net/9p/trans_virtio.c

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * The Virtio 9p transport driver
0004  *
0005  * This is a block based transport driver based on the lguest block driver
0006  * code.
0007  *
0008  *  Copyright (C) 2007, 2008 Eric Van Hensbergen, IBM Corporation
0009  *
0010  *  Based on virtio console driver
0011  *  Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation
0012  */
0013
0014 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
0015
0016 #include <linux/in.h>
0017 #include <linux/module.h>
0018 #include <linux/net.h>
0019 #include <linux/ipv6.h>
0020 #include <linux/errno.h>
0021 #include <linux/kernel.h>
0022 #include <linux/un.h>
0023 #include <linux/uaccess.h>
0024 #include <linux/inet.h>
0025 #include <linux/idr.h>
0026 #include <linux/file.h>
0027 #include <linux/highmem.h>
0028 #include <linux/slab.h>
0029 #include <net/9p/9p.h>
0030 #include <linux/parser.h>
0031 #include <net/9p/client.h>
0032 #include <net/9p/transport.h>
0033 #include <linux/scatterlist.h>
0034 #include <linux/swap.h>
0035 #include <linux/virtio.h>
0036 #include <linux/virtio_9p.h>
0037 #include "trans_common.h"
0038
0039 #define VIRTQUEUE_NUM   128
0040
0041 /* a single mutex to manage channel initialization and attachment */
0042 static DEFINE_MUTEX(virtio_9p_lock);
0043 static DECLARE_WAIT_QUEUE_HEAD(vp_wq);
0044 static atomic_t vp_pinned = ATOMIC_INIT(0);
0045
0046 /**
0047  * struct virtio_chan - per-instance transport information
0048  * @inuse: whether the channel is in use
0049  * @lock: protects multiple elements within this structure
0050  * @client: client instance
0051  * @vdev: virtio dev associated with this channel
0052  * @vq: virtio queue associated with this channel
0053  * @ring_bufs_avail: flag to indicate there is some available in the ring buf
0054  * @vc_wq: wait queue for waiting for thing to be added to ring buf
0055  * @p9_max_pages: maximum number of pinned pages
0056  * @sg: scatter gather list which is used to pack a request (protected?)
0057  * @chan_list: linked list of channels
0058  *
0059  * We keep all per-channel information in a structure.
0060  * This structure is allocated within the devices dev->mem space.
0061  * A pointer to the structure will get put in the transport private.
0062  *
0063  */
0064
0065 struct virtio_chan {
0066     bool inuse;
0067
0068     spinlock_t lock;
0069
0070     struct p9_client *client;
0071     struct virtio_device *vdev;
0072     struct virtqueue *vq;
0073     int ring_bufs_avail;
0074     wait_queue_head_t *vc_wq;
0075     /* This is global limit. Since we don't have a global structure,
0076      * will be placing it in each channel.
0077      */
0078     unsigned long p9_max_pages;
0079     /* Scatterlist: can be too big for stack. */
0080     struct scatterlist sg[VIRTQUEUE_NUM];
0081     /**
0082      * @tag: name to identify a mount null terminated
0083      */
0084     char *tag;
0085
0086     struct list_head chan_list;
0087 };
0088
0089 static struct list_head virtio_chan_list;
0090
0091 /* How many bytes left in this page. */
0092 static unsigned int rest_of_page(void *data)
0093 {
0094     return PAGE_SIZE - offset_in_page(data);
0095 }
0096
0097 /**
0098  * p9_virtio_close - reclaim resources of a channel
0099  * @client: client instance
0100  *
0101  * This reclaims a channel by freeing its resources and
0102  * resetting its inuse flag.
0103  *
0104  */
0105
0106 static void p9_virtio_close(struct p9_client *client)
0107 {
0108     struct virtio_chan *chan = client->trans;
0109
0110     mutex_lock(&virtio_9p_lock);
0111     if (chan)
0112         chan->inuse = false;
0113     mutex_unlock(&virtio_9p_lock);
0114 }
0115
0116 /**
0117  * req_done - callback which signals activity from the server
0118  * @vq: virtio queue activity was received on
0119  *
0120  * This notifies us that the server has triggered some activity
0121  * on the virtio channel - most likely a response to request we
0122  * sent.  Figure out which requests now have responses and wake up
0123  * those threads.
0124  *
0125  * Bugs: could do with some additional sanity checking, but appears to work.
0126  *
0127  */
0128
0129 static void req_done(struct virtqueue *vq)
0130 {
0131     struct virtio_chan *chan = vq->vdev->priv;
0132     unsigned int len;
0133     struct p9_req_t *req;
0134     bool need_wakeup = false;
0135     unsigned long flags;
0136
0137     p9_debug(P9_DEBUG_TRANS, ": request done\n");
0138
0139     spin_lock_irqsave(&chan->lock, flags);
0140     while ((req = virtqueue_get_buf(chan->vq, &len)) != NULL) {
0141         if (!chan->ring_bufs_avail) {
0142             chan->ring_bufs_avail = 1;
0143             need_wakeup = true;
0144         }
0145
0146         if (len) {
0147             req->rc.size = len;
0148             p9_client_cb(chan->client, req, REQ_STATUS_RCVD);
0149         }
0150     }
0151     spin_unlock_irqrestore(&chan->lock, flags);
0152     /* Wakeup if anyone waiting for VirtIO ring space. */
0153     if (need_wakeup)
0154         wake_up(chan->vc_wq);
0155 }
0156
0157 /**
0158  * pack_sg_list - pack a scatter gather list from a linear buffer
0159  * @sg: scatter/gather list to pack into
0160  * @start: which segment of the sg_list to start at
0161  * @limit: maximum segment to pack data to
0162  * @data: data to pack into scatter/gather list
0163  * @count: amount of data to pack into the scatter/gather list
0164  *
0165  * sg_lists have multiple segments of various sizes.  This will pack
0166  * arbitrary data into an existing scatter gather list, segmenting the
0167  * data as necessary within constraints.
0168  *
0169  */
0170
0171 static int pack_sg_list(struct scatterlist *sg, int start,
0172             int limit, char *data, int count)
0173 {
0174     int s;
0175     int index = start;
0176
0177     while (count) {
0178         s = rest_of_page(data);
0179         if (s > count)
0180             s = count;
0181         BUG_ON(index >= limit);
0182         /* Make sure we don't terminate early. */
0183         sg_unmark_end(&sg[index]);
0184         sg_set_buf(&sg[index++], data, s);
0185         count -= s;
0186         data += s;
0187     }
0188     if (index-start)
0189         sg_mark_end(&sg[index - 1]);
0190     return index-start;
0191 }
0192
0193 /* We don't currently allow canceling of virtio requests */
0194 static int p9_virtio_cancel(struct p9_client *client, struct p9_req_t *req)
0195 {
0196     return 1;
0197 }
0198
0199 /* Reply won't come, so drop req ref */
0200 static int p9_virtio_cancelled(struct p9_client *client, struct p9_req_t *req)
0201 {
0202     p9_req_put(client, req);
0203     return 0;
0204 }
0205
0206 /**
0207  * pack_sg_list_p - Just like pack_sg_list. Instead of taking a buffer,
0208  * this takes a list of pages.
0209  * @sg: scatter/gather list to pack into
0210  * @start: which segment of the sg_list to start at
0211  * @limit: maximum number of pages in sg list.
0212  * @pdata: a list of pages to add into sg.
0213  * @nr_pages: number of pages to pack into the scatter/gather list
0214  * @offs: amount of data in the beginning of first page _not_ to pack
0215  * @count: amount of data to pack into the scatter/gather list
0216  */
0217 static int
0218 pack_sg_list_p(struct scatterlist *sg, int start, int limit,
0219            struct page **pdata, int nr_pages, size_t offs, int count)
0220 {
0221     int i = 0, s;
0222     int data_off = offs;
0223     int index = start;
0224
0225     BUG_ON(nr_pages > (limit - start));
0226     /*
0227      * if the first page doesn't start at
0228      * page boundary find the offset
0229      */
0230     while (nr_pages) {
0231         s = PAGE_SIZE - data_off;
0232         if (s > count)
0233             s = count;
0234         BUG_ON(index >= limit);
0235         /* Make sure we don't terminate early. */
0236         sg_unmark_end(&sg[index]);
0237         sg_set_page(&sg[index++], pdata[i++], s, data_off);
0238         data_off = 0;
0239         count -= s;
0240         nr_pages--;
0241     }
0242
0243     if (index-start)
0244         sg_mark_end(&sg[index - 1]);
0245     return index - start;
0246 }
0247
0248 /**
0249  * p9_virtio_request - issue a request
0250  * @client: client instance issuing the request
0251  * @req: request to be issued
0252  *
0253  */
0254
0255 static int
0256 p9_virtio_request(struct p9_client *client, struct p9_req_t *req)
0257 {
0258     int err;
0259     int in, out, out_sgs, in_sgs;
0260     unsigned long flags;
0261     struct virtio_chan *chan = client->trans;
0262     struct scatterlist *sgs[2];
0263
0264     p9_debug(P9_DEBUG_TRANS, "9p debug: virtio request\n");
0265
0266     req->status = REQ_STATUS_SENT;
0267 req_retry:
0268     spin_lock_irqsave(&chan->lock, flags);
0269
0270     out_sgs = in_sgs = 0;
0271     /* Handle out VirtIO ring buffers */
0272     out = pack_sg_list(chan->sg, 0,
0273                VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
0274     if (out)
0275         sgs[out_sgs++] = chan->sg;
0276
0277     in = pack_sg_list(chan->sg, out,
0278               VIRTQUEUE_NUM, req->rc.sdata, req->rc.capacity);
0279     if (in)
0280         sgs[out_sgs + in_sgs++] = chan->sg + out;
0281
0282     err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
0283                 GFP_ATOMIC);
0284     if (err < 0) {
0285         if (err == -ENOSPC) {
0286             chan->ring_bufs_avail = 0;
0287             spin_unlock_irqrestore(&chan->lock, flags);
0288             err = wait_event_killable(*chan->vc_wq,
0289                           chan->ring_bufs_avail);
0290             if (err  == -ERESTARTSYS)
0291                 return err;
0292
0293             p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
0294             goto req_retry;
0295         } else {
0296             spin_unlock_irqrestore(&chan->lock, flags);
0297             p9_debug(P9_DEBUG_TRANS,
0298                  "virtio rpc add_sgs returned failure\n");
0299             return -EIO;
0300         }
0301     }
0302     virtqueue_kick(chan->vq);
0303     spin_unlock_irqrestore(&chan->lock, flags);
0304
0305     p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
0306     return 0;
0307 }
0308
0309 static int p9_get_mapped_pages(struct virtio_chan *chan,
0310                    struct page ***pages,
0311                    struct iov_iter *data,
0312                    int count,
0313                    size_t *offs,
0314                    int *need_drop)
0315 {
0316     int nr_pages;
0317     int err;
0318
0319     if (!iov_iter_count(data))
0320         return 0;
0321
0322     if (!iov_iter_is_kvec(data)) {
0323         int n;
0324         /*
0325          * We allow only p9_max_pages pinned. We wait for the
0326          * Other zc request to finish here
0327          */
0328         if (atomic_read(&vp_pinned) >= chan->p9_max_pages) {
0329             err = wait_event_killable(vp_wq,
0330                   (atomic_read(&vp_pinned) < chan->p9_max_pages));
0331             if (err == -ERESTARTSYS)
0332                 return err;
0333         }
0334         n = iov_iter_get_pages_alloc2(data, pages, count, offs);
0335         if (n < 0)
0336             return n;
0337         *need_drop = 1;
0338         nr_pages = DIV_ROUND_UP(n + *offs, PAGE_SIZE);
0339         atomic_add(nr_pages, &vp_pinned);
0340         return n;
0341     } else {
0342         /* kernel buffer, no need to pin pages */
0343         int index;
0344         size_t len;
0345         void *p;
0346
0347         /* we'd already checked that it's non-empty */
0348         while (1) {
0349             len = iov_iter_single_seg_count(data);
0350             if (likely(len)) {
0351                 p = data->kvec->iov_base + data->iov_offset;
0352                 break;
0353             }
0354             iov_iter_advance(data, 0);
0355         }
0356         if (len > count)
0357             len = count;
0358
0359         nr_pages = DIV_ROUND_UP((unsigned long)p + len, PAGE_SIZE) -
0360                (unsigned long)p / PAGE_SIZE;
0361
0362         *pages = kmalloc_array(nr_pages, sizeof(struct page *),
0363                        GFP_NOFS);
0364         if (!*pages)
0365             return -ENOMEM;
0366
0367         *need_drop = 0;
0368         p -= (*offs = offset_in_page(p));
0369         for (index = 0; index < nr_pages; index++) {
0370             if (is_vmalloc_addr(p))
0371                 (*pages)[index] = vmalloc_to_page(p);
0372             else
0373                 (*pages)[index] = kmap_to_page(p);
0374             p += PAGE_SIZE;
0375         }
0376         iov_iter_advance(data, len);
0377         return len;
0378     }
0379 }
0380
0381 static void handle_rerror(struct p9_req_t *req, int in_hdr_len,
0382               size_t offs, struct page **pages)
0383 {
0384     unsigned size, n;
0385     void *to = req->rc.sdata + in_hdr_len;
0386
0387     // Fits entirely into the static data?  Nothing to do.
0388     if (req->rc.size < in_hdr_len)
0389         return;
0390
0391     // Really long error message?  Tough, truncate the reply.  Might get
0392     // rejected (we can't be arsed to adjust the size encoded in header,
0393     // or string size for that matter), but it wouldn't be anything valid
0394     // anyway.
0395     if (unlikely(req->rc.size > P9_ZC_HDR_SZ))
0396         req->rc.size = P9_ZC_HDR_SZ;
0397
0398     // data won't span more than two pages
0399     size = req->rc.size - in_hdr_len;
0400     n = PAGE_SIZE - offs;
0401     if (size > n) {
0402         memcpy_from_page(to, *pages++, offs, n);
0403         offs = 0;
0404         to += n;
0405         size -= n;
0406     }
0407     memcpy_from_page(to, *pages, offs, size);
0408 }
0409
0410 /**
0411  * p9_virtio_zc_request - issue a zero copy request
0412  * @client: client instance issuing the request
0413  * @req: request to be issued
0414  * @uidata: user buffer that should be used for zero copy read
0415  * @uodata: user buffer that should be used for zero copy write
0416  * @inlen: read buffer size
0417  * @outlen: write buffer size
0418  * @in_hdr_len: reader header size, This is the size of response protocol data
0419  *
0420  */
0421 static int
0422 p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req,
0423              struct iov_iter *uidata, struct iov_iter *uodata,
0424              int inlen, int outlen, int in_hdr_len)
0425 {
0426     int in, out, err, out_sgs, in_sgs;
0427     unsigned long flags;
0428     int in_nr_pages = 0, out_nr_pages = 0;
0429     struct page **in_pages = NULL, **out_pages = NULL;
0430     struct virtio_chan *chan = client->trans;
0431     struct scatterlist *sgs[4];
0432     size_t offs;
0433     int need_drop = 0;
0434     int kicked = 0;
0435
0436     p9_debug(P9_DEBUG_TRANS, "virtio request\n");
0437
0438     if (uodata) {
0439         __le32 sz;
0440         int n = p9_get_mapped_pages(chan, &out_pages, uodata,
0441                         outlen, &offs, &need_drop);
0442         if (n < 0) {
0443             err = n;
0444             goto err_out;
0445         }
0446         out_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
0447         if (n != outlen) {
0448             __le32 v = cpu_to_le32(n);
0449             memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
0450             outlen = n;
0451         }
0452         /* The size field of the message must include the length of the
0453          * header and the length of the data.  We didn't actually know
0454          * the length of the data until this point so add it in now.
0455          */
0456         sz = cpu_to_le32(req->tc.size + outlen);
0457         memcpy(&req->tc.sdata[0], &sz, sizeof(sz));
0458     } else if (uidata) {
0459         int n = p9_get_mapped_pages(chan, &in_pages, uidata,
0460                         inlen, &offs, &need_drop);
0461         if (n < 0) {
0462             err = n;
0463             goto err_out;
0464         }
0465         in_nr_pages = DIV_ROUND_UP(n + offs, PAGE_SIZE);
0466         if (n != inlen) {
0467             __le32 v = cpu_to_le32(n);
0468             memcpy(&req->tc.sdata[req->tc.size - 4], &v, 4);
0469             inlen = n;
0470         }
0471     }
0472     req->status = REQ_STATUS_SENT;
0473 req_retry_pinned:
0474     spin_lock_irqsave(&chan->lock, flags);
0475
0476     out_sgs = in_sgs = 0;
0477
0478     /* out data */
0479     out = pack_sg_list(chan->sg, 0,
0480                VIRTQUEUE_NUM, req->tc.sdata, req->tc.size);
0481
0482     if (out)
0483         sgs[out_sgs++] = chan->sg;
0484
0485     if (out_pages) {
0486         sgs[out_sgs++] = chan->sg + out;
0487         out += pack_sg_list_p(chan->sg, out, VIRTQUEUE_NUM,
0488                       out_pages, out_nr_pages, offs, outlen);
0489     }
0490
0491     /*
0492      * Take care of in data
0493      * For example TREAD have 11.
0494      * 11 is the read/write header = PDU Header(7) + IO Size (4).
0495      * Arrange in such a way that server places header in the
0496      * allocated memory and payload onto the user buffer.
0497      */
0498     in = pack_sg_list(chan->sg, out,
0499               VIRTQUEUE_NUM, req->rc.sdata, in_hdr_len);
0500     if (in)
0501         sgs[out_sgs + in_sgs++] = chan->sg + out;
0502
0503     if (in_pages) {
0504         sgs[out_sgs + in_sgs++] = chan->sg + out + in;
0505         in += pack_sg_list_p(chan->sg, out + in, VIRTQUEUE_NUM,
0506                      in_pages, in_nr_pages, offs, inlen);
0507     }
0508
0509     BUG_ON(out_sgs + in_sgs > ARRAY_SIZE(sgs));
0510     err = virtqueue_add_sgs(chan->vq, sgs, out_sgs, in_sgs, req,
0511                 GFP_ATOMIC);
0512     if (err < 0) {
0513         if (err == -ENOSPC) {
0514             chan->ring_bufs_avail = 0;
0515             spin_unlock_irqrestore(&chan->lock, flags);
0516             err = wait_event_killable(*chan->vc_wq,
0517                           chan->ring_bufs_avail);
0518             if (err  == -ERESTARTSYS)
0519                 goto err_out;
0520
0521             p9_debug(P9_DEBUG_TRANS, "Retry virtio request\n");
0522             goto req_retry_pinned;
0523         } else {
0524             spin_unlock_irqrestore(&chan->lock, flags);
0525             p9_debug(P9_DEBUG_TRANS,
0526                  "virtio rpc add_sgs returned failure\n");
0527             err = -EIO;
0528             goto err_out;
0529         }
0530     }
0531     virtqueue_kick(chan->vq);
0532     spin_unlock_irqrestore(&chan->lock, flags);
0533     kicked = 1;
0534     p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n");
0535     err = wait_event_killable(req->wq, req->status >= REQ_STATUS_RCVD);
0536     // RERROR needs reply (== error string) in static data
0537     if (req->status == REQ_STATUS_RCVD &&
0538         unlikely(req->rc.sdata[4] == P9_RERROR))
0539         handle_rerror(req, in_hdr_len, offs, in_pages);
0540
0541     /*
0542      * Non kernel buffers are pinned, unpin them
0543      */
0544 err_out:
0545     if (need_drop) {
0546         if (in_pages) {
0547             p9_release_pages(in_pages, in_nr_pages);
0548             atomic_sub(in_nr_pages, &vp_pinned);
0549         }
0550         if (out_pages) {
0551             p9_release_pages(out_pages, out_nr_pages);
0552             atomic_sub(out_nr_pages, &vp_pinned);
0553         }
0554         /* wakeup anybody waiting for slots to pin pages */
0555         wake_up(&vp_wq);
0556     }
0557     kvfree(in_pages);
0558     kvfree(out_pages);
0559     if (!kicked) {
0560         /* reply won't come */
0561         p9_req_put(client, req);
0562     }
0563     return err;
0564 }
0565
0566 static ssize_t p9_mount_tag_show(struct device *dev,
0567                 struct device_attribute *attr, char *buf)
0568 {
0569     struct virtio_chan *chan;
0570     struct virtio_device *vdev;
0571     int tag_len;
0572
0573     vdev = dev_to_virtio(dev);
0574     chan = vdev->priv;
0575     tag_len = strlen(chan->tag);
0576
0577     memcpy(buf, chan->tag, tag_len + 1);
0578
0579     return tag_len + 1;
0580 }
0581
0582 static DEVICE_ATTR(mount_tag, 0444, p9_mount_tag_show, NULL);
0583
0584 /**
0585  * p9_virtio_probe - probe for existence of 9P virtio channels
0586  * @vdev: virtio device to probe
0587  *
0588  * This probes for existing virtio channels.
0589  *
0590  */
0591
0592 static int p9_virtio_probe(struct virtio_device *vdev)
0593 {
0594     __u16 tag_len;
0595     char *tag;
0596     int err;
0597     struct virtio_chan *chan;
0598
0599     if (!vdev->config->get) {
0600         dev_err(&vdev->dev, "%s failure: config access disabled\n",
0601             __func__);
0602         return -EINVAL;
0603     }
0604
0605     chan = kmalloc(sizeof(struct virtio_chan), GFP_KERNEL);
0606     if (!chan) {
0607         pr_err("Failed to allocate virtio 9P channel\n");
0608         err = -ENOMEM;
0609         goto fail;
0610     }
0611
0612     chan->vdev = vdev;
0613
0614     /* We expect one virtqueue, for requests. */
0615     chan->vq = virtio_find_single_vq(vdev, req_done, "requests");
0616     if (IS_ERR(chan->vq)) {
0617         err = PTR_ERR(chan->vq);
0618         goto out_free_chan;
0619     }
0620     chan->vq->vdev->priv = chan;
0621     spin_lock_init(&chan->lock);
0622
0623     sg_init_table(chan->sg, VIRTQUEUE_NUM);
0624
0625     chan->inuse = false;
0626     if (virtio_has_feature(vdev, VIRTIO_9P_MOUNT_TAG)) {
0627         virtio_cread(vdev, struct virtio_9p_config, tag_len, &tag_len);
0628     } else {
0629         err = -EINVAL;
0630         goto out_free_vq;
0631     }
0632     tag = kzalloc(tag_len + 1, GFP_KERNEL);
0633     if (!tag) {
0634         err = -ENOMEM;
0635         goto out_free_vq;
0636     }
0637
0638     virtio_cread_bytes(vdev, offsetof(struct virtio_9p_config, tag),
0639                tag, tag_len);
0640     chan->tag = tag;
0641     err = sysfs_create_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
0642     if (err) {
0643         goto out_free_tag;
0644     }
0645     chan->vc_wq = kmalloc(sizeof(wait_queue_head_t), GFP_KERNEL);
0646     if (!chan->vc_wq) {
0647         err = -ENOMEM;
0648         goto out_remove_file;
0649     }
0650     init_waitqueue_head(chan->vc_wq);
0651     chan->ring_bufs_avail = 1;
0652     /* Ceiling limit to avoid denial of service attacks */
0653     chan->p9_max_pages = nr_free_buffer_pages()/4;
0654
0655     virtio_device_ready(vdev);
0656
0657     mutex_lock(&virtio_9p_lock);
0658     list_add_tail(&chan->chan_list, &virtio_chan_list);
0659     mutex_unlock(&virtio_9p_lock);
0660
0661     /* Let udev rules use the new mount_tag attribute. */
0662     kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
0663
0664     return 0;
0665
0666 out_remove_file:
0667     sysfs_remove_file(&vdev->dev.kobj, &dev_attr_mount_tag.attr);
0668 out_free_tag:
0669     kfree(tag);
0670 out_free_vq:
0671     vdev->config->del_vqs(vdev);
0672 out_free_chan:
0673     kfree(chan);
0674 fail:
0675     return err;
0676 }
0677
0678
0679 /**
0680  * p9_virtio_create - allocate a new virtio channel
0681  * @client: client instance invoking this transport
0682  * @devname: string identifying the channel to connect to (unused)
0683  * @args: args passed from sys_mount() for per-transport options (unused)
0684  *
0685  * This sets up a transport channel for 9p communication.  Right now
0686  * we only match the first available channel, but eventually we could look up
0687  * alternate channels by matching devname versus a virtio_config entry.
0688  * We use a simple reference count mechanism to ensure that only a single
0689  * mount has a channel open at a time.
0690  *
0691  */
0692
0693 static int
0694 p9_virtio_create(struct p9_client *client, const char *devname, char *args)
0695 {
0696     struct virtio_chan *chan;
0697     int ret = -ENOENT;
0698     int found = 0;
0699
0700     if (devname == NULL)
0701         return -EINVAL;
0702
0703     mutex_lock(&virtio_9p_lock);
0704     list_for_each_entry(chan, &virtio_chan_list, chan_list) {
0705         if (!strcmp(devname, chan->tag)) {
0706             if (!chan->inuse) {
0707                 chan->inuse = true;
0708                 found = 1;
0709                 break;
0710             }
0711             ret = -EBUSY;
0712         }
0713     }
0714     mutex_unlock(&virtio_9p_lock);
0715
0716     if (!found) {
0717         pr_err("no channels available for device %s\n", devname);
0718         return ret;
0719     }
0720
0721     client->trans = (void *)chan;
0722     client->status = Connected;
0723     chan->client = client;
0724
0725     return 0;
0726 }
0727
0728 /**
0729  * p9_virtio_remove - clean up resources associated with a virtio device
0730  * @vdev: virtio device to remove
0731  *
0732  */
0733
0734 static void p9_virtio_remove(struct virtio_device *vdev)
0735 {
0736     struct virtio_chan *chan = vdev->priv;
0737     unsigned long warning_time;
0738
0739     mutex_lock(&virtio_9p_lock);
0740
0741     /* Remove self from list so we don't get new users. */
0742     list_del(&chan->chan_list);
0743     warning_time = jiffies;
0744
0745     /* Wait for existing users to close. */
0746     while (chan->inuse) {
0747         mutex_unlock(&virtio_9p_lock);
0748         msleep(250);
0749         if (time_after(jiffies, warning_time + 10 * HZ)) {
0750             dev_emerg(&vdev->dev,
0751                   "p9_virtio_remove: waiting for device in use.\n");
0752             warning_time = jiffies;
0753         }
0754         mutex_lock(&virtio_9p_lock);
0755     }
0756
0757     mutex_unlock(&virtio_9p_lock);
0758
0759     virtio_reset_device(vdev);
0760     vdev->config->del_vqs(vdev);
0761
0762     sysfs_remove_file(&(vdev->dev.kobj), &dev_attr_mount_tag.attr);
0763     kobject_uevent(&(vdev->dev.kobj), KOBJ_CHANGE);
0764     kfree(chan->tag);
0765     kfree(chan->vc_wq);
0766     kfree(chan);
0767
0768 }
0769
0770 static struct virtio_device_id id_table[] = {
0771     { VIRTIO_ID_9P, VIRTIO_DEV_ANY_ID },
0772     { 0 },
0773 };
0774
0775 static unsigned int features[] = {
0776     VIRTIO_9P_MOUNT_TAG,
0777 };
0778
0779 /* The standard "struct lguest_driver": */
0780 static struct virtio_driver p9_virtio_drv = {
0781     .feature_table  = features,
0782     .feature_table_size = ARRAY_SIZE(features),
0783     .driver.name    = KBUILD_MODNAME,
0784     .driver.owner   = THIS_MODULE,
0785     .id_table   = id_table,
0786     .probe      = p9_virtio_probe,
0787     .remove     = p9_virtio_remove,
0788 };
0789
0790 static struct p9_trans_module p9_virtio_trans = {
0791     .name = "virtio",
0792     .create = p9_virtio_create,
0793     .close = p9_virtio_close,
0794     .request = p9_virtio_request,
0795     .zc_request = p9_virtio_zc_request,
0796     .cancel = p9_virtio_cancel,
0797     .cancelled = p9_virtio_cancelled,
0798     /*
0799      * We leave one entry for input and one entry for response
0800      * headers. We also skip one more entry to accommodate, address
0801      * that are not at page boundary, that can result in an extra
0802      * page in zero copy.
0803      */
0804     .maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
0805     .def = 1,
0806     .owner = THIS_MODULE,
0807 };
0808
0809 /* The standard init function */
0810 static int __init p9_virtio_init(void)
0811 {
0812     int rc;
0813
0814     INIT_LIST_HEAD(&virtio_chan_list);
0815
0816     v9fs_register_trans(&p9_virtio_trans);
0817     rc = register_virtio_driver(&p9_virtio_drv);
0818     if (rc)
0819         v9fs_unregister_trans(&p9_virtio_trans);
0820
0821     return rc;
0822 }
0823
0824 static void __exit p9_virtio_cleanup(void)
0825 {
0826     unregister_virtio_driver(&p9_virtio_drv);
0827     v9fs_unregister_trans(&p9_virtio_trans);
0828 }
0829
0830 module_init(p9_virtio_init);
0831 module_exit(p9_virtio_cleanup);
0832 MODULE_ALIAS_9P("virtio");
0833
0834 MODULE_DEVICE_TABLE(virtio, id_table);
0835 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
0836 MODULE_DESCRIPTION("Virtio 9p Transport");
0837 MODULE_LICENSE("GPL");