679 | /** | 
|---|
| 680 | * p9_virtio_create - allocate a new virtio channel | 
|---|
| 681 | * @client: client instance invoking this transport | 
|---|
| 682 | * @devname: string identifying the channel to connect to (unused) | 
|---|
| 683 | * @args: args passed from sys_mount() for per-transport options (unused) | 
|---|
| 684 | * | 
|---|
| 685 | * This sets up a transport channel for 9p communication.  Right now | 
|---|
| 686 | * we only match the first available channel, but eventually we could look up | 
|---|
| 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 | 
|---|
| 689 | * mount has a channel open at a time. | 
|---|
| 690 | * | 
|---|
| 691 | */ | 
|---|
| 692 |  | 
|---|
| 693 | static int | 
|---|
| 694 | p9_virtio_create(struct p9_client *client, const char *devname, char *args) | 
|---|
| 695 | { | 
|---|
| 696 | struct virtio_chan *chan; | 
|---|
| 697 | int ret = -ENOENT; | 
|---|
| 698 | int found = 0; | 
|---|
| 699 |  | 
|---|
| 700 | if (devname == NULL) | 
|---|
| 701 | return -EINVAL; | 
|---|
| 702 |  | 
|---|
| 703 | mutex_lock(lock: &virtio_9p_lock); | 
|---|
| 704 | list_for_each_entry(chan, &virtio_chan_list, chan_list) { | 
|---|
| 705 | if (!strcmp(devname, chan->tag)) { | 
|---|
| 706 | if (!chan->inuse) { | 
|---|
| 707 | chan->inuse = true; | 
|---|
| 708 | found = 1; | 
|---|
| 709 | break; | 
|---|
| 710 | } | 
|---|
| 711 | ret = -EBUSY; | 
|---|
| 712 | } | 
|---|
| 713 | } | 
|---|
| 714 | mutex_unlock(lock: &virtio_9p_lock); | 
|---|
| 715 |  | 
|---|
| 716 | if (!found) { | 
|---|
| 717 | pr_err( "no channels available for device %s\n", devname); | 
|---|
| 718 | return ret; | 
|---|
| 719 | } | 
|---|
| 720 |  | 
|---|
| 721 | client->trans = (void *)chan; | 
|---|
| 722 | client->status = Connected; | 
|---|
| 723 | chan->client = client; | 
|---|
| 724 |  | 
|---|
| 725 | return 0; | 
|---|
| 726 | } | 
|---|
| 727 |  | 
|---|
| 728 | /** | 
|---|
| 729 | * p9_virtio_remove - clean up resources associated with a virtio device | 
|---|
| 730 | * @vdev: virtio device to remove | 
|---|
| 731 | * | 
|---|
| 732 | */ | 
|---|
| 733 |  | 
|---|
| 734 | static void p9_virtio_remove(struct virtio_device *vdev) | 
|---|
| 735 | { | 
|---|
| 736 | struct virtio_chan *chan = vdev->priv; | 
|---|
| 737 | unsigned long warning_time; | 
|---|
| 738 |  | 
|---|
| 739 | mutex_lock(lock: &virtio_9p_lock); | 
|---|
| 740 |  | 
|---|
| 741 | /* Remove self from list so we don't get new users. */ | 
|---|
| 742 | list_del(entry: &chan->chan_list); | 
|---|
| 743 | warning_time = jiffies; | 
|---|
| 744 |  | 
|---|
| 745 | /* Wait for existing users to close. */ | 
|---|
| 746 | while (chan->inuse) { | 
|---|
| 747 | mutex_unlock(lock: &virtio_9p_lock); | 
|---|
| 748 | msleep(msecs: 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(lock: &virtio_9p_lock); | 
|---|
| 755 | } | 
|---|
| 756 |  | 
|---|
| 757 | mutex_unlock(lock: &virtio_9p_lock); | 
|---|
| 758 |  | 
|---|
| 759 | virtio_reset_device(dev: vdev); | 
|---|
| 760 | vdev->config->del_vqs(vdev); | 
|---|
| 761 |  | 
|---|
| 762 | sysfs_remove_file(kobj: &(vdev->dev.kobj), attr: &dev_attr_mount_tag.attr); | 
|---|
| 763 | kobject_uevent(kobj: &(vdev->dev.kobj), action: KOBJ_CHANGE); | 
|---|
| 764 | kfree(objp: chan->tag); | 
|---|
| 765 | kfree(objp: chan->vc_wq); | 
|---|
| 766 | kfree(objp: chan); | 
|---|
| 767 |  | 
|---|
| 768 | } | 
|---|
| 769 |  | 
|---|
| 770 | static struct virtio_device_id id_table[] = { | 
|---|
| 771 | { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID }, | 
|---|
| 772 | { 0 }, | 
|---|
| 773 | }; | 
|---|
| 774 |  | 
|---|
| 775 | static unsigned int features[] = { | 
|---|
| 776 | VIRTIO_9P_MOUNT_TAG, | 
|---|
| 777 | }; | 
|---|
| 778 |  | 
|---|
| 779 | /* The standard "struct lguest_driver": */ | 
|---|
| 780 | static struct virtio_driver p9_virtio_drv = { | 
|---|
| 781 | .feature_table  = features, | 
|---|
| 782 | .feature_table_size = ARRAY_SIZE(features), | 
|---|
| 783 | .driver.name    = KBUILD_MODNAME, | 
|---|
| 784 | .id_table	= id_table, | 
|---|
| 785 | .probe		= p9_virtio_probe, | 
|---|
| 786 | .remove		= p9_virtio_remove, | 
|---|
| 787 | }; | 
|---|
| 788 |  | 
|---|
| 789 | static struct p9_trans_module p9_virtio_trans = { | 
|---|
| 790 | .name = "virtio", | 
|---|
| 791 | .create = p9_virtio_create, | 
|---|
| 792 | .close = p9_virtio_close, | 
|---|
| 793 | .request = p9_virtio_request, | 
|---|
| 794 | .zc_request = p9_virtio_zc_request, | 
|---|
| 795 | .cancel = p9_virtio_cancel, | 
|---|
| 796 | .cancelled = p9_virtio_cancelled, | 
|---|
| 797 | /* | 
|---|
| 798 | * We leave one entry for input and one entry for response | 
|---|
| 799 | * headers. We also skip one more entry to accommodate, address | 
|---|
| 800 | * that are not at page boundary, that can result in an extra | 
|---|
| 801 | * page in zero copy. | 
|---|
| 802 | */ | 
|---|
| 803 | .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3), | 
|---|
| 804 | .pooled_rbuffers = false, | 
|---|
| 805 | .def = 1, | 
|---|
| 806 | .owner = THIS_MODULE, | 
|---|
| 807 | }; | 
|---|
| 808 |  | 
|---|
| 809 | /* The standard init function */ | 
|---|
| 810 | static int __init p9_virtio_init(void) | 
|---|
| 811 | { | 
|---|
| 812 | int rc; | 
|---|
| 813 |  | 
|---|
| 814 | INIT_LIST_HEAD(list: &virtio_chan_list); | 
|---|
| 815 |  | 
|---|
| 816 | v9fs_register_trans(m: &p9_virtio_trans); | 
|---|
| 817 | rc = register_virtio_driver(&p9_virtio_drv); | 
|---|
| 818 | if (rc) | 
|---|
| 819 | v9fs_unregister_trans(m: &p9_virtio_trans); | 
|---|
| 820 |  | 
|---|
| 821 | return rc; | 
|---|
| 822 | } | 
|---|
| 823 |  | 
|---|
| 824 | static void __exit p9_virtio_cleanup(void) | 
|---|
| 825 | { | 
|---|
| 826 | unregister_virtio_driver(drv: &p9_virtio_drv); | 
|---|
| 827 | v9fs_unregister_trans(m: &p9_virtio_trans); | 
|---|
| 828 | } | 
|---|
| 829 |  | 
|---|
| 830 | module_init(p9_virtio_init); | 
|---|
| 831 | module_exit(p9_virtio_cleanup); | 
|---|
| 832 | MODULE_ALIAS_9P( "virtio"); | 
|---|
| 833 |  | 
|---|
| 834 | MODULE_DEVICE_TABLE(virtio, id_table); | 
|---|
| 835 | MODULE_AUTHOR( "Eric Van Hensbergen <ericvh@gmail.com>"); | 
|---|
| 836 | MODULE_DESCRIPTION( "Virtio 9p Transport"); | 
|---|
| 837 | MODULE_LICENSE( "GPL"); | 
|---|
| 838 |  | 
|---|