drivers/block/xen-blkback/common.h

   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/hypervisor.h>
  40 #include <xen/grant_table.h>
  41 #include <xen/page.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 extern unsigned int xen_blkif_max_ring_order;
  48 extern unsigned int xenblk_max_queues;
  49 /*
  50  * This is the maximum number of segments that would be allowed in indirect
  51  * requests. This value will also be passed to the frontend.
  52  */
  53 #define MAX_INDIRECT_SEGMENTS 256
  54
  55 /*
  56  * Xen use 4K pages. The guest may use different page size (4K or 64K)
  57  * Number of Xen pages per segment
  58  */
  59 #define XEN_PAGES_PER_SEGMENT   (PAGE_SIZE / XEN_PAGE_SIZE)
  60
  61 #define XEN_PAGES_PER_INDIRECT_FRAME \
  62         (XEN_PAGE_SIZE/sizeof(struct blkif_request_segment))
  63 #define SEGS_PER_INDIRECT_FRAME \
  64         (XEN_PAGES_PER_INDIRECT_FRAME / XEN_PAGES_PER_SEGMENT)
  65
  66 #define MAX_INDIRECT_PAGES \
  67         ((MAX_INDIRECT_SEGMENTS + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
  68 #define INDIRECT_PAGES(_segs) DIV_ROUND_UP(_segs, XEN_PAGES_PER_INDIRECT_FRAME)
  69
  70 /* Not a real protocol.  Used to generate ring structs which contain
  71  * the elements common to all protocols only.  This way we get a
  72  * compiler-checkable way to use common struct elements, so we can
  73  * avoid using switch(protocol) in a number of places.  */
  74 struct blkif_common_request {
  75         char dummy;
  76 };
  77
  78 /* i386 protocol version */
  79
  80 struct blkif_x86_32_request_rw {
  81         uint8_t        nr_segments;  /* number of segments                   */
  82         blkif_vdev_t   handle;       /* only for read/write requests         */
  83         uint64_t       id;           /* private guest value, echoed in resp  */
  84         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
  85         struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  86 } __attribute__((__packed__));
  87
  88 struct blkif_x86_32_request_discard {
  89         uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
  90         blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
  91         uint64_t       id;           /* private guest value, echoed in resp  */
  92         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
  93         uint64_t       nr_sectors;
  94 } __attribute__((__packed__));
  95
  96 struct blkif_x86_32_request_other {
  97         uint8_t        _pad1;
  98         blkif_vdev_t   _pad2;
  99         uint64_t       id;           /* private guest value, echoed in resp  */
 100 } __attribute__((__packed__));
 101
 102 struct blkif_x86_32_request_indirect {
 103         uint8_t        indirect_op;
 104         uint16_t       nr_segments;
 105         uint64_t       id;
 106         blkif_sector_t sector_number;
 107         blkif_vdev_t   handle;
 108         uint16_t       _pad1;
 109         grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
 110         /*
 111          * The maximum number of indirect segments (and pages) that will
 112          * be used is determined by MAX_INDIRECT_SEGMENTS, this value
 113          * is also exported to the guest (via xenstore
 114          * feature-max-indirect-segments entry), so the frontend knows how
 115          * many indirect segments the backend supports.
 116          */
 117         uint64_t       _pad2;        /* make it 64 byte aligned */
 118 } __attribute__((__packed__));
 119
 120 struct blkif_x86_32_request {
 121         uint8_t        operation;    /* BLKIF_OP_???                         */
 122         union {
 123                 struct blkif_x86_32_request_rw rw;
 124                 struct blkif_x86_32_request_discard discard;
 125                 struct blkif_x86_32_request_other other;
 126                 struct blkif_x86_32_request_indirect indirect;
 127         } u;
 128 } __attribute__((__packed__));
 129
 130 /* x86_64 protocol version */
 131
 132 struct blkif_x86_64_request_rw {
 133         uint8_t        nr_segments;  /* number of segments                   */
 134         blkif_vdev_t   handle;       /* only for read/write requests         */
 135         uint32_t       _pad1;        /* offsetof(blkif_reqest..,u.rw.id)==8  */
 136         uint64_t       id;
 137         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
 138         struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 139 } __attribute__((__packed__));
 140
 141 struct blkif_x86_64_request_discard {
 142         uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
 143         blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
 144         uint32_t       _pad2;        /* offsetof(blkif_..,u.discard.id)==8   */
 145         uint64_t       id;
 146         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
 147         uint64_t       nr_sectors;
 148 } __attribute__((__packed__));
 149
 150 struct blkif_x86_64_request_other {
 151         uint8_t        _pad1;
 152         blkif_vdev_t   _pad2;
 153         uint32_t       _pad3;        /* offsetof(blkif_..,u.discard.id)==8   */
 154         uint64_t       id;           /* private guest value, echoed in resp  */
 155 } __attribute__((__packed__));
 156
 157 struct blkif_x86_64_request_indirect {
 158         uint8_t        indirect_op;
 159         uint16_t       nr_segments;
 160         uint32_t       _pad1;        /* offsetof(blkif_..,u.indirect.id)==8   */
 161         uint64_t       id;
 162         blkif_sector_t sector_number;
 163         blkif_vdev_t   handle;
 164         uint16_t       _pad2;
 165         grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
 166         /*
 167          * The maximum number of indirect segments (and pages) that will
 168          * be used is determined by MAX_INDIRECT_SEGMENTS, this value
 169          * is also exported to the guest (via xenstore
 170          * feature-max-indirect-segments entry), so the frontend knows how
 171          * many indirect segments the backend supports.
 172          */
 173         uint32_t       _pad3;        /* make it 64 byte aligned */
 174 } __attribute__((__packed__));
 175
 176 struct blkif_x86_64_request {
 177         uint8_t        operation;    /* BLKIF_OP_???                         */
 178         union {
 179                 struct blkif_x86_64_request_rw rw;
 180                 struct blkif_x86_64_request_discard discard;
 181                 struct blkif_x86_64_request_other other;
 182                 struct blkif_x86_64_request_indirect indirect;
 183         } u;
 184 } __attribute__((__packed__));
 185
 186 DEFINE_RING_TYPES(blkif_common, struct blkif_common_request,
 187                   struct blkif_response);
 188 DEFINE_RING_TYPES(blkif_x86_32, struct blkif_x86_32_request,
 189                   struct blkif_response __packed);
 190 DEFINE_RING_TYPES(blkif_x86_64, struct blkif_x86_64_request,
 191                   struct blkif_response);
 192
 193 union blkif_back_rings {
 194         struct blkif_back_ring        native;
 195         struct blkif_common_back_ring common;
 196         struct blkif_x86_32_back_ring x86_32;
 197         struct blkif_x86_64_back_ring x86_64;
 198 };
 199
 200 enum blkif_protocol {
 201         BLKIF_PROTOCOL_NATIVE = 1,
 202         BLKIF_PROTOCOL_X86_32 = 2,
 203         BLKIF_PROTOCOL_X86_64 = 3,
 204 };
 205
 206 /*
 207  * Default protocol if the frontend doesn't specify one.
 208  */
 209 #ifdef CONFIG_X86
 210 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_X86_32
 211 #else
 212 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_NATIVE
 213 #endif
 214
 215 struct xen_vbd {
 216         /* What the domain refers to this vbd as. */
 217         blkif_vdev_t            handle;
 218         /* Non-zero -> read-only */
 219         unsigned char           readonly;
 220         /* VDISK_xxx */
 221         unsigned char           type;
 222         /* phys device that this vbd maps to. */
 223         u32                     pdevice;
 224         struct block_device     *bdev;
 225         /* Cached size parameter. */
 226         sector_t                size;
 227         unsigned int            flush_support:1;
 228         unsigned int            discard_secure:1;
 229         /* Connect-time cached feature_persistent parameter value */
 230         unsigned int            feature_gnt_persistent_parm:1;
 231         /* Persistent grants feature negotiation result */
 232         unsigned int            feature_gnt_persistent:1;
 233         unsigned int            overflow_max_grants:1;
 234 };
 235
 236 struct backend_info;
 237
 238 /* Number of requests that we can fit in a ring */
 239 #define XEN_BLKIF_REQS_PER_PAGE         32
 240
 241 struct persistent_gnt {
 242         struct page *page;
 243         grant_ref_t gnt;
 244         grant_handle_t handle;
 245         unsigned long last_used;
 246         bool active;
 247         struct rb_node node;
 248         struct list_head remove_node;
 249 };
 250
 251 /* Per-ring information. */
 252 struct xen_blkif_ring {
 253         /* Physical parameters of the comms window. */
 254         unsigned int            irq;
 255         union blkif_back_rings  blk_rings;
 256         void                    *blk_ring;
 257         /* Private fields. */
 258         spinlock_t              blk_ring_lock;
 259
 260         wait_queue_head_t       wq;
 261         atomic_t                inflight;
 262         bool                    active;
 263         /* One thread per blkif ring. */
 264         struct task_struct      *xenblkd;
 265         unsigned int            waiting_reqs;
 266
 267         /* List of all 'pending_req' available */
 268         struct list_head        pending_free;
 269         /* And its spinlock. */
 270         spinlock_t              pending_free_lock;
 271         wait_queue_head_t       pending_free_wq;
 272
 273         /* Tree to store persistent grants. */
 274         struct rb_root          persistent_gnts;
 275         unsigned int            persistent_gnt_c;
 276         atomic_t                persistent_gnt_in_use;
 277         unsigned long           next_lru;
 278
 279         /* Statistics. */
 280         unsigned long           st_print;
 281         unsigned long long      st_rd_req;
 282         unsigned long long      st_wr_req;
 283         unsigned long long      st_oo_req;
 284         unsigned long long      st_f_req;
 285         unsigned long long      st_ds_req;
 286         unsigned long long      st_rd_sect;
 287         unsigned long long      st_wr_sect;
 288
 289         /* Used by the kworker that offload work from the persistent purge. */
 290         struct list_head        persistent_purge_list;
 291         struct work_struct      persistent_purge_work;
 292
 293         /* Buffer of free pages to map grant refs. */
 294         struct gnttab_page_cache free_pages;
 295
 296         struct work_struct      free_work;
 297         /* Thread shutdown wait queue. */
 298         wait_queue_head_t       shutdown_wq;
 299         struct xen_blkif        *blkif;
 300 };
 301
 302 struct xen_blkif {
 303         /* Unique identifier for this interface. */
 304         domid_t                 domid;
 305         unsigned int            handle;
 306         /* Comms information. */
 307         enum blkif_protocol     blk_protocol;
 308         /* The VBD attached to this interface. */
 309         struct xen_vbd          vbd;
 310         /* Back pointer to the backend_info. */
 311         struct backend_info     *be;
 312         atomic_t                refcnt;
 313         /* for barrier (drain) requests */
 314         struct completion       drain_complete;
 315         atomic_t                drain;
 316
 317         struct work_struct      free_work;
 318         unsigned int            nr_ring_pages;
 319         bool                    multi_ref;
 320         /* All rings for this device. */
 321         struct xen_blkif_ring   *rings;
 322         unsigned int            nr_rings;
 323         unsigned long           buffer_squeeze_end;
 324 };
 325
 326 struct seg_buf {
 327         unsigned long offset;
 328         unsigned int nsec;
 329 };
 330
 331 struct grant_page {
 332         struct page             *page;
 333         struct persistent_gnt   *persistent_gnt;
 334         grant_handle_t          handle;
 335         grant_ref_t             gref;
 336 };
 337
 338 /*
 339  * Each outstanding request that we've passed to the lower device layers has a
 340  * 'pending_req' allocated to it. Each buffer_head that completes decrements
 341  * the pendcnt towards zero. When it hits zero, the specified domain has a
 342  * response queued for it, with the saved 'id' passed back.
 343  */
 344 struct pending_req {
 345         struct xen_blkif_ring   *ring;
 346         u64                     id;
 347         int                     nr_segs;
 348         atomic_t                pendcnt;
 349         unsigned short          operation;
 350         int                     status;
 351         struct list_head        free_list;
 352         struct grant_page       *segments[MAX_INDIRECT_SEGMENTS];
 353         /* Indirect descriptors */
 354         struct grant_page       *indirect_pages[MAX_INDIRECT_PAGES];
 355         struct seg_buf          seg[MAX_INDIRECT_SEGMENTS];
 356         struct bio              *biolist[MAX_INDIRECT_SEGMENTS];
 357         struct gnttab_unmap_grant_ref unmap[MAX_INDIRECT_SEGMENTS];
 358         struct page                   *unmap_pages[MAX_INDIRECT_SEGMENTS];
 359         struct gntab_unmap_queue_data gnttab_unmap_data;
 360 };
 361
 362
 363 #define vbd_sz(_v)      ((_v)->bdev->bd_part ? \
 364                          (_v)->bdev->bd_part->nr_sects : \
 365                           get_capacity((_v)->bdev->bd_disk))
 366
 367 #define xen_blkif_get(_b) (atomic_inc(&(_b)->refcnt))
 368 #define xen_blkif_put(_b)                               \
 369         do {                                            \
 370                 if (atomic_dec_and_test(&(_b)->refcnt)) \
 371                         schedule_work(&(_b)->free_work);\
 372         } while (0)
 373
 374 struct phys_req {
 375         unsigned short          dev;
 376         blkif_sector_t          nr_sects;
 377         struct block_device     *bdev;
 378         blkif_sector_t          sector_number;
 379 };
 380
 381 int xen_blkif_interface_init(void);
 382 void xen_blkif_interface_fini(void);
 383
 384 int xen_blkif_xenbus_init(void);
 385 void xen_blkif_xenbus_fini(void);
 386
 387 irqreturn_t xen_blkif_be_int(int irq, void *dev_id);
 388 int xen_blkif_schedule(void *arg);
 389 int xen_blkif_purge_persistent(void *arg);
 390 void xen_blkbk_free_caches(struct xen_blkif_ring *ring);
 391
 392 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 393                               struct backend_info *be, int state);
 394
 395 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 396                       struct backend_info *be, int state);
 397 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be);
 398 void xen_blkbk_unmap_purged_grants(struct work_struct *work);
 399
 400 static inline void blkif_get_x86_32_req(struct blkif_request *dst,
 401                                         struct blkif_x86_32_request *src)
 402 {
 403         int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
 404         dst->operation = READ_ONCE(src->operation);
 405         switch (dst->operation) {
 406         case BLKIF_OP_READ:
 407         case BLKIF_OP_WRITE:
 408         case BLKIF_OP_WRITE_BARRIER:
 409         case BLKIF_OP_FLUSH_DISKCACHE:
 410                 dst->u.rw.nr_segments = src->u.rw.nr_segments;
 411                 dst->u.rw.handle = src->u.rw.handle;
 412                 dst->u.rw.id = src->u.rw.id;
 413                 dst->u.rw.sector_number = src->u.rw.sector_number;
 414                 barrier();
 415                 if (n > dst->u.rw.nr_segments)
 416                         n = dst->u.rw.nr_segments;
 417                 for (i = 0; i < n; i++)
 418                         dst->u.rw.seg[i] = src->u.rw.seg[i];
 419                 break;
 420         case BLKIF_OP_DISCARD:
 421                 dst->u.discard.flag = src->u.discard.flag;
 422                 dst->u.discard.id = src->u.discard.id;
 423                 dst->u.discard.sector_number = src->u.discard.sector_number;
 424                 dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
 425                 break;
 426         case BLKIF_OP_INDIRECT:
 427                 dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
 428                 dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
 429                 dst->u.indirect.handle = src->u.indirect.handle;
 430                 dst->u.indirect.id = src->u.indirect.id;
 431                 dst->u.indirect.sector_number = src->u.indirect.sector_number;
 432                 barrier();
 433                 j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
 434                 for (i = 0; i < j; i++)
 435                         dst->u.indirect.indirect_grefs[i] =
 436                                 src->u.indirect.indirect_grefs[i];
 437                 break;
 438         default:
 439                 /*
 440                  * Don't know how to translate this op. Only get the
 441                  * ID so failure can be reported to the frontend.
 442                  */
 443                 dst->u.other.id = src->u.other.id;
 444                 break;
 445         }
 446 }
 447
 448 static inline void blkif_get_x86_64_req(struct blkif_request *dst,
 449                                         struct blkif_x86_64_request *src)
 450 {
 451         int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
 452         dst->operation = READ_ONCE(src->operation);
 453         switch (dst->operation) {
 454         case BLKIF_OP_READ:
 455         case BLKIF_OP_WRITE:
 456         case BLKIF_OP_WRITE_BARRIER:
 457         case BLKIF_OP_FLUSH_DISKCACHE:
 458                 dst->u.rw.nr_segments = src->u.rw.nr_segments;
 459                 dst->u.rw.handle = src->u.rw.handle;
 460                 dst->u.rw.id = src->u.rw.id;
 461                 dst->u.rw.sector_number = src->u.rw.sector_number;
 462                 barrier();
 463                 if (n > dst->u.rw.nr_segments)
 464                         n = dst->u.rw.nr_segments;
 465                 for (i = 0; i < n; i++)
 466                         dst->u.rw.seg[i] = src->u.rw.seg[i];
 467                 break;
 468         case BLKIF_OP_DISCARD:
 469                 dst->u.discard.flag = src->u.discard.flag;
 470                 dst->u.discard.id = src->u.discard.id;
 471                 dst->u.discard.sector_number = src->u.discard.sector_number;
 472                 dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
 473                 break;
 474         case BLKIF_OP_INDIRECT:
 475                 dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
 476                 dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
 477                 dst->u.indirect.handle = src->u.indirect.handle;
 478                 dst->u.indirect.id = src->u.indirect.id;
 479                 dst->u.indirect.sector_number = src->u.indirect.sector_number;
 480                 barrier();
 481                 j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
 482                 for (i = 0; i < j; i++)
 483                         dst->u.indirect.indirect_grefs[i] =
 484                                 src->u.indirect.indirect_grefs[i];
 485                 break;
 486         default:
 487                 /*
 488                  * Don't know how to translate this op. Only get the
 489                  * ID so failure can be reported to the frontend.
 490                  */
 491                 dst->u.other.id = src->u.other.id;
 492                 break;
 493         }
 494 }
 495
 496 #endif /* __XEN_BLKIF__BACKEND__COMMON_H__ */