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 int v9fs_do_open(V9fsState *s, V9fsString *path, int flags)
  60 {
  61     return s->ops->open(&s->ctx, path->data, flags);
  62 }
  63
  64 static DIR *v9fs_do_opendir(V9fsState *s, V9fsString *path)
  65 {
  66     return s->ops->opendir(&s->ctx, path->data);
  67 }
  68
  69 static void v9fs_do_rewinddir(V9fsState *s, DIR *dir)
  70 {
  71     return s->ops->rewinddir(&s->ctx, dir);
  72 }
  73
  74 static off_t v9fs_do_telldir(V9fsState *s, DIR *dir)
  75 {
  76     return s->ops->telldir(&s->ctx, dir);
  77 }
  78
  79 static struct dirent *v9fs_do_readdir(V9fsState *s, DIR *dir)
  80 {
  81     return s->ops->readdir(&s->ctx, dir);
  82 }
  83
  84 static void v9fs_do_seekdir(V9fsState *s, DIR *dir, off_t off)
  85 {
  86     return s->ops->seekdir(&s->ctx, dir, off);
  87 }
  88
  89 static int v9fs_do_readv(V9fsState *s, int fd, const struct iovec *iov,
  90                             int iovcnt)
  91 {
  92     return s->ops->readv(&s->ctx, fd, iov, iovcnt);
  93 }
  94
  95 static off_t v9fs_do_lseek(V9fsState *s, int fd, off_t offset, int whence)
  96 {
  97     return s->ops->lseek(&s->ctx, fd, offset, whence);
  98 }
  99
 100 static int v9fs_do_writev(V9fsState *s, int fd, const struct iovec *iov,
 101                        int iovcnt)
 102 {
 103     return s->ops->writev(&s->ctx, fd, iov, iovcnt);
 104 }
 105
 106 static int v9fs_do_chmod(V9fsState *s, V9fsString *path, mode_t mode)
 107 {
 108     return s->ops->chmod(&s->ctx, path->data, mode);
 109 }
 110
 111 static int v9fs_do_mknod(V9fsState *s, V9fsString *path, mode_t mode, dev_t dev)
 112 {
 113     return s->ops->mknod(&s->ctx, path->data, mode, dev);
 114 }
 115
 116 static int v9fs_do_mksock(V9fsState *s, V9fsString *path)
 117 {
 118     return s->ops->mksock(&s->ctx, path->data);
 119 }
 120
 121 static int v9fs_do_mkdir(V9fsState *s, V9fsString *path, mode_t mode)
 122 {
 123     return s->ops->mkdir(&s->ctx, path->data, mode);
 124 }
 125
 126 static int v9fs_do_fstat(V9fsState *s, int fd, struct stat *stbuf)
 127 {
 128     return s->ops->fstat(&s->ctx, fd, stbuf);
 129 }
 130
 131 static int v9fs_do_open2(V9fsState *s, V9fsString *path, int flags, mode_t mode)
 132 {
 133     return s->ops->open2(&s->ctx, path->data, flags, mode);
 134 }
 135
 136 static int v9fs_do_symlink(V9fsState *s, V9fsString *oldpath,
 137                             V9fsString *newpath)
 138 {
 139     return s->ops->symlink(&s->ctx, oldpath->data, newpath->data);
 140 }
 141
 142 static int v9fs_do_link(V9fsState *s, V9fsString *oldpath, V9fsString *newpath)
 143 {
 144     return s->ops->link(&s->ctx, oldpath->data, newpath->data);
 145 }
 146
 147 static void v9fs_string_init(V9fsString *str)
 148 {
 149     str->data = NULL;
 150     str->size = 0;
 151 }
 152
 153 static void v9fs_string_free(V9fsString *str)
 154 {
 155     qemu_free(str->data);
 156     str->data = NULL;
 157     str->size = 0;
 158 }
 159
 160 static void v9fs_string_null(V9fsString *str)
 161 {
 162     v9fs_string_free(str);
 163 }
 164
 165 static int number_to_string(void *arg, char type)
 166 {
 167     unsigned int ret = 0;
 168
 169     switch (type) {
 170     case 'u': {
 171         unsigned int num = *(unsigned int *)arg;
 172
 173         do {
 174             ret++;
 175             num = num/10;
 176         } while (num);
 177         break;
 178     }
 179     default:
 180         printf("Number_to_string: Unknown number format\n");
 181         return -1;
 182     }
 183
 184     return ret;
 185 }
 186
 187 static int v9fs_string_alloc_printf(char **strp, const char *fmt, va_list ap)
 188 {
 189     va_list ap2;
 190     char *iter = (char *)fmt;
 191     int len = 0;
 192     int nr_args = 0;
 193     char *arg_char_ptr;
 194     unsigned int arg_uint;
 195
 196     /* Find the number of %'s that denotes an argument */
 197     for (iter = strstr(iter, "%"); iter; iter = strstr(iter, "%")) {
 198         nr_args++;
 199         iter++;
 200     }
 201
 202     len = strlen(fmt) - 2*nr_args;
 203
 204     if (!nr_args) {
 205         goto alloc_print;
 206     }
 207
 208     va_copy(ap2, ap);
 209
 210     iter = (char *)fmt;
 211
 212     /* Now parse the format string */
 213     for (iter = strstr(iter, "%"); iter; iter = strstr(iter, "%")) {
 214         iter++;
 215         switch (*iter) {
 216         case 'u':
 217             arg_uint = va_arg(ap2, unsigned int);
 218             len += number_to_string((void *)&arg_uint, 'u');
 219             break;
 220         case 's':
 221             arg_char_ptr = va_arg(ap2, char *);
 222             len += strlen(arg_char_ptr);
 223             break;
 224         case 'c':
 225             len += 1;
 226             break;
 227         default:
 228             fprintf(stderr,
 229                     "v9fs_string_alloc_printf:Incorrect format %c", *iter);
 230             return -1;
 231         }
 232         iter++;
 233     }
 234
 235 alloc_print:
 236     *strp = qemu_malloc((len + 1) * sizeof(**strp));
 237
 238     return vsprintf(*strp, fmt, ap);
 239 }
 240
 241 static void v9fs_string_sprintf(V9fsString *str, const char *fmt, ...)
 242 {
 243     va_list ap;
 244     int err;
 245
 246     v9fs_string_free(str);
 247
 248     va_start(ap, fmt);
 249     err = v9fs_string_alloc_printf(&str->data, fmt, ap);
 250     BUG_ON(err == -1);
 251     va_end(ap);
 252
 253     str->size = err;
 254 }
 255
 256 static void v9fs_string_copy(V9fsString *lhs, V9fsString *rhs)
 257 {
 258     v9fs_string_free(lhs);
 259     v9fs_string_sprintf(lhs, "%s", rhs->data);
 260 }
 261
 262 static size_t v9fs_string_size(V9fsString *str)
 263 {
 264     return str->size;
 265 }
 266
 267 static V9fsFidState *lookup_fid(V9fsState *s, int32_t fid)
 268 {
 269     V9fsFidState *f;
 270
 271     for (f = s->fid_list; f; f = f->next) {
 272         if (f->fid == fid) {
 273             v9fs_do_setuid(s, f->uid);
 274             return f;
 275         }
 276     }
 277
 278     return NULL;
 279 }
 280
 281 static V9fsFidState *alloc_fid(V9fsState *s, int32_t fid)
 282 {
 283     V9fsFidState *f;
 284
 285     f = lookup_fid(s, fid);
 286     if (f) {
 287         return NULL;
 288     }
 289
 290     f = qemu_mallocz(sizeof(V9fsFidState));
 291
 292     f->fid = fid;
 293     f->fd = -1;
 294     f->dir = NULL;
 295
 296     f->next = s->fid_list;
 297     s->fid_list = f;
 298
 299     return f;
 300 }
 301
 302 static int free_fid(V9fsState *s, int32_t fid)
 303 {
 304     V9fsFidState **fidpp, *fidp;
 305
 306     for (fidpp = &s->fid_list; *fidpp; fidpp = &(*fidpp)->next) {
 307         if ((*fidpp)->fid == fid) {
 308             break;
 309         }
 310     }
 311
 312     if (*fidpp == NULL) {
 313         return -ENOENT;
 314     }
 315
 316     fidp = *fidpp;
 317     *fidpp = fidp->next;
 318
 319     if (fidp->fd != -1) {
 320         v9fs_do_close(s, fidp->fd);
 321     }
 322     if (fidp->dir) {
 323         v9fs_do_closedir(s, fidp->dir);
 324     }
 325     v9fs_string_free(&fidp->path);
 326     qemu_free(fidp);
 327
 328     return 0;
 329 }
 330
 331 #define P9_QID_TYPE_DIR         0x80
 332 #define P9_QID_TYPE_SYMLINK     0x02
 333
 334 #define P9_STAT_MODE_DIR        0x80000000
 335 #define P9_STAT_MODE_APPEND     0x40000000
 336 #define P9_STAT_MODE_EXCL       0x20000000
 337 #define P9_STAT_MODE_MOUNT      0x10000000
 338 #define P9_STAT_MODE_AUTH       0x08000000
 339 #define P9_STAT_MODE_TMP        0x04000000
 340 #define P9_STAT_MODE_SYMLINK    0x02000000
 341 #define P9_STAT_MODE_LINK       0x01000000
 342 #define P9_STAT_MODE_DEVICE     0x00800000
 343 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
 344 #define P9_STAT_MODE_SOCKET     0x00100000
 345 #define P9_STAT_MODE_SETUID     0x00080000
 346 #define P9_STAT_MODE_SETGID     0x00040000
 347 #define P9_STAT_MODE_SETVTX     0x00010000
 348
 349 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR |          \
 350                                 P9_STAT_MODE_SYMLINK |      \
 351                                 P9_STAT_MODE_LINK |         \
 352                                 P9_STAT_MODE_DEVICE |       \
 353                                 P9_STAT_MODE_NAMED_PIPE |   \
 354                                 P9_STAT_MODE_SOCKET)
 355
 356 /* This is the algorithm from ufs in spfs */
 357 static void stat_to_qid(const struct stat *stbuf, V9fsQID *qidp)
 358 {
 359     size_t size;
 360
 361     size = MIN(sizeof(stbuf->st_ino), sizeof(qidp->path));
 362     memcpy(&qidp->path, &stbuf->st_ino, size);
 363     qidp->version = stbuf->st_mtime ^ (stbuf->st_size << 8);
 364     qidp->type = 0;
 365     if (S_ISDIR(stbuf->st_mode)) {
 366         qidp->type |= P9_QID_TYPE_DIR;
 367     }
 368     if (S_ISLNK(stbuf->st_mode)) {
 369         qidp->type |= P9_QID_TYPE_SYMLINK;
 370     }
 371 }
 372
 373 static int fid_to_qid(V9fsState *s, V9fsFidState *fidp, V9fsQID *qidp)
 374 {
 375     struct stat stbuf;
 376     int err;
 377
 378     err = v9fs_do_lstat(s, &fidp->path, &stbuf);
 379     if (err) {
 380         return err;
 381     }
 382
 383     stat_to_qid(&stbuf, qidp);
 384     return 0;
 385 }
 386
 387 static V9fsPDU *alloc_pdu(V9fsState *s)
 388 {
 389     V9fsPDU *pdu = NULL;
 390
 391     if (!QLIST_EMPTY(&s->free_list)) {
 392         pdu = QLIST_FIRST(&s->free_list);
 393         QLIST_REMOVE(pdu, next);
 394     }
 395     return pdu;
 396 }
 397
 398 static void free_pdu(V9fsState *s, V9fsPDU *pdu)
 399 {
 400     if (pdu) {
 401         QLIST_INSERT_HEAD(&s->free_list, pdu, next);
 402     }
 403 }
 404
 405 size_t pdu_packunpack(void *addr, struct iovec *sg, int sg_count,
 406                         size_t offset, size_t size, int pack)
 407 {
 408     int i = 0;
 409     size_t copied = 0;
 410
 411     for (i = 0; size && i < sg_count; i++) {
 412         size_t len;
 413         if (offset >= sg[i].iov_len) {
 414             /* skip this sg */
 415             offset -= sg[i].iov_len;
 416             continue;
 417         } else {
 418             len = MIN(sg[i].iov_len - offset, size);
 419             if (pack) {
 420                 memcpy(sg[i].iov_base + offset, addr, len);
 421             } else {
 422                 memcpy(addr, sg[i].iov_base + offset, len);
 423             }
 424             size -= len;
 425             copied += len;
 426             addr += len;
 427             if (size) {
 428                 offset = 0;
 429                 continue;
 430             }
 431         }
 432     }
 433
 434     return copied;
 435 }
 436
 437 static size_t pdu_unpack(void *dst, V9fsPDU *pdu, size_t offset, size_t size)
 438 {
 439     return pdu_packunpack(dst, pdu->elem.out_sg, pdu->elem.out_num,
 440                          offset, size, 0);
 441 }
 442
 443 static size_t pdu_pack(V9fsPDU *pdu, size_t offset, const void *src,
 444                         size_t size)
 445 {
 446     return pdu_packunpack((void *)src, pdu->elem.in_sg, pdu->elem.in_num,
 447                              offset, size, 1);
 448 }
 449
 450 static int pdu_copy_sg(V9fsPDU *pdu, size_t offset, int rx, struct iovec *sg)
 451 {
 452     size_t pos = 0;
 453     int i, j;
 454     struct iovec *src_sg;
 455     unsigned int num;
 456
 457     if (rx) {
 458         src_sg = pdu->elem.in_sg;
 459         num = pdu->elem.in_num;
 460     } else {
 461         src_sg = pdu->elem.out_sg;
 462         num = pdu->elem.out_num;
 463     }
 464
 465     j = 0;
 466     for (i = 0; i < num; i++) {
 467         if (offset <= pos) {
 468             sg[j].iov_base = src_sg[i].iov_base;
 469             sg[j].iov_len = src_sg[i].iov_len;
 470             j++;
 471         } else if (offset < (src_sg[i].iov_len + pos)) {
 472             sg[j].iov_base = src_sg[i].iov_base;
 473             sg[j].iov_len = src_sg[i].iov_len;
 474             sg[j].iov_base += (offset - pos);
 475             sg[j].iov_len -= (offset - pos);
 476             j++;
 477         }
 478         pos += src_sg[i].iov_len;
 479     }
 480
 481     return j;
 482 }
 483
 484 static size_t pdu_unmarshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
 485 {
 486     size_t old_offset = offset;
 487     va_list ap;
 488     int i;
 489
 490     va_start(ap, fmt);
 491     for (i = 0; fmt[i]; i++) {
 492         switch (fmt[i]) {
 493         case 'b': {
 494             uint8_t *valp = va_arg(ap, uint8_t *);
 495             offset += pdu_unpack(valp, pdu, offset, sizeof(*valp));
 496             break;
 497         }
 498         case 'w': {
 499             uint16_t val, *valp;
 500             valp = va_arg(ap, uint16_t *);
 501             val = le16_to_cpupu(valp);
 502             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 503             *valp = val;
 504             break;
 505         }
 506         case 'd': {
 507             uint32_t val, *valp;
 508             valp = va_arg(ap, uint32_t *);
 509             val = le32_to_cpupu(valp);
 510             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 511             *valp = val;
 512             break;
 513         }
 514         case 'q': {
 515             uint64_t val, *valp;
 516             valp = va_arg(ap, uint64_t *);
 517             val = le64_to_cpup(valp);
 518             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 519             *valp = val;
 520             break;
 521         }
 522         case 'v': {
 523             struct iovec *iov = va_arg(ap, struct iovec *);
 524             int *iovcnt = va_arg(ap, int *);
 525             *iovcnt = pdu_copy_sg(pdu, offset, 0, iov);
 526             break;
 527         }
 528         case 's': {
 529             V9fsString *str = va_arg(ap, V9fsString *);
 530             offset += pdu_unmarshal(pdu, offset, "w", &str->size);
 531             /* FIXME: sanity check str->size */
 532             str->data = qemu_malloc(str->size + 1);
 533             offset += pdu_unpack(str->data, pdu, offset, str->size);
 534             str->data[str->size] = 0;
 535             break;
 536         }
 537         case 'Q': {
 538             V9fsQID *qidp = va_arg(ap, V9fsQID *);
 539             offset += pdu_unmarshal(pdu, offset, "bdq",
 540                         &qidp->type, &qidp->version, &qidp->path);
 541             break;
 542         }
 543         case 'S': {
 544             V9fsStat *statp = va_arg(ap, V9fsStat *);
 545             offset += pdu_unmarshal(pdu, offset, "wwdQdddqsssssddd",
 546                         &statp->size, &statp->type, &statp->dev,
 547                         &statp->qid, &statp->mode, &statp->atime,
 548                         &statp->mtime, &statp->length,
 549                         &statp->name, &statp->uid, &statp->gid,
 550                         &statp->muid, &statp->extension,
 551                         &statp->n_uid, &statp->n_gid,
 552                         &statp->n_muid);
 553             break;
 554         }
 555         default:
 556             break;
 557         }
 558     }
 559
 560     va_end(ap);
 561
 562     return offset - old_offset;
 563 }
 564
 565 static size_t pdu_marshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
 566 {
 567     size_t old_offset = offset;
 568     va_list ap;
 569     int i;
 570
 571     va_start(ap, fmt);
 572     for (i = 0; fmt[i]; i++) {
 573         switch (fmt[i]) {
 574         case 'b': {
 575             uint8_t val = va_arg(ap, int);
 576             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 577             break;
 578         }
 579         case 'w': {
 580             uint16_t val;
 581             cpu_to_le16w(&val, va_arg(ap, int));
 582             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 583             break;
 584         }
 585         case 'd': {
 586             uint32_t val;
 587             cpu_to_le32w(&val, va_arg(ap, uint32_t));
 588             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 589             break;
 590         }
 591         case 'q': {
 592             uint64_t val;
 593             cpu_to_le64w(&val, va_arg(ap, uint64_t));
 594             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 595             break;
 596         }
 597         case 'v': {
 598             struct iovec *iov = va_arg(ap, struct iovec *);
 599             int *iovcnt = va_arg(ap, int *);
 600             *iovcnt = pdu_copy_sg(pdu, offset, 1, iov);
 601             break;
 602         }
 603         case 's': {
 604             V9fsString *str = va_arg(ap, V9fsString *);
 605             offset += pdu_marshal(pdu, offset, "w", str->size);
 606             offset += pdu_pack(pdu, offset, str->data, str->size);
 607             break;
 608         }
 609         case 'Q': {
 610             V9fsQID *qidp = va_arg(ap, V9fsQID *);
 611             offset += pdu_marshal(pdu, offset, "bdq",
 612                         qidp->type, qidp->version, qidp->path);
 613             break;
 614         }
 615         case 'S': {
 616             V9fsStat *statp = va_arg(ap, V9fsStat *);
 617             offset += pdu_marshal(pdu, offset, "wwdQdddqsssssddd",
 618                         statp->size, statp->type, statp->dev,
 619                         &statp->qid, statp->mode, statp->atime,
 620                         statp->mtime, statp->length, &statp->name,
 621                         &statp->uid, &statp->gid, &statp->muid,
 622                         &statp->extension, statp->n_uid,
 623                         statp->n_gid, statp->n_muid);
 624             break;
 625         }
 626         default:
 627             break;
 628         }
 629     }
 630     va_end(ap);
 631
 632     return offset - old_offset;
 633 }
 634
 635 static void complete_pdu(V9fsState *s, V9fsPDU *pdu, ssize_t len)
 636 {
 637     int8_t id = pdu->id + 1; /* Response */
 638
 639     if (len < 0) {
 640         V9fsString str;
 641         int err = -len;
 642
 643         str.data = strerror(err);
 644         str.size = strlen(str.data);
 645
 646         len = 7;
 647         len += pdu_marshal(pdu, len, "s", &str);
 648         if (dotu) {
 649             len += pdu_marshal(pdu, len, "d", err);
 650         }
 651
 652         id = P9_RERROR;
 653     }
 654
 655     /* fill out the header */
 656     pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag);
 657
 658     /* keep these in sync */
 659     pdu->size = len;
 660     pdu->id = id;
 661
 662     /* push onto queue and notify */
 663     virtqueue_push(s->vq, &pdu->elem, len);
 664
 665     /* FIXME: we should batch these completions */
 666     virtio_notify(&s->vdev, s->vq);
 667
 668     free_pdu(s, pdu);
 669 }
 670
 671 static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
 672 {
 673     mode_t ret;
 674
 675     ret = mode & 0777;
 676     if (mode & P9_STAT_MODE_DIR) {
 677         ret |= S_IFDIR;
 678     }
 679
 680     if (dotu) {
 681         if (mode & P9_STAT_MODE_SYMLINK) {
 682             ret |= S_IFLNK;
 683         }
 684         if (mode & P9_STAT_MODE_SOCKET) {
 685             ret |= S_IFSOCK;
 686         }
 687         if (mode & P9_STAT_MODE_NAMED_PIPE) {
 688             ret |= S_IFIFO;
 689         }
 690         if (mode & P9_STAT_MODE_DEVICE) {
 691             if (extension && extension->data[0] == 'c') {
 692                 ret |= S_IFCHR;
 693             } else {
 694                 ret |= S_IFBLK;
 695             }
 696         }
 697     }
 698
 699     if (!(ret&~0777)) {
 700         ret |= S_IFREG;
 701     }
 702
 703     if (mode & P9_STAT_MODE_SETUID) {
 704         ret |= S_ISUID;
 705     }
 706     if (mode & P9_STAT_MODE_SETGID) {
 707         ret |= S_ISGID;
 708     }
 709     if (mode & P9_STAT_MODE_SETVTX) {
 710         ret |= S_ISVTX;
 711     }
 712
 713     return ret;
 714 }
 715
 716 static int donttouch_stat(V9fsStat *stat)
 717 {
 718     if (stat->type == -1 &&
 719         stat->dev == -1 &&
 720         stat->qid.type == -1 &&
 721         stat->qid.version == -1 &&
 722         stat->qid.path == -1 &&
 723         stat->mode == -1 &&
 724         stat->atime == -1 &&
 725         stat->mtime == -1 &&
 726         stat->length == -1 &&
 727         !stat->name.size &&
 728         !stat->uid.size &&
 729         !stat->gid.size &&
 730         !stat->muid.size &&
 731         stat->n_uid == -1 &&
 732         stat->n_gid == -1 &&
 733         stat->n_muid == -1) {
 734         return 1;
 735     }
 736
 737     return 0;
 738 }
 739
 740 static void v9fs_stat_free(V9fsStat *stat)
 741 {
 742     v9fs_string_free(&stat->name);
 743     v9fs_string_free(&stat->uid);
 744     v9fs_string_free(&stat->gid);
 745     v9fs_string_free(&stat->muid);
 746     v9fs_string_free(&stat->extension);
 747 }
 748
 749 static uint32_t stat_to_v9mode(const struct stat *stbuf)
 750 {
 751     uint32_t mode;
 752
 753     mode = stbuf->st_mode & 0777;
 754     if (S_ISDIR(stbuf->st_mode)) {
 755         mode |= P9_STAT_MODE_DIR;
 756     }
 757
 758     if (dotu) {
 759         if (S_ISLNK(stbuf->st_mode)) {
 760             mode |= P9_STAT_MODE_SYMLINK;
 761         }
 762
 763         if (S_ISSOCK(stbuf->st_mode)) {
 764             mode |= P9_STAT_MODE_SOCKET;
 765         }
 766
 767         if (S_ISFIFO(stbuf->st_mode)) {
 768             mode |= P9_STAT_MODE_NAMED_PIPE;
 769         }
 770
 771         if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
 772             mode |= P9_STAT_MODE_DEVICE;
 773         }
 774
 775         if (stbuf->st_mode & S_ISUID) {
 776             mode |= P9_STAT_MODE_SETUID;
 777         }
 778
 779         if (stbuf->st_mode & S_ISGID) {
 780             mode |= P9_STAT_MODE_SETGID;
 781         }
 782
 783         if (stbuf->st_mode & S_ISVTX) {
 784             mode |= P9_STAT_MODE_SETVTX;
 785         }
 786     }
 787
 788     return mode;
 789 }
 790
 791 static int stat_to_v9stat(V9fsState *s, V9fsString *name,
 792                             const struct stat *stbuf,
 793                             V9fsStat *v9stat)
 794 {
 795     int err;
 796     const char *str;
 797
 798     memset(v9stat, 0, sizeof(*v9stat));
 799
 800     stat_to_qid(stbuf, &v9stat->qid);
 801     v9stat->mode = stat_to_v9mode(stbuf);
 802     v9stat->atime = stbuf->st_atime;
 803     v9stat->mtime = stbuf->st_mtime;
 804     v9stat->length = stbuf->st_size;
 805
 806     v9fs_string_null(&v9stat->uid);
 807     v9fs_string_null(&v9stat->gid);
 808     v9fs_string_null(&v9stat->muid);
 809
 810     if (dotu) {
 811         v9stat->n_uid = stbuf->st_uid;
 812         v9stat->n_gid = stbuf->st_gid;
 813         v9stat->n_muid = 0;
 814
 815         v9fs_string_null(&v9stat->extension);
 816
 817         if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
 818             err = v9fs_do_readlink(s, name, &v9stat->extension);
 819             if (err == -1) {
 820                 err = -errno;
 821                 return err;
 822             }
 823             v9stat->extension.data[err] = 0;
 824             v9stat->extension.size = err;
 825         } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
 826             v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
 827                     S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
 828                     major(stbuf->st_rdev), minor(stbuf->st_rdev));
 829         } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
 830             v9fs_string_sprintf(&v9stat->extension, "%s %u",
 831                     "HARDLINKCOUNT", stbuf->st_nlink);
 832         }
 833     }
 834
 835     str = strrchr(name->data, '/');
 836     if (str) {
 837         str += 1;
 838     } else {
 839         str = name->data;
 840     }
 841
 842     v9fs_string_sprintf(&v9stat->name, "%s", str);
 843
 844     v9stat->size = 61 +
 845         v9fs_string_size(&v9stat->name) +
 846         v9fs_string_size(&v9stat->uid) +
 847         v9fs_string_size(&v9stat->gid) +
 848         v9fs_string_size(&v9stat->muid) +
 849         v9fs_string_size(&v9stat->extension);
 850     return 0;
 851 }
 852
 853 static struct iovec *adjust_sg(struct iovec *sg, int len, int *iovcnt)
 854 {
 855     while (len && *iovcnt) {
 856         if (len < sg->iov_len) {
 857             sg->iov_len -= len;
 858             sg->iov_base += len;
 859             len = 0;
 860         } else {
 861             len -= sg->iov_len;
 862             sg++;
 863             *iovcnt -= 1;
 864         }
 865     }
 866
 867     return sg;
 868 }
 869
 870 static struct iovec *cap_sg(struct iovec *sg, int cap, int *cnt)
 871 {
 872     int i;
 873     int total = 0;
 874
 875     for (i = 0; i < *cnt; i++) {
 876         if ((total + sg[i].iov_len) > cap) {
 877             sg[i].iov_len -= ((total + sg[i].iov_len) - cap);
 878             i++;
 879             break;
 880         }
 881         total += sg[i].iov_len;
 882     }
 883
 884     *cnt = i;
 885
 886     return sg;
 887 }
 888
 889 static void print_sg(struct iovec *sg, int cnt)
 890 {
 891     int i;
 892
 893     printf("sg[%d]: {", cnt);
 894     for (i = 0; i < cnt; i++) {
 895         if (i) {
 896             printf(", ");
 897         }
 898         printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
 899     }
 900     printf("}\n");
 901 }
 902
 903 static void v9fs_dummy(V9fsState *s, V9fsPDU *pdu)
 904 {
 905     /* Note: The following have been added to prevent GCC from complaining
 906      * They will be removed in the subsequent patches */
 907     (void)pdu_unmarshal;
 908     (void) complete_pdu;
 909     (void) v9fs_string_init;
 910     (void) v9fs_string_free;
 911     (void) v9fs_string_null;
 912     (void) v9fs_string_sprintf;
 913     (void) v9fs_string_copy;
 914     (void) v9fs_string_size;
 915     (void) v9fs_do_lstat;
 916     (void) v9fs_do_setuid;
 917     (void) v9fs_do_readlink;
 918     (void) v9fs_do_close;
 919     (void) v9fs_do_closedir;
 920     (void) alloc_fid;
 921     (void) free_fid;
 922     (void) fid_to_qid;
 923     (void) v9mode_to_mode;
 924     (void) donttouch_stat;
 925     (void) v9fs_stat_free;
 926     (void) stat_to_v9stat;
 927     (void) adjust_sg;
 928     (void) cap_sg;
 929     (void) print_sg;
 930 }
 931
 932 static void v9fs_version(V9fsState *s, V9fsPDU *pdu)
 933 {
 934     int32_t msize;
 935     V9fsString version;
 936     size_t offset = 7;
 937
 938     pdu_unmarshal(pdu, offset, "ds", &msize, &version);
 939
 940     if (strcmp(version.data, "9P2000.u")) {
 941         v9fs_string_sprintf(&version, "unknown");
 942     }
 943
 944     offset += pdu_marshal(pdu, offset, "ds", msize, &version);
 945     complete_pdu(s, pdu, offset);
 946
 947     v9fs_string_free(&version);
 948 }
 949
 950 static void v9fs_attach(V9fsState *s, V9fsPDU *pdu)
 951 {
 952     int32_t fid, afid, n_uname;
 953     V9fsString uname, aname;
 954     V9fsFidState *fidp;
 955     V9fsQID qid;
 956     size_t offset = 7;
 957     ssize_t err;
 958
 959     pdu_unmarshal(pdu, offset, "ddssd", &fid, &afid, &uname, &aname, &n_uname);
 960
 961     fidp = alloc_fid(s, fid);
 962     if (fidp == NULL) {
 963         err = -EINVAL;
 964         goto out;
 965     }
 966
 967     fidp->uid = n_uname;
 968
 969     v9fs_string_sprintf(&fidp->path, "%s", "/");
 970     err = fid_to_qid(s, fidp, &qid);
 971     if (err) {
 972         err = -EINVAL;
 973         free_fid(s, fid);
 974         goto out;
 975     }
 976
 977     offset += pdu_marshal(pdu, offset, "Q", &qid);
 978
 979     err = offset;
 980 out:
 981     complete_pdu(s, pdu, err);
 982     v9fs_string_free(&uname);
 983     v9fs_string_free(&aname);
 984 }
 985
 986 typedef struct V9fsStatState {
 987     V9fsPDU *pdu;
 988     size_t offset;
 989     V9fsStat v9stat;
 990     V9fsFidState *fidp;
 991     struct stat stbuf;
 992 } V9fsStatState;
 993
 994 static void v9fs_stat_post_lstat(V9fsState *s, V9fsStatState *vs, int err)
 995 {
 996     if (err == -1) {
 997         err = -errno;
 998         goto out;
 999     }
1000
1001     err = stat_to_v9stat(s, &vs->fidp->path, &vs->stbuf, &vs->v9stat);
1002     if (err) {
1003         goto out;
1004     }
1005     vs->offset += pdu_marshal(vs->pdu, vs->offset, "wS", 0, &vs->v9stat);
1006     err = vs->offset;
1007
1008 out:
1009     complete_pdu(s, vs->pdu, err);
1010     v9fs_stat_free(&vs->v9stat);
1011     qemu_free(vs);
1012 }
1013
1014 static void v9fs_stat(V9fsState *s, V9fsPDU *pdu)
1015 {
1016     int32_t fid;
1017     V9fsStatState *vs;
1018     ssize_t err = 0;
1019
1020     vs = qemu_malloc(sizeof(*vs));
1021     vs->pdu = pdu;
1022     vs->offset = 7;
1023
1024     memset(&vs->v9stat, 0, sizeof(vs->v9stat));
1025
1026     pdu_unmarshal(vs->pdu, vs->offset, "d", &fid);
1027
1028     vs->fidp = lookup_fid(s, fid);
1029     if (vs->fidp == NULL) {
1030         err = -ENOENT;
1031         goto out;
1032     }
1033
1034     err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1035     v9fs_stat_post_lstat(s, vs, err);
1036     return;
1037
1038 out:
1039     complete_pdu(s, vs->pdu, err);
1040     v9fs_stat_free(&vs->v9stat);
1041     qemu_free(vs);
1042 }
1043
1044 typedef struct V9fsWalkState {
1045     V9fsPDU *pdu;
1046     size_t offset;
1047     int16_t nwnames;
1048     int name_idx;
1049     V9fsQID *qids;
1050     V9fsFidState *fidp;
1051     V9fsFidState *newfidp;
1052     V9fsString path;
1053     V9fsString *wnames;
1054     struct stat stbuf;
1055 } V9fsWalkState;
1056
1057 static void v9fs_walk_complete(V9fsState *s, V9fsWalkState *vs, int err)
1058 {
1059     complete_pdu(s, vs->pdu, err);
1060
1061     if (vs->nwnames) {
1062         for (vs->name_idx = 0; vs->name_idx < vs->nwnames; vs->name_idx++) {
1063             v9fs_string_free(&vs->wnames[vs->name_idx]);
1064         }
1065
1066         qemu_free(vs->wnames);
1067         qemu_free(vs->qids);
1068     }
1069 }
1070
1071 static void v9fs_walk_marshal(V9fsWalkState *vs)
1072 {
1073     int i;
1074     vs->offset = 7;
1075     vs->offset += pdu_marshal(vs->pdu, vs->offset, "w", vs->nwnames);
1076
1077     for (i = 0; i < vs->nwnames; i++) {
1078         vs->offset += pdu_marshal(vs->pdu, vs->offset, "Q", &vs->qids[i]);
1079     }
1080 }
1081
1082 static void v9fs_walk_post_newfid_lstat(V9fsState *s, V9fsWalkState *vs,
1083                                                                 int err)
1084 {
1085     if (err == -1) {
1086         free_fid(s, vs->newfidp->fid);
1087         v9fs_string_free(&vs->path);
1088         err = -ENOENT;
1089         goto out;
1090     }
1091
1092     stat_to_qid(&vs->stbuf, &vs->qids[vs->name_idx]);
1093
1094     vs->name_idx++;
1095     if (vs->name_idx < vs->nwnames) {
1096         v9fs_string_sprintf(&vs->path, "%s/%s", vs->newfidp->path.data,
1097                                             vs->wnames[vs->name_idx].data);
1098         v9fs_string_copy(&vs->newfidp->path, &vs->path);
1099
1100         err = v9fs_do_lstat(s, &vs->newfidp->path, &vs->stbuf);
1101         v9fs_walk_post_newfid_lstat(s, vs, err);
1102         return;
1103     }
1104
1105     v9fs_string_free(&vs->path);
1106     v9fs_walk_marshal(vs);
1107     err = vs->offset;
1108 out:
1109     v9fs_walk_complete(s, vs, err);
1110 }
1111
1112 static void v9fs_walk_post_oldfid_lstat(V9fsState *s, V9fsWalkState *vs,
1113         int err)
1114 {
1115     if (err == -1) {
1116         v9fs_string_free(&vs->path);
1117         err = -ENOENT;
1118         goto out;
1119     }
1120
1121     stat_to_qid(&vs->stbuf, &vs->qids[vs->name_idx]);
1122     vs->name_idx++;
1123     if (vs->name_idx < vs->nwnames) {
1124
1125         v9fs_string_sprintf(&vs->path, "%s/%s",
1126                 vs->fidp->path.data, vs->wnames[vs->name_idx].data);
1127         v9fs_string_copy(&vs->fidp->path, &vs->path);
1128
1129         err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1130         v9fs_walk_post_oldfid_lstat(s, vs, err);
1131         return;
1132     }
1133
1134     v9fs_string_free(&vs->path);
1135     v9fs_walk_marshal(vs);
1136     err = vs->offset;
1137 out:
1138     v9fs_walk_complete(s, vs, err);
1139 }
1140
1141 static void v9fs_walk(V9fsState *s, V9fsPDU *pdu)
1142 {
1143     int32_t fid, newfid;
1144     V9fsWalkState *vs;
1145     int err = 0;
1146     int i;
1147
1148     vs = qemu_malloc(sizeof(*vs));
1149     vs->pdu = pdu;
1150     vs->wnames = NULL;
1151     vs->qids = NULL;
1152     vs->offset = 7;
1153
1154     vs->offset += pdu_unmarshal(vs->pdu, vs->offset, "ddw", &fid,
1155                                             &newfid, &vs->nwnames);
1156
1157     if (vs->nwnames) {
1158         vs->wnames = qemu_mallocz(sizeof(vs->wnames[0]) * vs->nwnames);
1159
1160         vs->qids = qemu_mallocz(sizeof(vs->qids[0]) * vs->nwnames);
1161
1162         for (i = 0; i < vs->nwnames; i++) {
1163             vs->offset += pdu_unmarshal(vs->pdu, vs->offset, "s",
1164                                             &vs->wnames[i]);
1165         }
1166     }
1167
1168     vs->fidp = lookup_fid(s, fid);
1169     if (vs->fidp == NULL) {
1170         err = -ENOENT;
1171         goto out;
1172     }
1173
1174     /* FIXME: is this really valid? */
1175     if (fid == newfid) {
1176
1177         BUG_ON(vs->fidp->fd != -1);
1178         BUG_ON(vs->fidp->dir);
1179         v9fs_string_init(&vs->path);
1180         vs->name_idx = 0;
1181
1182         if (vs->name_idx < vs->nwnames) {
1183             v9fs_string_sprintf(&vs->path, "%s/%s",
1184                 vs->fidp->path.data, vs->wnames[vs->name_idx].data);
1185             v9fs_string_copy(&vs->fidp->path, &vs->path);
1186
1187             err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1188             v9fs_walk_post_oldfid_lstat(s, vs, err);
1189             return;
1190         }
1191     } else {
1192         vs->newfidp = alloc_fid(s, newfid);
1193         if (vs->newfidp == NULL) {
1194             err = -EINVAL;
1195             goto out;
1196         }
1197
1198         vs->newfidp->uid = vs->fidp->uid;
1199         v9fs_string_init(&vs->path);
1200         vs->name_idx = 0;
1201         v9fs_string_copy(&vs->newfidp->path, &vs->fidp->path);
1202
1203         if (vs->name_idx < vs->nwnames) {
1204             v9fs_string_sprintf(&vs->path, "%s/%s", vs->newfidp->path.data,
1205                                 vs->wnames[vs->name_idx].data);
1206             v9fs_string_copy(&vs->newfidp->path, &vs->path);
1207
1208             err = v9fs_do_lstat(s, &vs->newfidp->path, &vs->stbuf);
1209             v9fs_walk_post_newfid_lstat(s, vs, err);
1210             return;
1211         }
1212     }
1213
1214     v9fs_walk_marshal(vs);
1215     err = vs->offset;
1216 out:
1217     v9fs_walk_complete(s, vs, err);
1218 }
1219
1220 typedef struct V9fsOpenState {
1221     V9fsPDU *pdu;
1222     size_t offset;
1223     int8_t mode;
1224     V9fsFidState *fidp;
1225     V9fsQID qid;
1226     struct stat stbuf;
1227
1228 } V9fsOpenState;
1229
1230 enum {
1231     Oread   = 0x00,
1232     Owrite  = 0x01,
1233     Ordwr   = 0x02,
1234     Oexec   = 0x03,
1235     Oexcl   = 0x04,
1236     Otrunc  = 0x10,
1237     Orexec  = 0x20,
1238     Orclose = 0x40,
1239     Oappend = 0x80,
1240 };
1241
1242 static int omode_to_uflags(int8_t mode)
1243 {
1244     int ret = 0;
1245
1246     switch (mode & 3) {
1247     case Oread:
1248         ret = O_RDONLY;
1249         break;
1250     case Ordwr:
1251         ret = O_RDWR;
1252         break;
1253     case Owrite:
1254         ret = O_WRONLY;
1255         break;
1256     case Oexec:
1257         ret = O_RDONLY;
1258         break;
1259     }
1260
1261     if (mode & Otrunc) {
1262         ret |= O_TRUNC;
1263     }
1264
1265     if (mode & Oappend) {
1266         ret |= O_APPEND;
1267     }
1268
1269     if (mode & Oexcl) {
1270         ret |= O_EXCL;
1271     }
1272
1273     return ret;
1274 }
1275
1276 static void v9fs_open_post_opendir(V9fsState *s, V9fsOpenState *vs, int err)
1277 {
1278     if (vs->fidp->dir == NULL) {
1279         err = -errno;
1280         goto out;
1281     }
1282
1283     vs->offset += pdu_marshal(vs->pdu, vs->offset, "Qd", &vs->qid, 0);
1284     err = vs->offset;
1285 out:
1286     complete_pdu(s, vs->pdu, err);
1287     qemu_free(vs);
1288
1289 }
1290
1291 static void v9fs_open_post_open(V9fsState *s, V9fsOpenState *vs, int err)
1292 {
1293     if (vs->fidp->fd == -1) {
1294         err = -errno;
1295         goto out;
1296     }
1297
1298     vs->offset += pdu_marshal(vs->pdu, vs->offset, "Qd", &vs->qid, 0);
1299     err = vs->offset;
1300 out:
1301     complete_pdu(s, vs->pdu, err);
1302     qemu_free(vs);
1303 }
1304
1305 static void v9fs_open_post_lstat(V9fsState *s, V9fsOpenState *vs, int err)
1306 {
1307     if (err) {
1308         err = -errno;
1309         goto out;
1310     }
1311
1312     stat_to_qid(&vs->stbuf, &vs->qid);
1313
1314     if (S_ISDIR(vs->stbuf.st_mode)) {
1315         vs->fidp->dir = v9fs_do_opendir(s, &vs->fidp->path);
1316         v9fs_open_post_opendir(s, vs, err);
1317     } else {
1318         vs->fidp->fd = v9fs_do_open(s, &vs->fidp->path,
1319                                     omode_to_uflags(vs->mode));
1320         v9fs_open_post_open(s, vs, err);
1321     }
1322     return;
1323 out:
1324     complete_pdu(s, vs->pdu, err);
1325     qemu_free(vs);
1326 }
1327
1328 static void v9fs_open(V9fsState *s, V9fsPDU *pdu)
1329 {
1330     int32_t fid;
1331     V9fsOpenState *vs;
1332     ssize_t err = 0;
1333
1334
1335     vs = qemu_malloc(sizeof(*vs));
1336     vs->pdu = pdu;
1337     vs->offset = 7;
1338
1339     pdu_unmarshal(vs->pdu, vs->offset, "db", &fid, &vs->mode);
1340
1341     vs->fidp = lookup_fid(s, fid);
1342     if (vs->fidp == NULL) {
1343         err = -ENOENT;
1344         goto out;
1345     }
1346
1347     BUG_ON(vs->fidp->fd != -1);
1348     BUG_ON(vs->fidp->dir);
1349
1350     err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1351
1352     v9fs_open_post_lstat(s, vs, err);
1353     return;
1354 out:
1355     complete_pdu(s, pdu, err);
1356     qemu_free(vs);
1357 }
1358
1359 static void v9fs_clunk(V9fsState *s, V9fsPDU *pdu)
1360 {
1361     int32_t fid;
1362     size_t offset = 7;
1363     int err;
1364
1365     pdu_unmarshal(pdu, offset, "d", &fid);
1366
1367     err = free_fid(s, fid);
1368     if (err < 0) {
1369         goto out;
1370     }
1371
1372     offset = 7;
1373     err = offset;
1374 out:
1375     complete_pdu(s, pdu, err);
1376 }
1377
1378 typedef struct V9fsReadState {
1379     V9fsPDU *pdu;
1380     size_t offset;
1381     int32_t count;
1382     int32_t total;
1383     int64_t off;
1384     V9fsFidState *fidp;
1385     struct iovec iov[128]; /* FIXME: bad, bad, bad */
1386     struct iovec *sg;
1387     off_t dir_pos;
1388     struct dirent *dent;
1389     struct stat stbuf;
1390     V9fsString name;
1391     V9fsStat v9stat;
1392     int32_t len;
1393     int32_t cnt;
1394     int32_t max_count;
1395 } V9fsReadState;
1396
1397 static void v9fs_read_post_readdir(V9fsState *, V9fsReadState *, ssize_t);
1398
1399 static void v9fs_read_post_seekdir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1400 {
1401     if (err) {
1402         goto out;
1403     }
1404     v9fs_stat_free(&vs->v9stat);
1405     v9fs_string_free(&vs->name);
1406     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1407     vs->offset += vs->count;
1408     err = vs->offset;
1409 out:
1410     complete_pdu(s, vs->pdu, err);
1411     qemu_free(vs);
1412     return;
1413 }
1414
1415 static void v9fs_read_post_dir_lstat(V9fsState *s, V9fsReadState *vs,
1416                                     ssize_t err)
1417 {
1418     if (err) {
1419         err = -errno;
1420         goto out;
1421     }
1422     err = stat_to_v9stat(s, &vs->name, &vs->stbuf, &vs->v9stat);
1423     if (err) {
1424         goto out;
1425     }
1426
1427     vs->len = pdu_marshal(vs->pdu, vs->offset + 4 + vs->count, "S",
1428                             &vs->v9stat);
1429     if ((vs->len != (vs->v9stat.size + 2)) ||
1430             ((vs->count + vs->len) > vs->max_count)) {
1431         v9fs_do_seekdir(s, vs->fidp->dir, vs->dir_pos);
1432         v9fs_read_post_seekdir(s, vs, err);
1433         return;
1434     }
1435     vs->count += vs->len;
1436     v9fs_stat_free(&vs->v9stat);
1437     v9fs_string_free(&vs->name);
1438     vs->dir_pos = vs->dent->d_off;
1439     vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1440     v9fs_read_post_readdir(s, vs, err);
1441     return;
1442 out:
1443     v9fs_do_seekdir(s, vs->fidp->dir, vs->dir_pos);
1444     v9fs_read_post_seekdir(s, vs, err);
1445     return;
1446
1447 }
1448
1449 static void v9fs_read_post_readdir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1450 {
1451     if (vs->dent) {
1452         memset(&vs->v9stat, 0, sizeof(vs->v9stat));
1453         v9fs_string_init(&vs->name);
1454         v9fs_string_sprintf(&vs->name, "%s/%s", vs->fidp->path.data,
1455                             vs->dent->d_name);
1456         err = v9fs_do_lstat(s, &vs->name, &vs->stbuf);
1457         v9fs_read_post_dir_lstat(s, vs, err);
1458         return;
1459     }
1460
1461     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1462     vs->offset += vs->count;
1463     err = vs->offset;
1464     complete_pdu(s, vs->pdu, err);
1465     qemu_free(vs);
1466     return;
1467 }
1468
1469 static void v9fs_read_post_telldir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1470 {
1471     vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1472     v9fs_read_post_readdir(s, vs, err);
1473     return;
1474 }
1475
1476 static void v9fs_read_post_rewinddir(V9fsState *s, V9fsReadState *vs,
1477                                        ssize_t err)
1478 {
1479     vs->dir_pos = v9fs_do_telldir(s, vs->fidp->dir);
1480     v9fs_read_post_telldir(s, vs, err);
1481     return;
1482 }
1483
1484 static void v9fs_read_post_readv(V9fsState *s, V9fsReadState *vs, ssize_t err)
1485 {
1486     if (err  < 0) {
1487         /* IO error return the error */
1488         err = -errno;
1489         goto out;
1490     }
1491     vs->total += vs->len;
1492     vs->sg = adjust_sg(vs->sg, vs->len, &vs->cnt);
1493     if (vs->total < vs->count && vs->len > 0) {
1494         do {
1495             if (0) {
1496                 print_sg(vs->sg, vs->cnt);
1497             }
1498             vs->len = v9fs_do_readv(s, vs->fidp->fd, vs->sg, vs->cnt);
1499         } while (vs->len == -1 && errno == EINTR);
1500         if (vs->len == -1) {
1501             err  = -errno;
1502         }
1503         v9fs_read_post_readv(s, vs, err);
1504         return;
1505     }
1506     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->total);
1507     vs->offset += vs->count;
1508     err = vs->offset;
1509
1510 out:
1511     complete_pdu(s, vs->pdu, err);
1512     qemu_free(vs);
1513 }
1514
1515 static void v9fs_read_post_lseek(V9fsState *s, V9fsReadState *vs, ssize_t err)
1516 {
1517     if (err == -1) {
1518         err = -errno;
1519         goto out;
1520     }
1521     vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
1522
1523     if (vs->total < vs->count) {
1524         do {
1525             if (0) {
1526                 print_sg(vs->sg, vs->cnt);
1527             }
1528             vs->len = v9fs_do_readv(s, vs->fidp->fd, vs->sg, vs->cnt);
1529         } while (vs->len == -1 && errno == EINTR);
1530         if (vs->len == -1) {
1531             err  = -errno;
1532         }
1533         v9fs_read_post_readv(s, vs, err);
1534         return;
1535     }
1536 out:
1537     complete_pdu(s, vs->pdu, err);
1538     qemu_free(vs);
1539 }
1540
1541 static void v9fs_read(V9fsState *s, V9fsPDU *pdu)
1542 {
1543     int32_t fid;
1544     V9fsReadState *vs;
1545     ssize_t err = 0;
1546
1547     vs = qemu_malloc(sizeof(*vs));
1548     vs->pdu = pdu;
1549     vs->offset = 7;
1550     vs->total = 0;
1551     vs->len = 0;
1552     vs->count = 0;
1553
1554     pdu_unmarshal(vs->pdu, vs->offset, "dqd", &fid, &vs->off, &vs->count);
1555
1556     vs->fidp = lookup_fid(s, fid);
1557     if (vs->fidp == NULL) {
1558         err = -EINVAL;
1559         goto out;
1560     }
1561
1562     if (vs->fidp->dir) {
1563         vs->max_count = vs->count;
1564         vs->count = 0;
1565         if (vs->off == 0) {
1566             v9fs_do_rewinddir(s, vs->fidp->dir);
1567         }
1568         v9fs_read_post_rewinddir(s, vs, err);
1569         return;
1570     } else if (vs->fidp->fd != -1) {
1571         vs->sg = vs->iov;
1572         pdu_marshal(vs->pdu, vs->offset + 4, "v", vs->sg, &vs->cnt);
1573         err = v9fs_do_lseek(s, vs->fidp->fd, vs->off, SEEK_SET);
1574         v9fs_read_post_lseek(s, vs, err);
1575         return;
1576     } else {
1577         err = -EINVAL;
1578     }
1579 out:
1580     complete_pdu(s, pdu, err);
1581     qemu_free(vs);
1582 }
1583
1584 typedef struct V9fsWriteState {
1585     V9fsPDU *pdu;
1586     size_t offset;
1587     int32_t len;
1588     int32_t count;
1589     int32_t total;
1590     int64_t off;
1591     V9fsFidState *fidp;
1592     struct iovec iov[128]; /* FIXME: bad, bad, bad */
1593     struct iovec *sg;
1594     int cnt;
1595 } V9fsWriteState;
1596
1597 static void v9fs_write_post_writev(V9fsState *s, V9fsWriteState *vs,
1598                                    ssize_t err)
1599 {
1600     if (err  < 0) {
1601         /* IO error return the error */
1602         err = -errno;
1603         goto out;
1604     }
1605     vs->total += vs->len;
1606     vs->sg = adjust_sg(vs->sg, vs->len, &vs->cnt);
1607     if (vs->total < vs->count && vs->len > 0) {
1608         do {
1609             if (0) {
1610                 print_sg(vs->sg, vs->cnt);
1611             }
1612             vs->len =  v9fs_do_writev(s, vs->fidp->fd, vs->sg, vs->cnt);
1613         } while (vs->len == -1 && errno == EINTR);
1614         if (vs->len == -1) {
1615             err  = -errno;
1616         }
1617         v9fs_write_post_writev(s, vs, err);
1618         return;
1619     }
1620     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->total);
1621
1622     err = vs->offset;
1623 out:
1624     complete_pdu(s, vs->pdu, err);
1625     qemu_free(vs);
1626 }
1627
1628 static void v9fs_write_post_lseek(V9fsState *s, V9fsWriteState *vs, ssize_t err)
1629 {
1630     if (err == -1) {
1631         err = -errno;
1632         goto out;
1633     }
1634     vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
1635
1636     if (vs->total < vs->count) {
1637         do {
1638             if (0) {
1639                 print_sg(vs->sg, vs->cnt);
1640             }
1641             vs->len = v9fs_do_writev(s, vs->fidp->fd, vs->sg, vs->cnt);
1642         } while (vs->len == -1 && errno == EINTR);
1643         if (vs->len == -1) {
1644             err  = -errno;
1645         }
1646         v9fs_write_post_writev(s, vs, err);
1647         return;
1648     }
1649
1650 out:
1651     complete_pdu(s, vs->pdu, err);
1652     qemu_free(vs);
1653 }
1654
1655 static void v9fs_write(V9fsState *s, V9fsPDU *pdu)
1656 {
1657     int32_t fid;
1658     V9fsWriteState *vs;
1659     ssize_t err;
1660
1661     vs = qemu_malloc(sizeof(*vs));
1662
1663     vs->pdu = pdu;
1664     vs->offset = 7;
1665     vs->sg = vs->iov;
1666     vs->total = 0;
1667     vs->len = 0;
1668
1669     pdu_unmarshal(vs->pdu, vs->offset, "dqdv", &fid, &vs->off, &vs->count,
1670                     vs->sg, &vs->cnt);
1671
1672     vs->fidp = lookup_fid(s, fid);
1673     if (vs->fidp == NULL) {
1674         err = -EINVAL;
1675         goto out;
1676     }
1677
1678     if (vs->fidp->fd == -1) {
1679         err = -EINVAL;
1680         goto out;
1681     }
1682
1683     err = v9fs_do_lseek(s, vs->fidp->fd, vs->off, SEEK_SET);
1684
1685     v9fs_write_post_lseek(s, vs, err);
1686     return;
1687
1688 out:
1689     complete_pdu(s, vs->pdu, err);
1690     qemu_free(vs);
1691 }
1692
1693 typedef struct V9fsCreateState {
1694     V9fsPDU *pdu;
1695     size_t offset;
1696     V9fsFidState *fidp;
1697     V9fsQID qid;
1698     int32_t perm;
1699     int8_t mode;
1700     struct stat stbuf;
1701     V9fsString name;
1702     V9fsString extension;
1703     V9fsString fullname;
1704 } V9fsCreateState;
1705
1706 static void v9fs_post_create(V9fsState *s, V9fsCreateState *vs, int err)
1707 {
1708     if (err == 0) {
1709         v9fs_string_copy(&vs->fidp->path, &vs->fullname);
1710         stat_to_qid(&vs->stbuf, &vs->qid);
1711
1712         vs->offset += pdu_marshal(vs->pdu, vs->offset, "Qd", &vs->qid, 0);
1713
1714         err = vs->offset;
1715     }
1716
1717     complete_pdu(s, vs->pdu, err);
1718     v9fs_string_free(&vs->name);
1719     v9fs_string_free(&vs->extension);
1720     v9fs_string_free(&vs->fullname);
1721     qemu_free(vs);
1722 }
1723
1724 static void v9fs_create_post_perms(V9fsState *s, V9fsCreateState *vs, int err)
1725 {
1726     if (err) {
1727         err = -errno;
1728     }
1729     v9fs_post_create(s, vs, err);
1730 }
1731
1732 static void v9fs_create_post_opendir(V9fsState *s, V9fsCreateState *vs,
1733                                                                     int err)
1734 {
1735     if (!vs->fidp->dir) {
1736         err = -errno;
1737     }
1738     v9fs_post_create(s, vs, err);
1739 }
1740
1741 static void v9fs_create_post_dir_lstat(V9fsState *s, V9fsCreateState *vs,
1742                                                                     int err)
1743 {
1744     if (err) {
1745         err = -errno;
1746         goto out;
1747     }
1748
1749     vs->fidp->dir = v9fs_do_opendir(s, &vs->fullname);
1750     v9fs_create_post_opendir(s, vs, err);
1751     return;
1752
1753 out:
1754     v9fs_post_create(s, vs, err);
1755 }
1756
1757 static void v9fs_create_post_mkdir(V9fsState *s, V9fsCreateState *vs, int err)
1758 {
1759     if (err) {
1760         err = -errno;
1761         goto out;
1762     }
1763
1764     err = v9fs_do_lstat(s, &vs->fullname, &vs->stbuf);
1765     v9fs_create_post_dir_lstat(s, vs, err);
1766     return;
1767
1768 out:
1769     v9fs_post_create(s, vs, err);
1770 }
1771
1772 static void v9fs_create_post_mksock(V9fsState *s, V9fsCreateState *vs,
1773                                                                 int err)
1774 {
1775     if (err) {
1776         err = -errno;
1777         goto out;
1778     }
1779
1780     err = v9fs_do_chmod(s, &vs->fullname, vs->perm & 0777);
1781     v9fs_create_post_perms(s, vs, err);
1782     return;
1783
1784 out:
1785     v9fs_post_create(s, vs, err);
1786 }
1787
1788 static void v9fs_create_post_fstat(V9fsState *s, V9fsCreateState *vs, int err)
1789 {
1790     if (err) {
1791         vs->fidp->fd = -1;
1792         err = -errno;
1793     }
1794
1795     v9fs_post_create(s, vs, err);
1796     return;
1797 }
1798
1799 static void v9fs_create_post_open2(V9fsState *s, V9fsCreateState *vs, int err)
1800 {
1801     if (vs->fidp->fd == -1) {
1802         err = -errno;
1803         goto out;
1804     }
1805
1806     err = v9fs_do_fstat(s, vs->fidp->fd, &vs->stbuf);
1807     v9fs_create_post_fstat(s, vs, err);
1808
1809     return;
1810
1811 out:
1812     v9fs_post_create(s, vs, err);
1813
1814 }
1815
1816 static void v9fs_create_post_lstat(V9fsState *s, V9fsCreateState *vs, int err)
1817 {
1818
1819     if (err == 0 || errno != ENOENT) {
1820         err = -errno;
1821         goto out;
1822     }
1823
1824     if (vs->perm & P9_STAT_MODE_DIR) {
1825         err = v9fs_do_mkdir(s, &vs->fullname, vs->perm & 0777);
1826         v9fs_create_post_mkdir(s, vs, err);
1827     } else if (vs->perm & P9_STAT_MODE_SYMLINK) {
1828         err = v9fs_do_symlink(s, &vs->extension, &vs->fullname);
1829         v9fs_create_post_perms(s, vs, err);
1830     } else if (vs->perm & P9_STAT_MODE_LINK) {
1831         int32_t nfid = atoi(vs->extension.data);
1832         V9fsFidState *nfidp = lookup_fid(s, nfid);
1833         if (nfidp == NULL) {
1834             err = -errno;
1835             v9fs_post_create(s, vs, err);
1836         }
1837         err = v9fs_do_link(s, &nfidp->path, &vs->fullname);
1838         v9fs_create_post_perms(s, vs, err);
1839     } else if (vs->perm & P9_STAT_MODE_DEVICE) {
1840         char ctype;
1841         uint32_t major, minor;
1842         mode_t nmode = 0;
1843
1844         if (sscanf(vs->extension.data, "%c %u %u", &ctype, &major,
1845                                         &minor) != 3) {
1846             err = -errno;
1847             v9fs_post_create(s, vs, err);
1848         }
1849
1850         switch (ctype) {
1851         case 'c':
1852             nmode = S_IFCHR;
1853             break;
1854         case 'b':
1855             nmode = S_IFBLK;
1856             break;
1857         default:
1858             err = -EIO;
1859             v9fs_post_create(s, vs, err);
1860         }
1861
1862         nmode |= vs->perm & 0777;
1863         err = v9fs_do_mknod(s, &vs->fullname, nmode, makedev(major, minor));
1864         v9fs_create_post_perms(s, vs, err);
1865     } else if (vs->perm & P9_STAT_MODE_NAMED_PIPE) {
1866         err = v9fs_do_mknod(s, &vs->fullname, S_IFIFO | (vs->mode & 0777), 0);
1867         v9fs_post_create(s, vs, err);
1868     } else if (vs->perm & P9_STAT_MODE_SOCKET) {
1869         err = v9fs_do_mksock(s, &vs->fullname);
1870         v9fs_create_post_mksock(s, vs, err);
1871     } else {
1872         vs->fidp->fd = v9fs_do_open2(s, &vs->fullname,
1873                                 omode_to_uflags(vs->mode) | O_CREAT,
1874                                 vs->perm & 0777);
1875         v9fs_create_post_open2(s, vs, err);
1876     }
1877
1878     return;
1879
1880 out:
1881     v9fs_post_create(s, vs, err);
1882 }
1883
1884 static void v9fs_create(V9fsState *s, V9fsPDU *pdu)
1885 {
1886     int32_t fid;
1887     V9fsCreateState *vs;
1888     int err = 0;
1889
1890     vs = qemu_malloc(sizeof(*vs));
1891     vs->pdu = pdu;
1892     vs->offset = 7;
1893
1894     v9fs_string_init(&vs->fullname);
1895
1896     pdu_unmarshal(vs->pdu, vs->offset, "dsdbs", &fid, &vs->name,
1897                                 &vs->perm, &vs->mode, &vs->extension);
1898
1899     vs->fidp = lookup_fid(s, fid);
1900     if (vs->fidp == NULL) {
1901         err = -EINVAL;
1902         goto out;
1903     }
1904
1905     v9fs_string_sprintf(&vs->fullname, "%s/%s", vs->fidp->path.data,
1906                                                         vs->name.data);
1907
1908     err = v9fs_do_lstat(s, &vs->fullname, &vs->stbuf);
1909     v9fs_create_post_lstat(s, vs, err);
1910     return;
1911
1912 out:
1913     complete_pdu(s, vs->pdu, err);
1914     v9fs_string_free(&vs->name);
1915     v9fs_string_free(&vs->extension);
1916     qemu_free(vs);
1917 }
1918
1919 static void v9fs_flush(V9fsState *s, V9fsPDU *pdu)
1920 {
1921     v9fs_dummy(s, pdu);
1922     if (debug_9p_pdu) {
1923         pprint_pdu(pdu);
1924     }
1925 }
1926
1927 static void v9fs_remove(V9fsState *s, V9fsPDU *pdu)
1928 {
1929     if (debug_9p_pdu) {
1930         pprint_pdu(pdu);
1931     }
1932 }
1933
1934 static void v9fs_wstat(V9fsState *s, V9fsPDU *pdu)
1935 {
1936     if (debug_9p_pdu) {
1937         pprint_pdu(pdu);
1938     }
1939 }
1940
1941 typedef void (pdu_handler_t)(V9fsState *s, V9fsPDU *pdu);
1942
1943 static pdu_handler_t *pdu_handlers[] = {
1944     [P9_TVERSION] = v9fs_version,
1945     [P9_TATTACH] = v9fs_attach,
1946     [P9_TSTAT] = v9fs_stat,
1947     [P9_TWALK] = v9fs_walk,
1948     [P9_TCLUNK] = v9fs_clunk,
1949     [P9_TOPEN] = v9fs_open,
1950     [P9_TREAD] = v9fs_read,
1951 #if 0
1952     [P9_TAUTH] = v9fs_auth,
1953 #endif
1954     [P9_TFLUSH] = v9fs_flush,
1955     [P9_TCREATE] = v9fs_create,
1956     [P9_TWRITE] = v9fs_write,
1957     [P9_TWSTAT] = v9fs_wstat,
1958     [P9_TREMOVE] = v9fs_remove,
1959 };
1960
1961 static void submit_pdu(V9fsState *s, V9fsPDU *pdu)
1962 {
1963     pdu_handler_t *handler;
1964
1965     if (debug_9p_pdu) {
1966         pprint_pdu(pdu);
1967     }
1968
1969     BUG_ON(pdu->id >= ARRAY_SIZE(pdu_handlers));
1970
1971     handler = pdu_handlers[pdu->id];
1972     BUG_ON(handler == NULL);
1973
1974     handler(s, pdu);
1975 }
1976
1977 static void handle_9p_output(VirtIODevice *vdev, VirtQueue *vq)
1978 {
1979     V9fsState *s = (V9fsState *)vdev;
1980     V9fsPDU *pdu;
1981     ssize_t len;
1982
1983     while ((pdu = alloc_pdu(s)) &&
1984             (len = virtqueue_pop(vq, &pdu->elem)) != 0) {
1985         uint8_t *ptr;
1986
1987         BUG_ON(pdu->elem.out_num == 0 || pdu->elem.in_num == 0);
1988         BUG_ON(pdu->elem.out_sg[0].iov_len < 7);
1989
1990         ptr = pdu->elem.out_sg[0].iov_base;
1991
1992         memcpy(&pdu->size, ptr, 4);
1993         pdu->id = ptr[4];
1994         memcpy(&pdu->tag, ptr + 5, 2);
1995
1996         submit_pdu(s, pdu);
1997     }
1998
1999     free_pdu(s, pdu);
2000 }
2001
2002 static uint32_t virtio_9p_get_features(VirtIODevice *vdev, uint32_t features)
2003 {
2004     features |= 1 << VIRTIO_9P_MOUNT_TAG;
2005     return features;
2006 }
2007
2008 static V9fsState *to_virtio_9p(VirtIODevice *vdev)
2009 {
2010     return (V9fsState *)vdev;
2011 }
2012
2013 static void virtio_9p_get_config(VirtIODevice *vdev, uint8_t *config)
2014 {
2015     struct virtio_9p_config *cfg;
2016     V9fsState *s = to_virtio_9p(vdev);
2017
2018     cfg = qemu_mallocz(sizeof(struct virtio_9p_config) +
2019                         s->tag_len);
2020     stw_raw(&cfg->tag_len, s->tag_len);
2021     memcpy(cfg->tag, s->tag, s->tag_len);
2022     memcpy(config, cfg, s->config_size);
2023     qemu_free(cfg);
2024 }
2025
2026 VirtIODevice *virtio_9p_init(DeviceState *dev, V9fsConf *conf)
2027  {
2028     V9fsState *s;
2029     int i, len;
2030     struct stat stat;
2031     FsTypeEntry *fse;
2032
2033
2034     s = (V9fsState *)virtio_common_init("virtio-9p",
2035                                     VIRTIO_ID_9P,
2036                                     sizeof(struct virtio_9p_config)+
2037                                     MAX_TAG_LEN,
2038                                     sizeof(V9fsState));
2039
2040     /* initialize pdu allocator */
2041     QLIST_INIT(&s->free_list);
2042     for (i = 0; i < (MAX_REQ - 1); i++) {
2043         QLIST_INSERT_HEAD(&s->free_list, &s->pdus[i], next);
2044     }
2045
2046     s->vq = virtio_add_queue(&s->vdev, MAX_REQ, handle_9p_output);
2047
2048     fse = get_fsdev_fsentry(conf->fsdev_id);
2049
2050     if (!fse) {
2051         /* We don't have a fsdev identified by fsdev_id */
2052         fprintf(stderr, "Virtio-9p device couldn't find fsdev "
2053                     "with the id %s\n", conf->fsdev_id);
2054         exit(1);
2055     }
2056
2057     if (!fse->path || !conf->tag) {
2058         /* we haven't specified a mount_tag or the path */
2059         fprintf(stderr, "fsdev with id %s needs path "
2060                 "and Virtio-9p device needs mount_tag arguments\n",
2061                 conf->fsdev_id);
2062         exit(1);
2063     }
2064
2065     if (lstat(fse->path, &stat)) {
2066         fprintf(stderr, "share path %s does not exist\n", fse->path);
2067         exit(1);
2068     } else if (!S_ISDIR(stat.st_mode)) {
2069         fprintf(stderr, "share path %s is not a directory \n", fse->path);
2070         exit(1);
2071     }
2072
2073     s->ctx.fs_root = qemu_strdup(fse->path);
2074     len = strlen(conf->tag);
2075     if (len > MAX_TAG_LEN) {
2076         len = MAX_TAG_LEN;
2077     }
2078     /* s->tag is non-NULL terminated string */
2079     s->tag = qemu_malloc(len);
2080     memcpy(s->tag, conf->tag, len);
2081     s->tag_len = len;
2082     s->ctx.uid = -1;
2083
2084     s->ops = fse->ops;
2085     s->vdev.get_features = virtio_9p_get_features;
2086     s->config_size = sizeof(struct virtio_9p_config) +
2087                         s->tag_len;
2088     s->vdev.get_config = virtio_9p_get_config;
2089
2090     return &s->vdev;
2091 }