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/pgalloc.h>
  40 #include <asm/hypervisor.h>
  41 #include <xen/grant_table.h>
  42 #include <xen/page.h>
  43 #include <xen/xenbus.h>
  44 #include <xen/interface/io/ring.h>
  45 #include <xen/interface/io/blkif.h>
  46 #include <xen/interface/io/protocols.h>
  47
  48 extern unsigned int xen_blkif_max_ring_order;
  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 struct blkif_common_response {
  78         char dummy;
  79 };
  80
  81 struct blkif_x86_32_request_rw {
  82         uint8_t        nr_segments;  /* number of segments                   */
  83         blkif_vdev_t   handle;       /* only for read/write requests         */
  84         uint64_t       id;           /* private guest value, echoed in resp  */
  85         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
  86         struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  87 } __attribute__((__packed__));
  88
  89 struct blkif_x86_32_request_discard {
  90         uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
  91         blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
  92         uint64_t       id;           /* private guest value, echoed in resp  */
  93         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
  94         uint64_t       nr_sectors;
  95 } __attribute__((__packed__));
  96
  97 struct blkif_x86_32_request_other {
  98         uint8_t        _pad1;
  99         blkif_vdev_t   _pad2;
 100         uint64_t       id;           /* private guest value, echoed in resp  */
 101 } __attribute__((__packed__));
 102
 103 struct blkif_x86_32_request_indirect {
 104         uint8_t        indirect_op;
 105         uint16_t       nr_segments;
 106         uint64_t       id;
 107         blkif_sector_t sector_number;
 108         blkif_vdev_t   handle;
 109         uint16_t       _pad1;
 110         grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
 111         /*
 112          * The maximum number of indirect segments (and pages) that will
 113          * be used is determined by MAX_INDIRECT_SEGMENTS, this value
 114          * is also exported to the guest (via xenstore
 115          * feature-max-indirect-segments entry), so the frontend knows how
 116          * many indirect segments the backend supports.
 117          */
 118         uint64_t       _pad2;        /* make it 64 byte aligned */
 119 } __attribute__((__packed__));
 120
 121 struct blkif_x86_32_request {
 122         uint8_t        operation;    /* BLKIF_OP_???                         */
 123         union {
 124                 struct blkif_x86_32_request_rw rw;
 125                 struct blkif_x86_32_request_discard discard;
 126                 struct blkif_x86_32_request_other other;
 127                 struct blkif_x86_32_request_indirect indirect;
 128         } u;
 129 } __attribute__((__packed__));
 130
 131 /* i386 protocol version */
 132 #pragma pack(push, 4)
 133 struct blkif_x86_32_response {
 134         uint64_t        id;              /* copied from request */
 135         uint8_t         operation;       /* copied from request */
 136         int16_t         status;          /* BLKIF_RSP_???       */
 137 };
 138 #pragma pack(pop)
 139 /* x86_64 protocol version */
 140
 141 struct blkif_x86_64_request_rw {
 142         uint8_t        nr_segments;  /* number of segments                   */
 143         blkif_vdev_t   handle;       /* only for read/write requests         */
 144         uint32_t       _pad1;        /* offsetof(blkif_reqest..,u.rw.id)==8  */
 145         uint64_t       id;
 146         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
 147         struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 148 } __attribute__((__packed__));
 149
 150 struct blkif_x86_64_request_discard {
 151         uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
 152         blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
 153         uint32_t       _pad2;        /* offsetof(blkif_..,u.discard.id)==8   */
 154         uint64_t       id;
 155         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
 156         uint64_t       nr_sectors;
 157 } __attribute__((__packed__));
 158
 159 struct blkif_x86_64_request_other {
 160         uint8_t        _pad1;
 161         blkif_vdev_t   _pad2;
 162         uint32_t       _pad3;        /* offsetof(blkif_..,u.discard.id)==8   */
 163         uint64_t       id;           /* private guest value, echoed in resp  */
 164 } __attribute__((__packed__));
 165
 166 struct blkif_x86_64_request_indirect {
 167         uint8_t        indirect_op;
 168         uint16_t       nr_segments;
 169         uint32_t       _pad1;        /* offsetof(blkif_..,u.indirect.id)==8   */
 170         uint64_t       id;
 171         blkif_sector_t sector_number;
 172         blkif_vdev_t   handle;
 173         uint16_t       _pad2;
 174         grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
 175         /*
 176          * The maximum number of indirect segments (and pages) that will
 177          * be used is determined by MAX_INDIRECT_SEGMENTS, this value
 178          * is also exported to the guest (via xenstore
 179          * feature-max-indirect-segments entry), so the frontend knows how
 180          * many indirect segments the backend supports.
 181          */
 182         uint32_t       _pad3;        /* make it 64 byte aligned */
 183 } __attribute__((__packed__));
 184
 185 struct blkif_x86_64_request {
 186         uint8_t        operation;    /* BLKIF_OP_???                         */
 187         union {
 188                 struct blkif_x86_64_request_rw rw;
 189                 struct blkif_x86_64_request_discard discard;
 190                 struct blkif_x86_64_request_other other;
 191                 struct blkif_x86_64_request_indirect indirect;
 192         } u;
 193 } __attribute__((__packed__));
 194
 195 struct blkif_x86_64_response {
 196         uint64_t       __attribute__((__aligned__(8))) id;
 197         uint8_t         operation;       /* copied from request */
 198         int16_t         status;          /* BLKIF_RSP_???       */
 199 };
 200
 201 DEFINE_RING_TYPES(blkif_common, struct blkif_common_request,
 202                   struct blkif_common_response);
 203 DEFINE_RING_TYPES(blkif_x86_32, struct blkif_x86_32_request,
 204                   struct blkif_x86_32_response);
 205 DEFINE_RING_TYPES(blkif_x86_64, struct blkif_x86_64_request,
 206                   struct blkif_x86_64_response);
 207
 208 union blkif_back_rings {
 209         struct blkif_back_ring        native;
 210         struct blkif_common_back_ring common;
 211         struct blkif_x86_32_back_ring x86_32;
 212         struct blkif_x86_64_back_ring x86_64;
 213 };
 214
 215 enum blkif_protocol {
 216         BLKIF_PROTOCOL_NATIVE = 1,
 217         BLKIF_PROTOCOL_X86_32 = 2,
 218         BLKIF_PROTOCOL_X86_64 = 3,
 219 };
 220
 221 /*
 222  * Default protocol if the frontend doesn't specify one.
 223  */
 224 #ifdef CONFIG_X86
 225 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_X86_32
 226 #else
 227 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_NATIVE
 228 #endif
 229
 230 struct xen_vbd {
 231         /* What the domain refers to this vbd as. */
 232         blkif_vdev_t            handle;
 233         /* Non-zero -> read-only */
 234         unsigned char           readonly;
 235         /* VDISK_xxx */
 236         unsigned char           type;
 237         /* phys device that this vbd maps to. */
 238         u32                     pdevice;
 239         struct block_device     *bdev;
 240         /* Cached size parameter. */
 241         sector_t                size;
 242         unsigned int            flush_support:1;
 243         unsigned int            discard_secure:1;
 244         unsigned int            feature_gnt_persistent:1;
 245         unsigned int            overflow_max_grants:1;
 246 };
 247
 248 struct backend_info;
 249
 250 /* Number of available flags */
 251 #define PERSISTENT_GNT_FLAGS_SIZE       2
 252 /* This persistent grant is currently in use */
 253 #define PERSISTENT_GNT_ACTIVE           0
 254 /*
 255  * This persistent grant has been used, this flag is set when we remove the
 256  * PERSISTENT_GNT_ACTIVE, to know that this grant has been used recently.
 257  */
 258 #define PERSISTENT_GNT_WAS_ACTIVE       1
 259
 260 /* Number of requests that we can fit in a ring */
 261 #define XEN_BLKIF_REQS_PER_PAGE         32
 262
 263 struct persistent_gnt {
 264         struct page *page;
 265         grant_ref_t gnt;
 266         grant_handle_t handle;
 267         DECLARE_BITMAP(flags, PERSISTENT_GNT_FLAGS_SIZE);
 268         struct rb_node node;
 269         struct list_head remove_node;
 270 };
 271
 272 /* Per-ring information. */
 273 struct xen_blkif_ring {
 274         /* Physical parameters of the comms window. */
 275         unsigned int            irq;
 276         union blkif_back_rings  blk_rings;
 277         void                    *blk_ring;
 278         /* Private fields. */
 279         spinlock_t              blk_ring_lock;
 280
 281         wait_queue_head_t       wq;
 282         atomic_t                inflight;
 283         /* One thread per blkif ring. */
 284         struct task_struct      *xenblkd;
 285         unsigned int            waiting_reqs;
 286
 287         /* List of all 'pending_req' available */
 288         struct list_head        pending_free;
 289         /* And its spinlock. */
 290         spinlock_t              pending_free_lock;
 291         wait_queue_head_t       pending_free_wq;
 292
 293         struct work_struct      free_work;
 294         /* Thread shutdown wait queue. */
 295         wait_queue_head_t       shutdown_wq;
 296         struct xen_blkif        *blkif;
 297 };
 298
 299 struct xen_blkif {
 300         /* Unique identifier for this interface. */
 301         domid_t                 domid;
 302         unsigned int            handle;
 303         /* Comms information. */
 304         enum blkif_protocol     blk_protocol;
 305         /* The VBD attached to this interface. */
 306         struct xen_vbd          vbd;
 307         /* Back pointer to the backend_info. */
 308         struct backend_info     *be;
 309         atomic_t                refcnt;
 310         /* for barrier (drain) requests */
 311         struct completion       drain_complete;
 312         atomic_t                drain;
 313
 314         /* tree to store persistent grants */
 315         spinlock_t              pers_gnts_lock;
 316         struct rb_root          persistent_gnts;
 317         unsigned int            persistent_gnt_c;
 318         atomic_t                persistent_gnt_in_use;
 319         unsigned long           next_lru;
 320
 321         /* used by the kworker that offload work from the persistent purge */
 322         struct list_head        persistent_purge_list;
 323         struct work_struct      persistent_purge_work;
 324
 325         /* buffer of free pages to map grant refs */
 326         spinlock_t              free_pages_lock;
 327         int                     free_pages_num;
 328         struct list_head        free_pages;
 329
 330         /* statistics */
 331         unsigned long           st_print;
 332         unsigned long long                      st_rd_req;
 333         unsigned long long                      st_wr_req;
 334         unsigned long long                      st_oo_req;
 335         unsigned long long                      st_f_req;
 336         unsigned long long                      st_ds_req;
 337         unsigned long long                      st_rd_sect;
 338         unsigned long long                      st_wr_sect;
 339
 340         struct work_struct      free_work;
 341         unsigned int            nr_ring_pages;
 342         /* All rings for this device. */
 343         struct xen_blkif_ring   *rings;
 344         unsigned int            nr_rings;
 345 };
 346
 347 struct seg_buf {
 348         unsigned long offset;
 349         unsigned int nsec;
 350 };
 351
 352 struct grant_page {
 353         struct page             *page;
 354         struct persistent_gnt   *persistent_gnt;
 355         grant_handle_t          handle;
 356         grant_ref_t             gref;
 357 };
 358
 359 /*
 360  * Each outstanding request that we've passed to the lower device layers has a
 361  * 'pending_req' allocated to it. Each buffer_head that completes decrements
 362  * the pendcnt towards zero. When it hits zero, the specified domain has a
 363  * response queued for it, with the saved 'id' passed back.
 364  */
 365 struct pending_req {
 366         struct xen_blkif_ring   *ring;
 367         u64                     id;
 368         int                     nr_segs;
 369         atomic_t                pendcnt;
 370         unsigned short          operation;
 371         int                     status;
 372         struct list_head        free_list;
 373         struct grant_page       *segments[MAX_INDIRECT_SEGMENTS];
 374         /* Indirect descriptors */
 375         struct grant_page       *indirect_pages[MAX_INDIRECT_PAGES];
 376         struct seg_buf          seg[MAX_INDIRECT_SEGMENTS];
 377         struct bio              *biolist[MAX_INDIRECT_SEGMENTS];
 378         struct gnttab_unmap_grant_ref unmap[MAX_INDIRECT_SEGMENTS];
 379         struct page                   *unmap_pages[MAX_INDIRECT_SEGMENTS];
 380         struct gntab_unmap_queue_data gnttab_unmap_data;
 381 };
 382
 383
 384 #define vbd_sz(_v)      ((_v)->bdev->bd_part ? \
 385                          (_v)->bdev->bd_part->nr_sects : \
 386                           get_capacity((_v)->bdev->bd_disk))
 387
 388 #define xen_blkif_get(_b) (atomic_inc(&(_b)->refcnt))
 389 #define xen_blkif_put(_b)                               \
 390         do {                                            \
 391                 if (atomic_dec_and_test(&(_b)->refcnt)) \
 392                         schedule_work(&(_b)->free_work);\
 393         } while (0)
 394
 395 struct phys_req {
 396         unsigned short          dev;
 397         blkif_sector_t          nr_sects;
 398         struct block_device     *bdev;
 399         blkif_sector_t          sector_number;
 400 };
 401 int xen_blkif_interface_init(void);
 402
 403 int xen_blkif_xenbus_init(void);
 404
 405 irqreturn_t xen_blkif_be_int(int irq, void *dev_id);
 406 int xen_blkif_schedule(void *arg);
 407 int xen_blkif_purge_persistent(void *arg);
 408 void xen_blkbk_free_caches(struct xen_blkif_ring *ring);
 409
 410 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 411                               struct backend_info *be, int state);
 412
 413 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 414                       struct backend_info *be, int state);
 415 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be);
 416 void xen_blkbk_unmap_purged_grants(struct work_struct *work);
 417
 418 static inline void blkif_get_x86_32_req(struct blkif_request *dst,
 419                                         struct blkif_x86_32_request *src)
 420 {
 421         int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
 422         dst->operation = src->operation;
 423         switch (src->operation) {
 424         case BLKIF_OP_READ:
 425         case BLKIF_OP_WRITE:
 426         case BLKIF_OP_WRITE_BARRIER:
 427         case BLKIF_OP_FLUSH_DISKCACHE:
 428                 dst->u.rw.nr_segments = src->u.rw.nr_segments;
 429                 dst->u.rw.handle = src->u.rw.handle;
 430                 dst->u.rw.id = src->u.rw.id;
 431                 dst->u.rw.sector_number = src->u.rw.sector_number;
 432                 barrier();
 433                 if (n > dst->u.rw.nr_segments)
 434                         n = dst->u.rw.nr_segments;
 435                 for (i = 0; i < n; i++)
 436                         dst->u.rw.seg[i] = src->u.rw.seg[i];
 437                 break;
 438         case BLKIF_OP_DISCARD:
 439                 dst->u.discard.flag = src->u.discard.flag;
 440                 dst->u.discard.id = src->u.discard.id;
 441                 dst->u.discard.sector_number = src->u.discard.sector_number;
 442                 dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
 443                 break;
 444         case BLKIF_OP_INDIRECT:
 445                 dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
 446                 dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
 447                 dst->u.indirect.handle = src->u.indirect.handle;
 448                 dst->u.indirect.id = src->u.indirect.id;
 449                 dst->u.indirect.sector_number = src->u.indirect.sector_number;
 450                 barrier();
 451                 j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
 452                 for (i = 0; i < j; i++)
 453                         dst->u.indirect.indirect_grefs[i] =
 454                                 src->u.indirect.indirect_grefs[i];
 455                 break;
 456         default:
 457                 /*
 458                  * Don't know how to translate this op. Only get the
 459                  * ID so failure can be reported to the frontend.
 460                  */
 461                 dst->u.other.id = src->u.other.id;
 462                 break;
 463         }
 464 }
 465
 466 static inline void blkif_get_x86_64_req(struct blkif_request *dst,
 467                                         struct blkif_x86_64_request *src)
 468 {
 469         int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
 470         dst->operation = src->operation;
 471         switch (src->operation) {
 472         case BLKIF_OP_READ:
 473         case BLKIF_OP_WRITE:
 474         case BLKIF_OP_WRITE_BARRIER:
 475         case BLKIF_OP_FLUSH_DISKCACHE:
 476                 dst->u.rw.nr_segments = src->u.rw.nr_segments;
 477                 dst->u.rw.handle = src->u.rw.handle;
 478                 dst->u.rw.id = src->u.rw.id;
 479                 dst->u.rw.sector_number = src->u.rw.sector_number;
 480                 barrier();
 481                 if (n > dst->u.rw.nr_segments)
 482                         n = dst->u.rw.nr_segments;
 483                 for (i = 0; i < n; i++)
 484                         dst->u.rw.seg[i] = src->u.rw.seg[i];
 485                 break;
 486         case BLKIF_OP_DISCARD:
 487                 dst->u.discard.flag = src->u.discard.flag;
 488                 dst->u.discard.id = src->u.discard.id;
 489                 dst->u.discard.sector_number = src->u.discard.sector_number;
 490                 dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
 491                 break;
 492         case BLKIF_OP_INDIRECT:
 493                 dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
 494                 dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
 495                 dst->u.indirect.handle = src->u.indirect.handle;
 496                 dst->u.indirect.id = src->u.indirect.id;
 497                 dst->u.indirect.sector_number = src->u.indirect.sector_number;
 498                 barrier();
 499                 j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
 500                 for (i = 0; i < j; i++)
 501                         dst->u.indirect.indirect_grefs[i] =
 502                                 src->u.indirect.indirect_grefs[i];
 503                 break;
 504         default:
 505                 /*
 506                  * Don't know how to translate this op. Only get the
 507                  * ID so failure can be reported to the frontend.
 508                  */
 509                 dst->u.other.id = src->u.other.id;
 510                 break;
 511         }
 512 }
 513
 514 #endif /* __XEN_BLKIF__BACKEND__COMMON_H__ */