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_utime(V9fsState *s, V9fsString *path,
 241                             const struct utimbuf *buf)
 242 {
 243     return s->ops->utime(&s->ctx, path->data, buf);
 244 }
 245
 246 static int v9fs_do_remove(V9fsState *s, V9fsString *path)
 247 {
 248     return s->ops->remove(&s->ctx, path->data);
 249 }
 250
 251 static int v9fs_do_fsync(V9fsState *s, int fd)
 252 {
 253     return s->ops->fsync(&s->ctx, fd);
 254 }
 255
 256 static void v9fs_string_init(V9fsString *str)
 257 {
 258     str->data = NULL;
 259     str->size = 0;
 260 }
 261
 262 static void v9fs_string_free(V9fsString *str)
 263 {
 264     qemu_free(str->data);
 265     str->data = NULL;
 266     str->size = 0;
 267 }
 268
 269 static void v9fs_string_null(V9fsString *str)
 270 {
 271     v9fs_string_free(str);
 272 }
 273
 274 static int number_to_string(void *arg, char type)
 275 {
 276     unsigned int ret = 0;
 277
 278     switch (type) {
 279     case 'u': {
 280         unsigned int num = *(unsigned int *)arg;
 281
 282         do {
 283             ret++;
 284             num = num/10;
 285         } while (num);
 286         break;
 287     }
 288     default:
 289         printf("Number_to_string: Unknown number format\n");
 290         return -1;
 291     }
 292
 293     return ret;
 294 }
 295
 296 static int v9fs_string_alloc_printf(char **strp, const char *fmt, va_list ap)
 297 {
 298     va_list ap2;
 299     char *iter = (char *)fmt;
 300     int len = 0;
 301     int nr_args = 0;
 302     char *arg_char_ptr;
 303     unsigned int arg_uint;
 304
 305     /* Find the number of %'s that denotes an argument */
 306     for (iter = strstr(iter, "%"); iter; iter = strstr(iter, "%")) {
 307         nr_args++;
 308         iter++;
 309     }
 310
 311     len = strlen(fmt) - 2*nr_args;
 312
 313     if (!nr_args) {
 314         goto alloc_print;
 315     }
 316
 317     va_copy(ap2, ap);
 318
 319     iter = (char *)fmt;
 320
 321     /* Now parse the format string */
 322     for (iter = strstr(iter, "%"); iter; iter = strstr(iter, "%")) {
 323         iter++;
 324         switch (*iter) {
 325         case 'u':
 326             arg_uint = va_arg(ap2, unsigned int);
 327             len += number_to_string((void *)&arg_uint, 'u');
 328             break;
 329         case 's':
 330             arg_char_ptr = va_arg(ap2, char *);
 331             len += strlen(arg_char_ptr);
 332             break;
 333         case 'c':
 334             len += 1;
 335             break;
 336         default:
 337             fprintf(stderr,
 338                     "v9fs_string_alloc_printf:Incorrect format %c", *iter);
 339             return -1;
 340         }
 341         iter++;
 342     }
 343
 344 alloc_print:
 345     *strp = qemu_malloc((len + 1) * sizeof(**strp));
 346
 347     return vsprintf(*strp, fmt, ap);
 348 }
 349
 350 static void v9fs_string_sprintf(V9fsString *str, const char *fmt, ...)
 351 {
 352     va_list ap;
 353     int err;
 354
 355     v9fs_string_free(str);
 356
 357     va_start(ap, fmt);
 358     err = v9fs_string_alloc_printf(&str->data, fmt, ap);
 359     BUG_ON(err == -1);
 360     va_end(ap);
 361
 362     str->size = err;
 363 }
 364
 365 static void v9fs_string_copy(V9fsString *lhs, V9fsString *rhs)
 366 {
 367     v9fs_string_free(lhs);
 368     v9fs_string_sprintf(lhs, "%s", rhs->data);
 369 }
 370
 371 static size_t v9fs_string_size(V9fsString *str)
 372 {
 373     return str->size;
 374 }
 375
 376 static V9fsFidState *lookup_fid(V9fsState *s, int32_t fid)
 377 {
 378     V9fsFidState *f;
 379
 380     for (f = s->fid_list; f; f = f->next) {
 381         if (f->fid == fid) {
 382             return f;
 383         }
 384     }
 385
 386     return NULL;
 387 }
 388
 389 static V9fsFidState *alloc_fid(V9fsState *s, int32_t fid)
 390 {
 391     V9fsFidState *f;
 392
 393     f = lookup_fid(s, fid);
 394     if (f) {
 395         return NULL;
 396     }
 397
 398     f = qemu_mallocz(sizeof(V9fsFidState));
 399
 400     f->fid = fid;
 401     f->fd = -1;
 402     f->dir = NULL;
 403
 404     f->next = s->fid_list;
 405     s->fid_list = f;
 406
 407     return f;
 408 }
 409
 410 static int free_fid(V9fsState *s, int32_t fid)
 411 {
 412     V9fsFidState **fidpp, *fidp;
 413
 414     for (fidpp = &s->fid_list; *fidpp; fidpp = &(*fidpp)->next) {
 415         if ((*fidpp)->fid == fid) {
 416             break;
 417         }
 418     }
 419
 420     if (*fidpp == NULL) {
 421         return -ENOENT;
 422     }
 423
 424     fidp = *fidpp;
 425     *fidpp = fidp->next;
 426
 427     if (fidp->fd != -1) {
 428         v9fs_do_close(s, fidp->fd);
 429     }
 430     if (fidp->dir) {
 431         v9fs_do_closedir(s, fidp->dir);
 432     }
 433     v9fs_string_free(&fidp->path);
 434     qemu_free(fidp);
 435
 436     return 0;
 437 }
 438
 439 #define P9_QID_TYPE_DIR         0x80
 440 #define P9_QID_TYPE_SYMLINK     0x02
 441
 442 #define P9_STAT_MODE_DIR        0x80000000
 443 #define P9_STAT_MODE_APPEND     0x40000000
 444 #define P9_STAT_MODE_EXCL       0x20000000
 445 #define P9_STAT_MODE_MOUNT      0x10000000
 446 #define P9_STAT_MODE_AUTH       0x08000000
 447 #define P9_STAT_MODE_TMP        0x04000000
 448 #define P9_STAT_MODE_SYMLINK    0x02000000
 449 #define P9_STAT_MODE_LINK       0x01000000
 450 #define P9_STAT_MODE_DEVICE     0x00800000
 451 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
 452 #define P9_STAT_MODE_SOCKET     0x00100000
 453 #define P9_STAT_MODE_SETUID     0x00080000
 454 #define P9_STAT_MODE_SETGID     0x00040000
 455 #define P9_STAT_MODE_SETVTX     0x00010000
 456
 457 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR |          \
 458                                 P9_STAT_MODE_SYMLINK |      \
 459                                 P9_STAT_MODE_LINK |         \
 460                                 P9_STAT_MODE_DEVICE |       \
 461                                 P9_STAT_MODE_NAMED_PIPE |   \
 462                                 P9_STAT_MODE_SOCKET)
 463
 464 /* This is the algorithm from ufs in spfs */
 465 static void stat_to_qid(const struct stat *stbuf, V9fsQID *qidp)
 466 {
 467     size_t size;
 468
 469     size = MIN(sizeof(stbuf->st_ino), sizeof(qidp->path));
 470     memcpy(&qidp->path, &stbuf->st_ino, size);
 471     qidp->version = stbuf->st_mtime ^ (stbuf->st_size << 8);
 472     qidp->type = 0;
 473     if (S_ISDIR(stbuf->st_mode)) {
 474         qidp->type |= P9_QID_TYPE_DIR;
 475     }
 476     if (S_ISLNK(stbuf->st_mode)) {
 477         qidp->type |= P9_QID_TYPE_SYMLINK;
 478     }
 479 }
 480
 481 static int fid_to_qid(V9fsState *s, V9fsFidState *fidp, V9fsQID *qidp)
 482 {
 483     struct stat stbuf;
 484     int err;
 485
 486     err = v9fs_do_lstat(s, &fidp->path, &stbuf);
 487     if (err) {
 488         return err;
 489     }
 490
 491     stat_to_qid(&stbuf, qidp);
 492     return 0;
 493 }
 494
 495 static V9fsPDU *alloc_pdu(V9fsState *s)
 496 {
 497     V9fsPDU *pdu = NULL;
 498
 499     if (!QLIST_EMPTY(&s->free_list)) {
 500         pdu = QLIST_FIRST(&s->free_list);
 501         QLIST_REMOVE(pdu, next);
 502     }
 503     return pdu;
 504 }
 505
 506 static void free_pdu(V9fsState *s, V9fsPDU *pdu)
 507 {
 508     if (pdu) {
 509         QLIST_INSERT_HEAD(&s->free_list, pdu, next);
 510     }
 511 }
 512
 513 size_t pdu_packunpack(void *addr, struct iovec *sg, int sg_count,
 514                         size_t offset, size_t size, int pack)
 515 {
 516     int i = 0;
 517     size_t copied = 0;
 518
 519     for (i = 0; size && i < sg_count; i++) {
 520         size_t len;
 521         if (offset >= sg[i].iov_len) {
 522             /* skip this sg */
 523             offset -= sg[i].iov_len;
 524             continue;
 525         } else {
 526             len = MIN(sg[i].iov_len - offset, size);
 527             if (pack) {
 528                 memcpy(sg[i].iov_base + offset, addr, len);
 529             } else {
 530                 memcpy(addr, sg[i].iov_base + offset, len);
 531             }
 532             size -= len;
 533             copied += len;
 534             addr += len;
 535             if (size) {
 536                 offset = 0;
 537                 continue;
 538             }
 539         }
 540     }
 541
 542     return copied;
 543 }
 544
 545 static size_t pdu_unpack(void *dst, V9fsPDU *pdu, size_t offset, size_t size)
 546 {
 547     return pdu_packunpack(dst, pdu->elem.out_sg, pdu->elem.out_num,
 548                          offset, size, 0);
 549 }
 550
 551 static size_t pdu_pack(V9fsPDU *pdu, size_t offset, const void *src,
 552                         size_t size)
 553 {
 554     return pdu_packunpack((void *)src, pdu->elem.in_sg, pdu->elem.in_num,
 555                              offset, size, 1);
 556 }
 557
 558 static int pdu_copy_sg(V9fsPDU *pdu, size_t offset, int rx, struct iovec *sg)
 559 {
 560     size_t pos = 0;
 561     int i, j;
 562     struct iovec *src_sg;
 563     unsigned int num;
 564
 565     if (rx) {
 566         src_sg = pdu->elem.in_sg;
 567         num = pdu->elem.in_num;
 568     } else {
 569         src_sg = pdu->elem.out_sg;
 570         num = pdu->elem.out_num;
 571     }
 572
 573     j = 0;
 574     for (i = 0; i < num; i++) {
 575         if (offset <= pos) {
 576             sg[j].iov_base = src_sg[i].iov_base;
 577             sg[j].iov_len = src_sg[i].iov_len;
 578             j++;
 579         } else if (offset < (src_sg[i].iov_len + pos)) {
 580             sg[j].iov_base = src_sg[i].iov_base;
 581             sg[j].iov_len = src_sg[i].iov_len;
 582             sg[j].iov_base += (offset - pos);
 583             sg[j].iov_len -= (offset - pos);
 584             j++;
 585         }
 586         pos += src_sg[i].iov_len;
 587     }
 588
 589     return j;
 590 }
 591
 592 static size_t pdu_unmarshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
 593 {
 594     size_t old_offset = offset;
 595     va_list ap;
 596     int i;
 597
 598     va_start(ap, fmt);
 599     for (i = 0; fmt[i]; i++) {
 600         switch (fmt[i]) {
 601         case 'b': {
 602             uint8_t *valp = va_arg(ap, uint8_t *);
 603             offset += pdu_unpack(valp, pdu, offset, sizeof(*valp));
 604             break;
 605         }
 606         case 'w': {
 607             uint16_t val, *valp;
 608             valp = va_arg(ap, uint16_t *);
 609             val = le16_to_cpupu(valp);
 610             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 611             *valp = val;
 612             break;
 613         }
 614         case 'd': {
 615             uint32_t val, *valp;
 616             valp = va_arg(ap, uint32_t *);
 617             val = le32_to_cpupu(valp);
 618             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 619             *valp = val;
 620             break;
 621         }
 622         case 'q': {
 623             uint64_t val, *valp;
 624             valp = va_arg(ap, uint64_t *);
 625             val = le64_to_cpup(valp);
 626             offset += pdu_unpack(&val, pdu, offset, sizeof(val));
 627             *valp = val;
 628             break;
 629         }
 630         case 'v': {
 631             struct iovec *iov = va_arg(ap, struct iovec *);
 632             int *iovcnt = va_arg(ap, int *);
 633             *iovcnt = pdu_copy_sg(pdu, offset, 0, iov);
 634             break;
 635         }
 636         case 's': {
 637             V9fsString *str = va_arg(ap, V9fsString *);
 638             offset += pdu_unmarshal(pdu, offset, "w", &str->size);
 639             /* FIXME: sanity check str->size */
 640             str->data = qemu_malloc(str->size + 1);
 641             offset += pdu_unpack(str->data, pdu, offset, str->size);
 642             str->data[str->size] = 0;
 643             break;
 644         }
 645         case 'Q': {
 646             V9fsQID *qidp = va_arg(ap, V9fsQID *);
 647             offset += pdu_unmarshal(pdu, offset, "bdq",
 648                         &qidp->type, &qidp->version, &qidp->path);
 649             break;
 650         }
 651         case 'S': {
 652             V9fsStat *statp = va_arg(ap, V9fsStat *);
 653             offset += pdu_unmarshal(pdu, offset, "wwdQdddqsssssddd",
 654                         &statp->size, &statp->type, &statp->dev,
 655                         &statp->qid, &statp->mode, &statp->atime,
 656                         &statp->mtime, &statp->length,
 657                         &statp->name, &statp->uid, &statp->gid,
 658                         &statp->muid, &statp->extension,
 659                         &statp->n_uid, &statp->n_gid,
 660                         &statp->n_muid);
 661             break;
 662         }
 663         default:
 664             break;
 665         }
 666     }
 667
 668     va_end(ap);
 669
 670     return offset - old_offset;
 671 }
 672
 673 static size_t pdu_marshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
 674 {
 675     size_t old_offset = offset;
 676     va_list ap;
 677     int i;
 678
 679     va_start(ap, fmt);
 680     for (i = 0; fmt[i]; i++) {
 681         switch (fmt[i]) {
 682         case 'b': {
 683             uint8_t val = va_arg(ap, int);
 684             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 685             break;
 686         }
 687         case 'w': {
 688             uint16_t val;
 689             cpu_to_le16w(&val, va_arg(ap, int));
 690             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 691             break;
 692         }
 693         case 'd': {
 694             uint32_t val;
 695             cpu_to_le32w(&val, va_arg(ap, uint32_t));
 696             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 697             break;
 698         }
 699         case 'q': {
 700             uint64_t val;
 701             cpu_to_le64w(&val, va_arg(ap, uint64_t));
 702             offset += pdu_pack(pdu, offset, &val, sizeof(val));
 703             break;
 704         }
 705         case 'v': {
 706             struct iovec *iov = va_arg(ap, struct iovec *);
 707             int *iovcnt = va_arg(ap, int *);
 708             *iovcnt = pdu_copy_sg(pdu, offset, 1, iov);
 709             break;
 710         }
 711         case 's': {
 712             V9fsString *str = va_arg(ap, V9fsString *);
 713             offset += pdu_marshal(pdu, offset, "w", str->size);
 714             offset += pdu_pack(pdu, offset, str->data, str->size);
 715             break;
 716         }
 717         case 'Q': {
 718             V9fsQID *qidp = va_arg(ap, V9fsQID *);
 719             offset += pdu_marshal(pdu, offset, "bdq",
 720                         qidp->type, qidp->version, qidp->path);
 721             break;
 722         }
 723         case 'S': {
 724             V9fsStat *statp = va_arg(ap, V9fsStat *);
 725             offset += pdu_marshal(pdu, offset, "wwdQdddqsssssddd",
 726                         statp->size, statp->type, statp->dev,
 727                         &statp->qid, statp->mode, statp->atime,
 728                         statp->mtime, statp->length, &statp->name,
 729                         &statp->uid, &statp->gid, &statp->muid,
 730                         &statp->extension, statp->n_uid,
 731                         statp->n_gid, statp->n_muid);
 732             break;
 733         }
 734         default:
 735             break;
 736         }
 737     }
 738     va_end(ap);
 739
 740     return offset - old_offset;
 741 }
 742
 743 static void complete_pdu(V9fsState *s, V9fsPDU *pdu, ssize_t len)
 744 {
 745     int8_t id = pdu->id + 1; /* Response */
 746
 747     if (len < 0) {
 748         V9fsString str;
 749         int err = -len;
 750
 751         str.data = strerror(err);
 752         str.size = strlen(str.data);
 753
 754         len = 7;
 755         len += pdu_marshal(pdu, len, "s", &str);
 756         if (dotu) {
 757             len += pdu_marshal(pdu, len, "d", err);
 758         }
 759
 760         id = P9_RERROR;
 761     }
 762
 763     /* fill out the header */
 764     pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag);
 765
 766     /* keep these in sync */
 767     pdu->size = len;
 768     pdu->id = id;
 769
 770     /* push onto queue and notify */
 771     virtqueue_push(s->vq, &pdu->elem, len);
 772
 773     /* FIXME: we should batch these completions */
 774     virtio_notify(&s->vdev, s->vq);
 775
 776     free_pdu(s, pdu);
 777 }
 778
 779 static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
 780 {
 781     mode_t ret;
 782
 783     ret = mode & 0777;
 784     if (mode & P9_STAT_MODE_DIR) {
 785         ret |= S_IFDIR;
 786     }
 787
 788     if (dotu) {
 789         if (mode & P9_STAT_MODE_SYMLINK) {
 790             ret |= S_IFLNK;
 791         }
 792         if (mode & P9_STAT_MODE_SOCKET) {
 793             ret |= S_IFSOCK;
 794         }
 795         if (mode & P9_STAT_MODE_NAMED_PIPE) {
 796             ret |= S_IFIFO;
 797         }
 798         if (mode & P9_STAT_MODE_DEVICE) {
 799             if (extension && extension->data[0] == 'c') {
 800                 ret |= S_IFCHR;
 801             } else {
 802                 ret |= S_IFBLK;
 803             }
 804         }
 805     }
 806
 807     if (!(ret&~0777)) {
 808         ret |= S_IFREG;
 809     }
 810
 811     if (mode & P9_STAT_MODE_SETUID) {
 812         ret |= S_ISUID;
 813     }
 814     if (mode & P9_STAT_MODE_SETGID) {
 815         ret |= S_ISGID;
 816     }
 817     if (mode & P9_STAT_MODE_SETVTX) {
 818         ret |= S_ISVTX;
 819     }
 820
 821     return ret;
 822 }
 823
 824 static int donttouch_stat(V9fsStat *stat)
 825 {
 826     if (stat->type == -1 &&
 827         stat->dev == -1 &&
 828         stat->qid.type == -1 &&
 829         stat->qid.version == -1 &&
 830         stat->qid.path == -1 &&
 831         stat->mode == -1 &&
 832         stat->atime == -1 &&
 833         stat->mtime == -1 &&
 834         stat->length == -1 &&
 835         !stat->name.size &&
 836         !stat->uid.size &&
 837         !stat->gid.size &&
 838         !stat->muid.size &&
 839         stat->n_uid == -1 &&
 840         stat->n_gid == -1 &&
 841         stat->n_muid == -1) {
 842         return 1;
 843     }
 844
 845     return 0;
 846 }
 847
 848 static void v9fs_stat_free(V9fsStat *stat)
 849 {
 850     v9fs_string_free(&stat->name);
 851     v9fs_string_free(&stat->uid);
 852     v9fs_string_free(&stat->gid);
 853     v9fs_string_free(&stat->muid);
 854     v9fs_string_free(&stat->extension);
 855 }
 856
 857 static uint32_t stat_to_v9mode(const struct stat *stbuf)
 858 {
 859     uint32_t mode;
 860
 861     mode = stbuf->st_mode & 0777;
 862     if (S_ISDIR(stbuf->st_mode)) {
 863         mode |= P9_STAT_MODE_DIR;
 864     }
 865
 866     if (dotu) {
 867         if (S_ISLNK(stbuf->st_mode)) {
 868             mode |= P9_STAT_MODE_SYMLINK;
 869         }
 870
 871         if (S_ISSOCK(stbuf->st_mode)) {
 872             mode |= P9_STAT_MODE_SOCKET;
 873         }
 874
 875         if (S_ISFIFO(stbuf->st_mode)) {
 876             mode |= P9_STAT_MODE_NAMED_PIPE;
 877         }
 878
 879         if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
 880             mode |= P9_STAT_MODE_DEVICE;
 881         }
 882
 883         if (stbuf->st_mode & S_ISUID) {
 884             mode |= P9_STAT_MODE_SETUID;
 885         }
 886
 887         if (stbuf->st_mode & S_ISGID) {
 888             mode |= P9_STAT_MODE_SETGID;
 889         }
 890
 891         if (stbuf->st_mode & S_ISVTX) {
 892             mode |= P9_STAT_MODE_SETVTX;
 893         }
 894     }
 895
 896     return mode;
 897 }
 898
 899 static int stat_to_v9stat(V9fsState *s, V9fsString *name,
 900                             const struct stat *stbuf,
 901                             V9fsStat *v9stat)
 902 {
 903     int err;
 904     const char *str;
 905
 906     memset(v9stat, 0, sizeof(*v9stat));
 907
 908     stat_to_qid(stbuf, &v9stat->qid);
 909     v9stat->mode = stat_to_v9mode(stbuf);
 910     v9stat->atime = stbuf->st_atime;
 911     v9stat->mtime = stbuf->st_mtime;
 912     v9stat->length = stbuf->st_size;
 913
 914     v9fs_string_null(&v9stat->uid);
 915     v9fs_string_null(&v9stat->gid);
 916     v9fs_string_null(&v9stat->muid);
 917
 918     if (dotu) {
 919         v9stat->n_uid = stbuf->st_uid;
 920         v9stat->n_gid = stbuf->st_gid;
 921         v9stat->n_muid = 0;
 922
 923         v9fs_string_null(&v9stat->extension);
 924
 925         if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
 926             err = v9fs_do_readlink(s, name, &v9stat->extension);
 927             if (err == -1) {
 928                 err = -errno;
 929                 return err;
 930             }
 931             v9stat->extension.data[err] = 0;
 932             v9stat->extension.size = err;
 933         } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
 934             v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
 935                     S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
 936                     major(stbuf->st_rdev), minor(stbuf->st_rdev));
 937         } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
 938             v9fs_string_sprintf(&v9stat->extension, "%s %u",
 939                     "HARDLINKCOUNT", stbuf->st_nlink);
 940         }
 941     }
 942
 943     str = strrchr(name->data, '/');
 944     if (str) {
 945         str += 1;
 946     } else {
 947         str = name->data;
 948     }
 949
 950     v9fs_string_sprintf(&v9stat->name, "%s", str);
 951
 952     v9stat->size = 61 +
 953         v9fs_string_size(&v9stat->name) +
 954         v9fs_string_size(&v9stat->uid) +
 955         v9fs_string_size(&v9stat->gid) +
 956         v9fs_string_size(&v9stat->muid) +
 957         v9fs_string_size(&v9stat->extension);
 958     return 0;
 959 }
 960
 961 static struct iovec *adjust_sg(struct iovec *sg, int len, int *iovcnt)
 962 {
 963     while (len && *iovcnt) {
 964         if (len < sg->iov_len) {
 965             sg->iov_len -= len;
 966             sg->iov_base += len;
 967             len = 0;
 968         } else {
 969             len -= sg->iov_len;
 970             sg++;
 971             *iovcnt -= 1;
 972         }
 973     }
 974
 975     return sg;
 976 }
 977
 978 static struct iovec *cap_sg(struct iovec *sg, int cap, int *cnt)
 979 {
 980     int i;
 981     int total = 0;
 982
 983     for (i = 0; i < *cnt; i++) {
 984         if ((total + sg[i].iov_len) > cap) {
 985             sg[i].iov_len -= ((total + sg[i].iov_len) - cap);
 986             i++;
 987             break;
 988         }
 989         total += sg[i].iov_len;
 990     }
 991
 992     *cnt = i;
 993
 994     return sg;
 995 }
 996
 997 static void print_sg(struct iovec *sg, int cnt)
 998 {
 999     int i;
1000
1001     printf("sg[%d]: {", cnt);
1002     for (i = 0; i < cnt; i++) {
1003         if (i) {
1004             printf(", ");
1005         }
1006         printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
1007     }
1008     printf("}\n");
1009 }
1010
1011 static void v9fs_fix_path(V9fsString *dst, V9fsString *src, int len)
1012 {
1013     V9fsString str;
1014     v9fs_string_init(&str);
1015     v9fs_string_copy(&str, dst);
1016     v9fs_string_sprintf(dst, "%s%s", src->data, str.data+len);
1017     v9fs_string_free(&str);
1018 }
1019
1020 static void v9fs_version(V9fsState *s, V9fsPDU *pdu)
1021 {
1022     int32_t msize;
1023     V9fsString version;
1024     size_t offset = 7;
1025
1026     pdu_unmarshal(pdu, offset, "ds", &msize, &version);
1027
1028     if (!strcmp(version.data, "9P2000.u")) {
1029         s->proto_version = V9FS_PROTO_2000U;
1030     } else if (!strcmp(version.data, "9P2000.L")) {
1031         s->proto_version = V9FS_PROTO_2000L;
1032     } else {
1033         v9fs_string_sprintf(&version, "unknown");
1034     }
1035
1036     offset += pdu_marshal(pdu, offset, "ds", msize, &version);
1037     complete_pdu(s, pdu, offset);
1038
1039     v9fs_string_free(&version);
1040 }
1041
1042 static void v9fs_attach(V9fsState *s, V9fsPDU *pdu)
1043 {
1044     int32_t fid, afid, n_uname;
1045     V9fsString uname, aname;
1046     V9fsFidState *fidp;
1047     V9fsQID qid;
1048     size_t offset = 7;
1049     ssize_t err;
1050
1051     pdu_unmarshal(pdu, offset, "ddssd", &fid, &afid, &uname, &aname, &n_uname);
1052
1053     fidp = alloc_fid(s, fid);
1054     if (fidp == NULL) {
1055         err = -EINVAL;
1056         goto out;
1057     }
1058
1059     fidp->uid = n_uname;
1060
1061     v9fs_string_sprintf(&fidp->path, "%s", "/");
1062     err = fid_to_qid(s, fidp, &qid);
1063     if (err) {
1064         err = -EINVAL;
1065         free_fid(s, fid);
1066         goto out;
1067     }
1068
1069     offset += pdu_marshal(pdu, offset, "Q", &qid);
1070
1071     err = offset;
1072 out:
1073     complete_pdu(s, pdu, err);
1074     v9fs_string_free(&uname);
1075     v9fs_string_free(&aname);
1076 }
1077
1078 static void v9fs_stat_post_lstat(V9fsState *s, V9fsStatState *vs, int err)
1079 {
1080     if (err == -1) {
1081         err = -errno;
1082         goto out;
1083     }
1084
1085     err = stat_to_v9stat(s, &vs->fidp->path, &vs->stbuf, &vs->v9stat);
1086     if (err) {
1087         goto out;
1088     }
1089     vs->offset += pdu_marshal(vs->pdu, vs->offset, "wS", 0, &vs->v9stat);
1090     err = vs->offset;
1091
1092 out:
1093     complete_pdu(s, vs->pdu, err);
1094     v9fs_stat_free(&vs->v9stat);
1095     qemu_free(vs);
1096 }
1097
1098 static void v9fs_stat(V9fsState *s, V9fsPDU *pdu)
1099 {
1100     int32_t fid;
1101     V9fsStatState *vs;
1102     ssize_t err = 0;
1103
1104     vs = qemu_malloc(sizeof(*vs));
1105     vs->pdu = pdu;
1106     vs->offset = 7;
1107
1108     memset(&vs->v9stat, 0, sizeof(vs->v9stat));
1109
1110     pdu_unmarshal(vs->pdu, vs->offset, "d", &fid);
1111
1112     vs->fidp = lookup_fid(s, fid);
1113     if (vs->fidp == NULL) {
1114         err = -ENOENT;
1115         goto out;
1116     }
1117
1118     err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1119     v9fs_stat_post_lstat(s, vs, err);
1120     return;
1121
1122 out:
1123     complete_pdu(s, vs->pdu, err);
1124     v9fs_stat_free(&vs->v9stat);
1125     qemu_free(vs);
1126 }
1127
1128 static void v9fs_walk_complete(V9fsState *s, V9fsWalkState *vs, int err)
1129 {
1130     complete_pdu(s, vs->pdu, err);
1131
1132     if (vs->nwnames) {
1133         for (vs->name_idx = 0; vs->name_idx < vs->nwnames; vs->name_idx++) {
1134             v9fs_string_free(&vs->wnames[vs->name_idx]);
1135         }
1136
1137         qemu_free(vs->wnames);
1138         qemu_free(vs->qids);
1139     }
1140 }
1141
1142 static void v9fs_walk_marshal(V9fsWalkState *vs)
1143 {
1144     int i;
1145     vs->offset = 7;
1146     vs->offset += pdu_marshal(vs->pdu, vs->offset, "w", vs->nwnames);
1147
1148     for (i = 0; i < vs->nwnames; i++) {
1149         vs->offset += pdu_marshal(vs->pdu, vs->offset, "Q", &vs->qids[i]);
1150     }
1151 }
1152
1153 static void v9fs_walk_post_newfid_lstat(V9fsState *s, V9fsWalkState *vs,
1154                                                                 int err)
1155 {
1156     if (err == -1) {
1157         free_fid(s, vs->newfidp->fid);
1158         v9fs_string_free(&vs->path);
1159         err = -ENOENT;
1160         goto out;
1161     }
1162
1163     stat_to_qid(&vs->stbuf, &vs->qids[vs->name_idx]);
1164
1165     vs->name_idx++;
1166     if (vs->name_idx < vs->nwnames) {
1167         v9fs_string_sprintf(&vs->path, "%s/%s", vs->newfidp->path.data,
1168                                             vs->wnames[vs->name_idx].data);
1169         v9fs_string_copy(&vs->newfidp->path, &vs->path);
1170
1171         err = v9fs_do_lstat(s, &vs->newfidp->path, &vs->stbuf);
1172         v9fs_walk_post_newfid_lstat(s, vs, err);
1173         return;
1174     }
1175
1176     v9fs_string_free(&vs->path);
1177     v9fs_walk_marshal(vs);
1178     err = vs->offset;
1179 out:
1180     v9fs_walk_complete(s, vs, err);
1181 }
1182
1183 static void v9fs_walk_post_oldfid_lstat(V9fsState *s, V9fsWalkState *vs,
1184         int err)
1185 {
1186     if (err == -1) {
1187         v9fs_string_free(&vs->path);
1188         err = -ENOENT;
1189         goto out;
1190     }
1191
1192     stat_to_qid(&vs->stbuf, &vs->qids[vs->name_idx]);
1193     vs->name_idx++;
1194     if (vs->name_idx < vs->nwnames) {
1195
1196         v9fs_string_sprintf(&vs->path, "%s/%s",
1197                 vs->fidp->path.data, vs->wnames[vs->name_idx].data);
1198         v9fs_string_copy(&vs->fidp->path, &vs->path);
1199
1200         err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1201         v9fs_walk_post_oldfid_lstat(s, vs, err);
1202         return;
1203     }
1204
1205     v9fs_string_free(&vs->path);
1206     v9fs_walk_marshal(vs);
1207     err = vs->offset;
1208 out:
1209     v9fs_walk_complete(s, vs, err);
1210 }
1211
1212 static void v9fs_walk(V9fsState *s, V9fsPDU *pdu)
1213 {
1214     int32_t fid, newfid;
1215     V9fsWalkState *vs;
1216     int err = 0;
1217     int i;
1218
1219     vs = qemu_malloc(sizeof(*vs));
1220     vs->pdu = pdu;
1221     vs->wnames = NULL;
1222     vs->qids = NULL;
1223     vs->offset = 7;
1224
1225     vs->offset += pdu_unmarshal(vs->pdu, vs->offset, "ddw", &fid,
1226                                             &newfid, &vs->nwnames);
1227
1228     if (vs->nwnames) {
1229         vs->wnames = qemu_mallocz(sizeof(vs->wnames[0]) * vs->nwnames);
1230
1231         vs->qids = qemu_mallocz(sizeof(vs->qids[0]) * vs->nwnames);
1232
1233         for (i = 0; i < vs->nwnames; i++) {
1234             vs->offset += pdu_unmarshal(vs->pdu, vs->offset, "s",
1235                                             &vs->wnames[i]);
1236         }
1237     }
1238
1239     vs->fidp = lookup_fid(s, fid);
1240     if (vs->fidp == NULL) {
1241         err = -ENOENT;
1242         goto out;
1243     }
1244
1245     /* FIXME: is this really valid? */
1246     if (fid == newfid) {
1247
1248         BUG_ON(vs->fidp->fd != -1);
1249         BUG_ON(vs->fidp->dir);
1250         v9fs_string_init(&vs->path);
1251         vs->name_idx = 0;
1252
1253         if (vs->name_idx < vs->nwnames) {
1254             v9fs_string_sprintf(&vs->path, "%s/%s",
1255                 vs->fidp->path.data, vs->wnames[vs->name_idx].data);
1256             v9fs_string_copy(&vs->fidp->path, &vs->path);
1257
1258             err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1259             v9fs_walk_post_oldfid_lstat(s, vs, err);
1260             return;
1261         }
1262     } else {
1263         vs->newfidp = alloc_fid(s, newfid);
1264         if (vs->newfidp == NULL) {
1265             err = -EINVAL;
1266             goto out;
1267         }
1268
1269         vs->newfidp->uid = vs->fidp->uid;
1270         v9fs_string_init(&vs->path);
1271         vs->name_idx = 0;
1272         v9fs_string_copy(&vs->newfidp->path, &vs->fidp->path);
1273
1274         if (vs->name_idx < vs->nwnames) {
1275             v9fs_string_sprintf(&vs->path, "%s/%s", vs->newfidp->path.data,
1276                                 vs->wnames[vs->name_idx].data);
1277             v9fs_string_copy(&vs->newfidp->path, &vs->path);
1278
1279             err = v9fs_do_lstat(s, &vs->newfidp->path, &vs->stbuf);
1280             v9fs_walk_post_newfid_lstat(s, vs, err);
1281             return;
1282         }
1283     }
1284
1285     v9fs_walk_marshal(vs);
1286     err = vs->offset;
1287 out:
1288     v9fs_walk_complete(s, vs, err);
1289 }
1290
1291 static void v9fs_open_post_opendir(V9fsState *s, V9fsOpenState *vs, int err)
1292 {
1293     if (vs->fidp->dir == NULL) {
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
1306 static void v9fs_open_post_open(V9fsState *s, V9fsOpenState *vs, int err)
1307 {
1308     if (vs->fidp->fd == -1) {
1309         err = -errno;
1310         goto out;
1311     }
1312
1313     vs->offset += pdu_marshal(vs->pdu, vs->offset, "Qd", &vs->qid, 0);
1314     err = vs->offset;
1315 out:
1316     complete_pdu(s, vs->pdu, err);
1317     qemu_free(vs);
1318 }
1319
1320 static void v9fs_open_post_lstat(V9fsState *s, V9fsOpenState *vs, int err)
1321 {
1322     if (err) {
1323         err = -errno;
1324         goto out;
1325     }
1326
1327     stat_to_qid(&vs->stbuf, &vs->qid);
1328
1329     if (S_ISDIR(vs->stbuf.st_mode)) {
1330         vs->fidp->dir = v9fs_do_opendir(s, &vs->fidp->path);
1331         v9fs_open_post_opendir(s, vs, err);
1332     } else {
1333         vs->fidp->fd = v9fs_do_open(s, &vs->fidp->path,
1334                                     omode_to_uflags(vs->mode));
1335         v9fs_open_post_open(s, vs, err);
1336     }
1337     return;
1338 out:
1339     complete_pdu(s, vs->pdu, err);
1340     qemu_free(vs);
1341 }
1342
1343 static void v9fs_open(V9fsState *s, V9fsPDU *pdu)
1344 {
1345     int32_t fid;
1346     V9fsOpenState *vs;
1347     ssize_t err = 0;
1348
1349
1350     vs = qemu_malloc(sizeof(*vs));
1351     vs->pdu = pdu;
1352     vs->offset = 7;
1353
1354     pdu_unmarshal(vs->pdu, vs->offset, "db", &fid, &vs->mode);
1355
1356     vs->fidp = lookup_fid(s, fid);
1357     if (vs->fidp == NULL) {
1358         err = -ENOENT;
1359         goto out;
1360     }
1361
1362     BUG_ON(vs->fidp->fd != -1);
1363     BUG_ON(vs->fidp->dir);
1364
1365     err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
1366
1367     v9fs_open_post_lstat(s, vs, err);
1368     return;
1369 out:
1370     complete_pdu(s, pdu, err);
1371     qemu_free(vs);
1372 }
1373
1374 static void v9fs_clunk(V9fsState *s, V9fsPDU *pdu)
1375 {
1376     int32_t fid;
1377     size_t offset = 7;
1378     int err;
1379
1380     pdu_unmarshal(pdu, offset, "d", &fid);
1381
1382     err = free_fid(s, fid);
1383     if (err < 0) {
1384         goto out;
1385     }
1386
1387     offset = 7;
1388     err = offset;
1389 out:
1390     complete_pdu(s, pdu, err);
1391 }
1392
1393 static void v9fs_read_post_readdir(V9fsState *, V9fsReadState *, ssize_t);
1394
1395 static void v9fs_read_post_seekdir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1396 {
1397     if (err) {
1398         goto out;
1399     }
1400     v9fs_stat_free(&vs->v9stat);
1401     v9fs_string_free(&vs->name);
1402     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1403     vs->offset += vs->count;
1404     err = vs->offset;
1405 out:
1406     complete_pdu(s, vs->pdu, err);
1407     qemu_free(vs);
1408     return;
1409 }
1410
1411 static void v9fs_read_post_dir_lstat(V9fsState *s, V9fsReadState *vs,
1412                                     ssize_t err)
1413 {
1414     if (err) {
1415         err = -errno;
1416         goto out;
1417     }
1418     err = stat_to_v9stat(s, &vs->name, &vs->stbuf, &vs->v9stat);
1419     if (err) {
1420         goto out;
1421     }
1422
1423     vs->len = pdu_marshal(vs->pdu, vs->offset + 4 + vs->count, "S",
1424                             &vs->v9stat);
1425     if ((vs->len != (vs->v9stat.size + 2)) ||
1426             ((vs->count + vs->len) > vs->max_count)) {
1427         v9fs_do_seekdir(s, vs->fidp->dir, vs->dir_pos);
1428         v9fs_read_post_seekdir(s, vs, err);
1429         return;
1430     }
1431     vs->count += vs->len;
1432     v9fs_stat_free(&vs->v9stat);
1433     v9fs_string_free(&vs->name);
1434     vs->dir_pos = vs->dent->d_off;
1435     vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1436     v9fs_read_post_readdir(s, vs, err);
1437     return;
1438 out:
1439     v9fs_do_seekdir(s, vs->fidp->dir, vs->dir_pos);
1440     v9fs_read_post_seekdir(s, vs, err);
1441     return;
1442
1443 }
1444
1445 static void v9fs_read_post_readdir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1446 {
1447     if (vs->dent) {
1448         memset(&vs->v9stat, 0, sizeof(vs->v9stat));
1449         v9fs_string_init(&vs->name);
1450         v9fs_string_sprintf(&vs->name, "%s/%s", vs->fidp->path.data,
1451                             vs->dent->d_name);
1452         err = v9fs_do_lstat(s, &vs->name, &vs->stbuf);
1453         v9fs_read_post_dir_lstat(s, vs, err);
1454         return;
1455     }
1456
1457     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1458     vs->offset += vs->count;
1459     err = vs->offset;
1460     complete_pdu(s, vs->pdu, err);
1461     qemu_free(vs);
1462     return;
1463 }
1464
1465 static void v9fs_read_post_telldir(V9fsState *s, V9fsReadState *vs, ssize_t err)
1466 {
1467     vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1468     v9fs_read_post_readdir(s, vs, err);
1469     return;
1470 }
1471
1472 static void v9fs_read_post_rewinddir(V9fsState *s, V9fsReadState *vs,
1473                                        ssize_t err)
1474 {
1475     vs->dir_pos = v9fs_do_telldir(s, vs->fidp->dir);
1476     v9fs_read_post_telldir(s, vs, err);
1477     return;
1478 }
1479
1480 static void v9fs_read_post_readv(V9fsState *s, V9fsReadState *vs, ssize_t err)
1481 {
1482     if (err  < 0) {
1483         /* IO error return the error */
1484         err = -errno;
1485         goto out;
1486     }
1487     vs->total += vs->len;
1488     vs->sg = adjust_sg(vs->sg, vs->len, &vs->cnt);
1489     if (vs->total < vs->count && vs->len > 0) {
1490         do {
1491             if (0) {
1492                 print_sg(vs->sg, vs->cnt);
1493             }
1494             vs->len = v9fs_do_readv(s, vs->fidp->fd, vs->sg, vs->cnt);
1495         } while (vs->len == -1 && errno == EINTR);
1496         if (vs->len == -1) {
1497             err  = -errno;
1498         }
1499         v9fs_read_post_readv(s, vs, err);
1500         return;
1501     }
1502     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->total);
1503     vs->offset += vs->count;
1504     err = vs->offset;
1505
1506 out:
1507     complete_pdu(s, vs->pdu, err);
1508     qemu_free(vs);
1509 }
1510
1511 static void v9fs_read_post_lseek(V9fsState *s, V9fsReadState *vs, ssize_t err)
1512 {
1513     if (err == -1) {
1514         err = -errno;
1515         goto out;
1516     }
1517     vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
1518
1519     if (vs->total < vs->count) {
1520         do {
1521             if (0) {
1522                 print_sg(vs->sg, vs->cnt);
1523             }
1524             vs->len = v9fs_do_readv(s, vs->fidp->fd, vs->sg, vs->cnt);
1525         } while (vs->len == -1 && errno == EINTR);
1526         if (vs->len == -1) {
1527             err  = -errno;
1528         }
1529         v9fs_read_post_readv(s, vs, err);
1530         return;
1531     }
1532 out:
1533     complete_pdu(s, vs->pdu, err);
1534     qemu_free(vs);
1535 }
1536
1537 static void v9fs_read(V9fsState *s, V9fsPDU *pdu)
1538 {
1539     int32_t fid;
1540     V9fsReadState *vs;
1541     ssize_t err = 0;
1542
1543     vs = qemu_malloc(sizeof(*vs));
1544     vs->pdu = pdu;
1545     vs->offset = 7;
1546     vs->total = 0;
1547     vs->len = 0;
1548     vs->count = 0;
1549
1550     pdu_unmarshal(vs->pdu, vs->offset, "dqd", &fid, &vs->off, &vs->count);
1551
1552     vs->fidp = lookup_fid(s, fid);
1553     if (vs->fidp == NULL) {
1554         err = -EINVAL;
1555         goto out;
1556     }
1557
1558     if (vs->fidp->dir) {
1559         vs->max_count = vs->count;
1560         vs->count = 0;
1561         if (vs->off == 0) {
1562             v9fs_do_rewinddir(s, vs->fidp->dir);
1563         }
1564         v9fs_read_post_rewinddir(s, vs, err);
1565         return;
1566     } else if (vs->fidp->fd != -1) {
1567         vs->sg = vs->iov;
1568         pdu_marshal(vs->pdu, vs->offset + 4, "v", vs->sg, &vs->cnt);
1569         err = v9fs_do_lseek(s, vs->fidp->fd, vs->off, SEEK_SET);
1570         v9fs_read_post_lseek(s, vs, err);
1571         return;
1572     } else {
1573         err = -EINVAL;
1574     }
1575 out:
1576     complete_pdu(s, pdu, err);
1577     qemu_free(vs);
1578 }
1579
1580 typedef struct V9fsReadDirState {
1581     V9fsPDU *pdu;
1582     V9fsFidState *fidp;
1583     V9fsQID qid;
1584     off_t saved_dir_pos;
1585     struct dirent *dent;
1586     int32_t count;
1587     int32_t max_count;
1588     size_t offset;
1589     int64_t initial_offset;
1590     V9fsString name;
1591 } V9fsReadDirState;
1592
1593 static void v9fs_readdir_post_seekdir(V9fsState *s, V9fsReadDirState *vs)
1594 {
1595     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1596     vs->offset += vs->count;
1597     complete_pdu(s, vs->pdu, vs->offset);
1598     qemu_free(vs);
1599     return;
1600 }
1601
1602 /* Size of each dirent on the wire: size of qid (13) + size of offset (8)
1603  * size of type (1) + size of name.size (2) + strlen(name.data)
1604  */
1605 #define V9_READDIR_DATA_SZ (24 + strlen(vs->name.data))
1606
1607 static void v9fs_readdir_post_readdir(V9fsState *s, V9fsReadDirState *vs)
1608 {
1609     int len;
1610     size_t size;
1611
1612     if (vs->dent) {
1613         v9fs_string_init(&vs->name);
1614         v9fs_string_sprintf(&vs->name, "%s", vs->dent->d_name);
1615
1616         if ((vs->count + V9_READDIR_DATA_SZ) > vs->max_count) {
1617             /* Ran out of buffer. Set dir back to old position and return */
1618             v9fs_do_seekdir(s, vs->fidp->dir, vs->saved_dir_pos);
1619             v9fs_readdir_post_seekdir(s, vs);
1620             return;
1621         }
1622
1623         /* Fill up just the path field of qid because the client uses
1624          * only that. To fill the entire qid structure we will have
1625          * to stat each dirent found, which is expensive
1626          */
1627         size = MIN(sizeof(vs->dent->d_ino), sizeof(vs->qid.path));
1628         memcpy(&vs->qid.path, &vs->dent->d_ino, size);
1629         /* Fill the other fields with dummy values */
1630         vs->qid.type = 0;
1631         vs->qid.version = 0;
1632
1633         len = pdu_marshal(vs->pdu, vs->offset+4+vs->count, "Qqbs",
1634                               &vs->qid, vs->dent->d_off,
1635                               vs->dent->d_type, &vs->name);
1636         vs->count += len;
1637         v9fs_string_free(&vs->name);
1638         vs->saved_dir_pos = vs->dent->d_off;
1639         vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1640         v9fs_readdir_post_readdir(s, vs);
1641         return;
1642     }
1643
1644     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->count);
1645     vs->offset += vs->count;
1646     complete_pdu(s, vs->pdu, vs->offset);
1647     qemu_free(vs);
1648     return;
1649 }
1650
1651 static void v9fs_readdir_post_telldir(V9fsState *s, V9fsReadDirState *vs)
1652 {
1653     vs->dent = v9fs_do_readdir(s, vs->fidp->dir);
1654     v9fs_readdir_post_readdir(s, vs);
1655     return;
1656 }
1657
1658 static void v9fs_readdir_post_setdir(V9fsState *s, V9fsReadDirState *vs)
1659 {
1660     vs->saved_dir_pos = v9fs_do_telldir(s, vs->fidp->dir);
1661     v9fs_readdir_post_telldir(s, vs);
1662     return;
1663 }
1664
1665 static void v9fs_readdir(V9fsState *s, V9fsPDU *pdu)
1666 {
1667     int32_t fid;
1668     V9fsReadDirState *vs;
1669     ssize_t err = 0;
1670     size_t offset = 7;
1671
1672     vs = qemu_malloc(sizeof(*vs));
1673     vs->pdu = pdu;
1674     vs->offset = 7;
1675     vs->count = 0;
1676
1677     pdu_unmarshal(vs->pdu, offset, "dqd", &fid, &vs->initial_offset,
1678                                                         &vs->max_count);
1679
1680     vs->fidp = lookup_fid(s, fid);
1681     if (vs->fidp == NULL || !(vs->fidp->dir)) {
1682         err = -EINVAL;
1683         goto out;
1684     }
1685
1686     if (vs->initial_offset == 0) {
1687         v9fs_do_rewinddir(s, vs->fidp->dir);
1688     } else {
1689         v9fs_do_seekdir(s, vs->fidp->dir, vs->initial_offset);
1690     }
1691
1692     v9fs_readdir_post_setdir(s, vs);
1693     return;
1694
1695 out:
1696     complete_pdu(s, pdu, err);
1697     qemu_free(vs);
1698     return;
1699 }
1700
1701 static void v9fs_write_post_writev(V9fsState *s, V9fsWriteState *vs,
1702                                    ssize_t err)
1703 {
1704     if (err  < 0) {
1705         /* IO error return the error */
1706         err = -errno;
1707         goto out;
1708     }
1709     vs->total += vs->len;
1710     vs->sg = adjust_sg(vs->sg, vs->len, &vs->cnt);
1711     if (vs->total < vs->count && vs->len > 0) {
1712         do {
1713             if (0) {
1714                 print_sg(vs->sg, vs->cnt);
1715             }
1716             vs->len =  v9fs_do_writev(s, vs->fidp->fd, vs->sg, vs->cnt);
1717         } while (vs->len == -1 && errno == EINTR);
1718         if (vs->len == -1) {
1719             err  = -errno;
1720         }
1721         v9fs_write_post_writev(s, vs, err);
1722         return;
1723     }
1724     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->total);
1725
1726     err = vs->offset;
1727 out:
1728     complete_pdu(s, vs->pdu, err);
1729     qemu_free(vs);
1730 }
1731
1732 static void v9fs_write_post_lseek(V9fsState *s, V9fsWriteState *vs, ssize_t err)
1733 {
1734     if (err == -1) {
1735         err = -errno;
1736         goto out;
1737     }
1738     vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
1739
1740     if (vs->total < vs->count) {
1741         do {
1742             if (0) {
1743                 print_sg(vs->sg, vs->cnt);
1744             }
1745             vs->len = v9fs_do_writev(s, vs->fidp->fd, vs->sg, vs->cnt);
1746         } while (vs->len == -1 && errno == EINTR);
1747         if (vs->len == -1) {
1748             err  = -errno;
1749         }
1750         v9fs_write_post_writev(s, vs, err);
1751         return;
1752     }
1753
1754 out:
1755     complete_pdu(s, vs->pdu, err);
1756     qemu_free(vs);
1757 }
1758
1759 static void v9fs_write(V9fsState *s, V9fsPDU *pdu)
1760 {
1761     int32_t fid;
1762     V9fsWriteState *vs;
1763     ssize_t err;
1764
1765     vs = qemu_malloc(sizeof(*vs));
1766
1767     vs->pdu = pdu;
1768     vs->offset = 7;
1769     vs->sg = vs->iov;
1770     vs->total = 0;
1771     vs->len = 0;
1772
1773     pdu_unmarshal(vs->pdu, vs->offset, "dqdv", &fid, &vs->off, &vs->count,
1774                     vs->sg, &vs->cnt);
1775
1776     vs->fidp = lookup_fid(s, fid);
1777     if (vs->fidp == NULL) {
1778         err = -EINVAL;
1779         goto out;
1780     }
1781
1782     if (vs->fidp->fd == -1) {
1783         err = -EINVAL;
1784         goto out;
1785     }
1786
1787     err = v9fs_do_lseek(s, vs->fidp->fd, vs->off, SEEK_SET);
1788
1789     v9fs_write_post_lseek(s, vs, err);
1790     return;
1791
1792 out:
1793     complete_pdu(s, vs->pdu, err);
1794     qemu_free(vs);
1795 }
1796
1797 static void v9fs_post_create(V9fsState *s, V9fsCreateState *vs, int err)
1798 {
1799     if (err == 0) {
1800         v9fs_string_copy(&vs->fidp->path, &vs->fullname);
1801         stat_to_qid(&vs->stbuf, &vs->qid);
1802
1803         vs->offset += pdu_marshal(vs->pdu, vs->offset, "Qd", &vs->qid, 0);
1804
1805         err = vs->offset;
1806     }
1807
1808     complete_pdu(s, vs->pdu, err);
1809     v9fs_string_free(&vs->name);
1810     v9fs_string_free(&vs->extension);
1811     v9fs_string_free(&vs->fullname);
1812     qemu_free(vs);
1813 }
1814
1815 static void v9fs_create_post_perms(V9fsState *s, V9fsCreateState *vs, int err)
1816 {
1817     if (err) {
1818         err = -errno;
1819     }
1820     v9fs_post_create(s, vs, err);
1821 }
1822
1823 static void v9fs_create_post_opendir(V9fsState *s, V9fsCreateState *vs,
1824                                                                     int err)
1825 {
1826     if (!vs->fidp->dir) {
1827         err = -errno;
1828     }
1829     v9fs_post_create(s, vs, err);
1830 }
1831
1832 static void v9fs_create_post_dir_lstat(V9fsState *s, V9fsCreateState *vs,
1833                                                                     int err)
1834 {
1835     if (err) {
1836         err = -errno;
1837         goto out;
1838     }
1839
1840     vs->fidp->dir = v9fs_do_opendir(s, &vs->fullname);
1841     v9fs_create_post_opendir(s, vs, err);
1842     return;
1843
1844 out:
1845     v9fs_post_create(s, vs, err);
1846 }
1847
1848 static void v9fs_create_post_mkdir(V9fsState *s, V9fsCreateState *vs, int err)
1849 {
1850     if (err) {
1851         err = -errno;
1852         goto out;
1853     }
1854
1855     err = v9fs_do_lstat(s, &vs->fullname, &vs->stbuf);
1856     v9fs_create_post_dir_lstat(s, vs, err);
1857     return;
1858
1859 out:
1860     v9fs_post_create(s, vs, err);
1861 }
1862
1863 static void v9fs_create_post_fstat(V9fsState *s, V9fsCreateState *vs, int err)
1864 {
1865     if (err) {
1866         vs->fidp->fd = -1;
1867         err = -errno;
1868     }
1869
1870     v9fs_post_create(s, vs, err);
1871     return;
1872 }
1873
1874 static void v9fs_create_post_open2(V9fsState *s, V9fsCreateState *vs, int err)
1875 {
1876     if (vs->fidp->fd == -1) {
1877         err = -errno;
1878         goto out;
1879     }
1880
1881     err = v9fs_do_fstat(s, vs->fidp->fd, &vs->stbuf);
1882     v9fs_create_post_fstat(s, vs, err);
1883
1884     return;
1885
1886 out:
1887     v9fs_post_create(s, vs, err);
1888
1889 }
1890
1891 static void v9fs_create_post_lstat(V9fsState *s, V9fsCreateState *vs, int err)
1892 {
1893
1894     if (err == 0 || errno != ENOENT) {
1895         err = -errno;
1896         goto out;
1897     }
1898
1899     if (vs->perm & P9_STAT_MODE_DIR) {
1900         err = v9fs_do_mkdir(s, vs);
1901         v9fs_create_post_mkdir(s, vs, err);
1902     } else if (vs->perm & P9_STAT_MODE_SYMLINK) {
1903         err = v9fs_do_symlink(s, vs);
1904         v9fs_create_post_perms(s, vs, err);
1905     } else if (vs->perm & P9_STAT_MODE_LINK) {
1906         int32_t nfid = atoi(vs->extension.data);
1907         V9fsFidState *nfidp = lookup_fid(s, nfid);
1908         if (nfidp == NULL) {
1909             err = -errno;
1910             v9fs_post_create(s, vs, err);
1911         }
1912         err = v9fs_do_link(s, &nfidp->path, &vs->fullname);
1913         v9fs_create_post_perms(s, vs, err);
1914     } else if (vs->perm & P9_STAT_MODE_DEVICE) {
1915         char ctype;
1916         uint32_t major, minor;
1917         mode_t nmode = 0;
1918
1919         if (sscanf(vs->extension.data, "%c %u %u", &ctype, &major,
1920                                         &minor) != 3) {
1921             err = -errno;
1922             v9fs_post_create(s, vs, err);
1923         }
1924
1925         switch (ctype) {
1926         case 'c':
1927             nmode = S_IFCHR;
1928             break;
1929         case 'b':
1930             nmode = S_IFBLK;
1931             break;
1932         default:
1933             err = -EIO;
1934             v9fs_post_create(s, vs, err);
1935         }
1936
1937         nmode |= vs->perm & 0777;
1938         err = v9fs_do_mknod(s, vs, nmode, makedev(major, minor));
1939         v9fs_create_post_perms(s, vs, err);
1940     } else if (vs->perm & P9_STAT_MODE_NAMED_PIPE) {
1941         err = v9fs_do_mknod(s, vs, S_IFIFO | (vs->perm & 0777), 0);
1942         v9fs_post_create(s, vs, err);
1943     } else if (vs->perm & P9_STAT_MODE_SOCKET) {
1944         err = v9fs_do_mknod(s, vs, S_IFSOCK | (vs->perm & 0777), 0);
1945         v9fs_post_create(s, vs, err);
1946     } else {
1947         vs->fidp->fd = v9fs_do_open2(s, vs);
1948         v9fs_create_post_open2(s, vs, err);
1949     }
1950
1951     return;
1952
1953 out:
1954     v9fs_post_create(s, vs, err);
1955 }
1956
1957 static void v9fs_create(V9fsState *s, V9fsPDU *pdu)
1958 {
1959     int32_t fid;
1960     V9fsCreateState *vs;
1961     int err = 0;
1962
1963     vs = qemu_malloc(sizeof(*vs));
1964     vs->pdu = pdu;
1965     vs->offset = 7;
1966
1967     v9fs_string_init(&vs->fullname);
1968
1969     pdu_unmarshal(vs->pdu, vs->offset, "dsdbs", &fid, &vs->name,
1970                                 &vs->perm, &vs->mode, &vs->extension);
1971
1972     vs->fidp = lookup_fid(s, fid);
1973     if (vs->fidp == NULL) {
1974         err = -EINVAL;
1975         goto out;
1976     }
1977
1978     v9fs_string_sprintf(&vs->fullname, "%s/%s", vs->fidp->path.data,
1979                                                         vs->name.data);
1980
1981     err = v9fs_do_lstat(s, &vs->fullname, &vs->stbuf);
1982     v9fs_create_post_lstat(s, vs, err);
1983     return;
1984
1985 out:
1986     complete_pdu(s, vs->pdu, err);
1987     v9fs_string_free(&vs->name);
1988     v9fs_string_free(&vs->extension);
1989     qemu_free(vs);
1990 }
1991
1992 static void v9fs_flush(V9fsState *s, V9fsPDU *pdu)
1993 {
1994     /* A nop call with no return */
1995     complete_pdu(s, pdu, 7);
1996 }
1997
1998 static void v9fs_remove_post_remove(V9fsState *s, V9fsRemoveState *vs,
1999                                                                 int err)
2000 {
2001     if (err < 0) {
2002         err = -errno;
2003     } else {
2004         err = vs->offset;
2005     }
2006
2007     /* For TREMOVE we need to clunk the fid even on failed remove */
2008     free_fid(s, vs->fidp->fid);
2009
2010     complete_pdu(s, vs->pdu, err);
2011     qemu_free(vs);
2012 }
2013
2014 static void v9fs_remove(V9fsState *s, V9fsPDU *pdu)
2015 {
2016     int32_t fid;
2017     V9fsRemoveState *vs;
2018     int err = 0;
2019
2020     vs = qemu_malloc(sizeof(*vs));
2021     vs->pdu = pdu;
2022     vs->offset = 7;
2023
2024     pdu_unmarshal(vs->pdu, vs->offset, "d", &fid);
2025
2026     vs->fidp = lookup_fid(s, fid);
2027     if (vs->fidp == NULL) {
2028         err = -EINVAL;
2029         goto out;
2030     }
2031
2032     err = v9fs_do_remove(s, &vs->fidp->path);
2033     v9fs_remove_post_remove(s, vs, err);
2034     return;
2035
2036 out:
2037     complete_pdu(s, pdu, err);
2038     qemu_free(vs);
2039 }
2040
2041 static void v9fs_wstat_post_truncate(V9fsState *s, V9fsWstatState *vs, int err)
2042 {
2043     if (err < 0) {
2044         goto out;
2045     }
2046
2047     err = vs->offset;
2048
2049 out:
2050     v9fs_stat_free(&vs->v9stat);
2051     complete_pdu(s, vs->pdu, err);
2052     qemu_free(vs);
2053 }
2054
2055 static void v9fs_wstat_post_rename(V9fsState *s, V9fsWstatState *vs, int err)
2056 {
2057     if (err < 0) {
2058         goto out;
2059     }
2060
2061     if (vs->v9stat.name.size != 0) {
2062         v9fs_string_free(&vs->nname);
2063     }
2064
2065     if (vs->v9stat.length != -1) {
2066         if (v9fs_do_truncate(s, &vs->fidp->path, vs->v9stat.length) < 0) {
2067             err = -errno;
2068         }
2069     }
2070     v9fs_wstat_post_truncate(s, vs, err);
2071     return;
2072
2073 out:
2074     v9fs_stat_free(&vs->v9stat);
2075     complete_pdu(s, vs->pdu, err);
2076     qemu_free(vs);
2077 }
2078
2079 static void v9fs_wstat_post_chown(V9fsState *s, V9fsWstatState *vs, int err)
2080 {
2081     V9fsFidState *fidp;
2082     if (err < 0) {
2083         goto out;
2084     }
2085
2086     if (vs->v9stat.name.size != 0) {
2087         char *old_name, *new_name;
2088         char *end;
2089
2090         old_name = vs->fidp->path.data;
2091         end = strrchr(old_name, '/');
2092         if (end) {
2093             end++;
2094         } else {
2095             end = old_name;
2096         }
2097
2098         new_name = qemu_mallocz(end - old_name + vs->v9stat.name.size + 1);
2099
2100         memcpy(new_name, old_name, end - old_name);
2101         memcpy(new_name + (end - old_name), vs->v9stat.name.data,
2102                 vs->v9stat.name.size);
2103         vs->nname.data = new_name;
2104         vs->nname.size = strlen(new_name);
2105
2106         if (strcmp(new_name, vs->fidp->path.data) != 0) {
2107             if (v9fs_do_rename(s, &vs->fidp->path, &vs->nname)) {
2108                 err = -errno;
2109             } else {
2110                 /*
2111                  * Fixup fid's pointing to the old name to
2112                  * start pointing to the new name
2113                  */
2114                 for (fidp = s->fid_list; fidp; fidp = fidp->next) {
2115
2116                     if (vs->fidp == fidp) {
2117                         /*
2118                          * we replace name of this fid towards the end
2119                          * so that our below strcmp will work
2120                          */
2121                         continue;
2122                     }
2123                     if (!strncmp(vs->fidp->path.data, fidp->path.data,
2124                                  strlen(vs->fidp->path.data))) {
2125                         /* replace the name */
2126                         v9fs_fix_path(&fidp->path, &vs->nname,
2127                                       strlen(vs->fidp->path.data));
2128                     }
2129                 }
2130                 v9fs_string_copy(&vs->fidp->path, &vs->nname);
2131             }
2132         }
2133     }
2134     v9fs_wstat_post_rename(s, vs, err);
2135     return;
2136
2137 out:
2138     v9fs_stat_free(&vs->v9stat);
2139     complete_pdu(s, vs->pdu, err);
2140     qemu_free(vs);
2141 }
2142
2143 static void v9fs_wstat_post_utime(V9fsState *s, V9fsWstatState *vs, int err)
2144 {
2145     if (err < 0) {
2146         goto out;
2147     }
2148
2149     if (vs->v9stat.n_gid != -1 || vs->v9stat.n_uid != -1) {
2150         if (v9fs_do_chown(s, &vs->fidp->path, vs->v9stat.n_uid,
2151                     vs->v9stat.n_gid)) {
2152             err = -errno;
2153         }
2154     }
2155     v9fs_wstat_post_chown(s, vs, err);
2156     return;
2157
2158 out:
2159     v9fs_stat_free(&vs->v9stat);
2160     complete_pdu(s, vs->pdu, err);
2161     qemu_free(vs);
2162 }
2163
2164 static void v9fs_wstat_post_chmod(V9fsState *s, V9fsWstatState *vs, int err)
2165 {
2166     if (err < 0) {
2167         goto out;
2168     }
2169
2170     if (vs->v9stat.mtime != -1) {
2171         struct utimbuf tb;
2172         tb.actime = 0;
2173         tb.modtime = vs->v9stat.mtime;
2174         if (v9fs_do_utime(s, &vs->fidp->path, &tb)) {
2175             err = -errno;
2176         }
2177     }
2178
2179     v9fs_wstat_post_utime(s, vs, err);
2180     return;
2181
2182 out:
2183     v9fs_stat_free(&vs->v9stat);
2184     complete_pdu(s, vs->pdu, err);
2185     qemu_free(vs);
2186 }
2187
2188 static void v9fs_wstat_post_fsync(V9fsState *s, V9fsWstatState *vs, int err)
2189 {
2190     if (err == -1) {
2191         err = -errno;
2192     }
2193     v9fs_stat_free(&vs->v9stat);
2194     complete_pdu(s, vs->pdu, err);
2195     qemu_free(vs);
2196 }
2197
2198 static void v9fs_wstat_post_lstat(V9fsState *s, V9fsWstatState *vs, int err)
2199 {
2200     uint32_t v9_mode;
2201
2202     if (err == -1) {
2203         err = -errno;
2204         goto out;
2205     }
2206
2207     v9_mode = stat_to_v9mode(&vs->stbuf);
2208
2209     if ((vs->v9stat.mode & P9_STAT_MODE_TYPE_BITS) !=
2210         (v9_mode & P9_STAT_MODE_TYPE_BITS)) {
2211             /* Attempting to change the type */
2212             err = -EIO;
2213             goto out;
2214     }
2215
2216     if (v9fs_do_chmod(s, &vs->fidp->path, v9mode_to_mode(vs->v9stat.mode,
2217                     &vs->v9stat.extension))) {
2218             err = -errno;
2219      }
2220     v9fs_wstat_post_chmod(s, vs, err);
2221     return;
2222
2223 out:
2224     v9fs_stat_free(&vs->v9stat);
2225     complete_pdu(s, vs->pdu, err);
2226     qemu_free(vs);
2227 }
2228
2229 static void v9fs_wstat(V9fsState *s, V9fsPDU *pdu)
2230 {
2231     int32_t fid;
2232     V9fsWstatState *vs;
2233     int err = 0;
2234
2235     vs = qemu_malloc(sizeof(*vs));
2236     vs->pdu = pdu;
2237     vs->offset = 7;
2238
2239     pdu_unmarshal(pdu, vs->offset, "dwS", &fid, &vs->unused, &vs->v9stat);
2240
2241     vs->fidp = lookup_fid(s, fid);
2242     if (vs->fidp == NULL) {
2243         err = -EINVAL;
2244         goto out;
2245     }
2246
2247     /* do we need to sync the file? */
2248     if (donttouch_stat(&vs->v9stat)) {
2249         err = v9fs_do_fsync(s, vs->fidp->fd);
2250         v9fs_wstat_post_fsync(s, vs, err);
2251         return;
2252     }
2253
2254     if (vs->v9stat.mode != -1) {
2255         err = v9fs_do_lstat(s, &vs->fidp->path, &vs->stbuf);
2256         v9fs_wstat_post_lstat(s, vs, err);
2257         return;
2258     }
2259
2260     v9fs_wstat_post_chmod(s, vs, err);
2261     return;
2262
2263 out:
2264     v9fs_stat_free(&vs->v9stat);
2265     complete_pdu(s, vs->pdu, err);
2266     qemu_free(vs);
2267 }
2268
2269 static int v9fs_do_statfs(V9fsState *s, V9fsString *path, struct statfs *stbuf)
2270 {
2271     return s->ops->statfs(&s->ctx, path->data, stbuf);
2272 }
2273
2274 static void v9fs_statfs_post_statfs(V9fsState *s, V9fsStatfsState *vs, int err)
2275 {
2276     if (err) {
2277         err = -errno;
2278         goto out;
2279     }
2280
2281     vs->v9statfs.f_type = vs->stbuf.f_type;
2282     vs->v9statfs.f_bsize = vs->stbuf.f_bsize;
2283     vs->v9statfs.f_blocks = vs->stbuf.f_blocks;
2284     vs->v9statfs.f_bfree = vs->stbuf.f_bfree;
2285     vs->v9statfs.f_bavail = vs->stbuf.f_bavail;
2286     vs->v9statfs.f_files = vs->stbuf.f_files;
2287     vs->v9statfs.f_ffree = vs->stbuf.f_ffree;
2288     vs->v9statfs.fsid_val = (unsigned int) vs->stbuf.f_fsid.__val[0] |
2289                         (unsigned long long)vs->stbuf.f_fsid.__val[1] << 32;
2290     vs->v9statfs.f_namelen = vs->stbuf.f_namelen;
2291
2292     vs->offset += pdu_marshal(vs->pdu, vs->offset, "ddqqqqqqd",
2293          vs->v9statfs.f_type, vs->v9statfs.f_bsize, vs->v9statfs.f_blocks,
2294          vs->v9statfs.f_bfree, vs->v9statfs.f_bavail, vs->v9statfs.f_files,
2295          vs->v9statfs.f_ffree, vs->v9statfs.fsid_val,
2296          vs->v9statfs.f_namelen);
2297
2298 out:
2299     complete_pdu(s, vs->pdu, vs->offset);
2300     qemu_free(vs);
2301 }
2302
2303 static void v9fs_statfs(V9fsState *s, V9fsPDU *pdu)
2304 {
2305     V9fsStatfsState *vs;
2306     ssize_t err = 0;
2307
2308     vs = qemu_malloc(sizeof(*vs));
2309     vs->pdu = pdu;
2310     vs->offset = 7;
2311
2312     memset(&vs->v9statfs, 0, sizeof(vs->v9statfs));
2313
2314     pdu_unmarshal(vs->pdu, vs->offset, "d", &vs->fid);
2315
2316     vs->fidp = lookup_fid(s, vs->fid);
2317     if (vs->fidp == NULL) {
2318         err = -ENOENT;
2319         goto out;
2320     }
2321
2322     err = v9fs_do_statfs(s, &vs->fidp->path, &vs->stbuf);
2323     v9fs_statfs_post_statfs(s, vs, err);
2324     return;
2325
2326 out:
2327     complete_pdu(s, vs->pdu, err);
2328     qemu_free(vs);
2329 }
2330
2331 typedef void (pdu_handler_t)(V9fsState *s, V9fsPDU *pdu);
2332
2333 static pdu_handler_t *pdu_handlers[] = {
2334     [P9_TREADDIR] = v9fs_readdir,
2335     [P9_TSTATFS] = v9fs_statfs,
2336     [P9_TVERSION] = v9fs_version,
2337     [P9_TATTACH] = v9fs_attach,
2338     [P9_TSTAT] = v9fs_stat,
2339     [P9_TWALK] = v9fs_walk,
2340     [P9_TCLUNK] = v9fs_clunk,
2341     [P9_TOPEN] = v9fs_open,
2342     [P9_TREAD] = v9fs_read,
2343 #if 0
2344     [P9_TAUTH] = v9fs_auth,
2345 #endif
2346     [P9_TFLUSH] = v9fs_flush,
2347     [P9_TCREATE] = v9fs_create,
2348     [P9_TWRITE] = v9fs_write,
2349     [P9_TWSTAT] = v9fs_wstat,
2350     [P9_TREMOVE] = v9fs_remove,
2351 };
2352
2353 static void submit_pdu(V9fsState *s, V9fsPDU *pdu)
2354 {
2355     pdu_handler_t *handler;
2356
2357     if (debug_9p_pdu) {
2358         pprint_pdu(pdu);
2359     }
2360
2361     BUG_ON(pdu->id >= ARRAY_SIZE(pdu_handlers));
2362
2363     handler = pdu_handlers[pdu->id];
2364     BUG_ON(handler == NULL);
2365
2366     handler(s, pdu);
2367 }
2368
2369 static void handle_9p_output(VirtIODevice *vdev, VirtQueue *vq)
2370 {
2371     V9fsState *s = (V9fsState *)vdev;
2372     V9fsPDU *pdu;
2373     ssize_t len;
2374
2375     while ((pdu = alloc_pdu(s)) &&
2376             (len = virtqueue_pop(vq, &pdu->elem)) != 0) {
2377         uint8_t *ptr;
2378
2379         BUG_ON(pdu->elem.out_num == 0 || pdu->elem.in_num == 0);
2380         BUG_ON(pdu->elem.out_sg[0].iov_len < 7);
2381
2382         ptr = pdu->elem.out_sg[0].iov_base;
2383
2384         memcpy(&pdu->size, ptr, 4);
2385         pdu->id = ptr[4];
2386         memcpy(&pdu->tag, ptr + 5, 2);
2387
2388         submit_pdu(s, pdu);
2389     }
2390
2391     free_pdu(s, pdu);
2392 }
2393
2394 static uint32_t virtio_9p_get_features(VirtIODevice *vdev, uint32_t features)
2395 {
2396     features |= 1 << VIRTIO_9P_MOUNT_TAG;
2397     return features;
2398 }
2399
2400 static V9fsState *to_virtio_9p(VirtIODevice *vdev)
2401 {
2402     return (V9fsState *)vdev;
2403 }
2404
2405 static void virtio_9p_get_config(VirtIODevice *vdev, uint8_t *config)
2406 {
2407     struct virtio_9p_config *cfg;
2408     V9fsState *s = to_virtio_9p(vdev);
2409
2410     cfg = qemu_mallocz(sizeof(struct virtio_9p_config) +
2411                         s->tag_len);
2412     stw_raw(&cfg->tag_len, s->tag_len);
2413     memcpy(cfg->tag, s->tag, s->tag_len);
2414     memcpy(config, cfg, s->config_size);
2415     qemu_free(cfg);
2416 }
2417
2418 VirtIODevice *virtio_9p_init(DeviceState *dev, V9fsConf *conf)
2419  {
2420     V9fsState *s;
2421     int i, len;
2422     struct stat stat;
2423     FsTypeEntry *fse;
2424
2425
2426     s = (V9fsState *)virtio_common_init("virtio-9p",
2427                                     VIRTIO_ID_9P,
2428                                     sizeof(struct virtio_9p_config)+
2429                                     MAX_TAG_LEN,
2430                                     sizeof(V9fsState));
2431
2432     /* initialize pdu allocator */
2433     QLIST_INIT(&s->free_list);
2434     for (i = 0; i < (MAX_REQ - 1); i++) {
2435         QLIST_INSERT_HEAD(&s->free_list, &s->pdus[i], next);
2436     }
2437
2438     s->vq = virtio_add_queue(&s->vdev, MAX_REQ, handle_9p_output);
2439
2440     fse = get_fsdev_fsentry(conf->fsdev_id);
2441
2442     if (!fse) {
2443         /* We don't have a fsdev identified by fsdev_id */
2444         fprintf(stderr, "Virtio-9p device couldn't find fsdev "
2445                     "with the id %s\n", conf->fsdev_id);
2446         exit(1);
2447     }
2448
2449     if (!fse->path || !conf->tag) {
2450         /* we haven't specified a mount_tag or the path */
2451         fprintf(stderr, "fsdev with id %s needs path "
2452                 "and Virtio-9p device needs mount_tag arguments\n",
2453                 conf->fsdev_id);
2454         exit(1);
2455     }
2456
2457     if (!strcmp(fse->security_model, "passthrough")) {
2458         /* Files on the Fileserver set to client user credentials */
2459         s->ctx.fs_sm = SM_PASSTHROUGH;
2460     } else if (!strcmp(fse->security_model, "mapped")) {
2461         /* Files on the fileserver are set to QEMU credentials.
2462          * Client user credentials are saved in extended attributes.
2463          */
2464         s->ctx.fs_sm = SM_MAPPED;
2465     } else {
2466         /* user haven't specified a correct security option */
2467         fprintf(stderr, "one of the following must be specified as the"
2468                 "security option:\n\t security_model=passthrough \n\t "
2469                 "security_model=mapped\n");
2470         return NULL;
2471     }
2472
2473     if (lstat(fse->path, &stat)) {
2474         fprintf(stderr, "share path %s does not exist\n", fse->path);
2475         exit(1);
2476     } else if (!S_ISDIR(stat.st_mode)) {
2477         fprintf(stderr, "share path %s is not a directory \n", fse->path);
2478         exit(1);
2479     }
2480
2481     s->ctx.fs_root = qemu_strdup(fse->path);
2482     len = strlen(conf->tag);
2483     if (len > MAX_TAG_LEN) {
2484         len = MAX_TAG_LEN;
2485     }
2486     /* s->tag is non-NULL terminated string */
2487     s->tag = qemu_malloc(len);
2488     memcpy(s->tag, conf->tag, len);
2489     s->tag_len = len;
2490     s->ctx.uid = -1;
2491
2492     s->ops = fse->ops;
2493     s->vdev.get_features = virtio_9p_get_features;
2494     s->config_size = sizeof(struct virtio_9p_config) +
2495                         s->tag_len;
2496     s->vdev.get_config = virtio_9p_get_config;
2497
2498     return &s->vdev;
2499 }