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 enum {
  25     Oread   = 0x00,
  26     Owrite  = 0x01,
  27     Ordwr   = 0x02,
  28     Oexec   = 0x03,
  29     Oexcl   = 0x04,
  30     Otrunc  = 0x10,
  31     Orexec  = 0x20,
  32     Orclose = 0x40,
  33     Oappend = 0x80,
  34 };
  35
  36 static int omode_to_uflags(int8_t mode)
  37 {
  38     int ret = 0;
  39
  40     switch (mode & 3) {
  41     case Oread:
  42         ret = O_RDONLY;
  43         break;
  44     case Ordwr:
  45         ret = O_RDWR;
  46         break;
  47     case Owrite:
  48         ret = O_WRONLY;
  49         break;
  50     case Oexec:
  51         ret = O_RDONLY;
  52         break;
  53     }
  54
  55     if (mode & Otrunc) {
  56         ret |= O_TRUNC;
  57     }
  58
  59     if (mode & Oappend) {
  60         ret |= O_APPEND;
  61     }
  62
  63     if (mode & Oexcl) {
  64         ret |= O_EXCL;
  65     }
  66
  67     return ret;
  68 }
  69
  70 void cred_init(FsCred *credp)
  71 {
  72     credp->fc_uid = -1;
  73     credp->fc_gid = -1;
  74     credp->fc_mode = -1;
  75     credp->fc_rdev = -1;
  76 }
  77
  78 static int v9fs_do_lstat(V9fsState *s, V9fsString *path, struct stat *stbuf)
  79 {
  80     return s->ops->lstat(&s->ctx, path->data, stbuf);
  81 }
  82
  83 static ssize_t v9fs_do_readlink(V9fsState *s, V9fsString *path, V9fsString *buf)
  84 {
  85     ssize_t len;
  86
  87     buf->data = qemu_malloc(1024);
  88
  89     len = s->ops->readlink(&s->ctx, path->data, buf->data, 1024 - 1);
  90     if (len > -1) {
  91         buf->size = len;
  92         buf->data[len] = 0;
  93     }
  94
  95     return len;
  96 }
  97
  98 static int v9fs_do_close(V9fsState *s, int fd)
  99 {
 100     return s->ops->close(&s->ctx, fd);
 101 }
 102
 103 static int v9fs_do_closedir(V9fsState *s, DIR *dir)
 104 {
 105     return s->ops->closedir(&s->ctx, dir);
 106 }
 107
 108 static int v9fs_do_open(V9fsState *s, V9fsString *path, int flags)
 109 {
 110     return s->ops->open(&s->ctx, path->data, flags);
 111 }
 112
 113 static DIR *v9fs_do_opendir(V9fsState *s, V9fsString *path)
 114 {
 115     return s->ops->opendir(&s->ctx, path->data);
 116 }
 117
 118 static void v9fs_do_rewinddir(V9fsState *s, DIR *dir)
 119 {
 120     return s->ops->rewinddir(&s->ctx, dir);
 121 }
 122
 123 static off_t v9fs_do_telldir(V9fsState *s, DIR *dir)
 124 {
 125     return s->ops->telldir(&s->ctx, dir);
 126 }
 127
 128 static struct dirent *v9fs_do_readdir(V9fsState *s, DIR *dir)
 129 {
 130     return s->ops->readdir(&s->ctx, dir);
 131 }
 132
 133 static void v9fs_do_seekdir(V9fsState *s, DIR *dir, off_t off)
 134 {
 135     return s->ops->seekdir(&s->ctx, dir, off);
 136 }
 137
 138 static int v9fs_do_readv(V9fsState *s, int fd, const struct iovec *iov,
 139                             int iovcnt)
 140 {
 141     return s->ops->readv(&s->ctx, fd, iov, iovcnt);
 142 }
 143
 144 static off_t v9fs_do_lseek(V9fsState *s, int fd, off_t offset, int whence)
 145 {
 146     return s->ops->lseek(&s->ctx, fd, offset, whence);
 147 }
 148
 149 static int v9fs_do_writev(V9fsState *s, int fd, const struct iovec *iov,
 150                        int iovcnt)
 151 {
 152     return s->ops->writev(&s->ctx, fd, iov, iovcnt);
 153 }
 154
 155 static int v9fs_do_chmod(V9fsState *s, V9fsString *path, mode_t mode)
 156 {
 157     FsCred cred;
 158     cred_init(&cred);
 159     cred.fc_mode = mode;
 160     return s->ops->chmod(&s->ctx, path->data, &cred);
 161 }
 162
 163 static int v9fs_do_mknod(V9fsState *s, V9fsCreateState *vs, mode_t mode,
 164         dev_t dev)
 165 {
 166     FsCred cred;
 167     cred_init(&cred);
 168     cred.fc_uid = vs->fidp->uid;
 169     cred.fc_mode = mode;
 170     cred.fc_rdev = dev;
 171     return s->ops->mknod(&s->ctx, vs->fullname.data, &cred);
 172 }
 173
 174 static int v9fs_do_mkdir(V9fsState *s, V9fsCreateState *vs)
 175 {
 176     FsCred cred;
 177
 178     cred_init(&cred);
 179     cred.fc_uid = vs->fidp->uid;
 180     cred.fc_mode = vs->perm & 0777;
 181
 182     return s->ops->mkdir(&s->ctx, vs->fullname.data, &cred);
 183 }
 184
 185 static int v9fs_do_fstat(V9fsState *s, int fd, struct stat *stbuf)
 186 {
 187     return s->ops->fstat(&s->ctx, fd, stbuf);
 188 }
 189
 190 static int v9fs_do_open2(V9fsState *s, V9fsCreateState *vs)
 191 {
 192     FsCred cred;
 193     int flags;
 194
 195     cred_init(&cred);
 196     cred.fc_uid = vs->fidp->uid;
 197     cred.fc_mode = vs->perm & 0777;
 198     flags = omode_to_uflags(vs->mode) | O_CREAT;
 199
 200     return s->ops->open2(&s->ctx, vs->fullname.data, flags, &cred);
 201 }
 202
 203 static int v9fs_do_symlink(V9fsState *s, V9fsCreateState *vs)
 204 {
 205     FsCred cred;
 206     cred_init(&cred);
 207     cred.fc_uid = vs->fidp->uid;
 208     cred.fc_mode = vs->perm | 0777;
 209
 210     return s->ops->symlink(&s->ctx, vs->extension.data, vs->fullname.data,
 211             &cred);
 212 }
 213
 214 static int v9fs_do_link(V9fsState *s, V9fsString *oldpath, V9fsString *newpath)
 215 {
 216     return s->ops->link(&s->ctx, oldpath->data, newpath->data);
 217 }
 218
 219 static int v9fs_do_truncate(V9fsState *s, V9fsString *path, off_t size)
 220 {
 221     return s->ops->truncate(&s->ctx, path->data, size);
 222 }
 223
 224 static int v9fs_do_rename(V9fsState *s, V9fsString *oldpath,
 225                             V9fsString *newpath)
 226 {
 227     return s->ops->rename(&s->ctx, oldpath->data, newpath->data);
 228 }
 229
 230 static int v9fs_do_chown(V9fsState *s, V9fsString *path, uid_t uid, gid_t gid)
 231 {
 232     FsCred cred;
 233     cred_init(&cred);
 234     cred.fc_uid = uid;
 235     cred.fc_gid = gid;
 236
 237     return s->ops->chown(&s->ctx, path->data, &cred);
 238 }
 239
 240 static int v9fs_do_utimensat(V9fsState *s, V9fsString *path, V9fsStat v9stat)
 241 {
 242     struct timespec ts[2];
 243
 244     if (v9stat.atime != -1) {
 245         ts[0].tv_sec = v9stat.atime;
 246         ts[0].tv_nsec = 0;
 247     } else {
 248         ts[0].tv_nsec = UTIME_OMIT;
 249     }
 250
 251     if (v9stat.mtime != -1) {
 252         ts[1].tv_sec = v9stat.mtime;
 253         ts[1].tv_nsec = 0;
 254     } else {
 255         ts[1].tv_nsec = UTIME_OMIT;
 256     }
 257
 258     return s->ops->utimensat(&s->ctx, path->data, ts);
 259 }
 260
 261 static int v9fs_do_remove(V9fsState *s, V9fsString *path)
 262 {
 263     return s->ops->remove(&s->ctx, path->data);
 264 }
 265
 266 static int v9fs_do_fsync(V9fsState *s, int fd)
 267 {
 268     return s->ops->fsync(&s->ctx, fd);
 269 }
 270
 271 static int v9fs_do_statfs(V9fsState *s, V9fsString *path, struct statfs *stbuf)
 272 {
 273     return s->ops->statfs(&s->ctx, path->data, stbuf);
 274 }
 275
 276 static void v9fs_string_init(V9fsString *str)
 277 {
 278     str->data = NULL;
 279     str->size = 0;
 280 }
 281
 282 static void v9fs_string_free(V9fsString *str)
 283 {
 284     qemu_free(str->data);
 285     str->data = NULL;
 286     str->size = 0;
 287 }
 288
 289 static void v9fs_string_null(V9fsString *str)
 290 {
 291     v9fs_string_free(str);
 292 }
 293
 294 static int number_to_string(void *arg, char type)
 295 {
 296     unsigned int ret = 0;
 297
 298     switch (type) {
 299     case 'u': {
 300         unsigned int num = *(unsigned int *)arg;
 301
 302         do {
 303             ret++;
 304             num = num/10;
 305         } while (num);
 306         break;
 307     }
 308     default:
 309         printf("Number_to_string: Unknown number format\n");
 310         return -1;
 311     }
 312
 313     return ret;
 314 }
 315
 316 static int v9fs_string_alloc_printf(char **strp, const char *fmt, va_list ap)
 317 {
 318     va_list ap2;
 319     char *iter = (char *)fmt;
 320     int len = 0;
 321     int nr_args = 0;
 322     char *arg_char_ptr;
 323     unsigned int arg_uint;
 324
 325     /* Find the number of %'s that denotes an argument */
 326     for (iter = strstr(iter, "%"); iter; iter = strstr(iter, "%")) {
 327         nr_args++;
 328         iter++;
 329     }
 330
 331     len = strlen(fmt) - 2*nr_args;
 332
 333     if (!nr_args) {
 334         goto alloc_print;
 335     }
 336
 337     va_copy(ap2, ap);
 338
 339     iter = (char *)fmt;
 340
 341     /* Now parse the format string */
 342     for (iter = strstr(iter, "%"); iter; iter = strstr(iter, "%")) {
 343         iter++;
 344         switch (*iter) {
 345         case 'u':
 346             arg_uint = va_arg(ap2, unsigned int);
 347             len += number_to_string((void *)&arg_uint, 'u');
 348             break;
 349         case 's':
 350             arg_char_ptr = va_arg(ap2, char *);
 351             len += strlen(arg_char_ptr);
 352             break;
 353         case 'c':
 354             len += 1;
 355             break;
 356         default:
 357             fprintf(stderr,
 358                     "v9fs_string_alloc_printf:Incorrect format %c", *iter);
 359             return -1;
 360         }
 361         iter++;
 362     }
 363
 364 alloc_print:
 365     *strp = qemu_malloc((len + 1) * sizeof(**strp));
 366
 367     return vsprintf(*strp, fmt, ap);
 368 }
 369
 370 static void v9fs_string_sprintf(V9fsString *str, const char *fmt, ...)
 371 {
 372     va_list ap;
 373     int err;
 374
 375     v9fs_string_free(str);
 376
 377     va_start(ap, fmt);
 378     err = v9fs_string_alloc_printf(&str->data, fmt, ap);
 379     BUG_ON(err == -1);
 380     va_end(ap);
 381
 382     str->size = err;
 383 }
 384
 385 static void v9fs_string_copy(V9fsString *lhs, V9fsString *rhs)
 386 {
 387     v9fs_string_free(lhs);
 388     v9fs_string_sprintf(lhs, "%s", rhs->data);
 389 }
 390
 391 static size_t v9fs_string_size(V9fsString *str)
 392 {
 393     return str->size;
 394 }
 395
 396 static V9fsFidState *lookup_fid(V9fsState *s, int32_t fid)
 397 {
 398     V9fsFidState *f;
 399
 400     for (f = s->fid_list; f; f = f->next) {
 401         if (f->fid == fid) {
 402             return f;
 403         }
 404     }
 405
 406     return NULL;
 407 }
 408
 409 static V9fsFidState *alloc_fid(V9fsState *s, int32_t fid)
 410 {
 411     V9fsFidState *f;
 412
 413     f = lookup_fid(s, fid);
 414     if (f) {
 415         return NULL;
 416     }
 417
 418     f = qemu_mallocz(sizeof(V9fsFidState));
 419
 420     f->fid = fid;
 421     f->fd = -1;
 422     f->dir = NULL;
 423
 424     f->next = s->fid_list;
 425     s->fid_list = f;
 426
 427     return f;
 428 }
 429
 430 static int free_fid(V9fsState *s, int32_t fid)
 431 {
 432     V9fsFidState **fidpp, *fidp;
 433
 434     for (fidpp = &s->fid_list; *fidpp; fidpp = &(*fidpp)->next) {
 435         if ((*fidpp)->fid == fid) {
 436             break;
 437         }
 438     }
 439
 440     if (*fidpp == NULL) {
 441         return -ENOENT;
 442     }
 443
 444     fidp = *fidpp;
 445     *fidpp = fidp->next;
 446
 447     if (fidp->fd != -1) {
 448         v9fs_do_close(s, fidp->fd);
 449     }
 450     if (fidp->dir) {
 451         v9fs_do_closedir(s, fidp->dir);
 452     }
 453     v9fs_string_free(&fidp->path);
 454     qemu_free(fidp);
 455
 456     return 0;
 457 }
 458
 459 #define P9_QID_TYPE_DIR         0x80
 460 #define P9_QID_TYPE_SYMLINK     0x02
 461
 462 #define P9_STAT_MODE_DIR        0x80000000
 463 #define P9_STAT_MODE_APPEND     0x40000000
 464 #define P9_STAT_MODE_EXCL       0x20000000
 465 #define P9_STAT_MODE_MOUNT      0x10000000
 466 #define P9_STAT_MODE_AUTH       0x08000000
 467 #define P9_STAT_MODE_TMP        0x04000000
 468 #define P9_STAT_MODE_SYMLINK    0x02000000
 469 #define P9_STAT_MODE_LINK       0x01000000
 470 #define P9_STAT_MODE_DEVICE     0x00800000
 471 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
 472 #define P9_STAT_MODE_SOCKET     0x00100000
 473 #define P9_STAT_MODE_SETUID     0x00080000
 474 #define P9_STAT_MODE_SETGID     0x00040000
 475 #define P9_STAT_MODE_SETVTX     0x00010000
 476
 477 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR |          \
 478                                 P9_STAT_MODE_SYMLINK |      \
 479                                 P9_STAT_MODE_LINK |         \
 480                                 P9_STAT_MODE_DEVICE |       \
 481                                 P9_STAT_MODE_NAMED_PIPE |   \
 482                                 P9_STAT_MODE_SOCKET)
 483
 484 /* This is the algorithm from ufs in spfs */
 485 static void stat_to_qid(const struct stat *stbuf, V9fsQID *qidp)
 486 {
 487     size_t size;
 488
 489     size = MIN(sizeof(stbuf->st_ino), sizeof(qidp->path));
 490     memcpy(&qidp->path, &stbuf->st_ino, size);
 491     qidp->version = stbuf->st_mtime ^ (stbuf->st_size << 8);
 492     qidp->type = 0;
 493     if (S_ISDIR(stbuf->st_mode)) {
 494         qidp->type |= P9_QID_TYPE_DIR;
 495     }
 496     if (S_ISLNK(stbuf->st_mode)) {
 497         qidp->type |= P9_QID_TYPE_SYMLINK;
 498     }
 499 }
 500
 501 static int fid_to_qid(V9fsState *s, V9fsFidState *fidp, V9fsQID *qidp)
 502 {
 503     struct stat stbuf;
 504     int err;
 505
 506     err = v9fs_do_lstat(s, &fidp->path, &stbuf);
 507     if (err) {
 508         return err;
 509     }
 510
 511     stat_to_qid(&stbuf, qidp);
 512     return 0;
 513 }
 514
 515 static V9fsPDU *alloc_pdu(V9fsState *s)
 516 {
 517     V9fsPDU *pdu = NULL;
 518
 519     if (!QLIST_EMPTY(&s->free_list)) {
 520         pdu = QLIST_FIRST(&s->free_list);
 521         QLIST_REMOVE(pdu, next);
 522     }
 523     return pdu;
 524 }
 525
 526 static void free_pdu(V9fsState *s, V9fsPDU *pdu)
 527 {
 528     if (pdu) {
 529         QLIST_INSERT_HEAD(&s->free_list, pdu, next);
 530     }
 531 }
 532
 533 size_t pdu_packunpack(void *addr, struct iovec *sg, int sg_count,
 534                         size_t offset, size_t size, int pack)
 535 {
 536     int i = 0;
 537     size_t copied = 0;
 538
 539     for (i = 0; size && i < sg_count; i++) {
 540         size_t len;
 541         if (offset >= sg[i].iov_len) {
 542             /* skip this sg */
 543             offset -= sg[i].iov_len;
 544             continue;
 545         } else {
 546             len = MIN(sg[i].iov_len - offset, size);
 547             if (pack) {
 548                 memcpy(sg[i].iov_base + offset, addr, len);
 549             } else {
 550                 memcpy(addr, sg[i].iov_base + offset, len);
 551             }
 552             size -= len;
 553             copied += len;
 554             addr += len;
 555             if (size) {
 556                 offset = 0;
 557                 continue;
 558             }
 559         }
 560     }
 561
 562     return copied;
 563 }
 564
 565 static size_t pdu_unpack(void *dst, V9fsPDU *pdu, size_t offset, size_t size)
 566 {
 567     return pdu_packunpack(dst, pdu->elem.out_sg, pdu->elem.out_num,
 568                          offset, size, 0);
 569 }
 570
 571 static size_t pdu_pack(V9fsPDU *pdu, size_t offset, const void *src,
 572                         size_t size)
 573 {
 574     return pdu_packunpack((void *)src, pdu->elem.in_sg, pdu->elem.in_num,
 575                              offset, size, 1);
 576 }
 577
 578 static int pdu_copy_sg(V9fsPDU *pdu, size_t offset, int rx, struct iovec *sg)
 579 {
 580     size_t pos = 0;
 581     int i, j;
 582     struct iovec *src_sg;
 583     unsigned int num;
 584
 585     if (rx) {
 586         src_sg = pdu->elem.in_sg;
 587         num = pdu->elem.in_num;
 588     } else {
 589         src_sg = pdu->elem.out_sg;
 590         num = pdu->elem.out_num;
 591     }
 592
 593     j = 0;
 594     for (i = 0; i < num; i++) {
 595         if (offset <= pos) {
 596             sg[j].iov_base = src_sg[i].iov_base;
 597             sg[j].iov_len = src_sg[i].iov_len;
 598             j++;
 599         } else if (offset < (src_sg[i].iov_len + pos)) {
 600             sg[j].iov_base = src_sg[i].iov_base;
 601             sg[j].iov_len = src_sg[i].iov_len;
 602             sg[j].iov_base += (offset - pos);
 603             sg[j].iov_len -= (offset - pos);
 604             j++;
 605         }
 606         pos += src_sg[i].iov_len;
 607     }
 608
 609     return j;
 610 }
 611
 612 static size_t pdu_unmarshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
 613 {
 614     size_t old_offset = offset;
 615     va_list ap;
 616     int i;
 617
 618     va_start(ap, fmt);
 619     for (i = 0; fmt[i]; i++) {
 620         switch (fmt[i]) {
 621         case 'b': {
 622             uint8_t *valp = va_arg(ap, uint8_t *);
 623             offset += pdu_unpack(valp, pdu, offset, sizeof(*valp));
 624             break;
 625         }
 626         case 'w': {
 627             uint16_t val, *valp;
 628             valp = va_arg(ap, uint16_t *);
 629             val = le16_to_cpupu(valp);
 630             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 631             *valp = val;
 632             break;
 633         }
 634         case 'd': {
 635             uint32_t val, *valp;
 636             valp = va_arg(ap, uint32_t *);
 637             val = le32_to_cpupu(valp);
 638             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 639             *valp = val;
 640             break;
 641         }
 642         case 'q': {
 643             uint64_t val, *valp;
 644             valp = va_arg(ap, uint64_t *);
 645             val = le64_to_cpup(valp);
 646             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 647             *valp = val;
 648             break;
 649         }
 650         case 'v': {
 651             struct iovec *iov = va_arg(ap, struct iovec *);
 652             int *iovcnt = va_arg(ap, int *);
 653             *iovcnt = pdu_copy_sg(pdu, offset, 0, iov);
 654             break;
 655         }
 656         case 's': {
 657             V9fsString *str = va_arg(ap, V9fsString *);
 658             offset += pdu_unmarshal(pdu, offset, "w", &str->size);
 659             /* FIXME: sanity check str->size */
 660             str->data = qemu_malloc(str->size + 1);
 661             offset += pdu_unpack(str->data, pdu, offset, str->size);
 662             str->data[str->size] = 0;
 663             break;
 664         }
 665         case 'Q': {
 666             V9fsQID *qidp = va_arg(ap, V9fsQID *);
 667             offset += pdu_unmarshal(pdu, offset, "bdq",
 668                         &qidp->type, &qidp->version, &qidp->path);
 669             break;
 670         }
 671         case 'S': {
 672             V9fsStat *statp = va_arg(ap, V9fsStat *);
 673             offset += pdu_unmarshal(pdu, offset, "wwdQdddqsssssddd",
 674                         &statp->size, &statp->type, &statp->dev,
 675                         &statp->qid, &statp->mode, &statp->atime,
 676                         &statp->mtime, &statp->length,
 677                         &statp->name, &statp->uid, &statp->gid,
 678                         &statp->muid, &statp->extension,
 679                         &statp->n_uid, &statp->n_gid,
 680                         &statp->n_muid);
 681             break;
 682         }
 683         default:
 684             break;
 685         }
 686     }
 687
 688     va_end(ap);
 689
 690     return offset - old_offset;
 691 }
 692
 693 static size_t pdu_marshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
 694 {
 695     size_t old_offset = offset;
 696     va_list ap;
 697     int i;
 698
 699     va_start(ap, fmt);
 700     for (i = 0; fmt[i]; i++) {
 701         switch (fmt[i]) {
 702         case 'b': {
 703             uint8_t val = va_arg(ap, int);
 704             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 705             break;
 706         }
 707         case 'w': {
 708             uint16_t val;
 709             cpu_to_le16w(&val, va_arg(ap, int));
 710             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 711             break;
 712         }
 713         case 'd': {
 714             uint32_t val;
 715             cpu_to_le32w(&val, va_arg(ap, uint32_t));
 716             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 717             break;
 718         }
 719         case 'q': {
 720             uint64_t val;
 721             cpu_to_le64w(&val, va_arg(ap, uint64_t));
 722             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 723             break;
 724         }
 725         case 'v': {
 726             struct iovec *iov = va_arg(ap, struct iovec *);
 727             int *iovcnt = va_arg(ap, int *);
 728             *iovcnt = pdu_copy_sg(pdu, offset, 1, iov);
 729             break;
 730         }
 731         case 's': {
 732             V9fsString *str = va_arg(ap, V9fsString *);
 733             offset += pdu_marshal(pdu, offset, "w", str->size);
 734             offset += pdu_pack(pdu, offset, str->data, str->size);
 735             break;
 736         }
 737         case 'Q': {
 738             V9fsQID *qidp = va_arg(ap, V9fsQID *);
 739             offset += pdu_marshal(pdu, offset, "bdq",
 740                         qidp->type, qidp->version, qidp->path);
 741             break;
 742         }
 743         case 'S': {
 744             V9fsStat *statp = va_arg(ap, V9fsStat *);
 745             offset += pdu_marshal(pdu, offset, "wwdQdddqsssssddd",
 746                         statp->size, statp->type, statp->dev,
 747                         &statp->qid, statp->mode, statp->atime,
 748                         statp->mtime, statp->length, &statp->name,
 749                         &statp->uid, &statp->gid, &statp->muid,
 750                         &statp->extension, statp->n_uid,
 751                         statp->n_gid, statp->n_muid);
 752             break;
 753         }
 754         case 'A': {
 755             V9fsStatDotl *statp = va_arg(ap, V9fsStatDotl *);
 756             offset += pdu_marshal(pdu, offset, "qQdddqqqqqqqqqqqqqqq",
 757                         statp->st_result_mask,
 758                         &statp->qid, statp->st_mode,
 759                         statp->st_uid, statp->st_gid,
 760                         statp->st_nlink, statp->st_rdev,
 761                         statp->st_size, statp->st_blksize, statp->st_blocks,
 762                         statp->st_atime_sec, statp->st_atime_nsec,
 763                         statp->st_mtime_sec, statp->st_mtime_nsec,
 764                         statp->st_ctime_sec, statp->st_ctime_nsec,
 765                         statp->st_btime_sec, statp->st_btime_nsec,
 766                         statp->st_gen, statp->st_data_version);
 767             break;
 768         }
 769         default:
 770             break;
 771         }
 772     }
 773     va_end(ap);
 774
 775     return offset - old_offset;
 776 }
 777
 778 static void complete_pdu(V9fsState *s, V9fsPDU *pdu, ssize_t len)
 779 {
 780     int8_t id = pdu->id + 1; /* Response */
 781
 782     if (len < 0) {
 783         V9fsString str;
 784         int err = -len;
 785
 786         str.data = strerror(err);
 787         str.size = strlen(str.data);
 788
 789         len = 7;
 790         len += pdu_marshal(pdu, len, "s", &str);
 791         if (dotu) {
 792             len += pdu_marshal(pdu, len, "d", err);
 793         }
 794
 795         id = P9_RERROR;
 796     }
 797
 798     /* fill out the header */
 799     pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag);
 800
 801     /* keep these in sync */
 802     pdu->size = len;
 803     pdu->id = id;
 804
 805     /* push onto queue and notify */
 806     virtqueue_push(s->vq, &pdu->elem, len);
 807
 808     /* FIXME: we should batch these completions */
 809     virtio_notify(&s->vdev, s->vq);
 810
 811     free_pdu(s, pdu);
 812 }
 813
 814 static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
 815 {
 816     mode_t ret;
 817
 818     ret = mode & 0777;
 819     if (mode & P9_STAT_MODE_DIR) {
 820         ret |= S_IFDIR;
 821     }
 822
 823     if (dotu) {
 824         if (mode & P9_STAT_MODE_SYMLINK) {
 825             ret |= S_IFLNK;
 826         }
 827         if (mode & P9_STAT_MODE_SOCKET) {
 828             ret |= S_IFSOCK;
 829         }
 830         if (mode & P9_STAT_MODE_NAMED_PIPE) {
 831             ret |= S_IFIFO;
 832         }
 833         if (mode & P9_STAT_MODE_DEVICE) {
 834             if (extension && extension->data[0] == 'c') {
 835                 ret |= S_IFCHR;
 836             } else {
 837                 ret |= S_IFBLK;
 838             }
 839         }
 840     }
 841
 842     if (!(ret&~0777)) {
 843         ret |= S_IFREG;
 844     }
 845
 846     if (mode & P9_STAT_MODE_SETUID) {
 847         ret |= S_ISUID;
 848     }
 849     if (mode & P9_STAT_MODE_SETGID) {
 850         ret |= S_ISGID;
 851     }
 852     if (mode & P9_STAT_MODE_SETVTX) {
 853         ret |= S_ISVTX;
 854     }
 855
 856     return ret;
 857 }
 858
 859 static int donttouch_stat(V9fsStat *stat)
 860 {
 861     if (stat->type == -1 &&
 862         stat->dev == -1 &&
 863         stat->qid.type == -1 &&
 864         stat->qid.version == -1 &&
 865         stat->qid.path == -1 &&
 866         stat->mode == -1 &&
 867         stat->atime == -1 &&
 868         stat->mtime == -1 &&
 869         stat->length == -1 &&
 870         !stat->name.size &&
 871         !stat->uid.size &&
 872         !stat->gid.size &&
 873         !stat->muid.size &&
 874         stat->n_uid == -1 &&
 875         stat->n_gid == -1 &&
 876         stat->n_muid == -1) {
 877         return 1;
 878     }
 879
 880     return 0;
 881 }
 882
 883 static void v9fs_stat_free(V9fsStat *stat)
 884 {
 885     v9fs_string_free(&stat->name);
 886     v9fs_string_free(&stat->uid);
 887     v9fs_string_free(&stat->gid);
 888     v9fs_string_free(&stat->muid);
 889     v9fs_string_free(&stat->extension);
 890 }
 891
 892 static uint32_t stat_to_v9mode(const struct stat *stbuf)
 893 {
 894     uint32_t mode;
 895
 896     mode = stbuf->st_mode & 0777;
 897     if (S_ISDIR(stbuf->st_mode)) {
 898         mode |= P9_STAT_MODE_DIR;
 899     }
 900
 901     if (dotu) {
 902         if (S_ISLNK(stbuf->st_mode)) {
 903             mode |= P9_STAT_MODE_SYMLINK;
 904         }
 905
 906         if (S_ISSOCK(stbuf->st_mode)) {
 907             mode |= P9_STAT_MODE_SOCKET;
 908         }
 909
 910         if (S_ISFIFO(stbuf->st_mode)) {
 911             mode |= P9_STAT_MODE_NAMED_PIPE;
 912         }
 913
 914         if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
 915             mode |= P9_STAT_MODE_DEVICE;
 916         }
 917
 918         if (stbuf->st_mode & S_ISUID) {
 919             mode |= P9_STAT_MODE_SETUID;
 920         }
 921
 922         if (stbuf->st_mode & S_ISGID) {
 923             mode |= P9_STAT_MODE_SETGID;
 924         }
 925
 926         if (stbuf->st_mode & S_ISVTX) {
 927             mode |= P9_STAT_MODE_SETVTX;
 928         }
 929     }
 930
 931     return mode;
 932 }
 933
 934 static int stat_to_v9stat(V9fsState *s, V9fsString *name,
 935                             const struct stat *stbuf,
 936                             V9fsStat *v9stat)
 937 {
 938     int err;
 939     const char *str;
 940
 941     memset(v9stat, 0, sizeof(*v9stat));
 942
 943     stat_to_qid(stbuf, &v9stat->qid);
 944     v9stat->mode = stat_to_v9mode(stbuf);
 945     v9stat->atime = stbuf->st_atime;
 946     v9stat->mtime = stbuf->st_mtime;
 947     v9stat->length = stbuf->st_size;
 948
 949     v9fs_string_null(&v9stat->uid);
 950     v9fs_string_null(&v9stat->gid);
 951     v9fs_string_null(&v9stat->muid);
 952
 953     if (dotu) {
 954         v9stat->n_uid = stbuf->st_uid;
 955         v9stat->n_gid = stbuf->st_gid;
 956         v9stat->n_muid = 0;
 957
 958         v9fs_string_null(&v9stat->extension);
 959
 960         if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
 961             err = v9fs_do_readlink(s, name, &v9stat->extension);
 962             if (err == -1) {
 963                 err = -errno;
 964                 return err;
 965             }
 966             v9stat->extension.data[err] = 0;
 967             v9stat->extension.size = err;
 968         } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
 969             v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
 970                     S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
 971                     major(stbuf->st_rdev), minor(stbuf->st_rdev));
 972         } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
 973             v9fs_string_sprintf(&v9stat->extension, "%s %u",
 974                     "HARDLINKCOUNT", stbuf->st_nlink);
 975         }
 976     }
 977
 978     str = strrchr(name->data, '/');
 979     if (str) {
 980         str += 1;
 981     } else {
 982         str = name->data;
 983     }
 984
 985     v9fs_string_sprintf(&v9stat->name, "%s", str);
 986
 987     v9stat->size = 61 +
 988         v9fs_string_size(&v9stat->name) +
 989         v9fs_string_size(&v9stat->uid) +
 990         v9fs_string_size(&v9stat->gid) +
 991         v9fs_string_size(&v9stat->muid) +
 992         v9fs_string_size(&v9stat->extension);
 993     return 0;
 994 }
 995
 996 #define P9_STATS_MODE          0x00000001ULL
 997 #define P9_STATS_NLINK         0x00000002ULL
 998 #define P9_STATS_UID           0x00000004ULL
 999 #define P9_STATS_GID           0x00000008ULL
1000 #define P9_STATS_RDEV          0x00000010ULL
1001 #define P9_STATS_ATIME         0x00000020ULL
1002 #define P9_STATS_MTIME         0x00000040ULL
1003 #define P9_STATS_CTIME         0x00000080ULL
1004 #define P9_STATS_INO           0x00000100ULL
1005 #define P9_STATS_SIZE          0x00000200ULL
1006 #define P9_STATS_BLOCKS        0x00000400ULL
1007
1008 #define P9_STATS_BTIME         0x00000800ULL
1009 #define P9_STATS_GEN           0x00001000ULL
1010 #define P9_STATS_DATA_VERSION  0x00002000ULL
1011
1012 #define P9_STATS_BASIC         0x000007ffULL /* Mask for fields up to BLOCKS */
1013 #define P9_STATS_ALL           0x00003fffULL /* Mask for All fields above */
1014
1015
1016 static void stat_to_v9stat_dotl(V9fsState *s, const struct stat *stbuf,
1017                             V9fsStatDotl *v9lstat)
1018 {
1019     memset(v9lstat, 0, sizeof(*v9lstat));
1020
1021     v9lstat->st_mode = stbuf->st_mode;
1022     v9lstat->st_nlink = stbuf->st_nlink;
1023     v9lstat->st_uid = stbuf->st_uid;
1024     v9lstat->st_gid = stbuf->st_gid;
1025     v9lstat->st_rdev = stbuf->st_rdev;
1026     v9lstat->st_size = stbuf->st_size;
1027     v9lstat->st_blksize = stbuf->st_blksize;
1028     v9lstat->st_blocks = stbuf->st_blocks;
1029     v9lstat->st_atime_sec = stbuf->st_atime;
1030     v9lstat->st_atime_nsec = stbuf->st_atim.tv_nsec;
1031     v9lstat->st_mtime_sec = stbuf->st_mtime;
1032     v9lstat->st_mtime_nsec = stbuf->st_mtim.tv_nsec;
1033     v9lstat->st_ctime_sec = stbuf->st_ctime;
1034     v9lstat->st_ctime_nsec = stbuf->st_ctim.tv_nsec;
1035     /* Currently we only support BASIC fields in stat */
1036     v9lstat->st_result_mask = P9_STATS_BASIC;
1037
1038     stat_to_qid(stbuf, &v9lstat->qid);
1039 }
1040
1041 static struct iovec *adjust_sg(struct iovec *sg, int len, int *iovcnt)
1042 {
1043     while (len && *iovcnt) {
1044         if (len < sg->iov_len) {
1045             sg->iov_len -= len;
1046             sg->iov_base += len;
1047             len = 0;
1048         } else {
1049             len -= sg->iov_len;
1050             sg++;
1051             *iovcnt -= 1;
1052         }
1053     }
1054
1055     return sg;
1056 }
1057
1058 static struct iovec *cap_sg(struct iovec *sg, int cap, int *cnt)
1059 {
1060     int i;
1061     int total = 0;
1062
1063     for (i = 0; i < *cnt; i++) {
1064         if ((total + sg[i].iov_len) > cap) {
1065             sg[i].iov_len -= ((total + sg[i].iov_len) - cap);
1066             i++;
1067             break;
1068         }
1069         total += sg[i].iov_len;
1070     }
1071
1072     *cnt = i;
1073
1074     return sg;
1075 }
1076
1077 static void print_sg(struct iovec *sg, int cnt)
1078 {
1079     int i;
1080
1081     printf("sg[%d]: {", cnt);
1082     for (i = 0; i < cnt; i++) {
1083         if (i) {
1084             printf(", ");
1085         }
1086         printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
1087     }
1088     printf("}\n");
1089 }
1090
1091 static void v9fs_fix_path(V9fsString *dst, V9fsString *src, int len)
1092 {
1093     V9fsString str;
1094     v9fs_string_init(&str);
1095     v9fs_string_copy(&str, dst);
1096     v9fs_string_sprintf(dst, "%s%s", src->data, str.data+len);
1097     v9fs_string_free(&str);
1098 }
1099
1100 static void v9fs_version(V9fsState *s, V9fsPDU *pdu)
1101 {
1102     V9fsString version;
1103     size_t offset = 7;
1104
1105     pdu_unmarshal(pdu, offset, "ds", &s->msize, &version);
1106
1107     if (!strcmp(version.data, "9P2000.u")) {
1108         s->proto_version = V9FS_PROTO_2000U;
1109     } else if (!strcmp(version.data, "9P2000.L")) {
1110         s->proto_version = V9FS_PROTO_2000L;
1111     } else {
1112         v9fs_string_sprintf(&version, "unknown");
1113     }
1114
1115     offset += pdu_marshal(pdu, offset, "ds", s->msize, &version);
1116     complete_pdu(s, pdu, offset);
1117
1118     v9fs_string_free(&version);
1119 }
1120
1121 static void v9fs_attach(V9fsState *s, V9fsPDU *pdu)
1122 {
1123     int32_t fid, afid, n_uname;
1124     V9fsString uname, aname;
1125     V9fsFidState *fidp;
1126     V9fsQID qid;
1127     size_t offset = 7;
1128     ssize_t err;
1129
1130     pdu_unmarshal(pdu, offset, "ddssd", &fid, &afid, &uname, &aname, &n_uname);
1131
1132     fidp = alloc_fid(s, fid);
1133     if (fidp == NULL) {
1134         err = -EINVAL;
1135         goto out;
1136     }
1137
1138     fidp->uid = n_uname;
1139
1140     v9fs_string_sprintf(&fidp->path, "%s", "/");
1141     err = fid_to_qid(s, fidp, &qid);
1142     if (err) {
1143         err = -EINVAL;
1144         free_fid(s, fid);
1145         goto out;
1146     }
1147
1148     offset += pdu_marshal(pdu, offset, "Q", &qid);
1149
1150     err = offset;
1151 out:
1152     complete_pdu(s, pdu, err);
1153     v9fs_string_free(&uname);
1154     v9fs_string_free(&aname);
1155 }
1156
1157 static void v9fs_stat_post_lstat(V9fsState *s, V9fsStatState *vs, int err)
1158 {
1159     if (err == -1) {
1160         err = -errno;
1161         goto out;
1162     }
1163
1164     err = stat_to_v9stat(s, &vs->fidp->path, &vs->stbuf, &vs->v9stat);
1165     if (err) {
1166         goto out;
1167     }
1168     vs->offset += pdu_marshal(vs->pdu, vs->offset, "wS", 0, &vs->v9stat);
1169     err = vs->offset;
1170
1171 out:
1172     complete_pdu(s, vs->pdu, err);
1173     v9fs_stat_free(&vs->v9stat);
1174     qemu_free(vs);
1175 }
1176
1177 static void v9fs_stat(V9fsState *s, V9fsPDU *pdu)
1178 {
1179     int32_t fid;
1180     V9fsStatState *vs;
1181     ssize_t err = 0;
1182
1183     vs = qemu_malloc(sizeof(*vs));
1184     vs->pdu = pdu;
1185     vs->offset = 7;
1186
1187     memset(&vs->v9stat, 0, sizeof(vs->v9stat));
1188
1189     pdu_unmarshal(vs->pdu, vs->offset, "d", &fid);
1190
1191     vs->fidp = lookup_fid(s, fid);
1192     if (vs->fidp == NULL) {
1193         err = -ENOENT;
1194         goto out;
1195     }
1196
1197     err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1198     v9fs_stat_post_lstat(s, vs, err);
1199     return;
1200
1201 out:
1202     complete_pdu(s, vs->pdu, err);
1203     v9fs_stat_free(&vs->v9stat);
1204     qemu_free(vs);
1205 }
1206
1207 static void v9fs_getattr_post_lstat(V9fsState *s, V9fsStatStateDotl *vs,
1208                                                                 int err)
1209 {
1210     if (err == -1) {
1211         err = -errno;
1212         goto out;
1213     }
1214
1215     stat_to_v9stat_dotl(s, &vs->stbuf, &vs->v9stat_dotl);
1216     vs->offset += pdu_marshal(vs->pdu, vs->offset, "A", &vs->v9stat_dotl);
1217     err = vs->offset;
1218
1219 out:
1220     complete_pdu(s, vs->pdu, err);
1221     qemu_free(vs);
1222 }
1223
1224 static void v9fs_getattr(V9fsState *s, V9fsPDU *pdu)
1225 {
1226     int32_t fid;
1227     V9fsStatStateDotl *vs;
1228     ssize_t err = 0;
1229     V9fsFidState *fidp;
1230     uint64_t request_mask;
1231
1232     vs = qemu_malloc(sizeof(*vs));
1233     vs->pdu = pdu;
1234     vs->offset = 7;
1235
1236     memset(&vs->v9stat_dotl, 0, sizeof(vs->v9stat_dotl));
1237
1238     pdu_unmarshal(vs->pdu, vs->offset, "dq", &fid, &request_mask);
1239
1240     fidp = lookup_fid(s, fid);
1241     if (fidp == NULL) {
1242         err = -ENOENT;
1243         goto out;
1244     }
1245
1246     /* Currently we only support BASIC fields in stat, so there is no
1247      * need to look at request_mask.
1248      */
1249     err = v9fs_do_lstat(s, &fidp->path, &vs->stbuf);
1250     v9fs_getattr_post_lstat(s, vs, err);
1251     return;
1252
1253 out:
1254     complete_pdu(s, vs->pdu, err);
1255     qemu_free(vs);
1256 }
1257
1258 static void v9fs_walk_complete(V9fsState *s, V9fsWalkState *vs, int err)
1259 {
1260     complete_pdu(s, vs->pdu, err);
1261
1262     if (vs->nwnames) {
1263         for (vs->name_idx = 0; vs->name_idx < vs->nwnames; vs->name_idx++) {
1264             v9fs_string_free(&vs->wnames[vs->name_idx]);
1265         }
1266
1267         qemu_free(vs->wnames);
1268         qemu_free(vs->qids);
1269     }
1270 }
1271
1272 static void v9fs_walk_marshal(V9fsWalkState *vs)
1273 {
1274     int i;
1275     vs->offset = 7;
1276     vs->offset += pdu_marshal(vs->pdu, vs->offset, "w", vs->nwnames);
1277
1278     for (i = 0; i < vs->nwnames; i++) {
1279         vs->offset += pdu_marshal(vs->pdu, vs->offset, "Q", &vs->qids[i]);
1280     }
1281 }
1282
1283 static void v9fs_walk_post_newfid_lstat(V9fsState *s, V9fsWalkState *vs,
1284                                                                 int err)
1285 {
1286     if (err == -1) {
1287         free_fid(s, vs->newfidp->fid);
1288         v9fs_string_free(&vs->path);
1289         err = -ENOENT;
1290         goto out;
1291     }
1292
1293     stat_to_qid(&vs->stbuf, &vs->qids[vs->name_idx]);
1294
1295     vs->name_idx++;
1296     if (vs->name_idx < vs->nwnames) {
1297         v9fs_string_sprintf(&vs->path, "%s/%s", vs->newfidp->path.data,
1298                                             vs->wnames[vs->name_idx].data);
1299         v9fs_string_copy(&vs->newfidp->path, &vs->path);
1300
1301         err = v9fs_do_lstat(s, &vs->newfidp->path, &vs->stbuf);
1302         v9fs_walk_post_newfid_lstat(s, vs, err);
1303         return;
1304     }
1305
1306     v9fs_string_free(&vs->path);
1307     v9fs_walk_marshal(vs);
1308     err = vs->offset;
1309 out:
1310     v9fs_walk_complete(s, vs, err);
1311 }
1312
1313 static void v9fs_walk_post_oldfid_lstat(V9fsState *s, V9fsWalkState *vs,
1314         int err)
1315 {
1316     if (err == -1) {
1317         v9fs_string_free(&vs->path);
1318         err = -ENOENT;
1319         goto out;
1320     }
1321
1322     stat_to_qid(&vs->stbuf, &vs->qids[vs->name_idx]);
1323     vs->name_idx++;
1324     if (vs->name_idx < vs->nwnames) {
1325
1326         v9fs_string_sprintf(&vs->path, "%s/%s",
1327                 vs->fidp->path.data, vs->wnames[vs->name_idx].data);
1328         v9fs_string_copy(&vs->fidp->path, &vs->path);
1329
1330         err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1331         v9fs_walk_post_oldfid_lstat(s, vs, err);
1332         return;
1333     }
1334
1335     v9fs_string_free(&vs->path);
1336     v9fs_walk_marshal(vs);
1337     err = vs->offset;
1338 out:
1339     v9fs_walk_complete(s, vs, err);
1340 }
1341
1342 static void v9fs_walk(V9fsState *s, V9fsPDU *pdu)
1343 {
1344     int32_t fid, newfid;
1345     V9fsWalkState *vs;
1346     int err = 0;
1347     int i;
1348
1349     vs = qemu_malloc(sizeof(*vs));
1350     vs->pdu = pdu;
1351     vs->wnames = NULL;
1352     vs->qids = NULL;
1353     vs->offset = 7;
1354
1355     vs->offset += pdu_unmarshal(vs->pdu, vs->offset, "ddw", &fid,
1356                                             &newfid, &vs->nwnames);
1357
1358     if (vs->nwnames) {
1359         vs->wnames = qemu_mallocz(sizeof(vs->wnames[0]) * vs->nwnames);
1360
1361         vs->qids = qemu_mallocz(sizeof(vs->qids[0]) * vs->nwnames);
1362
1363         for (i = 0; i < vs->nwnames; i++) {
1364             vs->offset += pdu_unmarshal(vs->pdu, vs->offset, "s",
1365                                             &vs->wnames[i]);
1366         }
1367     }
1368
1369     vs->fidp = lookup_fid(s, fid);
1370     if (vs->fidp == NULL) {
1371         err = -ENOENT;
1372         goto out;
1373     }
1374
1375     /* FIXME: is this really valid? */
1376     if (fid == newfid) {
1377
1378         BUG_ON(vs->fidp->fd != -1);
1379         BUG_ON(vs->fidp->dir);
1380         v9fs_string_init(&vs->path);
1381         vs->name_idx = 0;
1382
1383         if (vs->name_idx < vs->nwnames) {
1384             v9fs_string_sprintf(&vs->path, "%s/%s",
1385                 vs->fidp->path.data, vs->wnames[vs->name_idx].data);
1386             v9fs_string_copy(&vs->fidp->path, &vs->path);
1387
1388             err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1389             v9fs_walk_post_oldfid_lstat(s, vs, err);
1390             return;
1391         }
1392     } else {
1393         vs->newfidp = alloc_fid(s, newfid);
1394         if (vs->newfidp == NULL) {
1395             err = -EINVAL;
1396             goto out;
1397         }
1398
1399         vs->newfidp->uid = vs->fidp->uid;
1400         v9fs_string_init(&vs->path);
1401         vs->name_idx = 0;
1402         v9fs_string_copy(&vs->newfidp->path, &vs->fidp->path);
1403
1404         if (vs->name_idx < vs->nwnames) {
1405             v9fs_string_sprintf(&vs->path, "%s/%s", vs->newfidp->path.data,
1406                                 vs->wnames[vs->name_idx].data);
1407             v9fs_string_copy(&vs->newfidp->path, &vs->path);
1408
1409             err = v9fs_do_lstat(s, &vs->newfidp->path, &vs->stbuf);
1410             v9fs_walk_post_newfid_lstat(s, vs, err);
1411             return;
1412         }
1413     }
1414
1415     v9fs_walk_marshal(vs);
1416     err = vs->offset;
1417 out:
1418     v9fs_walk_complete(s, vs, err);
1419 }
1420
1421 static int32_t get_iounit(V9fsState *s, V9fsString *name)
1422 {
1423     struct statfs stbuf;
1424     int32_t iounit = 0;
1425
1426     /*
1427      * iounit should be multiples of f_bsize (host filesystem block size
1428      * and as well as less than (client msize - P9_IOHDRSZ))
1429      */
1430     if (!v9fs_do_statfs(s, name, &stbuf)) {
1431         iounit = stbuf.f_bsize;
1432         iounit *= (s->msize - P9_IOHDRSZ)/stbuf.f_bsize;
1433     }
1434
1435     if (!iounit) {
1436         iounit = s->msize - P9_IOHDRSZ;
1437     }
1438     return iounit;
1439 }
1440
1441 static void v9fs_open_post_opendir(V9fsState *s, V9fsOpenState *vs, int err)
1442 {
1443     if (vs->fidp->dir == NULL) {
1444         err = -errno;
1445         goto out;
1446     }
1447
1448     vs->offset += pdu_marshal(vs->pdu, vs->offset, "Qd", &vs->qid, 0);
1449     err = vs->offset;
1450 out:
1451     complete_pdu(s, vs->pdu, err);
1452     qemu_free(vs);
1453
1454 }
1455
1456 static void v9fs_open_post_getiounit(V9fsState *s, V9fsOpenState *vs)
1457 {
1458     int err;
1459     vs->offset += pdu_marshal(vs->pdu, vs->offset, "Qd", &vs->qid, vs->iounit);
1460     err = vs->offset;
1461     complete_pdu(s, vs->pdu, err);
1462     qemu_free(vs);
1463 }
1464
1465 static void v9fs_open_post_open(V9fsState *s, V9fsOpenState *vs, int err)
1466 {
1467     if (vs->fidp->fd == -1) {
1468         err = -errno;
1469         goto out;
1470     }
1471
1472     vs->iounit = get_iounit(s, &vs->fidp->path);
1473     v9fs_open_post_getiounit(s, vs);
1474     return;
1475 out:
1476     complete_pdu(s, vs->pdu, err);
1477     qemu_free(vs);
1478 }
1479
1480 static void v9fs_open_post_lstat(V9fsState *s, V9fsOpenState *vs, int err)
1481 {
1482     if (err) {
1483         err = -errno;
1484         goto out;
1485     }
1486
1487     stat_to_qid(&vs->stbuf, &vs->qid);
1488
1489     if (S_ISDIR(vs->stbuf.st_mode)) {
1490         vs->fidp->dir = v9fs_do_opendir(s, &vs->fidp->path);
1491         v9fs_open_post_opendir(s, vs, err);
1492     } else {
1493         vs->fidp->fd = v9fs_do_open(s, &vs->fidp->path,
1494                                     omode_to_uflags(vs->mode));
1495         v9fs_open_post_open(s, vs, err);
1496     }
1497     return;
1498 out:
1499     complete_pdu(s, vs->pdu, err);
1500     qemu_free(vs);
1501 }
1502
1503 static void v9fs_open(V9fsState *s, V9fsPDU *pdu)
1504 {
1505     int32_t fid;
1506     V9fsOpenState *vs;
1507     ssize_t err = 0;
1508
1509
1510     vs = qemu_malloc(sizeof(*vs));
1511     vs->pdu = pdu;
1512     vs->offset = 7;
1513
1514     pdu_unmarshal(vs->pdu, vs->offset, "db", &fid, &vs->mode);
1515
1516     vs->fidp = lookup_fid(s, fid);
1517     if (vs->fidp == NULL) {
1518         err = -ENOENT;
1519         goto out;
1520     }
1521
1522     BUG_ON(vs->fidp->fd != -1);
1523     BUG_ON(vs->fidp->dir);
1524
1525     err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1526
1527     v9fs_open_post_lstat(s, vs, err);
1528     return;
1529 out:
1530     complete_pdu(s, pdu, err);
1531     qemu_free(vs);
1532 }
1533
1534 static void v9fs_clunk(V9fsState *s, V9fsPDU *pdu)
1535 {
1536     int32_t fid;
1537     size_t offset = 7;
1538     int err;
1539
1540     pdu_unmarshal(pdu, offset, "d", &fid);
1541
1542     err = free_fid(s, fid);
1543     if (err < 0) {
1544         goto out;
1545     }
1546
1547     offset = 7;
1548     err = offset;
1549 out:
1550     complete_pdu(s, pdu, err);
1551 }
1552
1553 static void v9fs_read_post_readdir(V9fsState *, V9fsReadState *, ssize_t);
1554
1555 static void v9fs_read_post_seekdir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1556 {
1557     if (err) {
1558         goto out;
1559     }
1560     v9fs_stat_free(&vs->v9stat);
1561     v9fs_string_free(&vs->name);
1562     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1563     vs->offset += vs->count;
1564     err = vs->offset;
1565 out:
1566     complete_pdu(s, vs->pdu, err);
1567     qemu_free(vs);
1568     return;
1569 }
1570
1571 static void v9fs_read_post_dir_lstat(V9fsState *s, V9fsReadState *vs,
1572                                     ssize_t err)
1573 {
1574     if (err) {
1575         err = -errno;
1576         goto out;
1577     }
1578     err = stat_to_v9stat(s, &vs->name, &vs->stbuf, &vs->v9stat);
1579     if (err) {
1580         goto out;
1581     }
1582
1583     vs->len = pdu_marshal(vs->pdu, vs->offset + 4 + vs->count, "S",
1584                             &vs->v9stat);
1585     if ((vs->len != (vs->v9stat.size + 2)) ||
1586             ((vs->count + vs->len) > vs->max_count)) {
1587         v9fs_do_seekdir(s, vs->fidp->dir, vs->dir_pos);
1588         v9fs_read_post_seekdir(s, vs, err);
1589         return;
1590     }
1591     vs->count += vs->len;
1592     v9fs_stat_free(&vs->v9stat);
1593     v9fs_string_free(&vs->name);
1594     vs->dir_pos = vs->dent->d_off;
1595     vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1596     v9fs_read_post_readdir(s, vs, err);
1597     return;
1598 out:
1599     v9fs_do_seekdir(s, vs->fidp->dir, vs->dir_pos);
1600     v9fs_read_post_seekdir(s, vs, err);
1601     return;
1602
1603 }
1604
1605 static void v9fs_read_post_readdir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1606 {
1607     if (vs->dent) {
1608         memset(&vs->v9stat, 0, sizeof(vs->v9stat));
1609         v9fs_string_init(&vs->name);
1610         v9fs_string_sprintf(&vs->name, "%s/%s", vs->fidp->path.data,
1611                             vs->dent->d_name);
1612         err = v9fs_do_lstat(s, &vs->name, &vs->stbuf);
1613         v9fs_read_post_dir_lstat(s, vs, err);
1614         return;
1615     }
1616
1617     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1618     vs->offset += vs->count;
1619     err = vs->offset;
1620     complete_pdu(s, vs->pdu, err);
1621     qemu_free(vs);
1622     return;
1623 }
1624
1625 static void v9fs_read_post_telldir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1626 {
1627     vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1628     v9fs_read_post_readdir(s, vs, err);
1629     return;
1630 }
1631
1632 static void v9fs_read_post_rewinddir(V9fsState *s, V9fsReadState *vs,
1633                                        ssize_t err)
1634 {
1635     vs->dir_pos = v9fs_do_telldir(s, vs->fidp->dir);
1636     v9fs_read_post_telldir(s, vs, err);
1637     return;
1638 }
1639
1640 static void v9fs_read_post_readv(V9fsState *s, V9fsReadState *vs, ssize_t err)
1641 {
1642     if (err  < 0) {
1643         /* IO error return the error */
1644         err = -errno;
1645         goto out;
1646     }
1647     vs->total += vs->len;
1648     vs->sg = adjust_sg(vs->sg, vs->len, &vs->cnt);
1649     if (vs->total < vs->count && vs->len > 0) {
1650         do {
1651             if (0) {
1652                 print_sg(vs->sg, vs->cnt);
1653             }
1654             vs->len = v9fs_do_readv(s, vs->fidp->fd, vs->sg, vs->cnt);
1655         } while (vs->len == -1 && errno == EINTR);
1656         if (vs->len == -1) {
1657             err  = -errno;
1658         }
1659         v9fs_read_post_readv(s, vs, err);
1660         return;
1661     }
1662     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->total);
1663     vs->offset += vs->count;
1664     err = vs->offset;
1665
1666 out:
1667     complete_pdu(s, vs->pdu, err);
1668     qemu_free(vs);
1669 }
1670
1671 static void v9fs_read_post_lseek(V9fsState *s, V9fsReadState *vs, ssize_t err)
1672 {
1673     if (err == -1) {
1674         err = -errno;
1675         goto out;
1676     }
1677     vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
1678
1679     if (vs->total < vs->count) {
1680         do {
1681             if (0) {
1682                 print_sg(vs->sg, vs->cnt);
1683             }
1684             vs->len = v9fs_do_readv(s, vs->fidp->fd, vs->sg, vs->cnt);
1685         } while (vs->len == -1 && errno == EINTR);
1686         if (vs->len == -1) {
1687             err  = -errno;
1688         }
1689         v9fs_read_post_readv(s, vs, err);
1690         return;
1691     }
1692 out:
1693     complete_pdu(s, vs->pdu, err);
1694     qemu_free(vs);
1695 }
1696
1697 static void v9fs_read(V9fsState *s, V9fsPDU *pdu)
1698 {
1699     int32_t fid;
1700     V9fsReadState *vs;
1701     ssize_t err = 0;
1702
1703     vs = qemu_malloc(sizeof(*vs));
1704     vs->pdu = pdu;
1705     vs->offset = 7;
1706     vs->total = 0;
1707     vs->len = 0;
1708     vs->count = 0;
1709
1710     pdu_unmarshal(vs->pdu, vs->offset, "dqd", &fid, &vs->off, &vs->count);
1711
1712     vs->fidp = lookup_fid(s, fid);
1713     if (vs->fidp == NULL) {
1714         err = -EINVAL;
1715         goto out;
1716     }
1717
1718     if (vs->fidp->dir) {
1719         vs->max_count = vs->count;
1720         vs->count = 0;
1721         if (vs->off == 0) {
1722             v9fs_do_rewinddir(s, vs->fidp->dir);
1723         }
1724         v9fs_read_post_rewinddir(s, vs, err);
1725         return;
1726     } else if (vs->fidp->fd != -1) {
1727         vs->sg = vs->iov;
1728         pdu_marshal(vs->pdu, vs->offset + 4, "v", vs->sg, &vs->cnt);
1729         err = v9fs_do_lseek(s, vs->fidp->fd, vs->off, SEEK_SET);
1730         v9fs_read_post_lseek(s, vs, err);
1731         return;
1732     } else {
1733         err = -EINVAL;
1734     }
1735 out:
1736     complete_pdu(s, pdu, err);
1737     qemu_free(vs);
1738 }
1739
1740 typedef struct V9fsReadDirState {
1741     V9fsPDU *pdu;
1742     V9fsFidState *fidp;
1743     V9fsQID qid;
1744     off_t saved_dir_pos;
1745     struct dirent *dent;
1746     int32_t count;
1747     int32_t max_count;
1748     size_t offset;
1749     int64_t initial_offset;
1750     V9fsString name;
1751 } V9fsReadDirState;
1752
1753 static void v9fs_readdir_post_seekdir(V9fsState *s, V9fsReadDirState *vs)
1754 {
1755     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1756     vs->offset += vs->count;
1757     complete_pdu(s, vs->pdu, vs->offset);
1758     qemu_free(vs);
1759     return;
1760 }
1761
1762 /* Size of each dirent on the wire: size of qid (13) + size of offset (8)
1763  * size of type (1) + size of name.size (2) + strlen(name.data)
1764  */
1765 #define V9_READDIR_DATA_SZ (24 + strlen(vs->name.data))
1766
1767 static void v9fs_readdir_post_readdir(V9fsState *s, V9fsReadDirState *vs)
1768 {
1769     int len;
1770     size_t size;
1771
1772     if (vs->dent) {
1773         v9fs_string_init(&vs->name);
1774         v9fs_string_sprintf(&vs->name, "%s", vs->dent->d_name);
1775
1776         if ((vs->count + V9_READDIR_DATA_SZ) > vs->max_count) {
1777             /* Ran out of buffer. Set dir back to old position and return */
1778             v9fs_do_seekdir(s, vs->fidp->dir, vs->saved_dir_pos);
1779             v9fs_readdir_post_seekdir(s, vs);
1780             return;
1781         }
1782
1783         /* Fill up just the path field of qid because the client uses
1784          * only that. To fill the entire qid structure we will have
1785          * to stat each dirent found, which is expensive
1786          */
1787         size = MIN(sizeof(vs->dent->d_ino), sizeof(vs->qid.path));
1788         memcpy(&vs->qid.path, &vs->dent->d_ino, size);
1789         /* Fill the other fields with dummy values */
1790         vs->qid.type = 0;
1791         vs->qid.version = 0;
1792
1793         len = pdu_marshal(vs->pdu, vs->offset+4+vs->count, "Qqbs",
1794                               &vs->qid, vs->dent->d_off,
1795                               vs->dent->d_type, &vs->name);
1796         vs->count += len;
1797         v9fs_string_free(&vs->name);
1798         vs->saved_dir_pos = vs->dent->d_off;
1799         vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1800         v9fs_readdir_post_readdir(s, vs);
1801         return;
1802     }
1803
1804     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1805     vs->offset += vs->count;
1806     complete_pdu(s, vs->pdu, vs->offset);
1807     qemu_free(vs);
1808     return;
1809 }
1810
1811 static void v9fs_readdir_post_telldir(V9fsState *s, V9fsReadDirState *vs)
1812 {
1813     vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1814     v9fs_readdir_post_readdir(s, vs);
1815     return;
1816 }
1817
1818 static void v9fs_readdir_post_setdir(V9fsState *s, V9fsReadDirState *vs)
1819 {
1820     vs->saved_dir_pos = v9fs_do_telldir(s, vs->fidp->dir);
1821     v9fs_readdir_post_telldir(s, vs);
1822     return;
1823 }
1824
1825 static void v9fs_readdir(V9fsState *s, V9fsPDU *pdu)
1826 {
1827     int32_t fid;
1828     V9fsReadDirState *vs;
1829     ssize_t err = 0;
1830     size_t offset = 7;
1831
1832     vs = qemu_malloc(sizeof(*vs));
1833     vs->pdu = pdu;
1834     vs->offset = 7;
1835     vs->count = 0;
1836
1837     pdu_unmarshal(vs->pdu, offset, "dqd", &fid, &vs->initial_offset,
1838                                                         &vs->max_count);
1839
1840     vs->fidp = lookup_fid(s, fid);
1841     if (vs->fidp == NULL || !(vs->fidp->dir)) {
1842         err = -EINVAL;
1843         goto out;
1844     }
1845
1846     if (vs->initial_offset == 0) {
1847         v9fs_do_rewinddir(s, vs->fidp->dir);
1848     } else {
1849         v9fs_do_seekdir(s, vs->fidp->dir, vs->initial_offset);
1850     }
1851
1852     v9fs_readdir_post_setdir(s, vs);
1853     return;
1854
1855 out:
1856     complete_pdu(s, pdu, err);
1857     qemu_free(vs);
1858     return;
1859 }
1860
1861 static void v9fs_write_post_writev(V9fsState *s, V9fsWriteState *vs,
1862                                    ssize_t err)
1863 {
1864     if (err  < 0) {
1865         /* IO error return the error */
1866         err = -errno;
1867         goto out;
1868     }
1869     vs->total += vs->len;
1870     vs->sg = adjust_sg(vs->sg, vs->len, &vs->cnt);
1871     if (vs->total < vs->count && vs->len > 0) {
1872         do {
1873             if (0) {
1874                 print_sg(vs->sg, vs->cnt);
1875             }
1876             vs->len =  v9fs_do_writev(s, vs->fidp->fd, vs->sg, vs->cnt);
1877         } while (vs->len == -1 && errno == EINTR);
1878         if (vs->len == -1) {
1879             err  = -errno;
1880         }
1881         v9fs_write_post_writev(s, vs, err);
1882         return;
1883     }
1884     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->total);
1885
1886     err = vs->offset;
1887 out:
1888     complete_pdu(s, vs->pdu, err);
1889     qemu_free(vs);
1890 }
1891
1892 static void v9fs_write_post_lseek(V9fsState *s, V9fsWriteState *vs, ssize_t err)
1893 {
1894     if (err == -1) {
1895         err = -errno;
1896         goto out;
1897     }
1898     vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
1899
1900     if (vs->total < vs->count) {
1901         do {
1902             if (0) {
1903                 print_sg(vs->sg, vs->cnt);
1904             }
1905             vs->len = v9fs_do_writev(s, vs->fidp->fd, vs->sg, vs->cnt);
1906         } while (vs->len == -1 && errno == EINTR);
1907         if (vs->len == -1) {
1908             err  = -errno;
1909         }
1910         v9fs_write_post_writev(s, vs, err);
1911         return;
1912     }
1913
1914 out:
1915     complete_pdu(s, vs->pdu, err);
1916     qemu_free(vs);
1917 }
1918
1919 static void v9fs_write(V9fsState *s, V9fsPDU *pdu)
1920 {
1921     int32_t fid;
1922     V9fsWriteState *vs;
1923     ssize_t err;
1924
1925     vs = qemu_malloc(sizeof(*vs));
1926
1927     vs->pdu = pdu;
1928     vs->offset = 7;
1929     vs->sg = vs->iov;
1930     vs->total = 0;
1931     vs->len = 0;
1932
1933     pdu_unmarshal(vs->pdu, vs->offset, "dqdv", &fid, &vs->off, &vs->count,
1934                     vs->sg, &vs->cnt);
1935
1936     vs->fidp = lookup_fid(s, fid);
1937     if (vs->fidp == NULL) {
1938         err = -EINVAL;
1939         goto out;
1940     }
1941
1942     if (vs->fidp->fd == -1) {
1943         err = -EINVAL;
1944         goto out;
1945     }
1946
1947     err = v9fs_do_lseek(s, vs->fidp->fd, vs->off, SEEK_SET);
1948
1949     v9fs_write_post_lseek(s, vs, err);
1950     return;
1951
1952 out:
1953     complete_pdu(s, vs->pdu, err);
1954     qemu_free(vs);
1955 }
1956
1957 static void v9fs_create_post_getiounit(V9fsState *s, V9fsCreateState *vs)
1958 {
1959     int err;
1960     v9fs_string_copy(&vs->fidp->path, &vs->fullname);
1961     stat_to_qid(&vs->stbuf, &vs->qid);
1962
1963     vs->offset += pdu_marshal(vs->pdu, vs->offset, "Qd", &vs->qid, vs->iounit);
1964     err = vs->offset;
1965
1966     complete_pdu(s, vs->pdu, err);
1967     v9fs_string_free(&vs->name);
1968     v9fs_string_free(&vs->extension);
1969     v9fs_string_free(&vs->fullname);
1970     qemu_free(vs);
1971 }
1972
1973 static void v9fs_post_create(V9fsState *s, V9fsCreateState *vs, int err)
1974 {
1975     if (err == 0) {
1976         vs->iounit = get_iounit(s, &vs->fidp->path);
1977         v9fs_create_post_getiounit(s, vs);
1978         return;
1979     }
1980
1981     complete_pdu(s, vs->pdu, err);
1982     v9fs_string_free(&vs->name);
1983     v9fs_string_free(&vs->extension);
1984     v9fs_string_free(&vs->fullname);
1985     qemu_free(vs);
1986 }
1987
1988 static void v9fs_create_post_perms(V9fsState *s, V9fsCreateState *vs, int err)
1989 {
1990     if (err) {
1991         err = -errno;
1992     }
1993     v9fs_post_create(s, vs, err);
1994 }
1995
1996 static void v9fs_create_post_opendir(V9fsState *s, V9fsCreateState *vs,
1997                                                                     int err)
1998 {
1999     if (!vs->fidp->dir) {
2000         err = -errno;
2001     }
2002     v9fs_post_create(s, vs, err);
2003 }
2004
2005 static void v9fs_create_post_dir_lstat(V9fsState *s, V9fsCreateState *vs,
2006                                                                     int err)
2007 {
2008     if (err) {
2009         err = -errno;
2010         goto out;
2011     }
2012
2013     vs->fidp->dir = v9fs_do_opendir(s, &vs->fullname);
2014     v9fs_create_post_opendir(s, vs, err);
2015     return;
2016
2017 out:
2018     v9fs_post_create(s, vs, err);
2019 }
2020
2021 static void v9fs_create_post_mkdir(V9fsState *s, V9fsCreateState *vs, int err)
2022 {
2023     if (err) {
2024         err = -errno;
2025         goto out;
2026     }
2027
2028     err = v9fs_do_lstat(s, &vs->fullname, &vs->stbuf);
2029     v9fs_create_post_dir_lstat(s, vs, err);
2030     return;
2031
2032 out:
2033     v9fs_post_create(s, vs, err);
2034 }
2035
2036 static void v9fs_create_post_fstat(V9fsState *s, V9fsCreateState *vs, int err)
2037 {
2038     if (err) {
2039         vs->fidp->fd = -1;
2040         err = -errno;
2041     }
2042
2043     v9fs_post_create(s, vs, err);
2044     return;
2045 }
2046
2047 static void v9fs_create_post_open2(V9fsState *s, V9fsCreateState *vs, int err)
2048 {
2049     if (vs->fidp->fd == -1) {
2050         err = -errno;
2051         goto out;
2052     }
2053
2054     err = v9fs_do_fstat(s, vs->fidp->fd, &vs->stbuf);
2055     v9fs_create_post_fstat(s, vs, err);
2056
2057     return;
2058
2059 out:
2060     v9fs_post_create(s, vs, err);
2061
2062 }
2063
2064 static void v9fs_create_post_lstat(V9fsState *s, V9fsCreateState *vs, int err)
2065 {
2066
2067     if (err == 0 || errno != ENOENT) {
2068         err = -errno;
2069         goto out;
2070     }
2071
2072     if (vs->perm & P9_STAT_MODE_DIR) {
2073         err = v9fs_do_mkdir(s, vs);
2074         v9fs_create_post_mkdir(s, vs, err);
2075     } else if (vs->perm & P9_STAT_MODE_SYMLINK) {
2076         err = v9fs_do_symlink(s, vs);
2077         v9fs_create_post_perms(s, vs, err);
2078     } else if (vs->perm & P9_STAT_MODE_LINK) {
2079         int32_t nfid = atoi(vs->extension.data);
2080         V9fsFidState *nfidp = lookup_fid(s, nfid);
2081         if (nfidp == NULL) {
2082             err = -errno;
2083             v9fs_post_create(s, vs, err);
2084         }
2085         err = v9fs_do_link(s, &nfidp->path, &vs->fullname);
2086         v9fs_create_post_perms(s, vs, err);
2087     } else if (vs->perm & P9_STAT_MODE_DEVICE) {
2088         char ctype;
2089         uint32_t major, minor;
2090         mode_t nmode = 0;
2091
2092         if (sscanf(vs->extension.data, "%c %u %u", &ctype, &major,
2093                                         &minor) != 3) {
2094             err = -errno;
2095             v9fs_post_create(s, vs, err);
2096         }
2097
2098         switch (ctype) {
2099         case 'c':
2100             nmode = S_IFCHR;
2101             break;
2102         case 'b':
2103             nmode = S_IFBLK;
2104             break;
2105         default:
2106             err = -EIO;
2107             v9fs_post_create(s, vs, err);
2108         }
2109
2110         nmode |= vs->perm & 0777;
2111         err = v9fs_do_mknod(s, vs, nmode, makedev(major, minor));
2112         v9fs_create_post_perms(s, vs, err);
2113     } else if (vs->perm & P9_STAT_MODE_NAMED_PIPE) {
2114         err = v9fs_do_mknod(s, vs, S_IFIFO | (vs->perm & 0777), 0);
2115         v9fs_post_create(s, vs, err);
2116     } else if (vs->perm & P9_STAT_MODE_SOCKET) {
2117         err = v9fs_do_mknod(s, vs, S_IFSOCK | (vs->perm & 0777), 0);
2118         v9fs_post_create(s, vs, err);
2119     } else {
2120         vs->fidp->fd = v9fs_do_open2(s, vs);
2121         v9fs_create_post_open2(s, vs, err);
2122     }
2123
2124     return;
2125
2126 out:
2127     v9fs_post_create(s, vs, err);
2128 }
2129
2130 static void v9fs_create(V9fsState *s, V9fsPDU *pdu)
2131 {
2132     int32_t fid;
2133     V9fsCreateState *vs;
2134     int err = 0;
2135
2136     vs = qemu_malloc(sizeof(*vs));
2137     vs->pdu = pdu;
2138     vs->offset = 7;
2139
2140     v9fs_string_init(&vs->fullname);
2141
2142     pdu_unmarshal(vs->pdu, vs->offset, "dsdbs", &fid, &vs->name,
2143                                 &vs->perm, &vs->mode, &vs->extension);
2144
2145     vs->fidp = lookup_fid(s, fid);
2146     if (vs->fidp == NULL) {
2147         err = -EINVAL;
2148         goto out;
2149     }
2150
2151     v9fs_string_sprintf(&vs->fullname, "%s/%s", vs->fidp->path.data,
2152                                                         vs->name.data);
2153
2154     err = v9fs_do_lstat(s, &vs->fullname, &vs->stbuf);
2155     v9fs_create_post_lstat(s, vs, err);
2156     return;
2157
2158 out:
2159     complete_pdu(s, vs->pdu, err);
2160     v9fs_string_free(&vs->name);
2161     v9fs_string_free(&vs->extension);
2162     qemu_free(vs);
2163 }
2164
2165 static void v9fs_flush(V9fsState *s, V9fsPDU *pdu)
2166 {
2167     /* A nop call with no return */
2168     complete_pdu(s, pdu, 7);
2169 }
2170
2171 static void v9fs_remove_post_remove(V9fsState *s, V9fsRemoveState *vs,
2172                                                                 int err)
2173 {
2174     if (err < 0) {
2175         err = -errno;
2176     } else {
2177         err = vs->offset;
2178     }
2179
2180     /* For TREMOVE we need to clunk the fid even on failed remove */
2181     free_fid(s, vs->fidp->fid);
2182
2183     complete_pdu(s, vs->pdu, err);
2184     qemu_free(vs);
2185 }
2186
2187 static void v9fs_remove(V9fsState *s, V9fsPDU *pdu)
2188 {
2189     int32_t fid;
2190     V9fsRemoveState *vs;
2191     int err = 0;
2192
2193     vs = qemu_malloc(sizeof(*vs));
2194     vs->pdu = pdu;
2195     vs->offset = 7;
2196
2197     pdu_unmarshal(vs->pdu, vs->offset, "d", &fid);
2198
2199     vs->fidp = lookup_fid(s, fid);
2200     if (vs->fidp == NULL) {
2201         err = -EINVAL;
2202         goto out;
2203     }
2204
2205     err = v9fs_do_remove(s, &vs->fidp->path);
2206     v9fs_remove_post_remove(s, vs, err);
2207     return;
2208
2209 out:
2210     complete_pdu(s, pdu, err);
2211     qemu_free(vs);
2212 }
2213
2214 static void v9fs_wstat_post_truncate(V9fsState *s, V9fsWstatState *vs, int err)
2215 {
2216     if (err < 0) {
2217         goto out;
2218     }
2219
2220     err = vs->offset;
2221
2222 out:
2223     v9fs_stat_free(&vs->v9stat);
2224     complete_pdu(s, vs->pdu, err);
2225     qemu_free(vs);
2226 }
2227
2228 static void v9fs_wstat_post_rename(V9fsState *s, V9fsWstatState *vs, int err)
2229 {
2230     if (err < 0) {
2231         goto out;
2232     }
2233
2234     if (vs->v9stat.name.size != 0) {
2235         v9fs_string_free(&vs->nname);
2236     }
2237
2238     if (vs->v9stat.length != -1) {
2239         if (v9fs_do_truncate(s, &vs->fidp->path, vs->v9stat.length) < 0) {
2240             err = -errno;
2241         }
2242     }
2243     v9fs_wstat_post_truncate(s, vs, err);
2244     return;
2245
2246 out:
2247     v9fs_stat_free(&vs->v9stat);
2248     complete_pdu(s, vs->pdu, err);
2249     qemu_free(vs);
2250 }
2251
2252 static void v9fs_wstat_post_chown(V9fsState *s, V9fsWstatState *vs, int err)
2253 {
2254     V9fsFidState *fidp;
2255     if (err < 0) {
2256         goto out;
2257     }
2258
2259     if (vs->v9stat.name.size != 0) {
2260         char *old_name, *new_name;
2261         char *end;
2262
2263         old_name = vs->fidp->path.data;
2264         end = strrchr(old_name, '/');
2265         if (end) {
2266             end++;
2267         } else {
2268             end = old_name;
2269         }
2270
2271         new_name = qemu_mallocz(end - old_name + vs->v9stat.name.size + 1);
2272
2273         memcpy(new_name, old_name, end - old_name);
2274         memcpy(new_name + (end - old_name), vs->v9stat.name.data,
2275                 vs->v9stat.name.size);
2276         vs->nname.data = new_name;
2277         vs->nname.size = strlen(new_name);
2278
2279         if (strcmp(new_name, vs->fidp->path.data) != 0) {
2280             if (v9fs_do_rename(s, &vs->fidp->path, &vs->nname)) {
2281                 err = -errno;
2282             } else {
2283                 /*
2284                  * Fixup fid's pointing to the old name to
2285                  * start pointing to the new name
2286                  */
2287                 for (fidp = s->fid_list; fidp; fidp = fidp->next) {
2288
2289                     if (vs->fidp == fidp) {
2290                         /*
2291                          * we replace name of this fid towards the end
2292                          * so that our below strcmp will work
2293                          */
2294                         continue;
2295                     }
2296                     if (!strncmp(vs->fidp->path.data, fidp->path.data,
2297                                  strlen(vs->fidp->path.data))) {
2298                         /* replace the name */
2299                         v9fs_fix_path(&fidp->path, &vs->nname,
2300                                       strlen(vs->fidp->path.data));
2301                     }
2302                 }
2303                 v9fs_string_copy(&vs->fidp->path, &vs->nname);
2304             }
2305         }
2306     }
2307     v9fs_wstat_post_rename(s, vs, err);
2308     return;
2309
2310 out:
2311     v9fs_stat_free(&vs->v9stat);
2312     complete_pdu(s, vs->pdu, err);
2313     qemu_free(vs);
2314 }
2315
2316 static void v9fs_wstat_post_utime(V9fsState *s, V9fsWstatState *vs, int err)
2317 {
2318     if (err < 0) {
2319         goto out;
2320     }
2321
2322     if (vs->v9stat.n_gid != -1 || vs->v9stat.n_uid != -1) {
2323         if (v9fs_do_chown(s, &vs->fidp->path, vs->v9stat.n_uid,
2324                     vs->v9stat.n_gid)) {
2325             err = -errno;
2326         }
2327     }
2328     v9fs_wstat_post_chown(s, vs, err);
2329     return;
2330
2331 out:
2332     v9fs_stat_free(&vs->v9stat);
2333     complete_pdu(s, vs->pdu, err);
2334     qemu_free(vs);
2335 }
2336
2337 static void v9fs_wstat_post_chmod(V9fsState *s, V9fsWstatState *vs, int err)
2338 {
2339     if (err < 0) {
2340         goto out;
2341     }
2342
2343     if (vs->v9stat.mtime != -1 || vs->v9stat.atime != -1) {
2344         if (v9fs_do_utimensat(s, &vs->fidp->path, vs->v9stat)) {
2345             err = -errno;
2346         }
2347     }
2348
2349     v9fs_wstat_post_utime(s, vs, err);
2350     return;
2351
2352 out:
2353     v9fs_stat_free(&vs->v9stat);
2354     complete_pdu(s, vs->pdu, err);
2355     qemu_free(vs);
2356 }
2357
2358 static void v9fs_wstat_post_fsync(V9fsState *s, V9fsWstatState *vs, int err)
2359 {
2360     if (err == -1) {
2361         err = -errno;
2362     }
2363     v9fs_stat_free(&vs->v9stat);
2364     complete_pdu(s, vs->pdu, err);
2365     qemu_free(vs);
2366 }
2367
2368 static void v9fs_wstat_post_lstat(V9fsState *s, V9fsWstatState *vs, int err)
2369 {
2370     uint32_t v9_mode;
2371
2372     if (err == -1) {
2373         err = -errno;
2374         goto out;
2375     }
2376
2377     v9_mode = stat_to_v9mode(&vs->stbuf);
2378
2379     if ((vs->v9stat.mode & P9_STAT_MODE_TYPE_BITS) !=
2380         (v9_mode & P9_STAT_MODE_TYPE_BITS)) {
2381             /* Attempting to change the type */
2382             err = -EIO;
2383             goto out;
2384     }
2385
2386     if (v9fs_do_chmod(s, &vs->fidp->path, v9mode_to_mode(vs->v9stat.mode,
2387                     &vs->v9stat.extension))) {
2388             err = -errno;
2389      }
2390     v9fs_wstat_post_chmod(s, vs, err);
2391     return;
2392
2393 out:
2394     v9fs_stat_free(&vs->v9stat);
2395     complete_pdu(s, vs->pdu, err);
2396     qemu_free(vs);
2397 }
2398
2399 static void v9fs_wstat(V9fsState *s, V9fsPDU *pdu)
2400 {
2401     int32_t fid;
2402     V9fsWstatState *vs;
2403     int err = 0;
2404
2405     vs = qemu_malloc(sizeof(*vs));
2406     vs->pdu = pdu;
2407     vs->offset = 7;
2408
2409     pdu_unmarshal(pdu, vs->offset, "dwS", &fid, &vs->unused, &vs->v9stat);
2410
2411     vs->fidp = lookup_fid(s, fid);
2412     if (vs->fidp == NULL) {
2413         err = -EINVAL;
2414         goto out;
2415     }
2416
2417     /* do we need to sync the file? */
2418     if (donttouch_stat(&vs->v9stat)) {
2419         err = v9fs_do_fsync(s, vs->fidp->fd);
2420         v9fs_wstat_post_fsync(s, vs, err);
2421         return;
2422     }
2423
2424     if (vs->v9stat.mode != -1) {
2425         err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
2426         v9fs_wstat_post_lstat(s, vs, err);
2427         return;
2428     }
2429
2430     v9fs_wstat_post_chmod(s, vs, err);
2431     return;
2432
2433 out:
2434     v9fs_stat_free(&vs->v9stat);
2435     complete_pdu(s, vs->pdu, err);
2436     qemu_free(vs);
2437 }
2438
2439 static void v9fs_statfs_post_statfs(V9fsState *s, V9fsStatfsState *vs, int err)
2440 {
2441     int32_t bsize_factor;
2442
2443     if (err) {
2444         err = -errno;
2445         goto out;
2446     }
2447
2448     /*
2449      * compute bsize factor based on host file system block size
2450      * and client msize
2451      */
2452     bsize_factor = (s->msize - P9_IOHDRSZ)/vs->stbuf.f_bsize;
2453     if (!bsize_factor) {
2454         bsize_factor = 1;
2455     }
2456     vs->v9statfs.f_type = vs->stbuf.f_type;
2457     vs->v9statfs.f_bsize = vs->stbuf.f_bsize;
2458     vs->v9statfs.f_bsize *= bsize_factor;
2459     /*
2460      * f_bsize is adjusted(multiplied) by bsize factor, so we need to
2461      * adjust(divide) the number of blocks, free blocks and available
2462      * blocks by bsize factor
2463      */
2464     vs->v9statfs.f_blocks = vs->stbuf.f_blocks/bsize_factor;
2465     vs->v9statfs.f_bfree = vs->stbuf.f_bfree/bsize_factor;
2466     vs->v9statfs.f_bavail = vs->stbuf.f_bavail/bsize_factor;
2467     vs->v9statfs.f_files = vs->stbuf.f_files;
2468     vs->v9statfs.f_ffree = vs->stbuf.f_ffree;
2469     vs->v9statfs.fsid_val = (unsigned int) vs->stbuf.f_fsid.__val[0] |
2470                         (unsigned long long)vs->stbuf.f_fsid.__val[1] << 32;
2471     vs->v9statfs.f_namelen = vs->stbuf.f_namelen;
2472
2473     vs->offset += pdu_marshal(vs->pdu, vs->offset, "ddqqqqqqd",
2474          vs->v9statfs.f_type, vs->v9statfs.f_bsize, vs->v9statfs.f_blocks,
2475          vs->v9statfs.f_bfree, vs->v9statfs.f_bavail, vs->v9statfs.f_files,
2476          vs->v9statfs.f_ffree, vs->v9statfs.fsid_val,
2477          vs->v9statfs.f_namelen);
2478
2479 out:
2480     complete_pdu(s, vs->pdu, vs->offset);
2481     qemu_free(vs);
2482 }
2483
2484 static void v9fs_statfs(V9fsState *s, V9fsPDU *pdu)
2485 {
2486     V9fsStatfsState *vs;
2487     ssize_t err = 0;
2488
2489     vs = qemu_malloc(sizeof(*vs));
2490     vs->pdu = pdu;
2491     vs->offset = 7;
2492
2493     memset(&vs->v9statfs, 0, sizeof(vs->v9statfs));
2494
2495     pdu_unmarshal(vs->pdu, vs->offset, "d", &vs->fid);
2496
2497     vs->fidp = lookup_fid(s, vs->fid);
2498     if (vs->fidp == NULL) {
2499         err = -ENOENT;
2500         goto out;
2501     }
2502
2503     err = v9fs_do_statfs(s, &vs->fidp->path, &vs->stbuf);
2504     v9fs_statfs_post_statfs(s, vs, err);
2505     return;
2506
2507 out:
2508     complete_pdu(s, vs->pdu, err);
2509     qemu_free(vs);
2510 }
2511
2512 typedef void (pdu_handler_t)(V9fsState *s, V9fsPDU *pdu);
2513
2514 static pdu_handler_t *pdu_handlers[] = {
2515     [P9_TREADDIR] = v9fs_readdir,
2516     [P9_TSTATFS] = v9fs_statfs,
2517     [P9_TGETATTR] = v9fs_getattr,
2518     [P9_TVERSION] = v9fs_version,
2519     [P9_TATTACH] = v9fs_attach,
2520     [P9_TSTAT] = v9fs_stat,
2521     [P9_TWALK] = v9fs_walk,
2522     [P9_TCLUNK] = v9fs_clunk,
2523     [P9_TOPEN] = v9fs_open,
2524     [P9_TREAD] = v9fs_read,
2525 #if 0
2526     [P9_TAUTH] = v9fs_auth,
2527 #endif
2528     [P9_TFLUSH] = v9fs_flush,
2529     [P9_TCREATE] = v9fs_create,
2530     [P9_TWRITE] = v9fs_write,
2531     [P9_TWSTAT] = v9fs_wstat,
2532     [P9_TREMOVE] = v9fs_remove,
2533 };
2534
2535 static void submit_pdu(V9fsState *s, V9fsPDU *pdu)
2536 {
2537     pdu_handler_t *handler;
2538
2539     if (debug_9p_pdu) {
2540         pprint_pdu(pdu);
2541     }
2542
2543     BUG_ON(pdu->id >= ARRAY_SIZE(pdu_handlers));
2544
2545     handler = pdu_handlers[pdu->id];
2546     BUG_ON(handler == NULL);
2547
2548     handler(s, pdu);
2549 }
2550
2551 static void handle_9p_output(VirtIODevice *vdev, VirtQueue *vq)
2552 {
2553     V9fsState *s = (V9fsState *)vdev;
2554     V9fsPDU *pdu;
2555     ssize_t len;
2556
2557     while ((pdu = alloc_pdu(s)) &&
2558             (len = virtqueue_pop(vq, &pdu->elem)) != 0) {
2559         uint8_t *ptr;
2560
2561         BUG_ON(pdu->elem.out_num == 0 || pdu->elem.in_num == 0);
2562         BUG_ON(pdu->elem.out_sg[0].iov_len < 7);
2563
2564         ptr = pdu->elem.out_sg[0].iov_base;
2565
2566         memcpy(&pdu->size, ptr, 4);
2567         pdu->id = ptr[4];
2568         memcpy(&pdu->tag, ptr + 5, 2);
2569
2570         submit_pdu(s, pdu);
2571     }
2572
2573     free_pdu(s, pdu);
2574 }
2575
2576 static uint32_t virtio_9p_get_features(VirtIODevice *vdev, uint32_t features)
2577 {
2578     features |= 1 << VIRTIO_9P_MOUNT_TAG;
2579     return features;
2580 }
2581
2582 static V9fsState *to_virtio_9p(VirtIODevice *vdev)
2583 {
2584     return (V9fsState *)vdev;
2585 }
2586
2587 static void virtio_9p_get_config(VirtIODevice *vdev, uint8_t *config)
2588 {
2589     struct virtio_9p_config *cfg;
2590     V9fsState *s = to_virtio_9p(vdev);
2591
2592     cfg = qemu_mallocz(sizeof(struct virtio_9p_config) +
2593                         s->tag_len);
2594     stw_raw(&cfg->tag_len, s->tag_len);
2595     memcpy(cfg->tag, s->tag, s->tag_len);
2596     memcpy(config, cfg, s->config_size);
2597     qemu_free(cfg);
2598 }
2599
2600 VirtIODevice *virtio_9p_init(DeviceState *dev, V9fsConf *conf)
2601  {
2602     V9fsState *s;
2603     int i, len;
2604     struct stat stat;
2605     FsTypeEntry *fse;
2606
2607
2608     s = (V9fsState *)virtio_common_init("virtio-9p",
2609                                     VIRTIO_ID_9P,
2610                                     sizeof(struct virtio_9p_config)+
2611                                     MAX_TAG_LEN,
2612                                     sizeof(V9fsState));
2613
2614     /* initialize pdu allocator */
2615     QLIST_INIT(&s->free_list);
2616     for (i = 0; i < (MAX_REQ - 1); i++) {
2617         QLIST_INSERT_HEAD(&s->free_list, &s->pdus[i], next);
2618     }
2619
2620     s->vq = virtio_add_queue(&s->vdev, MAX_REQ, handle_9p_output);
2621
2622     fse = get_fsdev_fsentry(conf->fsdev_id);
2623
2624     if (!fse) {
2625         /* We don't have a fsdev identified by fsdev_id */
2626         fprintf(stderr, "Virtio-9p device couldn't find fsdev "
2627                     "with the id %s\n", conf->fsdev_id);
2628         exit(1);
2629     }
2630
2631     if (!fse->path || !conf->tag) {
2632         /* we haven't specified a mount_tag or the path */
2633         fprintf(stderr, "fsdev with id %s needs path "
2634                 "and Virtio-9p device needs mount_tag arguments\n",
2635                 conf->fsdev_id);
2636         exit(1);
2637     }
2638
2639     if (!strcmp(fse->security_model, "passthrough")) {
2640         /* Files on the Fileserver set to client user credentials */
2641         s->ctx.fs_sm = SM_PASSTHROUGH;
2642     } else if (!strcmp(fse->security_model, "mapped")) {
2643         /* Files on the fileserver are set to QEMU credentials.
2644          * Client user credentials are saved in extended attributes.
2645          */
2646         s->ctx.fs_sm = SM_MAPPED;
2647     } else {
2648         /* user haven't specified a correct security option */
2649         fprintf(stderr, "one of the following must be specified as the"
2650                 "security option:\n\t security_model=passthrough \n\t "
2651                 "security_model=mapped\n");
2652         return NULL;
2653     }
2654
2655     if (lstat(fse->path, &stat)) {
2656         fprintf(stderr, "share path %s does not exist\n", fse->path);
2657         exit(1);
2658     } else if (!S_ISDIR(stat.st_mode)) {
2659         fprintf(stderr, "share path %s is not a directory \n", fse->path);
2660         exit(1);
2661     }
2662
2663     s->ctx.fs_root = qemu_strdup(fse->path);
2664     len = strlen(conf->tag);
2665     if (len > MAX_TAG_LEN) {
2666         len = MAX_TAG_LEN;
2667     }
2668     /* s->tag is non-NULL terminated string */
2669     s->tag = qemu_malloc(len);
2670     memcpy(s->tag, conf->tag, len);
2671     s->tag_len = len;
2672     s->ctx.uid = -1;
2673
2674     s->ops = fse->ops;
2675     s->vdev.get_features = virtio_9p_get_features;
2676     s->config_size = sizeof(struct virtio_9p_config) +
2677                         s->tag_len;
2678     s->vdev.get_config = virtio_9p_get_config;
2679
2680     return &s->vdev;
2681 }