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 } 413