hw/virtio-9p.c

   1 /*
   2  * Virtio 9p backend
   3  *
   4  * Copyright IBM, Corp. 2010
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "virtio.h"
  15 #include "pc.h"
  16 #include "qemu_socket.h"
  17 #include "virtio-9p.h"
  18 #include "fsdev/qemu-fsdev.h"
  19 #include "virtio-9p-debug.h"
  20
  21 int dotu = 1;
  22 int debug_9p_pdu;
  23
  24 static int v9fs_do_lstat(V9fsState *s, V9fsString *path, struct stat *stbuf)
  25 {
  26     return s->ops->lstat(&s->ctx, path->data, stbuf);
  27 }
  28
  29 static int v9fs_do_setuid(V9fsState *s, uid_t uid)
  30 {
  31     return s->ops->setuid(&s->ctx, uid);
  32 }
  33
  34 static ssize_t v9fs_do_readlink(V9fsState *s, V9fsString *path, V9fsString *buf)
  35 {
  36     ssize_t len;
  37
  38     buf->data = qemu_malloc(1024);
  39
  40     len = s->ops->readlink(&s->ctx, path->data, buf->data, 1024 - 1);
  41     if (len > -1) {
  42         buf->size = len;
  43         buf->data[len] = 0;
  44     }
  45
  46     return len;
  47 }
  48
  49 static int v9fs_do_close(V9fsState *s, int fd)
  50 {
  51     return s->ops->close(&s->ctx, fd);
  52 }
  53
  54 static int v9fs_do_closedir(V9fsState *s, DIR *dir)
  55 {
  56     return s->ops->closedir(&s->ctx, dir);
  57 }
  58
  59 static void v9fs_string_init(V9fsString *str)
  60 {
  61     str->data = NULL;
  62     str->size = 0;
  63 }
  64
  65 static void v9fs_string_free(V9fsString *str)
  66 {
  67     qemu_free(str->data);
  68     str->data = NULL;
  69     str->size = 0;
  70 }
  71
  72 static void v9fs_string_null(V9fsString *str)
  73 {
  74     v9fs_string_free(str);
  75 }
  76
  77 static int number_to_string(void *arg, char type)
  78 {
  79     unsigned int ret = 0;
  80
  81     switch (type) {
  82     case 'u': {
  83         unsigned int num = *(unsigned int *)arg;
  84
  85         do {
  86             ret++;
  87             num = num/10;
  88         } while (num);
  89         break;
  90     }
  91     default:
  92         printf("Number_to_string: Unknown number format\n");
  93         return -1;
  94     }
  95
  96     return ret;
  97 }
  98
  99 static int v9fs_string_alloc_printf(char **strp, const char *fmt, va_list ap)
 100 {
 101     va_list ap2;
 102     char *iter = (char *)fmt;
 103     int len = 0;
 104     int nr_args = 0;
 105     char *arg_char_ptr;
 106     unsigned int arg_uint;
 107
 108     /* Find the number of %'s that denotes an argument */
 109     for (iter = strstr(iter, "%"); iter; iter = strstr(iter, "%")) {
 110         nr_args++;
 111         iter++;
 112     }
 113
 114     len = strlen(fmt) - 2*nr_args;
 115
 116     if (!nr_args) {
 117         goto alloc_print;
 118     }
 119
 120     va_copy(ap2, ap);
 121
 122     iter = (char *)fmt;
 123
 124     /* Now parse the format string */
 125     for (iter = strstr(iter, "%"); iter; iter = strstr(iter, "%")) {
 126         iter++;
 127         switch (*iter) {
 128         case 'u':
 129             arg_uint = va_arg(ap2, unsigned int);
 130             len += number_to_string((void *)&arg_uint, 'u');
 131             break;
 132         case 's':
 133             arg_char_ptr = va_arg(ap2, char *);
 134             len += strlen(arg_char_ptr);
 135             break;
 136         case 'c':
 137             len += 1;
 138             break;
 139         default:
 140             fprintf(stderr,
 141                     "v9fs_string_alloc_printf:Incorrect format %c", *iter);
 142             return -1;
 143         }
 144         iter++;
 145     }
 146
 147 alloc_print:
 148     *strp = qemu_malloc((len + 1) * sizeof(**strp));
 149
 150     return vsprintf(*strp, fmt, ap);
 151 }
 152
 153 static void v9fs_string_sprintf(V9fsString *str, const char *fmt, ...)
 154 {
 155     va_list ap;
 156     int err;
 157
 158     v9fs_string_free(str);
 159
 160     va_start(ap, fmt);
 161     err = v9fs_string_alloc_printf(&str->data, fmt, ap);
 162     BUG_ON(err == -1);
 163     va_end(ap);
 164
 165     str->size = err;
 166 }
 167
 168 static void v9fs_string_copy(V9fsString *lhs, V9fsString *rhs)
 169 {
 170     v9fs_string_free(lhs);
 171     v9fs_string_sprintf(lhs, "%s", rhs->data);
 172 }
 173
 174 static size_t v9fs_string_size(V9fsString *str)
 175 {
 176     return str->size;
 177 }
 178
 179 static V9fsFidState *lookup_fid(V9fsState *s, int32_t fid)
 180 {
 181     V9fsFidState *f;
 182
 183     for (f = s->fid_list; f; f = f->next) {
 184         if (f->fid == fid) {
 185             v9fs_do_setuid(s, f->uid);
 186             return f;
 187         }
 188     }
 189
 190     return NULL;
 191 }
 192
 193 static V9fsFidState *alloc_fid(V9fsState *s, int32_t fid)
 194 {
 195     V9fsFidState *f;
 196
 197     f = lookup_fid(s, fid);
 198     if (f) {
 199         return NULL;
 200     }
 201
 202     f = qemu_mallocz(sizeof(V9fsFidState));
 203
 204     f->fid = fid;
 205     f->fd = -1;
 206     f->dir = NULL;
 207
 208     f->next = s->fid_list;
 209     s->fid_list = f;
 210
 211     return f;
 212 }
 213
 214 static int free_fid(V9fsState *s, int32_t fid)
 215 {
 216     V9fsFidState **fidpp, *fidp;
 217
 218     for (fidpp = &s->fid_list; *fidpp; fidpp = &(*fidpp)->next) {
 219         if ((*fidpp)->fid == fid) {
 220             break;
 221         }
 222     }
 223
 224     if (*fidpp == NULL) {
 225         return -ENOENT;
 226     }
 227
 228     fidp = *fidpp;
 229     *fidpp = fidp->next;
 230
 231     if (fidp->fd != -1) {
 232         v9fs_do_close(s, fidp->fd);
 233     }
 234     if (fidp->dir) {
 235         v9fs_do_closedir(s, fidp->dir);
 236     }
 237     v9fs_string_free(&fidp->path);
 238     qemu_free(fidp);
 239
 240     return 0;
 241 }
 242
 243 #define P9_QID_TYPE_DIR         0x80
 244 #define P9_QID_TYPE_SYMLINK     0x02
 245
 246 #define P9_STAT_MODE_DIR        0x80000000
 247 #define P9_STAT_MODE_APPEND     0x40000000
 248 #define P9_STAT_MODE_EXCL       0x20000000
 249 #define P9_STAT_MODE_MOUNT      0x10000000
 250 #define P9_STAT_MODE_AUTH       0x08000000
 251 #define P9_STAT_MODE_TMP        0x04000000
 252 #define P9_STAT_MODE_SYMLINK    0x02000000
 253 #define P9_STAT_MODE_LINK       0x01000000
 254 #define P9_STAT_MODE_DEVICE     0x00800000
 255 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
 256 #define P9_STAT_MODE_SOCKET     0x00100000
 257 #define P9_STAT_MODE_SETUID     0x00080000
 258 #define P9_STAT_MODE_SETGID     0x00040000
 259 #define P9_STAT_MODE_SETVTX     0x00010000
 260
 261 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR |          \
 262                                 P9_STAT_MODE_SYMLINK |      \
 263                                 P9_STAT_MODE_LINK |         \
 264                                 P9_STAT_MODE_DEVICE |       \
 265                                 P9_STAT_MODE_NAMED_PIPE |   \
 266                                 P9_STAT_MODE_SOCKET)
 267
 268 /* This is the algorithm from ufs in spfs */
 269 static void stat_to_qid(const struct stat *stbuf, V9fsQID *qidp)
 270 {
 271     size_t size;
 272
 273     size = MIN(sizeof(stbuf->st_ino), sizeof(qidp->path));
 274     memcpy(&qidp->path, &stbuf->st_ino, size);
 275     qidp->version = stbuf->st_mtime ^ (stbuf->st_size << 8);
 276     qidp->type = 0;
 277     if (S_ISDIR(stbuf->st_mode)) {
 278         qidp->type |= P9_QID_TYPE_DIR;
 279     }
 280     if (S_ISLNK(stbuf->st_mode)) {
 281         qidp->type |= P9_QID_TYPE_SYMLINK;
 282     }
 283 }
 284
 285 static int fid_to_qid(V9fsState *s, V9fsFidState *fidp, V9fsQID *qidp)
 286 {
 287     struct stat stbuf;
 288     int err;
 289
 290     err = v9fs_do_lstat(s, &fidp->path, &stbuf);
 291     if (err) {
 292         return err;
 293     }
 294
 295     stat_to_qid(&stbuf, qidp);
 296     return 0;
 297 }
 298
 299 static V9fsPDU *alloc_pdu(V9fsState *s)
 300 {
 301     V9fsPDU *pdu = NULL;
 302
 303     if (!QLIST_EMPTY(&s->free_list)) {
 304         pdu = QLIST_FIRST(&s->free_list);
 305         QLIST_REMOVE(pdu, next);
 306     }
 307     return pdu;
 308 }
 309
 310 static void free_pdu(V9fsState *s, V9fsPDU *pdu)
 311 {
 312     if (pdu) {
 313         QLIST_INSERT_HEAD(&s->free_list, pdu, next);
 314     }
 315 }
 316
 317 size_t pdu_packunpack(void *addr, struct iovec *sg, int sg_count,
 318                         size_t offset, size_t size, int pack)
 319 {
 320     int i = 0;
 321     size_t copied = 0;
 322
 323     for (i = 0; size && i < sg_count; i++) {
 324         size_t len;
 325         if (offset >= sg[i].iov_len) {
 326             /* skip this sg */
 327             offset -= sg[i].iov_len;
 328             continue;
 329         } else {
 330             len = MIN(sg[i].iov_len - offset, size);
 331             if (pack) {
 332                 memcpy(sg[i].iov_base + offset, addr, len);
 333             } else {
 334                 memcpy(addr, sg[i].iov_base + offset, len);
 335             }
 336             size -= len;
 337             copied += len;
 338             addr += len;
 339             if (size) {
 340                 offset = 0;
 341                 continue;
 342             }
 343         }
 344     }
 345
 346     return copied;
 347 }
 348
 349 static size_t pdu_unpack(void *dst, V9fsPDU *pdu, size_t offset, size_t size)
 350 {
 351     return pdu_packunpack(dst, pdu->elem.out_sg, pdu->elem.out_num,
 352                          offset, size, 0);
 353 }
 354
 355 static size_t pdu_pack(V9fsPDU *pdu, size_t offset, const void *src,
 356                         size_t size)
 357 {
 358     return pdu_packunpack((void *)src, pdu->elem.in_sg, pdu->elem.in_num,
 359                              offset, size, 1);
 360 }
 361
 362 static int pdu_copy_sg(V9fsPDU *pdu, size_t offset, int rx, struct iovec *sg)
 363 {
 364     size_t pos = 0;
 365     int i, j;
 366     struct iovec *src_sg;
 367     unsigned int num;
 368
 369     if (rx) {
 370         src_sg = pdu->elem.in_sg;
 371         num = pdu->elem.in_num;
 372     } else {
 373         src_sg = pdu->elem.out_sg;
 374         num = pdu->elem.out_num;
 375     }
 376
 377     j = 0;
 378     for (i = 0; i < num; i++) {
 379         if (offset <= pos) {
 380             sg[j].iov_base = src_sg[i].iov_base;
 381             sg[j].iov_len = src_sg[i].iov_len;
 382             j++;
 383         } else if (offset < (src_sg[i].iov_len + pos)) {
 384             sg[j].iov_base = src_sg[i].iov_base;
 385             sg[j].iov_len = src_sg[i].iov_len;
 386             sg[j].iov_base += (offset - pos);
 387             sg[j].iov_len -= (offset - pos);
 388             j++;
 389         }
 390         pos += src_sg[i].iov_len;
 391     }
 392
 393     return j;
 394 }
 395
 396 static size_t pdu_unmarshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
 397 {
 398     size_t old_offset = offset;
 399     va_list ap;
 400     int i;
 401
 402     va_start(ap, fmt);
 403     for (i = 0; fmt[i]; i++) {
 404         switch (fmt[i]) {
 405         case 'b': {
 406             uint8_t *valp = va_arg(ap, uint8_t *);
 407             offset += pdu_unpack(valp, pdu, offset, sizeof(*valp));
 408             break;
 409         }
 410         case 'w': {
 411             uint16_t val, *valp;
 412             valp = va_arg(ap, uint16_t *);
 413             val = le16_to_cpupu(valp);
 414             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 415             *valp = val;
 416             break;
 417         }
 418         case 'd': {
 419             uint32_t val, *valp;
 420             valp = va_arg(ap, uint32_t *);
 421             val = le32_to_cpupu(valp);
 422             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 423             *valp = val;
 424             break;
 425         }
 426         case 'q': {
 427             uint64_t val, *valp;
 428             valp = va_arg(ap, uint64_t *);
 429             val = le64_to_cpup(valp);
 430             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 431             *valp = val;
 432             break;
 433         }
 434         case 'v': {
 435             struct iovec *iov = va_arg(ap, struct iovec *);
 436             int *iovcnt = va_arg(ap, int *);
 437             *iovcnt = pdu_copy_sg(pdu, offset, 0, iov);
 438             break;
 439         }
 440         case 's': {
 441             V9fsString *str = va_arg(ap, V9fsString *);
 442             offset += pdu_unmarshal(pdu, offset, "w", &str->size);
 443             /* FIXME: sanity check str->size */
 444             str->data = qemu_malloc(str->size + 1);
 445             offset += pdu_unpack(str->data, pdu, offset, str->size);
 446             str->data[str->size] = 0;
 447             break;
 448         }
 449         case 'Q': {
 450             V9fsQID *qidp = va_arg(ap, V9fsQID *);
 451             offset += pdu_unmarshal(pdu, offset, "bdq",
 452                         &qidp->type, &qidp->version, &qidp->path);
 453             break;
 454         }
 455         case 'S': {
 456             V9fsStat *statp = va_arg(ap, V9fsStat *);
 457             offset += pdu_unmarshal(pdu, offset, "wwdQdddqsssssddd",
 458                         &statp->size, &statp->type, &statp->dev,
 459                         &statp->qid, &statp->mode, &statp->atime,
 460                         &statp->mtime, &statp->length,
 461                         &statp->name, &statp->uid, &statp->gid,
 462                         &statp->muid, &statp->extension,
 463                         &statp->n_uid, &statp->n_gid,
 464                         &statp->n_muid);
 465             break;
 466         }
 467         default:
 468             break;
 469         }
 470     }
 471
 472     va_end(ap);
 473
 474     return offset - old_offset;
 475 }
 476
 477 static size_t pdu_marshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
 478 {
 479     size_t old_offset = offset;
 480     va_list ap;
 481     int i;
 482
 483     va_start(ap, fmt);
 484     for (i = 0; fmt[i]; i++) {
 485         switch (fmt[i]) {
 486         case 'b': {
 487             uint8_t val = va_arg(ap, int);
 488             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 489             break;
 490         }
 491         case 'w': {
 492             uint16_t val;
 493             cpu_to_le16w(&val, va_arg(ap, int));
 494             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 495             break;
 496         }
 497         case 'd': {
 498             uint32_t val;
 499             cpu_to_le32w(&val, va_arg(ap, uint32_t));
 500             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 501             break;
 502         }
 503         case 'q': {
 504             uint64_t val;
 505             cpu_to_le64w(&val, va_arg(ap, uint64_t));
 506             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 507             break;
 508         }
 509         case 'v': {
 510             struct iovec *iov = va_arg(ap, struct iovec *);
 511             int *iovcnt = va_arg(ap, int *);
 512             *iovcnt = pdu_copy_sg(pdu, offset, 1, iov);
 513             break;
 514         }
 515         case 's': {
 516             V9fsString *str = va_arg(ap, V9fsString *);
 517             offset += pdu_marshal(pdu, offset, "w", str->size);
 518             offset += pdu_pack(pdu, offset, str->data, str->size);
 519             break;
 520         }
 521         case 'Q': {
 522             V9fsQID *qidp = va_arg(ap, V9fsQID *);
 523             offset += pdu_marshal(pdu, offset, "bdq",
 524                         qidp->type, qidp->version, qidp->path);
 525             break;
 526         }
 527         case 'S': {
 528             V9fsStat *statp = va_arg(ap, V9fsStat *);
 529             offset += pdu_marshal(pdu, offset, "wwdQdddqsssssddd",
 530                         statp->size, statp->type, statp->dev,
 531                         &statp->qid, statp->mode, statp->atime,
 532                         statp->mtime, statp->length, &statp->name,
 533                         &statp->uid, &statp->gid, &statp->muid,
 534                         &statp->extension, statp->n_uid,
 535                         statp->n_gid, statp->n_muid);
 536             break;
 537         }
 538         default:
 539             break;
 540         }
 541     }
 542     va_end(ap);
 543
 544     return offset - old_offset;
 545 }
 546
 547 static void complete_pdu(V9fsState *s, V9fsPDU *pdu, ssize_t len)
 548 {
 549     int8_t id = pdu->id + 1; /* Response */
 550
 551     if (len < 0) {
 552         V9fsString str;
 553         int err = -len;
 554
 555         str.data = strerror(err);
 556         str.size = strlen(str.data);
 557
 558         len = 7;
 559         len += pdu_marshal(pdu, len, "s", &str);
 560         if (dotu) {
 561             len += pdu_marshal(pdu, len, "d", err);
 562         }
 563
 564         id = P9_RERROR;
 565     }
 566
 567     /* fill out the header */
 568     pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag);
 569
 570     /* keep these in sync */
 571     pdu->size = len;
 572     pdu->id = id;
 573
 574     /* push onto queue and notify */
 575     virtqueue_push(s->vq, &pdu->elem, len);
 576
 577     /* FIXME: we should batch these completions */
 578     virtio_notify(&s->vdev, s->vq);
 579
 580     free_pdu(s, pdu);
 581 }
 582
 583 static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
 584 {
 585     mode_t ret;
 586
 587     ret = mode & 0777;
 588     if (mode & P9_STAT_MODE_DIR) {
 589         ret |= S_IFDIR;
 590     }
 591
 592     if (dotu) {
 593         if (mode & P9_STAT_MODE_SYMLINK) {
 594             ret |= S_IFLNK;
 595         }
 596         if (mode & P9_STAT_MODE_SOCKET) {
 597             ret |= S_IFSOCK;
 598         }
 599         if (mode & P9_STAT_MODE_NAMED_PIPE) {
 600             ret |= S_IFIFO;
 601         }
 602         if (mode & P9_STAT_MODE_DEVICE) {
 603             if (extension && extension->data[0] == 'c') {
 604                 ret |= S_IFCHR;
 605             } else {
 606                 ret |= S_IFBLK;
 607             }
 608         }
 609     }
 610
 611     if (!(ret&~0777)) {
 612         ret |= S_IFREG;
 613     }
 614
 615     if (mode & P9_STAT_MODE_SETUID) {
 616         ret |= S_ISUID;
 617     }
 618     if (mode & P9_STAT_MODE_SETGID) {
 619         ret |= S_ISGID;
 620     }
 621     if (mode & P9_STAT_MODE_SETVTX) {
 622         ret |= S_ISVTX;
 623     }
 624
 625     return ret;
 626 }
 627
 628 static int donttouch_stat(V9fsStat *stat)
 629 {
 630     if (stat->type == -1 &&
 631         stat->dev == -1 &&
 632         stat->qid.type == -1 &&
 633         stat->qid.version == -1 &&
 634         stat->qid.path == -1 &&
 635         stat->mode == -1 &&
 636         stat->atime == -1 &&
 637         stat->mtime == -1 &&
 638         stat->length == -1 &&
 639         !stat->name.size &&
 640         !stat->uid.size &&
 641         !stat->gid.size &&
 642         !stat->muid.size &&
 643         stat->n_uid == -1 &&
 644         stat->n_gid == -1 &&
 645         stat->n_muid == -1) {
 646         return 1;
 647     }
 648
 649     return 0;
 650 }
 651
 652 static void v9fs_stat_free(V9fsStat *stat)
 653 {
 654     v9fs_string_free(&stat->name);
 655     v9fs_string_free(&stat->uid);
 656     v9fs_string_free(&stat->gid);
 657     v9fs_string_free(&stat->muid);
 658     v9fs_string_free(&stat->extension);
 659 }
 660
 661 static uint32_t stat_to_v9mode(const struct stat *stbuf)
 662 {
 663     uint32_t mode;
 664
 665     mode = stbuf->st_mode & 0777;
 666     if (S_ISDIR(stbuf->st_mode)) {
 667         mode |= P9_STAT_MODE_DIR;
 668     }
 669
 670     if (dotu) {
 671         if (S_ISLNK(stbuf->st_mode)) {
 672             mode |= P9_STAT_MODE_SYMLINK;
 673         }
 674
 675         if (S_ISSOCK(stbuf->st_mode)) {
 676             mode |= P9_STAT_MODE_SOCKET;
 677         }
 678
 679         if (S_ISFIFO(stbuf->st_mode)) {
 680             mode |= P9_STAT_MODE_NAMED_PIPE;
 681         }
 682
 683         if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
 684             mode |= P9_STAT_MODE_DEVICE;
 685         }
 686
 687         if (stbuf->st_mode & S_ISUID) {
 688             mode |= P9_STAT_MODE_SETUID;
 689         }
 690
 691         if (stbuf->st_mode & S_ISGID) {
 692             mode |= P9_STAT_MODE_SETGID;
 693         }
 694
 695         if (stbuf->st_mode & S_ISVTX) {
 696             mode |= P9_STAT_MODE_SETVTX;
 697         }
 698     }
 699
 700     return mode;
 701 }
 702
 703 static int stat_to_v9stat(V9fsState *s, V9fsString *name,
 704                             const struct stat *stbuf,
 705                             V9fsStat *v9stat)
 706 {
 707     int err;
 708     const char *str;
 709
 710     memset(v9stat, 0, sizeof(*v9stat));
 711
 712     stat_to_qid(stbuf, &v9stat->qid);
 713     v9stat->mode = stat_to_v9mode(stbuf);
 714     v9stat->atime = stbuf->st_atime;
 715     v9stat->mtime = stbuf->st_mtime;
 716     v9stat->length = stbuf->st_size;
 717
 718     v9fs_string_null(&v9stat->uid);
 719     v9fs_string_null(&v9stat->gid);
 720     v9fs_string_null(&v9stat->muid);
 721
 722     if (dotu) {
 723         v9stat->n_uid = stbuf->st_uid;
 724         v9stat->n_gid = stbuf->st_gid;
 725         v9stat->n_muid = 0;
 726
 727         v9fs_string_null(&v9stat->extension);
 728
 729         if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
 730             err = v9fs_do_readlink(s, name, &v9stat->extension);
 731             if (err == -1) {
 732                 err = -errno;
 733                 return err;
 734             }
 735             v9stat->extension.data[err] = 0;
 736             v9stat->extension.size = err;
 737         } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
 738             v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
 739                     S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
 740                     major(stbuf->st_rdev), minor(stbuf->st_rdev));
 741         } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
 742             v9fs_string_sprintf(&v9stat->extension, "%s %u",
 743                     "HARDLINKCOUNT", stbuf->st_nlink);
 744         }
 745     }
 746
 747     str = strrchr(name->data, '/');
 748     if (str) {
 749         str += 1;
 750     } else {
 751         str = name->data;
 752     }
 753
 754     v9fs_string_sprintf(&v9stat->name, "%s", str);
 755
 756     v9stat->size = 61 +
 757         v9fs_string_size(&v9stat->name) +
 758         v9fs_string_size(&v9stat->uid) +
 759         v9fs_string_size(&v9stat->gid) +
 760         v9fs_string_size(&v9stat->muid) +
 761         v9fs_string_size(&v9stat->extension);
 762     return 0;
 763 }
 764
 765 static struct iovec *adjust_sg(struct iovec *sg, int len, int *iovcnt)
 766 {
 767     while (len && *iovcnt) {
 768         if (len < sg->iov_len) {
 769             sg->iov_len -= len;
 770             sg->iov_base += len;
 771             len = 0;
 772         } else {
 773             len -= sg->iov_len;
 774             sg++;
 775             *iovcnt -= 1;
 776         }
 777     }
 778
 779     return sg;
 780 }
 781
 782 static struct iovec *cap_sg(struct iovec *sg, int cap, int *cnt)
 783 {
 784     int i;
 785     int total = 0;
 786
 787     for (i = 0; i < *cnt; i++) {
 788         if ((total + sg[i].iov_len) > cap) {
 789             sg[i].iov_len -= ((total + sg[i].iov_len) - cap);
 790             i++;
 791             break;
 792         }
 793         total += sg[i].iov_len;
 794     }
 795
 796     *cnt = i;
 797
 798     return sg;
 799 }
 800
 801 static void print_sg(struct iovec *sg, int cnt)
 802 {
 803     int i;
 804
 805     printf("sg[%d]: {", cnt);
 806     for (i = 0; i < cnt; i++) {
 807         if (i) {
 808             printf(", ");
 809         }
 810         printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
 811     }
 812     printf("}\n");
 813 }
 814
 815 static void v9fs_dummy(V9fsState *s, V9fsPDU *pdu)
 816 {
 817     /* Note: The following have been added to prevent GCC from complaining
 818      * They will be removed in the subsequent patches */
 819     (void)pdu_unmarshal;
 820     (void) complete_pdu;
 821     (void) v9fs_string_init;
 822     (void) v9fs_string_free;
 823     (void) v9fs_string_null;
 824     (void) v9fs_string_sprintf;
 825     (void) v9fs_string_copy;
 826     (void) v9fs_string_size;
 827     (void) v9fs_do_lstat;
 828     (void) v9fs_do_setuid;
 829     (void) v9fs_do_readlink;
 830     (void) v9fs_do_close;
 831     (void) v9fs_do_closedir;
 832     (void) alloc_fid;
 833     (void) free_fid;
 834     (void) fid_to_qid;
 835     (void) v9mode_to_mode;
 836     (void) donttouch_stat;
 837     (void) v9fs_stat_free;
 838     (void) stat_to_v9stat;
 839     (void) adjust_sg;
 840     (void) cap_sg;
 841     (void) print_sg;
 842 }
 843
 844 static void v9fs_version(V9fsState *s, V9fsPDU *pdu)
 845 {
 846     int32_t msize;
 847     V9fsString version;
 848     size_t offset = 7;
 849
 850     pdu_unmarshal(pdu, offset, "ds", &msize, &version);
 851
 852     if (strcmp(version.data, "9P2000.u")) {
 853         v9fs_string_sprintf(&version, "unknown");
 854     }
 855
 856     offset += pdu_marshal(pdu, offset, "ds", msize, &version);
 857     complete_pdu(s, pdu, offset);
 858
 859     v9fs_string_free(&version);
 860 }
 861
 862 static void v9fs_attach(V9fsState *s, V9fsPDU *pdu)
 863 {
 864     int32_t fid, afid, n_uname;
 865     V9fsString uname, aname;
 866     V9fsFidState *fidp;
 867     V9fsQID qid;
 868     size_t offset = 7;
 869     ssize_t err;
 870
 871     pdu_unmarshal(pdu, offset, "ddssd", &fid, &afid, &uname, &aname, &n_uname);
 872
 873     fidp = alloc_fid(s, fid);
 874     if (fidp == NULL) {
 875         err = -EINVAL;
 876         goto out;
 877     }
 878
 879     fidp->uid = n_uname;
 880
 881     v9fs_string_sprintf(&fidp->path, "%s", "/");
 882     err = fid_to_qid(s, fidp, &qid);
 883     if (err) {
 884         err = -EINVAL;
 885         free_fid(s, fid);
 886         goto out;
 887     }
 888
 889     offset += pdu_marshal(pdu, offset, "Q", &qid);
 890
 891     err = offset;
 892 out:
 893     complete_pdu(s, pdu, err);
 894     v9fs_string_free(&uname);
 895     v9fs_string_free(&aname);
 896 }
 897
 898 typedef struct V9fsStatState {
 899     V9fsPDU *pdu;
 900     size_t offset;
 901     V9fsStat v9stat;
 902     V9fsFidState *fidp;
 903     struct stat stbuf;
 904 } V9fsStatState;
 905
 906 static void v9fs_stat_post_lstat(V9fsState *s, V9fsStatState *vs, int err)
 907 {
 908     if (err == -1) {
 909         err = -errno;
 910         goto out;
 911     }
 912
 913     err = stat_to_v9stat(s, &vs->fidp->path, &vs->stbuf, &vs->v9stat);
 914     if (err) {
 915         goto out;
 916     }
 917     vs->offset += pdu_marshal(vs->pdu, vs->offset, "wS", 0, &vs->v9stat);
 918     err = vs->offset;
 919
 920 out:
 921     complete_pdu(s, vs->pdu, err);
 922     v9fs_stat_free(&vs->v9stat);
 923     qemu_free(vs);
 924 }
 925
 926 static void v9fs_stat(V9fsState *s, V9fsPDU *pdu)
 927 {
 928     int32_t fid;
 929     V9fsStatState *vs;
 930     ssize_t err = 0;
 931
 932     vs = qemu_malloc(sizeof(*vs));
 933     vs->pdu = pdu;
 934     vs->offset = 7;
 935
 936     memset(&vs->v9stat, 0, sizeof(vs->v9stat));
 937
 938     pdu_unmarshal(vs->pdu, vs->offset, "d", &fid);
 939
 940     vs->fidp = lookup_fid(s, fid);
 941     if (vs->fidp == NULL) {
 942         err = -ENOENT;
 943         goto out;
 944     }
 945
 946     err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
 947     v9fs_stat_post_lstat(s, vs, err);
 948     return;
 949
 950 out:
 951     complete_pdu(s, vs->pdu, err);
 952     v9fs_stat_free(&vs->v9stat);
 953     qemu_free(vs);
 954 }
 955
 956 static void v9fs_walk(V9fsState *s, V9fsPDU *pdu)
 957 {
 958     if (debug_9p_pdu) {
 959         pprint_pdu(pdu);
 960     }
 961 }
 962
 963 static void v9fs_clunk(V9fsState *s, V9fsPDU *pdu)
 964 {
 965     if (debug_9p_pdu) {
 966         pprint_pdu(pdu);
 967     }
 968 }
 969
 970 static void v9fs_open(V9fsState *s, V9fsPDU *pdu)
 971 {    if (debug_9p_pdu) {
 972         pprint_pdu(pdu);
 973      }
 974 }
 975
 976 static void v9fs_read(V9fsState *s, V9fsPDU *pdu)
 977 {
 978     if (debug_9p_pdu) {
 979         pprint_pdu(pdu);
 980     }
 981 }
 982
 983 static void v9fs_write(V9fsState *s, V9fsPDU *pdu)
 984 {
 985     if (debug_9p_pdu) {
 986         pprint_pdu(pdu);
 987     }
 988 }
 989
 990 static void v9fs_create(V9fsState *s, V9fsPDU *pdu)
 991 {
 992     if (debug_9p_pdu) {
 993         pprint_pdu(pdu);
 994     }
 995 }
 996
 997 static void v9fs_flush(V9fsState *s, V9fsPDU *pdu)
 998 {
 999     v9fs_dummy(s, pdu);
1000     if (debug_9p_pdu) {
1001         pprint_pdu(pdu);
1002     }
1003 }
1004
1005 static void v9fs_remove(V9fsState *s, V9fsPDU *pdu)
1006 {
1007     if (debug_9p_pdu) {
1008         pprint_pdu(pdu);
1009     }
1010 }
1011
1012 static void v9fs_wstat(V9fsState *s, V9fsPDU *pdu)
1013 {
1014     if (debug_9p_pdu) {
1015         pprint_pdu(pdu);
1016     }
1017 }
1018
1019 typedef void (pdu_handler_t)(V9fsState *s, V9fsPDU *pdu);
1020
1021 static pdu_handler_t *pdu_handlers[] = {
1022     [P9_TVERSION] = v9fs_version,
1023     [P9_TATTACH] = v9fs_attach,
1024     [P9_TSTAT] = v9fs_stat,
1025     [P9_TWALK] = v9fs_walk,
1026     [P9_TCLUNK] = v9fs_clunk,
1027     [P9_TOPEN] = v9fs_open,
1028     [P9_TREAD] = v9fs_read,
1029 #if 0
1030     [P9_TAUTH] = v9fs_auth,
1031 #endif
1032     [P9_TFLUSH] = v9fs_flush,
1033     [P9_TCREATE] = v9fs_create,
1034     [P9_TWRITE] = v9fs_write,
1035     [P9_TWSTAT] = v9fs_wstat,
1036     [P9_TREMOVE] = v9fs_remove,
1037 };
1038
1039 static void submit_pdu(V9fsState *s, V9fsPDU *pdu)
1040 {
1041     pdu_handler_t *handler;
1042
1043     if (debug_9p_pdu) {
1044         pprint_pdu(pdu);
1045     }
1046
1047     BUG_ON(pdu->id >= ARRAY_SIZE(pdu_handlers));
1048
1049     handler = pdu_handlers[pdu->id];
1050     BUG_ON(handler == NULL);
1051
1052     handler(s, pdu);
1053 }
1054
1055 static void handle_9p_output(VirtIODevice *vdev, VirtQueue *vq)
1056 {
1057     V9fsState *s = (V9fsState *)vdev;
1058     V9fsPDU *pdu;
1059     ssize_t len;
1060
1061     while ((pdu = alloc_pdu(s)) &&
1062             (len = virtqueue_pop(vq, &pdu->elem)) != 0) {
1063         uint8_t *ptr;
1064
1065         BUG_ON(pdu->elem.out_num == 0 || pdu->elem.in_num == 0);
1066         BUG_ON(pdu->elem.out_sg[0].iov_len < 7);
1067
1068         ptr = pdu->elem.out_sg[0].iov_base;
1069
1070         memcpy(&pdu->size, ptr, 4);
1071         pdu->id = ptr[4];
1072         memcpy(&pdu->tag, ptr + 5, 2);
1073
1074         submit_pdu(s, pdu);
1075     }
1076
1077     free_pdu(s, pdu);
1078 }
1079
1080 static uint32_t virtio_9p_get_features(VirtIODevice *vdev, uint32_t features)
1081 {
1082     features |= 1 << VIRTIO_9P_MOUNT_TAG;
1083     return features;
1084 }
1085
1086 static V9fsState *to_virtio_9p(VirtIODevice *vdev)
1087 {
1088     return (V9fsState *)vdev;
1089 }
1090
1091 static void virtio_9p_get_config(VirtIODevice *vdev, uint8_t *config)
1092 {
1093     struct virtio_9p_config *cfg;
1094     V9fsState *s = to_virtio_9p(vdev);
1095
1096     cfg = qemu_mallocz(sizeof(struct virtio_9p_config) +
1097                         s->tag_len);
1098     stw_raw(&cfg->tag_len, s->tag_len);
1099     memcpy(cfg->tag, s->tag, s->tag_len);
1100     memcpy(config, cfg, s->config_size);
1101     qemu_free(cfg);
1102 }
1103
1104 VirtIODevice *virtio_9p_init(DeviceState *dev, V9fsConf *conf)
1105  {
1106     V9fsState *s;
1107     int i, len;
1108     struct stat stat;
1109     FsTypeEntry *fse;
1110
1111
1112     s = (V9fsState *)virtio_common_init("virtio-9p",
1113                                     VIRTIO_ID_9P,
1114                                     sizeof(struct virtio_9p_config)+
1115                                     MAX_TAG_LEN,
1116                                     sizeof(V9fsState));
1117
1118     /* initialize pdu allocator */
1119     QLIST_INIT(&s->free_list);
1120     for (i = 0; i < (MAX_REQ - 1); i++) {
1121         QLIST_INSERT_HEAD(&s->free_list, &s->pdus[i], next);
1122     }
1123
1124     s->vq = virtio_add_queue(&s->vdev, MAX_REQ, handle_9p_output);
1125
1126     fse = get_fsdev_fsentry(conf->fsdev_id);
1127
1128     if (!fse) {
1129         /* We don't have a fsdev identified by fsdev_id */
1130         fprintf(stderr, "Virtio-9p device couldn't find fsdev "
1131                     "with the id %s\n", conf->fsdev_id);
1132         exit(1);
1133     }
1134
1135     if (!fse->path || !conf->tag) {
1136         /* we haven't specified a mount_tag or the path */
1137         fprintf(stderr, "fsdev with id %s needs path "
1138                 "and Virtio-9p device needs mount_tag arguments\n",
1139                 conf->fsdev_id);
1140         exit(1);
1141     }
1142
1143     if (lstat(fse->path, &stat)) {
1144         fprintf(stderr, "share path %s does not exist\n", fse->path);
1145         exit(1);
1146     } else if (!S_ISDIR(stat.st_mode)) {
1147         fprintf(stderr, "share path %s is not a directory \n", fse->path);
1148         exit(1);
1149     }
1150
1151     s->ctx.fs_root = qemu_strdup(fse->path);
1152     len = strlen(conf->tag);
1153     if (len > MAX_TAG_LEN) {
1154         len = MAX_TAG_LEN;
1155     }
1156     /* s->tag is non-NULL terminated string */
1157     s->tag = qemu_malloc(len);
1158     memcpy(s->tag, conf->tag, len);
1159     s->tag_len = len;
1160     s->ctx.uid = -1;
1161
1162     s->ops = fse->ops;
1163     s->vdev.get_features = virtio_9p_get_features;
1164     s->config_size = sizeof(struct virtio_9p_config) +
1165                         s->tag_len;
1166     s->vdev.get_config = virtio_9p_get_config;
1167
1168     return &s->vdev;
1169 }