1 /* 2 * This program is free software; you can redistribute it and/or 3 * modify it under the terms of the GNU General Public License version 2 4 * as published by the Free Software Foundation; or, when distributed 5 * separately from the Linux kernel or incorporated into other 6 * software packages, subject to the following license: 7 * 8 * Permission is hereby granted, free of charge, to any person obtaining a copy 9 * of this source file (the "Software"), to deal in the Software without 10 * restriction, including without limitation the rights to use, copy, modify, 11 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 12 * and to permit persons to whom the Software is furnished to do so, subject to 13 * the following conditions: 14 * 15 * The above copyright notice and this permission notice shall be included in 16 * all copies or substantial portions of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 21 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 23 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 24 * IN THE SOFTWARE. 25 */ 26 27 #ifndef __XEN_BLKIF__BACKEND__COMMON_H__ 28 #define __XEN_BLKIF__BACKEND__COMMON_H__ 29 30 #include <linux/module.h> 31 #include <linux/interrupt.h> 32 #include <linux/slab.h> 33 #include <linux/blkdev.h> 34 #include <linux/vmalloc.h> 35 #include <linux/wait.h> 36 #include <linux/io.h> 37 #include <linux/rbtree.h> 38 #include <asm/setup.h> 39 #include <asm/pgalloc.h> 40 #include <asm/hypervisor.h> 41 #include <xen/grant_table.h> 42 #include <xen/xenbus.h> 43 #include <xen/interface/io/ring.h> 44 #include <xen/interface/io/blkif.h> 45 #include <xen/interface/io/protocols.h> 46 47 #define DRV_PFX "xen-blkback:" 48 #define DPRINTK(fmt, args...) \ 49 pr_debug(DRV_PFX "(%s:%d) " fmt ".\n", \ 50 __func__, __LINE__, ##args) 51 52 53 /* 54 * This is the maximum number of segments that would be allowed in indirect 55 * requests. This value will also be passed to the frontend. 56 */ 57 #define MAX_INDIRECT_SEGMENTS 256 58 59 #define SEGS_PER_INDIRECT_FRAME \ 60 (PAGE_SIZE/sizeof(struct blkif_request_segment_aligned)) 61 #define MAX_INDIRECT_PAGES \ 62 ((MAX_INDIRECT_SEGMENTS + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME) 63 #define INDIRECT_PAGES(_segs) \ 64 ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME) 65 66 /* Not a real protocol. Used to generate ring structs which contain 67 * the elements common to all protocols only. This way we get a 68 * compiler-checkable way to use common struct elements, so we can 69 * avoid using switch(protocol) in a number of places. */ 70 struct blkif_common_request { 71 char dummy; 72 }; 73 struct blkif_common_response { 74 char dummy; 75 }; 76 77 struct blkif_x86_32_request_rw { 78 uint8_t nr_segments; /* number of segments */ 79 blkif_vdev_t handle; /* only for read/write requests */ 80 uint64_t id; /* private guest value, echoed in resp */ 81 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 82 struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 83 } __attribute__((__packed__)); 84 85 struct blkif_x86_32_request_discard { 86 uint8_t flag; /* BLKIF_DISCARD_SECURE or zero */ 87 blkif_vdev_t _pad1; /* was "handle" for read/write requests */ 88 uint64_t id; /* private guest value, echoed in resp */ 89 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 90 uint64_t nr_sectors; 91 } __attribute__((__packed__)); 92 93 struct blkif_x86_32_request_other { 94 uint8_t _pad1; 95 blkif_vdev_t _pad2; 96 uint64_t id; /* private guest value, echoed in resp */ 97 } __attribute__((__packed__)); 98 99 struct blkif_x86_32_request_indirect { 100 uint8_t indirect_op; 101 uint16_t nr_segments; 102 uint64_t id; 103 blkif_sector_t sector_number; 104 blkif_vdev_t handle; 105 uint16_t _pad1; 106 grant_ref_t indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST]; 107 /* 108 * The maximum number of indirect segments (and pages) that will 109 * be used is determined by MAX_INDIRECT_SEGMENTS, this value 110 * is also exported to the guest (via xenstore 111 * feature-max-indirect-segments entry), so the frontend knows how 112 * many indirect segments the backend supports. 113 */ 114 uint64_t _pad2; /* make it 64 byte aligned */ 115 } __attribute__((__packed__)); 116 117 struct blkif_x86_32_request { 118 uint8_t operation; /* BLKIF_OP_??? */ 119 union { 120 struct blkif_x86_32_request_rw rw; 121 struct blkif_x86_32_request_discard discard; 122 struct blkif_x86_32_request_other other; 123 struct blkif_x86_32_request_indirect indirect; 124 } u; 125 } __attribute__((__packed__)); 126 127 /* i386 protocol version */ 128 #pragma pack(push, 4) 129 struct blkif_x86_32_response { 130 uint64_t id; /* copied from request */ 131 uint8_t operation; /* copied from request */ 132 int16_t status; /* BLKIF_RSP_??? */ 133 }; 134 #pragma pack(pop) 135 /* x86_64 protocol version */ 136 137 struct blkif_x86_64_request_rw { 138 uint8_t nr_segments; /* number of segments */ 139 blkif_vdev_t handle; /* only for read/write requests */ 140 uint32_t _pad1; /* offsetof(blkif_reqest..,u.rw.id)==8 */ 141 uint64_t id; 142 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 143 struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 144 } __attribute__((__packed__)); 145 146 struct blkif_x86_64_request_discard { 147 uint8_t flag; /* BLKIF_DISCARD_SECURE or zero */ 148 blkif_vdev_t _pad1; /* was "handle" for read/write requests */ 149 uint32_t _pad2; /* offsetof(blkif_..,u.discard.id)==8 */ 150 uint64_t id; 151 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 152 uint64_t nr_sectors; 153 } __attribute__((__packed__)); 154 155 struct blkif_x86_64_request_other { 156 uint8_t _pad1; 157 blkif_vdev_t _pad2; 158 uint32_t _pad3; /* offsetof(blkif_..,u.discard.id)==8 */ 159 uint64_t id; /* private guest value, echoed in resp */ 160 } __attribute__((__packed__)); 161 162 struct blkif_x86_64_request_indirect { 163 uint8_t indirect_op; 164 uint16_t nr_segments; 165 uint32_t _pad1; /* offsetof(blkif_..,u.indirect.id)==8 */ 166 uint64_t id; 167 blkif_sector_t sector_number; 168 blkif_vdev_t handle; 169 uint16_t _pad2; 170 grant_ref_t indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST]; 171 /* 172 * The maximum number of indirect segments (and pages) that will 173 * be used is determined by MAX_INDIRECT_SEGMENTS, this value 174 * is also exported to the guest (via xenstore 175 * feature-max-indirect-segments entry), so the frontend knows how 176 * many indirect segments the backend supports. 177 */ 178 uint32_t _pad3; /* make it 64 byte aligned */ 179 } __attribute__((__packed__)); 180 181 struct blkif_x86_64_request { 182 uint8_t operation; /* BLKIF_OP_??? */ 183 union { 184 struct blkif_x86_64_request_rw rw; 185 struct blkif_x86_64_request_discard discard; 186 struct blkif_x86_64_request_other other; 187 struct blkif_x86_64_request_indirect indirect; 188 } u; 189 } __attribute__((__packed__)); 190 191 struct blkif_x86_64_response { 192 uint64_t __attribute__((__aligned__(8))) id; 193 uint8_t operation; /* copied from request */ 194 int16_t status; /* BLKIF_RSP_??? */ 195 }; 196 197 DEFINE_RING_TYPES(blkif_common, struct blkif_common_request, 198 struct blkif_common_response); 199 DEFINE_RING_TYPES(blkif_x86_32, struct blkif_x86_32_request, 200 struct blkif_x86_32_response); 201 DEFINE_RING_TYPES(blkif_x86_64, struct blkif_x86_64_request, 202 struct blkif_x86_64_response); 203 204 union blkif_back_rings { 205 struct blkif_back_ring native; 206 struct blkif_common_back_ring common; 207 struct blkif_x86_32_back_ring x86_32; 208 struct blkif_x86_64_back_ring x86_64; 209 }; 210 211 enum blkif_protocol { 212 BLKIF_PROTOCOL_NATIVE = 1, 213 BLKIF_PROTOCOL_X86_32 = 2, 214 BLKIF_PROTOCOL_X86_64 = 3, 215 }; 216 217 struct xen_vbd { 218 /* What the domain refers to this vbd as. */ 219 blkif_vdev_t handle; 220 /* Non-zero -> read-only */ 221 unsigned char readonly; 222 /* VDISK_xxx */ 223 unsigned char type; 224 /* phys device that this vbd maps to. */ 225 u32 pdevice; 226 struct block_device *bdev; 227 /* Cached size parameter. */ 228 sector_t size; 229 unsigned int flush_support:1; 230 unsigned int discard_secure:1; 231 unsigned int feature_gnt_persistent:1; 232 unsigned int overflow_max_grants:1; 233 }; 234 235 struct backend_info; 236 237 /* Number of available flags */ 238 #define PERSISTENT_GNT_FLAGS_SIZE 2 239 /* This persistent grant is currently in use */ 240 #define PERSISTENT_GNT_ACTIVE 0 241 /* 242 * This persistent grant has been used, this flag is set when we remove the 243 * PERSISTENT_GNT_ACTIVE, to know that this grant has been used recently. 244 */ 245 #define PERSISTENT_GNT_WAS_ACTIVE 1 246 247 /* Number of requests that we can fit in a ring */ 248 #define XEN_BLKIF_REQS 32 249 250 struct persistent_gnt { 251 struct page *page; 252 grant_ref_t gnt; 253 grant_handle_t handle; 254 DECLARE_BITMAP(flags, PERSISTENT_GNT_FLAGS_SIZE); 255 struct rb_node node; 256 struct list_head remove_node; 257 }; 258 259 struct xen_blkif { 260 /* Unique identifier for this interface. */ 261 domid_t domid; 262 unsigned int handle; 263 /* Physical parameters of the comms window. */ 264 unsigned int irq; 265 /* Comms information. */ 266 enum blkif_protocol blk_protocol; 267 union blkif_back_rings blk_rings; 268 void *blk_ring; 269 /* The VBD attached to this interface. */ 270 struct xen_vbd vbd; 271 /* Back pointer to the backend_info. */ 272 struct backend_info *be; 273 /* Private fields. */ 274 spinlock_t blk_ring_lock; 275 atomic_t refcnt; 276 277 wait_queue_head_t wq; 278 /* for barrier (drain) requests */ 279 struct completion drain_complete; 280 atomic_t drain; 281 /* One thread per one blkif. */ 282 struct task_struct *xenblkd; 283 unsigned int waiting_reqs; 284 285 /* tree to store persistent grants */ 286 struct rb_root persistent_gnts; 287 unsigned int persistent_gnt_c; 288 atomic_t persistent_gnt_in_use; 289 unsigned long next_lru; 290 291 /* used by the kworker that offload work from the persistent purge */ 292 struct list_head persistent_purge_list; 293 struct work_struct persistent_purge_work; 294 295 /* buffer of free pages to map grant refs */ 296 spinlock_t free_pages_lock; 297 int free_pages_num; 298 struct list_head free_pages; 299 300 /* List of all 'pending_req' available */ 301 struct list_head pending_free; 302 /* And its spinlock. */ 303 spinlock_t pending_free_lock; 304 wait_queue_head_t pending_free_wq; 305 306 /* statistics */ 307 unsigned long st_print; 308 unsigned long long st_rd_req; 309 unsigned long long st_wr_req; 310 unsigned long long st_oo_req; 311 unsigned long long st_f_req; 312 unsigned long long st_ds_req; 313 unsigned long long st_rd_sect; 314 unsigned long long st_wr_sect; 315 316 wait_queue_head_t waiting_to_free; 317 /* Thread shutdown wait queue. */ 318 wait_queue_head_t shutdown_wq; 319 }; 320 321 struct seg_buf { 322 unsigned long offset; 323 unsigned int nsec; 324 }; 325 326 struct grant_page { 327 struct page *page; 328 struct persistent_gnt *persistent_gnt; 329 grant_handle_t handle; 330 grant_ref_t gref; 331 }; 332 333 /* 334 * Each outstanding request that we've passed to the lower device layers has a 335 * 'pending_req' allocated to it. Each buffer_head that completes decrements 336 * the pendcnt towards zero. When it hits zero, the specified domain has a 337 * response queued for it, with the saved 'id' passed back. 338 */ 339 struct pending_req { 340 struct xen_blkif *blkif; 341 u64 id; 342 int nr_pages; 343 atomic_t pendcnt; 344 unsigned short operation; 345 int status; 346 struct list_head free_list; 347 struct grant_page *segments[MAX_INDIRECT_SEGMENTS]; 348 /* Indirect descriptors */ 349 struct grant_page *indirect_pages[MAX_INDIRECT_PAGES]; 350 struct seg_buf seg[MAX_INDIRECT_SEGMENTS]; 351 struct bio *biolist[MAX_INDIRECT_SEGMENTS]; 352 }; 353 354 355 #define vbd_sz(_v) ((_v)->bdev->bd_part ? \ 356 (_v)->bdev->bd_part->nr_sects : \ 357 get_capacity((_v)->bdev->bd_disk)) 358 359 #define xen_blkif_get(_b) (atomic_inc(&(_b)->refcnt)) 360 #define xen_blkif_put(_b) \ 361 do { \ 362 if (atomic_dec_and_test(&(_b)->refcnt)) \ 363 wake_up(&(_b)->waiting_to_free);\ 364 } while (0) 365 366 struct phys_req { 367 unsigned short dev; 368 blkif_sector_t nr_sects; 369 struct block_device *bdev; 370 blkif_sector_t sector_number; 371 }; 372 int xen_blkif_interface_init(void); 373 374 int xen_blkif_xenbus_init(void); 375 376 irqreturn_t xen_blkif_be_int(int irq, void *dev_id); 377 int xen_blkif_schedule(void *arg); 378 int xen_blkif_purge_persistent(void *arg); 379 380 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt, 381 struct backend_info *be, int state); 382 383 int xen_blkbk_barrier(struct xenbus_transaction xbt, 384 struct backend_info *be, int state); 385 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be); 386 387 static inline void blkif_get_x86_32_req(struct blkif_request *dst, 388 struct blkif_x86_32_request *src) 389 { 390 int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j; 391 dst->operation = src->operation; 392 switch (src->operation) { 393 case BLKIF_OP_READ: 394 case BLKIF_OP_WRITE: 395 case BLKIF_OP_WRITE_BARRIER: 396 case BLKIF_OP_FLUSH_DISKCACHE: 397 dst->u.rw.nr_segments = src->u.rw.nr_segments; 398 dst->u.rw.handle = src->u.rw.handle; 399 dst->u.rw.id = src->u.rw.id; 400 dst->u.rw.sector_number = src->u.rw.sector_number; 401 barrier(); 402 if (n > dst->u.rw.nr_segments) 403 n = dst->u.rw.nr_segments; 404 for (i = 0; i < n; i++) 405 dst->u.rw.seg[i] = src->u.rw.seg[i]; 406 break; 407 case BLKIF_OP_DISCARD: 408 dst->u.discard.flag = src->u.discard.flag; 409 dst->u.discard.id = src->u.discard.id; 410 dst->u.discard.sector_number = src->u.discard.sector_number; 411 dst->u.discard.nr_sectors = src->u.discard.nr_sectors; 412 break; 413 case BLKIF_OP_INDIRECT: 414 dst->u.indirect.indirect_op = src->u.indirect.indirect_op; 415 dst->u.indirect.nr_segments = src->u.indirect.nr_segments; 416 dst->u.indirect.handle = src->u.indirect.handle; 417 dst->u.indirect.id = src->u.indirect.id; 418 dst->u.indirect.sector_number = src->u.indirect.sector_number; 419 barrier(); 420 j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments)); 421 for (i = 0; i < j; i++) 422 dst->u.indirect.indirect_grefs[i] = 423 src->u.indirect.indirect_grefs[i]; 424 break; 425 default: 426 /* 427 * Don't know how to translate this op. Only get the 428 * ID so failure can be reported to the frontend. 429 */ 430 dst->u.other.id = src->u.other.id; 431 break; 432 } 433 } 434 435 static inline void blkif_get_x86_64_req(struct blkif_request *dst, 436 struct blkif_x86_64_request *src) 437 { 438 int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j; 439 dst->operation = src->operation; 440 switch (src->operation) { 441 case BLKIF_OP_READ: 442 case BLKIF_OP_WRITE: 443 case BLKIF_OP_WRITE_BARRIER: 444 case BLKIF_OP_FLUSH_DISKCACHE: 445 dst->u.rw.nr_segments = src->u.rw.nr_segments; 446 dst->u.rw.handle = src->u.rw.handle; 447 dst->u.rw.id = src->u.rw.id; 448 dst->u.rw.sector_number = src->u.rw.sector_number; 449 barrier(); 450 if (n > dst->u.rw.nr_segments) 451 n = dst->u.rw.nr_segments; 452 for (i = 0; i < n; i++) 453 dst->u.rw.seg[i] = src->u.rw.seg[i]; 454 break; 455 case BLKIF_OP_DISCARD: 456 dst->u.discard.flag = src->u.discard.flag; 457 dst->u.discard.id = src->u.discard.id; 458 dst->u.discard.sector_number = src->u.discard.sector_number; 459 dst->u.discard.nr_sectors = src->u.discard.nr_sectors; 460 break; 461 case BLKIF_OP_INDIRECT: 462 dst->u.indirect.indirect_op = src->u.indirect.indirect_op; 463 dst->u.indirect.nr_segments = src->u.indirect.nr_segments; 464 dst->u.indirect.handle = src->u.indirect.handle; 465 dst->u.indirect.id = src->u.indirect.id; 466 dst->u.indirect.sector_number = src->u.indirect.sector_number; 467 barrier(); 468 j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments)); 469 for (i = 0; i < j; i++) 470 dst->u.indirect.indirect_grefs[i] = 471 src->u.indirect.indirect_grefs[i]; 472 break; 473 default: 474 /* 475 * Don't know how to translate this op. Only get the 476 * ID so failure can be reported to the frontend. 477 */ 478 dst->u.other.id = src->u.other.id; 479 break; 480 } 481 } 482 483 #endif /* __XEN_BLKIF__BACKEND__COMMON_H__ */ 484