drivers/hv/ring_buffer.c

   1 /*
   2  *
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along with
  15  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  16  * Place - Suite 330, Boston, MA 02111-1307 USA.
  17  *
  18  * Authors:
  19  *   Haiyang Zhang <haiyangz@microsoft.com>
  20  *   Hank Janssen  <hjanssen@microsoft.com>
  21  *   K. Y. Srinivasan <kys@microsoft.com>
  22  *
  23  */
  24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  25
  26 #include <linux/kernel.h>
  27 #include <linux/mm.h>
  28 #include <linux/hyperv.h>
  29 #include <linux/uio.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/slab.h>
  32
  33 #include "hyperv_vmbus.h"
  34
  35 /*
  36  * When we write to the ring buffer, check if the host needs to
  37  * be signaled. Here is the details of this protocol:
  38  *
  39  *      1. The host guarantees that while it is draining the
  40  *         ring buffer, it will set the interrupt_mask to
  41  *         indicate it does not need to be interrupted when
  42  *         new data is placed.
  43  *
  44  *      2. The host guarantees that it will completely drain
  45  *         the ring buffer before exiting the read loop. Further,
  46  *         once the ring buffer is empty, it will clear the
  47  *         interrupt_mask and re-check to see if new data has
  48  *         arrived.
  49  *
  50  * KYS: Oct. 30, 2016:
  51  * It looks like Windows hosts have logic to deal with DOS attacks that
  52  * can be triggered if it receives interrupts when it is not expecting
  53  * the interrupt. The host expects interrupts only when the ring
  54  * transitions from empty to non-empty (or full to non full on the guest
  55  * to host ring).
  56  * So, base the signaling decision solely on the ring state until the
  57  * host logic is fixed.
  58  */
  59
  60 static void hv_signal_on_write(u32 old_write, struct vmbus_channel *channel)
  61 {
  62         struct hv_ring_buffer_info *rbi = &channel->outbound;
  63
  64         virt_mb();
  65         if (READ_ONCE(rbi->ring_buffer->interrupt_mask))
  66                 return;
  67
  68         /* check interrupt_mask before read_index */
  69         virt_rmb();
  70         /*
  71          * This is the only case we need to signal when the
  72          * ring transitions from being empty to non-empty.
  73          */
  74         if (old_write == READ_ONCE(rbi->ring_buffer->read_index))
  75                 vmbus_setevent(channel);
  76
  77         return;
  78 }
  79
  80 /* Get the next write location for the specified ring buffer. */
  81 static inline u32
  82 hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
  83 {
  84         u32 next = ring_info->ring_buffer->write_index;
  85
  86         return next;
  87 }
  88
  89 /* Set the next write location for the specified ring buffer. */
  90 static inline void
  91 hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
  92                      u32 next_write_location)
  93 {
  94         ring_info->ring_buffer->write_index = next_write_location;
  95 }
  96
  97 /* Get the next read location for the specified ring buffer. */
  98 static inline u32
  99 hv_get_next_read_location(const struct hv_ring_buffer_info *ring_info)
 100 {
 101         return ring_info->ring_buffer->read_index;
 102 }
 103
 104 /*
 105  * Get the next read location + offset for the specified ring buffer.
 106  * This allows the caller to skip.
 107  */
 108 static inline u32
 109 hv_get_next_readlocation_withoffset(const struct hv_ring_buffer_info *ring_info,
 110                                     u32 offset)
 111 {
 112         u32 next = ring_info->ring_buffer->read_index;
 113
 114         next += offset;
 115         if (next >= ring_info->ring_datasize)
 116                 next -= ring_info->ring_datasize;
 117
 118         return next;
 119 }
 120
 121 /* Set the next read location for the specified ring buffer. */
 122 static inline void
 123 hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
 124                     u32 next_read_location)
 125 {
 126         ring_info->ring_buffer->read_index = next_read_location;
 127         ring_info->priv_read_index = next_read_location;
 128 }
 129
 130 /* Get the size of the ring buffer. */
 131 static inline u32
 132 hv_get_ring_buffersize(const struct hv_ring_buffer_info *ring_info)
 133 {
 134         return ring_info->ring_datasize;
 135 }
 136
 137 /* Get the read and write indices as u64 of the specified ring buffer. */
 138 static inline u64
 139 hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
 140 {
 141         return (u64)ring_info->ring_buffer->write_index << 32;
 142 }
 143
 144 /*
 145  * Helper routine to copy to source from ring buffer.
 146  * Assume there is enough room. Handles wrap-around in src case only!!
 147  */
 148 static u32 hv_copyfrom_ringbuffer(
 149         const struct hv_ring_buffer_info *ring_info,
 150         void                            *dest,
 151         u32                             destlen,
 152         u32                             start_read_offset)
 153 {
 154         void *ring_buffer = hv_get_ring_buffer(ring_info);
 155         u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
 156
 157         memcpy(dest, ring_buffer + start_read_offset, destlen);
 158
 159         start_read_offset += destlen;
 160         if (start_read_offset >= ring_buffer_size)
 161                 start_read_offset -= ring_buffer_size;
 162
 163         return start_read_offset;
 164 }
 165
 166
 167 /*
 168  * Helper routine to copy from source to ring buffer.
 169  * Assume there is enough room. Handles wrap-around in dest case only!!
 170  */
 171 static u32 hv_copyto_ringbuffer(
 172         struct hv_ring_buffer_info      *ring_info,
 173         u32                             start_write_offset,
 174         const void                      *src,
 175         u32                             srclen)
 176 {
 177         void *ring_buffer = hv_get_ring_buffer(ring_info);
 178         u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
 179
 180         memcpy(ring_buffer + start_write_offset, src, srclen);
 181
 182         start_write_offset += srclen;
 183         if (start_write_offset >= ring_buffer_size)
 184                 start_write_offset -= ring_buffer_size;
 185
 186         return start_write_offset;
 187 }
 188
 189 /* Get various debug metrics for the specified ring buffer. */
 190 void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info *ring_info,
 191                                  struct hv_ring_buffer_debug_info *debug_info)
 192 {
 193         u32 bytes_avail_towrite;
 194         u32 bytes_avail_toread;
 195
 196         if (ring_info->ring_buffer) {
 197                 hv_get_ringbuffer_availbytes(ring_info,
 198                                         &bytes_avail_toread,
 199                                         &bytes_avail_towrite);
 200
 201                 debug_info->bytes_avail_toread = bytes_avail_toread;
 202                 debug_info->bytes_avail_towrite = bytes_avail_towrite;
 203                 debug_info->current_read_index =
 204                         ring_info->ring_buffer->read_index;
 205                 debug_info->current_write_index =
 206                         ring_info->ring_buffer->write_index;
 207                 debug_info->current_interrupt_mask =
 208                         ring_info->ring_buffer->interrupt_mask;
 209         }
 210 }
 211
 212 /* Initialize the ring buffer. */
 213 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
 214                        struct page *pages, u32 page_cnt)
 215 {
 216         int i;
 217         struct page **pages_wraparound;
 218
 219         BUILD_BUG_ON((sizeof(struct hv_ring_buffer) != PAGE_SIZE));
 220
 221         memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
 222
 223         /*
 224          * First page holds struct hv_ring_buffer, do wraparound mapping for
 225          * the rest.
 226          */
 227         pages_wraparound = kzalloc(sizeof(struct page *) * (page_cnt * 2 - 1),
 228                                    GFP_KERNEL);
 229         if (!pages_wraparound)
 230                 return -ENOMEM;
 231
 232         pages_wraparound[0] = pages;
 233         for (i = 0; i < 2 * (page_cnt - 1); i++)
 234                 pages_wraparound[i + 1] = &pages[i % (page_cnt - 1) + 1];
 235
 236         ring_info->ring_buffer = (struct hv_ring_buffer *)
 237                 vmap(pages_wraparound, page_cnt * 2 - 1, VM_MAP, PAGE_KERNEL);
 238
 239         kfree(pages_wraparound);
 240
 241
 242         if (!ring_info->ring_buffer)
 243                 return -ENOMEM;
 244
 245         ring_info->ring_buffer->read_index =
 246                 ring_info->ring_buffer->write_index = 0;
 247
 248         /* Set the feature bit for enabling flow control. */
 249         ring_info->ring_buffer->feature_bits.value = 1;
 250
 251         ring_info->ring_size = page_cnt << PAGE_SHIFT;
 252         ring_info->ring_datasize = ring_info->ring_size -
 253                 sizeof(struct hv_ring_buffer);
 254
 255         spin_lock_init(&ring_info->ring_lock);
 256
 257         return 0;
 258 }
 259
 260 /* Cleanup the ring buffer. */
 261 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
 262 {
 263         vunmap(ring_info->ring_buffer);
 264 }
 265
 266 /* Write to the ring buffer. */
 267 int hv_ringbuffer_write(struct vmbus_channel *channel,
 268                         const struct kvec *kv_list, u32 kv_count)
 269 {
 270         int i = 0;
 271         u32 bytes_avail_towrite;
 272         u32 totalbytes_towrite = 0;
 273
 274         u32 next_write_location;
 275         u32 old_write;
 276         u64 prev_indices = 0;
 277         unsigned long flags = 0;
 278         struct hv_ring_buffer_info *outring_info = &channel->outbound;
 279
 280         if (channel->rescind)
 281                 return -ENODEV;
 282
 283         for (i = 0; i < kv_count; i++)
 284                 totalbytes_towrite += kv_list[i].iov_len;
 285
 286         totalbytes_towrite += sizeof(u64);
 287
 288         spin_lock_irqsave(&outring_info->ring_lock, flags);
 289
 290         bytes_avail_towrite = hv_get_bytes_to_write(outring_info);
 291
 292         /*
 293          * If there is only room for the packet, assume it is full.
 294          * Otherwise, the next time around, we think the ring buffer
 295          * is empty since the read index == write index.
 296          */
 297         if (bytes_avail_towrite <= totalbytes_towrite) {
 298                 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 299                 return -EAGAIN;
 300         }
 301
 302         /* Write to the ring buffer */
 303         next_write_location = hv_get_next_write_location(outring_info);
 304
 305         old_write = next_write_location;
 306
 307         for (i = 0; i < kv_count; i++) {
 308                 next_write_location = hv_copyto_ringbuffer(outring_info,
 309                                                      next_write_location,
 310                                                      kv_list[i].iov_base,
 311                                                      kv_list[i].iov_len);
 312         }
 313
 314         /* Set previous packet start */
 315         prev_indices = hv_get_ring_bufferindices(outring_info);
 316
 317         next_write_location = hv_copyto_ringbuffer(outring_info,
 318                                              next_write_location,
 319                                              &prev_indices,
 320                                              sizeof(u64));
 321
 322         /* Issue a full memory barrier before updating the write index */
 323         virt_mb();
 324
 325         /* Now, update the write location */
 326         hv_set_next_write_location(outring_info, next_write_location);
 327
 328
 329         spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 330
 331         hv_signal_on_write(old_write, channel);
 332
 333         if (channel->rescind)
 334                 return -ENODEV;
 335
 336         return 0;
 337 }
 338
 339 int hv_ringbuffer_read(struct vmbus_channel *channel,
 340                        void *buffer, u32 buflen, u32 *buffer_actual_len,
 341                        u64 *requestid, bool raw)
 342 {
 343         u32 bytes_avail_toread;
 344         u32 next_read_location = 0;
 345         u64 prev_indices = 0;
 346         struct vmpacket_descriptor desc;
 347         u32 offset;
 348         u32 packetlen;
 349         int ret = 0;
 350         struct hv_ring_buffer_info *inring_info = &channel->inbound;
 351
 352         if (buflen <= 0)
 353                 return -EINVAL;
 354
 355
 356         *buffer_actual_len = 0;
 357         *requestid = 0;
 358
 359         bytes_avail_toread = hv_get_bytes_to_read(inring_info);
 360         /* Make sure there is something to read */
 361         if (bytes_avail_toread < sizeof(desc)) {
 362                 /*
 363                  * No error is set when there is even no header, drivers are
 364                  * supposed to analyze buffer_actual_len.
 365                  */
 366                 return ret;
 367         }
 368
 369         init_cached_read_index(channel);
 370         next_read_location = hv_get_next_read_location(inring_info);
 371         next_read_location = hv_copyfrom_ringbuffer(inring_info, &desc,
 372                                                     sizeof(desc),
 373                                                     next_read_location);
 374
 375         offset = raw ? 0 : (desc.offset8 << 3);
 376         packetlen = (desc.len8 << 3) - offset;
 377         *buffer_actual_len = packetlen;
 378         *requestid = desc.trans_id;
 379
 380         if (bytes_avail_toread < packetlen + offset)
 381                 return -EAGAIN;
 382
 383         if (packetlen > buflen)
 384                 return -ENOBUFS;
 385
 386         next_read_location =
 387                 hv_get_next_readlocation_withoffset(inring_info, offset);
 388
 389         next_read_location = hv_copyfrom_ringbuffer(inring_info,
 390                                                 buffer,
 391                                                 packetlen,
 392                                                 next_read_location);
 393
 394         next_read_location = hv_copyfrom_ringbuffer(inring_info,
 395                                                 &prev_indices,
 396                                                 sizeof(u64),
 397                                                 next_read_location);
 398
 399         /*
 400          * Make sure all reads are done before we update the read index since
 401          * the writer may start writing to the read area once the read index
 402          * is updated.
 403          */
 404         virt_mb();
 405
 406         /* Update the read index */
 407         hv_set_next_read_location(inring_info, next_read_location);
 408
 409         hv_signal_on_read(channel);
 410
 411         return ret;
 412 }