qemu/hw/9pfs/9p.c
<<
>>
Prefs
   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 "qemu/osdep.h"
  15#include <glib/gprintf.h>
  16#include "hw/virtio/virtio.h"
  17#include "qapi/error.h"
  18#include "qemu/error-report.h"
  19#include "qemu/iov.h"
  20#include "qemu/sockets.h"
  21#include "virtio-9p.h"
  22#include "fsdev/qemu-fsdev.h"
  23#include "9p-xattr.h"
  24#include "coth.h"
  25#include "trace.h"
  26#include "migration/blocker.h"
  27#include "sysemu/qtest.h"
  28
  29int open_fd_hw;
  30int total_open_fd;
  31static int open_fd_rc;
  32
  33enum {
  34    Oread   = 0x00,
  35    Owrite  = 0x01,
  36    Ordwr   = 0x02,
  37    Oexec   = 0x03,
  38    Oexcl   = 0x04,
  39    Otrunc  = 0x10,
  40    Orexec  = 0x20,
  41    Orclose = 0x40,
  42    Oappend = 0x80,
  43};
  44
  45static ssize_t pdu_marshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
  46{
  47    ssize_t ret;
  48    va_list ap;
  49
  50    va_start(ap, fmt);
  51    ret = pdu->s->transport->pdu_vmarshal(pdu, offset, fmt, ap);
  52    va_end(ap);
  53
  54    return ret;
  55}
  56
  57static ssize_t pdu_unmarshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
  58{
  59    ssize_t ret;
  60    va_list ap;
  61
  62    va_start(ap, fmt);
  63    ret = pdu->s->transport->pdu_vunmarshal(pdu, offset, fmt, ap);
  64    va_end(ap);
  65
  66    return ret;
  67}
  68
  69static int omode_to_uflags(int8_t mode)
  70{
  71    int ret = 0;
  72
  73    switch (mode & 3) {
  74    case Oread:
  75        ret = O_RDONLY;
  76        break;
  77    case Ordwr:
  78        ret = O_RDWR;
  79        break;
  80    case Owrite:
  81        ret = O_WRONLY;
  82        break;
  83    case Oexec:
  84        ret = O_RDONLY;
  85        break;
  86    }
  87
  88    if (mode & Otrunc) {
  89        ret |= O_TRUNC;
  90    }
  91
  92    if (mode & Oappend) {
  93        ret |= O_APPEND;
  94    }
  95
  96    if (mode & Oexcl) {
  97        ret |= O_EXCL;
  98    }
  99
 100    return ret;
 101}
 102
 103typedef struct DotlOpenflagMap {
 104    int dotl_flag;
 105    int open_flag;
 106} DotlOpenflagMap;
 107
 108static int dotl_to_open_flags(int flags)
 109{
 110    int i;
 111    /*
 112     * We have same bits for P9_DOTL_READONLY, P9_DOTL_WRONLY
 113     * and P9_DOTL_NOACCESS
 114     */
 115    int oflags = flags & O_ACCMODE;
 116
 117    DotlOpenflagMap dotl_oflag_map[] = {
 118        { P9_DOTL_CREATE, O_CREAT },
 119        { P9_DOTL_EXCL, O_EXCL },
 120        { P9_DOTL_NOCTTY , O_NOCTTY },
 121        { P9_DOTL_TRUNC, O_TRUNC },
 122        { P9_DOTL_APPEND, O_APPEND },
 123        { P9_DOTL_NONBLOCK, O_NONBLOCK } ,
 124        { P9_DOTL_DSYNC, O_DSYNC },
 125        { P9_DOTL_FASYNC, FASYNC },
 126        { P9_DOTL_DIRECT, O_DIRECT },
 127        { P9_DOTL_LARGEFILE, O_LARGEFILE },
 128        { P9_DOTL_DIRECTORY, O_DIRECTORY },
 129        { P9_DOTL_NOFOLLOW, O_NOFOLLOW },
 130        { P9_DOTL_NOATIME, O_NOATIME },
 131        { P9_DOTL_SYNC, O_SYNC },
 132    };
 133
 134    for (i = 0; i < ARRAY_SIZE(dotl_oflag_map); i++) {
 135        if (flags & dotl_oflag_map[i].dotl_flag) {
 136            oflags |= dotl_oflag_map[i].open_flag;
 137        }
 138    }
 139
 140    return oflags;
 141}
 142
 143void cred_init(FsCred *credp)
 144{
 145    credp->fc_uid = -1;
 146    credp->fc_gid = -1;
 147    credp->fc_mode = -1;
 148    credp->fc_rdev = -1;
 149}
 150
 151static int get_dotl_openflags(V9fsState *s, int oflags)
 152{
 153    int flags;
 154    /*
 155     * Filter the client open flags
 156     */
 157    flags = dotl_to_open_flags(oflags);
 158    flags &= ~(O_NOCTTY | O_ASYNC | O_CREAT);
 159    /*
 160     * Ignore direct disk access hint until the server supports it.
 161     */
 162    flags &= ~O_DIRECT;
 163    return flags;
 164}
 165
 166void v9fs_path_init(V9fsPath *path)
 167{
 168    path->data = NULL;
 169    path->size = 0;
 170}
 171
 172void v9fs_path_free(V9fsPath *path)
 173{
 174    g_free(path->data);
 175    path->data = NULL;
 176    path->size = 0;
 177}
 178
 179
 180void GCC_FMT_ATTR(2, 3)
 181v9fs_path_sprintf(V9fsPath *path, const char *fmt, ...)
 182{
 183    va_list ap;
 184
 185    v9fs_path_free(path);
 186
 187    va_start(ap, fmt);
 188    /* Bump the size for including terminating NULL */
 189    path->size = g_vasprintf(&path->data, fmt, ap) + 1;
 190    va_end(ap);
 191}
 192
 193void v9fs_path_copy(V9fsPath *dst, const V9fsPath *src)
 194{
 195    v9fs_path_free(dst);
 196    dst->size = src->size;
 197    dst->data = g_memdup(src->data, src->size);
 198}
 199
 200int v9fs_name_to_path(V9fsState *s, V9fsPath *dirpath,
 201                      const char *name, V9fsPath *path)
 202{
 203    int err;
 204    err = s->ops->name_to_path(&s->ctx, dirpath, name, path);
 205    if (err < 0) {
 206        err = -errno;
 207    }
 208    return err;
 209}
 210
 211/*
 212 * Return TRUE if s1 is an ancestor of s2.
 213 *
 214 * E.g. "a/b" is an ancestor of "a/b/c" but not of "a/bc/d".
 215 * As a special case, We treat s1 as ancestor of s2 if they are same!
 216 */
 217static int v9fs_path_is_ancestor(V9fsPath *s1, V9fsPath *s2)
 218{
 219    if (!strncmp(s1->data, s2->data, s1->size - 1)) {
 220        if (s2->data[s1->size - 1] == '\0' || s2->data[s1->size - 1] == '/') {
 221            return 1;
 222        }
 223    }
 224    return 0;
 225}
 226
 227static size_t v9fs_string_size(V9fsString *str)
 228{
 229    return str->size;
 230}
 231
 232/*
 233 * returns 0 if fid got re-opened, 1 if not, < 0 on error */
 234static int coroutine_fn v9fs_reopen_fid(V9fsPDU *pdu, V9fsFidState *f)
 235{
 236    int err = 1;
 237    if (f->fid_type == P9_FID_FILE) {
 238        if (f->fs.fd == -1) {
 239            do {
 240                err = v9fs_co_open(pdu, f, f->open_flags);
 241            } while (err == -EINTR && !pdu->cancelled);
 242        }
 243    } else if (f->fid_type == P9_FID_DIR) {
 244        if (f->fs.dir.stream == NULL) {
 245            do {
 246                err = v9fs_co_opendir(pdu, f);
 247            } while (err == -EINTR && !pdu->cancelled);
 248        }
 249    }
 250    return err;
 251}
 252
 253static V9fsFidState *coroutine_fn get_fid(V9fsPDU *pdu, int32_t fid)
 254{
 255    int err;
 256    V9fsFidState *f;
 257    V9fsState *s = pdu->s;
 258
 259    for (f = s->fid_list; f; f = f->next) {
 260        BUG_ON(f->clunked);
 261        if (f->fid == fid) {
 262            /*
 263             * Update the fid ref upfront so that
 264             * we don't get reclaimed when we yield
 265             * in open later.
 266             */
 267            f->ref++;
 268            /*
 269             * check whether we need to reopen the
 270             * file. We might have closed the fd
 271             * while trying to free up some file
 272             * descriptors.
 273             */
 274            err = v9fs_reopen_fid(pdu, f);
 275            if (err < 0) {
 276                f->ref--;
 277                return NULL;
 278            }
 279            /*
 280             * Mark the fid as referenced so that the LRU
 281             * reclaim won't close the file descriptor
 282             */
 283            f->flags |= FID_REFERENCED;
 284            return f;
 285        }
 286    }
 287    return NULL;
 288}
 289
 290static V9fsFidState *alloc_fid(V9fsState *s, int32_t fid)
 291{
 292    V9fsFidState *f;
 293
 294    for (f = s->fid_list; f; f = f->next) {
 295        /* If fid is already there return NULL */
 296        BUG_ON(f->clunked);
 297        if (f->fid == fid) {
 298            return NULL;
 299        }
 300    }
 301    f = g_malloc0(sizeof(V9fsFidState));
 302    f->fid = fid;
 303    f->fid_type = P9_FID_NONE;
 304    f->ref = 1;
 305    /*
 306     * Mark the fid as referenced so that the LRU
 307     * reclaim won't close the file descriptor
 308     */
 309    f->flags |= FID_REFERENCED;
 310    f->next = s->fid_list;
 311    s->fid_list = f;
 312
 313    v9fs_readdir_init(&f->fs.dir);
 314    v9fs_readdir_init(&f->fs_reclaim.dir);
 315
 316    return f;
 317}
 318
 319static int coroutine_fn v9fs_xattr_fid_clunk(V9fsPDU *pdu, V9fsFidState *fidp)
 320{
 321    int retval = 0;
 322
 323    if (fidp->fs.xattr.xattrwalk_fid) {
 324        /* getxattr/listxattr fid */
 325        goto free_value;
 326    }
 327    /*
 328     * if this is fid for setxattr. clunk should
 329     * result in setxattr localcall
 330     */
 331    if (fidp->fs.xattr.len != fidp->fs.xattr.copied_len) {
 332        /* clunk after partial write */
 333        retval = -EINVAL;
 334        goto free_out;
 335    }
 336    if (fidp->fs.xattr.len) {
 337        retval = v9fs_co_lsetxattr(pdu, &fidp->path, &fidp->fs.xattr.name,
 338                                   fidp->fs.xattr.value,
 339                                   fidp->fs.xattr.len,
 340                                   fidp->fs.xattr.flags);
 341    } else {
 342        retval = v9fs_co_lremovexattr(pdu, &fidp->path, &fidp->fs.xattr.name);
 343    }
 344free_out:
 345    v9fs_string_free(&fidp->fs.xattr.name);
 346free_value:
 347    g_free(fidp->fs.xattr.value);
 348    return retval;
 349}
 350
 351static int coroutine_fn free_fid(V9fsPDU *pdu, V9fsFidState *fidp)
 352{
 353    int retval = 0;
 354
 355    if (fidp->fid_type == P9_FID_FILE) {
 356        /* If we reclaimed the fd no need to close */
 357        if (fidp->fs.fd != -1) {
 358            retval = v9fs_co_close(pdu, &fidp->fs);
 359        }
 360    } else if (fidp->fid_type == P9_FID_DIR) {
 361        if (fidp->fs.dir.stream != NULL) {
 362            retval = v9fs_co_closedir(pdu, &fidp->fs);
 363        }
 364    } else if (fidp->fid_type == P9_FID_XATTR) {
 365        retval = v9fs_xattr_fid_clunk(pdu, fidp);
 366    }
 367    v9fs_path_free(&fidp->path);
 368    g_free(fidp);
 369    return retval;
 370}
 371
 372static int coroutine_fn put_fid(V9fsPDU *pdu, V9fsFidState *fidp)
 373{
 374    BUG_ON(!fidp->ref);
 375    fidp->ref--;
 376    /*
 377     * Don't free the fid if it is in reclaim list
 378     */
 379    if (!fidp->ref && fidp->clunked) {
 380        if (fidp->fid == pdu->s->root_fid) {
 381            /*
 382             * if the clunked fid is root fid then we
 383             * have unmounted the fs on the client side.
 384             * delete the migration blocker. Ideally, this
 385             * should be hooked to transport close notification
 386             */
 387            if (pdu->s->migration_blocker) {
 388                migrate_del_blocker(pdu->s->migration_blocker);
 389                error_free(pdu->s->migration_blocker);
 390                pdu->s->migration_blocker = NULL;
 391            }
 392        }
 393        return free_fid(pdu, fidp);
 394    }
 395    return 0;
 396}
 397
 398static V9fsFidState *clunk_fid(V9fsState *s, int32_t fid)
 399{
 400    V9fsFidState **fidpp, *fidp;
 401
 402    for (fidpp = &s->fid_list; *fidpp; fidpp = &(*fidpp)->next) {
 403        if ((*fidpp)->fid == fid) {
 404            break;
 405        }
 406    }
 407    if (*fidpp == NULL) {
 408        return NULL;
 409    }
 410    fidp = *fidpp;
 411    *fidpp = fidp->next;
 412    fidp->clunked = 1;
 413    return fidp;
 414}
 415
 416void coroutine_fn v9fs_reclaim_fd(V9fsPDU *pdu)
 417{
 418    int reclaim_count = 0;
 419    V9fsState *s = pdu->s;
 420    V9fsFidState *f, *reclaim_list = NULL;
 421
 422    for (f = s->fid_list; f; f = f->next) {
 423        /*
 424         * Unlink fids cannot be reclaimed. Check
 425         * for them and skip them. Also skip fids
 426         * currently being operated on.
 427         */
 428        if (f->ref || f->flags & FID_NON_RECLAIMABLE) {
 429            continue;
 430        }
 431        /*
 432         * if it is a recently referenced fid
 433         * we leave the fid untouched and clear the
 434         * reference bit. We come back to it later
 435         * in the next iteration. (a simple LRU without
 436         * moving list elements around)
 437         */
 438        if (f->flags & FID_REFERENCED) {
 439            f->flags &= ~FID_REFERENCED;
 440            continue;
 441        }
 442        /*
 443         * Add fids to reclaim list.
 444         */
 445        if (f->fid_type == P9_FID_FILE) {
 446            if (f->fs.fd != -1) {
 447                /*
 448                 * Up the reference count so that
 449                 * a clunk request won't free this fid
 450                 */
 451                f->ref++;
 452                f->rclm_lst = reclaim_list;
 453                reclaim_list = f;
 454                f->fs_reclaim.fd = f->fs.fd;
 455                f->fs.fd = -1;
 456                reclaim_count++;
 457            }
 458        } else if (f->fid_type == P9_FID_DIR) {
 459            if (f->fs.dir.stream != NULL) {
 460                /*
 461                 * Up the reference count so that
 462                 * a clunk request won't free this fid
 463                 */
 464                f->ref++;
 465                f->rclm_lst = reclaim_list;
 466                reclaim_list = f;
 467                f->fs_reclaim.dir.stream = f->fs.dir.stream;
 468                f->fs.dir.stream = NULL;
 469                reclaim_count++;
 470            }
 471        }
 472        if (reclaim_count >= open_fd_rc) {
 473            break;
 474        }
 475    }
 476    /*
 477     * Now close the fid in reclaim list. Free them if they
 478     * are already clunked.
 479     */
 480    while (reclaim_list) {
 481        f = reclaim_list;
 482        reclaim_list = f->rclm_lst;
 483        if (f->fid_type == P9_FID_FILE) {
 484            v9fs_co_close(pdu, &f->fs_reclaim);
 485        } else if (f->fid_type == P9_FID_DIR) {
 486            v9fs_co_closedir(pdu, &f->fs_reclaim);
 487        }
 488        f->rclm_lst = NULL;
 489        /*
 490         * Now drop the fid reference, free it
 491         * if clunked.
 492         */
 493        put_fid(pdu, f);
 494    }
 495}
 496
 497static int coroutine_fn v9fs_mark_fids_unreclaim(V9fsPDU *pdu, V9fsPath *path)
 498{
 499    int err;
 500    V9fsState *s = pdu->s;
 501    V9fsFidState *fidp, head_fid;
 502
 503    head_fid.next = s->fid_list;
 504    for (fidp = s->fid_list; fidp; fidp = fidp->next) {
 505        if (fidp->path.size != path->size) {
 506            continue;
 507        }
 508        if (!memcmp(fidp->path.data, path->data, path->size)) {
 509            /* Mark the fid non reclaimable. */
 510            fidp->flags |= FID_NON_RECLAIMABLE;
 511
 512            /* reopen the file/dir if already closed */
 513            err = v9fs_reopen_fid(pdu, fidp);
 514            if (err < 0) {
 515                return err;
 516            }
 517            /*
 518             * Go back to head of fid list because
 519             * the list could have got updated when
 520             * switched to the worker thread
 521             */
 522            if (err == 0) {
 523                fidp = &head_fid;
 524            }
 525        }
 526    }
 527    return 0;
 528}
 529
 530static void coroutine_fn virtfs_reset(V9fsPDU *pdu)
 531{
 532    V9fsState *s = pdu->s;
 533    V9fsFidState *fidp;
 534
 535    /* Free all fids */
 536    while (s->fid_list) {
 537        /* Get fid */
 538        fidp = s->fid_list;
 539        fidp->ref++;
 540
 541        /* Clunk fid */
 542        s->fid_list = fidp->next;
 543        fidp->clunked = 1;
 544
 545        put_fid(pdu, fidp);
 546    }
 547}
 548
 549#define P9_QID_TYPE_DIR         0x80
 550#define P9_QID_TYPE_SYMLINK     0x02
 551
 552#define P9_STAT_MODE_DIR        0x80000000
 553#define P9_STAT_MODE_APPEND     0x40000000
 554#define P9_STAT_MODE_EXCL       0x20000000
 555#define P9_STAT_MODE_MOUNT      0x10000000
 556#define P9_STAT_MODE_AUTH       0x08000000
 557#define P9_STAT_MODE_TMP        0x04000000
 558#define P9_STAT_MODE_SYMLINK    0x02000000
 559#define P9_STAT_MODE_LINK       0x01000000
 560#define P9_STAT_MODE_DEVICE     0x00800000
 561#define P9_STAT_MODE_NAMED_PIPE 0x00200000
 562#define P9_STAT_MODE_SOCKET     0x00100000
 563#define P9_STAT_MODE_SETUID     0x00080000
 564#define P9_STAT_MODE_SETGID     0x00040000
 565#define P9_STAT_MODE_SETVTX     0x00010000
 566
 567#define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR |          \
 568                                P9_STAT_MODE_SYMLINK |      \
 569                                P9_STAT_MODE_LINK |         \
 570                                P9_STAT_MODE_DEVICE |       \
 571                                P9_STAT_MODE_NAMED_PIPE |   \
 572                                P9_STAT_MODE_SOCKET)
 573
 574/* This is the algorithm from ufs in spfs */
 575static void stat_to_qid(const struct stat *stbuf, V9fsQID *qidp)
 576{
 577    size_t size;
 578
 579    memset(&qidp->path, 0, sizeof(qidp->path));
 580    size = MIN(sizeof(stbuf->st_ino), sizeof(qidp->path));
 581    memcpy(&qidp->path, &stbuf->st_ino, size);
 582    qidp->version = stbuf->st_mtime ^ (stbuf->st_size << 8);
 583    qidp->type = 0;
 584    if (S_ISDIR(stbuf->st_mode)) {
 585        qidp->type |= P9_QID_TYPE_DIR;
 586    }
 587    if (S_ISLNK(stbuf->st_mode)) {
 588        qidp->type |= P9_QID_TYPE_SYMLINK;
 589    }
 590}
 591
 592static int coroutine_fn fid_to_qid(V9fsPDU *pdu, V9fsFidState *fidp,
 593                                   V9fsQID *qidp)
 594{
 595    struct stat stbuf;
 596    int err;
 597
 598    err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
 599    if (err < 0) {
 600        return err;
 601    }
 602    stat_to_qid(&stbuf, qidp);
 603    return 0;
 604}
 605
 606V9fsPDU *pdu_alloc(V9fsState *s)
 607{
 608    V9fsPDU *pdu = NULL;
 609
 610    if (!QLIST_EMPTY(&s->free_list)) {
 611        pdu = QLIST_FIRST(&s->free_list);
 612        QLIST_REMOVE(pdu, next);
 613        QLIST_INSERT_HEAD(&s->active_list, pdu, next);
 614    }
 615    return pdu;
 616}
 617
 618void pdu_free(V9fsPDU *pdu)
 619{
 620    V9fsState *s = pdu->s;
 621
 622    g_assert(!pdu->cancelled);
 623    QLIST_REMOVE(pdu, next);
 624    QLIST_INSERT_HEAD(&s->free_list, pdu, next);
 625}
 626
 627static void coroutine_fn pdu_complete(V9fsPDU *pdu, ssize_t len)
 628{
 629    int8_t id = pdu->id + 1; /* Response */
 630    V9fsState *s = pdu->s;
 631    int ret;
 632
 633    /*
 634     * The 9p spec requires that successfully cancelled pdus receive no reply.
 635     * Sending a reply would confuse clients because they would
 636     * assume that any EINTR is the actual result of the operation,
 637     * rather than a consequence of the cancellation. However, if
 638     * the operation completed (succesfully or with an error other
 639     * than caused be cancellation), we do send out that reply, both
 640     * for efficiency and to avoid confusing the rest of the state machine
 641     * that assumes passing a non-error here will mean a successful
 642     * transmission of the reply.
 643     */
 644    bool discard = pdu->cancelled && len == -EINTR;
 645    if (discard) {
 646        trace_v9fs_rcancel(pdu->tag, pdu->id);
 647        pdu->size = 0;
 648        goto out_notify;
 649    }
 650
 651    if (len < 0) {
 652        int err = -len;
 653        len = 7;
 654
 655        if (s->proto_version != V9FS_PROTO_2000L) {
 656            V9fsString str;
 657
 658            str.data = strerror(err);
 659            str.size = strlen(str.data);
 660
 661            ret = pdu_marshal(pdu, len, "s", &str);
 662            if (ret < 0) {
 663                goto out_notify;
 664            }
 665            len += ret;
 666            id = P9_RERROR;
 667        }
 668
 669        ret = pdu_marshal(pdu, len, "d", err);
 670        if (ret < 0) {
 671            goto out_notify;
 672        }
 673        len += ret;
 674
 675        if (s->proto_version == V9FS_PROTO_2000L) {
 676            id = P9_RLERROR;
 677        }
 678        trace_v9fs_rerror(pdu->tag, pdu->id, err); /* Trace ERROR */
 679    }
 680
 681    /* fill out the header */
 682    if (pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag) < 0) {
 683        goto out_notify;
 684    }
 685
 686    /* keep these in sync */
 687    pdu->size = len;
 688    pdu->id = id;
 689
 690out_notify:
 691    pdu->s->transport->push_and_notify(pdu);
 692
 693    /* Now wakeup anybody waiting in flush for this request */
 694    if (!qemu_co_queue_next(&pdu->complete)) {
 695        pdu_free(pdu);
 696    }
 697}
 698
 699static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
 700{
 701    mode_t ret;
 702
 703    ret = mode & 0777;
 704    if (mode & P9_STAT_MODE_DIR) {
 705        ret |= S_IFDIR;
 706    }
 707
 708    if (mode & P9_STAT_MODE_SYMLINK) {
 709        ret |= S_IFLNK;
 710    }
 711    if (mode & P9_STAT_MODE_SOCKET) {
 712        ret |= S_IFSOCK;
 713    }
 714    if (mode & P9_STAT_MODE_NAMED_PIPE) {
 715        ret |= S_IFIFO;
 716    }
 717    if (mode & P9_STAT_MODE_DEVICE) {
 718        if (extension->size && extension->data[0] == 'c') {
 719            ret |= S_IFCHR;
 720        } else {
 721            ret |= S_IFBLK;
 722        }
 723    }
 724
 725    if (!(ret&~0777)) {
 726        ret |= S_IFREG;
 727    }
 728
 729    if (mode & P9_STAT_MODE_SETUID) {
 730        ret |= S_ISUID;
 731    }
 732    if (mode & P9_STAT_MODE_SETGID) {
 733        ret |= S_ISGID;
 734    }
 735    if (mode & P9_STAT_MODE_SETVTX) {
 736        ret |= S_ISVTX;
 737    }
 738
 739    return ret;
 740}
 741
 742static int donttouch_stat(V9fsStat *stat)
 743{
 744    if (stat->type == -1 &&
 745        stat->dev == -1 &&
 746        stat->qid.type == -1 &&
 747        stat->qid.version == -1 &&
 748        stat->qid.path == -1 &&
 749        stat->mode == -1 &&
 750        stat->atime == -1 &&
 751        stat->mtime == -1 &&
 752        stat->length == -1 &&
 753        !stat->name.size &&
 754        !stat->uid.size &&
 755        !stat->gid.size &&
 756        !stat->muid.size &&
 757        stat->n_uid == -1 &&
 758        stat->n_gid == -1 &&
 759        stat->n_muid == -1) {
 760        return 1;
 761    }
 762
 763    return 0;
 764}
 765
 766static void v9fs_stat_init(V9fsStat *stat)
 767{
 768    v9fs_string_init(&stat->name);
 769    v9fs_string_init(&stat->uid);
 770    v9fs_string_init(&stat->gid);
 771    v9fs_string_init(&stat->muid);
 772    v9fs_string_init(&stat->extension);
 773}
 774
 775static void v9fs_stat_free(V9fsStat *stat)
 776{
 777    v9fs_string_free(&stat->name);
 778    v9fs_string_free(&stat->uid);
 779    v9fs_string_free(&stat->gid);
 780    v9fs_string_free(&stat->muid);
 781    v9fs_string_free(&stat->extension);
 782}
 783
 784static uint32_t stat_to_v9mode(const struct stat *stbuf)
 785{
 786    uint32_t mode;
 787
 788    mode = stbuf->st_mode & 0777;
 789    if (S_ISDIR(stbuf->st_mode)) {
 790        mode |= P9_STAT_MODE_DIR;
 791    }
 792
 793    if (S_ISLNK(stbuf->st_mode)) {
 794        mode |= P9_STAT_MODE_SYMLINK;
 795    }
 796
 797    if (S_ISSOCK(stbuf->st_mode)) {
 798        mode |= P9_STAT_MODE_SOCKET;
 799    }
 800
 801    if (S_ISFIFO(stbuf->st_mode)) {
 802        mode |= P9_STAT_MODE_NAMED_PIPE;
 803    }
 804
 805    if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
 806        mode |= P9_STAT_MODE_DEVICE;
 807    }
 808
 809    if (stbuf->st_mode & S_ISUID) {
 810        mode |= P9_STAT_MODE_SETUID;
 811    }
 812
 813    if (stbuf->st_mode & S_ISGID) {
 814        mode |= P9_STAT_MODE_SETGID;
 815    }
 816
 817    if (stbuf->st_mode & S_ISVTX) {
 818        mode |= P9_STAT_MODE_SETVTX;
 819    }
 820
 821    return mode;
 822}
 823
 824static int coroutine_fn stat_to_v9stat(V9fsPDU *pdu, V9fsPath *path,
 825                                       const char *basename,
 826                                       const struct stat *stbuf,
 827                                       V9fsStat *v9stat)
 828{
 829    int err;
 830
 831    memset(v9stat, 0, sizeof(*v9stat));
 832
 833    stat_to_qid(stbuf, &v9stat->qid);
 834    v9stat->mode = stat_to_v9mode(stbuf);
 835    v9stat->atime = stbuf->st_atime;
 836    v9stat->mtime = stbuf->st_mtime;
 837    v9stat->length = stbuf->st_size;
 838
 839    v9fs_string_free(&v9stat->uid);
 840    v9fs_string_free(&v9stat->gid);
 841    v9fs_string_free(&v9stat->muid);
 842
 843    v9stat->n_uid = stbuf->st_uid;
 844    v9stat->n_gid = stbuf->st_gid;
 845    v9stat->n_muid = 0;
 846
 847    v9fs_string_free(&v9stat->extension);
 848
 849    if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
 850        err = v9fs_co_readlink(pdu, path, &v9stat->extension);
 851        if (err < 0) {
 852            return err;
 853        }
 854    } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
 855        v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
 856                S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
 857                major(stbuf->st_rdev), minor(stbuf->st_rdev));
 858    } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
 859        v9fs_string_sprintf(&v9stat->extension, "%s %lu",
 860                "HARDLINKCOUNT", (unsigned long)stbuf->st_nlink);
 861    }
 862
 863    v9fs_string_sprintf(&v9stat->name, "%s", basename);
 864
 865    v9stat->size = 61 +
 866        v9fs_string_size(&v9stat->name) +
 867        v9fs_string_size(&v9stat->uid) +
 868        v9fs_string_size(&v9stat->gid) +
 869        v9fs_string_size(&v9stat->muid) +
 870        v9fs_string_size(&v9stat->extension);
 871    return 0;
 872}
 873
 874#define P9_STATS_MODE          0x00000001ULL
 875#define P9_STATS_NLINK         0x00000002ULL
 876#define P9_STATS_UID           0x00000004ULL
 877#define P9_STATS_GID           0x00000008ULL
 878#define P9_STATS_RDEV          0x00000010ULL
 879#define P9_STATS_ATIME         0x00000020ULL
 880#define P9_STATS_MTIME         0x00000040ULL
 881#define P9_STATS_CTIME         0x00000080ULL
 882#define P9_STATS_INO           0x00000100ULL
 883#define P9_STATS_SIZE          0x00000200ULL
 884#define P9_STATS_BLOCKS        0x00000400ULL
 885
 886#define P9_STATS_BTIME         0x00000800ULL
 887#define P9_STATS_GEN           0x00001000ULL
 888#define P9_STATS_DATA_VERSION  0x00002000ULL
 889
 890#define P9_STATS_BASIC         0x000007ffULL /* Mask for fields up to BLOCKS */
 891#define P9_STATS_ALL           0x00003fffULL /* Mask for All fields above */
 892
 893
 894static void stat_to_v9stat_dotl(V9fsState *s, const struct stat *stbuf,
 895                                V9fsStatDotl *v9lstat)
 896{
 897    memset(v9lstat, 0, sizeof(*v9lstat));
 898
 899    v9lstat->st_mode = stbuf->st_mode;
 900    v9lstat->st_nlink = stbuf->st_nlink;
 901    v9lstat->st_uid = stbuf->st_uid;
 902    v9lstat->st_gid = stbuf->st_gid;
 903    v9lstat->st_rdev = stbuf->st_rdev;
 904    v9lstat->st_size = stbuf->st_size;
 905    v9lstat->st_blksize = stbuf->st_blksize;
 906    v9lstat->st_blocks = stbuf->st_blocks;
 907    v9lstat->st_atime_sec = stbuf->st_atime;
 908    v9lstat->st_atime_nsec = stbuf->st_atim.tv_nsec;
 909    v9lstat->st_mtime_sec = stbuf->st_mtime;
 910    v9lstat->st_mtime_nsec = stbuf->st_mtim.tv_nsec;
 911    v9lstat->st_ctime_sec = stbuf->st_ctime;
 912    v9lstat->st_ctime_nsec = stbuf->st_ctim.tv_nsec;
 913    /* Currently we only support BASIC fields in stat */
 914    v9lstat->st_result_mask = P9_STATS_BASIC;
 915
 916    stat_to_qid(stbuf, &v9lstat->qid);
 917}
 918
 919static void print_sg(struct iovec *sg, int cnt)
 920{
 921    int i;
 922
 923    printf("sg[%d]: {", cnt);
 924    for (i = 0; i < cnt; i++) {
 925        if (i) {
 926            printf(", ");
 927        }
 928        printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
 929    }
 930    printf("}\n");
 931}
 932
 933/* Will call this only for path name based fid */
 934static void v9fs_fix_path(V9fsPath *dst, V9fsPath *src, int len)
 935{
 936    V9fsPath str;
 937    v9fs_path_init(&str);
 938    v9fs_path_copy(&str, dst);
 939    v9fs_path_sprintf(dst, "%s%s", src->data, str.data + len);
 940    v9fs_path_free(&str);
 941}
 942
 943static inline bool is_ro_export(FsContext *ctx)
 944{
 945    return ctx->export_flags & V9FS_RDONLY;
 946}
 947
 948static void coroutine_fn v9fs_version(void *opaque)
 949{
 950    ssize_t err;
 951    V9fsPDU *pdu = opaque;
 952    V9fsState *s = pdu->s;
 953    V9fsString version;
 954    size_t offset = 7;
 955
 956    v9fs_string_init(&version);
 957    err = pdu_unmarshal(pdu, offset, "ds", &s->msize, &version);
 958    if (err < 0) {
 959        goto out;
 960    }
 961    trace_v9fs_version(pdu->tag, pdu->id, s->msize, version.data);
 962
 963    virtfs_reset(pdu);
 964
 965    if (!strcmp(version.data, "9P2000.u")) {
 966        s->proto_version = V9FS_PROTO_2000U;
 967    } else if (!strcmp(version.data, "9P2000.L")) {
 968        s->proto_version = V9FS_PROTO_2000L;
 969    } else {
 970        v9fs_string_sprintf(&version, "unknown");
 971    }
 972
 973    err = pdu_marshal(pdu, offset, "ds", s->msize, &version);
 974    if (err < 0) {
 975        goto out;
 976    }
 977    err += offset;
 978    trace_v9fs_version_return(pdu->tag, pdu->id, s->msize, version.data);
 979out:
 980    pdu_complete(pdu, err);
 981    v9fs_string_free(&version);
 982}
 983
 984static void coroutine_fn v9fs_attach(void *opaque)
 985{
 986    V9fsPDU *pdu = opaque;
 987    V9fsState *s = pdu->s;
 988    int32_t fid, afid, n_uname;
 989    V9fsString uname, aname;
 990    V9fsFidState *fidp;
 991    size_t offset = 7;
 992    V9fsQID qid;
 993    ssize_t err;
 994    Error *local_err = NULL;
 995
 996    v9fs_string_init(&uname);
 997    v9fs_string_init(&aname);
 998    err = pdu_unmarshal(pdu, offset, "ddssd", &fid,
 999                        &afid, &uname, &aname, &n_uname);
1000    if (err < 0) {
1001        goto out_nofid;
1002    }
1003    trace_v9fs_attach(pdu->tag, pdu->id, fid, afid, uname.data, aname.data);
1004
1005    fidp = alloc_fid(s, fid);
1006    if (fidp == NULL) {
1007        err = -EINVAL;
1008        goto out_nofid;
1009    }
1010    fidp->uid = n_uname;
1011    err = v9fs_co_name_to_path(pdu, NULL, "/", &fidp->path);
1012    if (err < 0) {
1013        err = -EINVAL;
1014        clunk_fid(s, fid);
1015        goto out;
1016    }
1017    err = fid_to_qid(pdu, fidp, &qid);
1018    if (err < 0) {
1019        err = -EINVAL;
1020        clunk_fid(s, fid);
1021        goto out;
1022    }
1023
1024    /*
1025     * disable migration if we haven't done already.
1026     * attach could get called multiple times for the same export.
1027     */
1028    if (!s->migration_blocker) {
1029        error_setg(&s->migration_blocker,
1030                   "Migration is disabled when VirtFS export path '%s' is mounted in the guest using mount_tag '%s'",
1031                   s->ctx.fs_root ? s->ctx.fs_root : "NULL", s->tag);
1032        err = migrate_add_blocker(s->migration_blocker, &local_err);
1033        if (local_err) {
1034            error_free(local_err);
1035            error_free(s->migration_blocker);
1036            s->migration_blocker = NULL;
1037            clunk_fid(s, fid);
1038            goto out;
1039        }
1040        s->root_fid = fid;
1041    }
1042
1043    err = pdu_marshal(pdu, offset, "Q", &qid);
1044    if (err < 0) {
1045        clunk_fid(s, fid);
1046        goto out;
1047    }
1048    err += offset;
1049
1050    memcpy(&s->root_qid, &qid, sizeof(qid));
1051    trace_v9fs_attach_return(pdu->tag, pdu->id,
1052                             qid.type, qid.version, qid.path);
1053out:
1054    put_fid(pdu, fidp);
1055out_nofid:
1056    pdu_complete(pdu, err);
1057    v9fs_string_free(&uname);
1058    v9fs_string_free(&aname);
1059}
1060
1061static void coroutine_fn v9fs_stat(void *opaque)
1062{
1063    int32_t fid;
1064    V9fsStat v9stat;
1065    ssize_t err = 0;
1066    size_t offset = 7;
1067    struct stat stbuf;
1068    V9fsFidState *fidp;
1069    V9fsPDU *pdu = opaque;
1070    char *basename;
1071
1072    err = pdu_unmarshal(pdu, offset, "d", &fid);
1073    if (err < 0) {
1074        goto out_nofid;
1075    }
1076    trace_v9fs_stat(pdu->tag, pdu->id, fid);
1077
1078    fidp = get_fid(pdu, fid);
1079    if (fidp == NULL) {
1080        err = -ENOENT;
1081        goto out_nofid;
1082    }
1083    err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1084    if (err < 0) {
1085        goto out;
1086    }
1087    basename = g_path_get_basename(fidp->path.data);
1088    err = stat_to_v9stat(pdu, &fidp->path, basename, &stbuf, &v9stat);
1089    g_free(basename);
1090    if (err < 0) {
1091        goto out;
1092    }
1093    err = pdu_marshal(pdu, offset, "wS", 0, &v9stat);
1094    if (err < 0) {
1095        v9fs_stat_free(&v9stat);
1096        goto out;
1097    }
1098    trace_v9fs_stat_return(pdu->tag, pdu->id, v9stat.mode,
1099                           v9stat.atime, v9stat.mtime, v9stat.length);
1100    err += offset;
1101    v9fs_stat_free(&v9stat);
1102out:
1103    put_fid(pdu, fidp);
1104out_nofid:
1105    pdu_complete(pdu, err);
1106}
1107
1108static void coroutine_fn v9fs_getattr(void *opaque)
1109{
1110    int32_t fid;
1111    size_t offset = 7;
1112    ssize_t retval = 0;
1113    struct stat stbuf;
1114    V9fsFidState *fidp;
1115    uint64_t request_mask;
1116    V9fsStatDotl v9stat_dotl;
1117    V9fsPDU *pdu = opaque;
1118    V9fsState *s = pdu->s;
1119
1120    retval = pdu_unmarshal(pdu, offset, "dq", &fid, &request_mask);
1121    if (retval < 0) {
1122        goto out_nofid;
1123    }
1124    trace_v9fs_getattr(pdu->tag, pdu->id, fid, request_mask);
1125
1126    fidp = get_fid(pdu, fid);
1127    if (fidp == NULL) {
1128        retval = -ENOENT;
1129        goto out_nofid;
1130    }
1131    /*
1132     * Currently we only support BASIC fields in stat, so there is no
1133     * need to look at request_mask.
1134     */
1135    retval = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1136    if (retval < 0) {
1137        goto out;
1138    }
1139    stat_to_v9stat_dotl(s, &stbuf, &v9stat_dotl);
1140
1141    /*  fill st_gen if requested and supported by underlying fs */
1142    if (request_mask & P9_STATS_GEN) {
1143        retval = v9fs_co_st_gen(pdu, &fidp->path, stbuf.st_mode, &v9stat_dotl);
1144        switch (retval) {
1145        case 0:
1146            /* we have valid st_gen: update result mask */
1147            v9stat_dotl.st_result_mask |= P9_STATS_GEN;
1148            break;
1149        case -EINTR:
1150            /* request cancelled, e.g. by Tflush */
1151            goto out;
1152        default:
1153            /* failed to get st_gen: not fatal, ignore */
1154            break;
1155        }
1156    }
1157    retval = pdu_marshal(pdu, offset, "A", &v9stat_dotl);
1158    if (retval < 0) {
1159        goto out;
1160    }
1161    retval += offset;
1162    trace_v9fs_getattr_return(pdu->tag, pdu->id, v9stat_dotl.st_result_mask,
1163                              v9stat_dotl.st_mode, v9stat_dotl.st_uid,
1164                              v9stat_dotl.st_gid);
1165out:
1166    put_fid(pdu, fidp);
1167out_nofid:
1168    pdu_complete(pdu, retval);
1169}
1170
1171/* Attribute flags */
1172#define P9_ATTR_MODE       (1 << 0)
1173#define P9_ATTR_UID        (1 << 1)
1174#define P9_ATTR_GID        (1 << 2)
1175#define P9_ATTR_SIZE       (1 << 3)
1176#define P9_ATTR_ATIME      (1 << 4)
1177#define P9_ATTR_MTIME      (1 << 5)
1178#define P9_ATTR_CTIME      (1 << 6)
1179#define P9_ATTR_ATIME_SET  (1 << 7)
1180#define P9_ATTR_MTIME_SET  (1 << 8)
1181
1182#define P9_ATTR_MASK    127
1183
1184static void coroutine_fn v9fs_setattr(void *opaque)
1185{
1186    int err = 0;
1187    int32_t fid;
1188    V9fsFidState *fidp;
1189    size_t offset = 7;
1190    V9fsIattr v9iattr;
1191    V9fsPDU *pdu = opaque;
1192
1193    err = pdu_unmarshal(pdu, offset, "dI", &fid, &v9iattr);
1194    if (err < 0) {
1195        goto out_nofid;
1196    }
1197
1198    trace_v9fs_setattr(pdu->tag, pdu->id, fid,
1199                       v9iattr.valid, v9iattr.mode, v9iattr.uid, v9iattr.gid,
1200                       v9iattr.size, v9iattr.atime_sec, v9iattr.mtime_sec);
1201
1202    fidp = get_fid(pdu, fid);
1203    if (fidp == NULL) {
1204        err = -EINVAL;
1205        goto out_nofid;
1206    }
1207    if (v9iattr.valid & P9_ATTR_MODE) {
1208        err = v9fs_co_chmod(pdu, &fidp->path, v9iattr.mode);
1209        if (err < 0) {
1210            goto out;
1211        }
1212    }
1213    if (v9iattr.valid & (P9_ATTR_ATIME | P9_ATTR_MTIME)) {
1214        struct timespec times[2];
1215        if (v9iattr.valid & P9_ATTR_ATIME) {
1216            if (v9iattr.valid & P9_ATTR_ATIME_SET) {
1217                times[0].tv_sec = v9iattr.atime_sec;
1218                times[0].tv_nsec = v9iattr.atime_nsec;
1219            } else {
1220                times[0].tv_nsec = UTIME_NOW;
1221            }
1222        } else {
1223            times[0].tv_nsec = UTIME_OMIT;
1224        }
1225        if (v9iattr.valid & P9_ATTR_MTIME) {
1226            if (v9iattr.valid & P9_ATTR_MTIME_SET) {
1227                times[1].tv_sec = v9iattr.mtime_sec;
1228                times[1].tv_nsec = v9iattr.mtime_nsec;
1229            } else {
1230                times[1].tv_nsec = UTIME_NOW;
1231            }
1232        } else {
1233            times[1].tv_nsec = UTIME_OMIT;
1234        }
1235        err = v9fs_co_utimensat(pdu, &fidp->path, times);
1236        if (err < 0) {
1237            goto out;
1238        }
1239    }
1240    /*
1241     * If the only valid entry in iattr is ctime we can call
1242     * chown(-1,-1) to update the ctime of the file
1243     */
1244    if ((v9iattr.valid & (P9_ATTR_UID | P9_ATTR_GID)) ||
1245        ((v9iattr.valid & P9_ATTR_CTIME)
1246         && !((v9iattr.valid & P9_ATTR_MASK) & ~P9_ATTR_CTIME))) {
1247        if (!(v9iattr.valid & P9_ATTR_UID)) {
1248            v9iattr.uid = -1;
1249        }
1250        if (!(v9iattr.valid & P9_ATTR_GID)) {
1251            v9iattr.gid = -1;
1252        }
1253        err = v9fs_co_chown(pdu, &fidp->path, v9iattr.uid,
1254                            v9iattr.gid);
1255        if (err < 0) {
1256            goto out;
1257        }
1258    }
1259    if (v9iattr.valid & (P9_ATTR_SIZE)) {
1260        err = v9fs_co_truncate(pdu, &fidp->path, v9iattr.size);
1261        if (err < 0) {
1262            goto out;
1263        }
1264    }
1265    err = offset;
1266    trace_v9fs_setattr_return(pdu->tag, pdu->id);
1267out:
1268    put_fid(pdu, fidp);
1269out_nofid:
1270    pdu_complete(pdu, err);
1271}
1272
1273static int v9fs_walk_marshal(V9fsPDU *pdu, uint16_t nwnames, V9fsQID *qids)
1274{
1275    int i;
1276    ssize_t err;
1277    size_t offset = 7;
1278
1279    err = pdu_marshal(pdu, offset, "w", nwnames);
1280    if (err < 0) {
1281        return err;
1282    }
1283    offset += err;
1284    for (i = 0; i < nwnames; i++) {
1285        err = pdu_marshal(pdu, offset, "Q", &qids[i]);
1286        if (err < 0) {
1287            return err;
1288        }
1289        offset += err;
1290    }
1291    return offset;
1292}
1293
1294static bool name_is_illegal(const char *name)
1295{
1296    return !*name || strchr(name, '/') != NULL;
1297}
1298
1299static bool not_same_qid(const V9fsQID *qid1, const V9fsQID *qid2)
1300{
1301    return
1302        qid1->type != qid2->type ||
1303        qid1->version != qid2->version ||
1304        qid1->path != qid2->path;
1305}
1306
1307static void coroutine_fn v9fs_walk(void *opaque)
1308{
1309    int name_idx;
1310    V9fsQID *qids = NULL;
1311    int i, err = 0;
1312    V9fsPath dpath, path;
1313    uint16_t nwnames;
1314    struct stat stbuf;
1315    size_t offset = 7;
1316    int32_t fid, newfid;
1317    V9fsString *wnames = NULL;
1318    V9fsFidState *fidp;
1319    V9fsFidState *newfidp = NULL;
1320    V9fsPDU *pdu = opaque;
1321    V9fsState *s = pdu->s;
1322    V9fsQID qid;
1323
1324    err = pdu_unmarshal(pdu, offset, "ddw", &fid, &newfid, &nwnames);
1325    if (err < 0) {
1326        pdu_complete(pdu, err);
1327        return ;
1328    }
1329    offset += err;
1330
1331    trace_v9fs_walk(pdu->tag, pdu->id, fid, newfid, nwnames);
1332
1333    if (nwnames && nwnames <= P9_MAXWELEM) {
1334        wnames = g_new0(V9fsString, nwnames);
1335        qids   = g_new0(V9fsQID, nwnames);
1336        for (i = 0; i < nwnames; i++) {
1337            err = pdu_unmarshal(pdu, offset, "s", &wnames[i]);
1338            if (err < 0) {
1339                goto out_nofid;
1340            }
1341            if (name_is_illegal(wnames[i].data)) {
1342                err = -ENOENT;
1343                goto out_nofid;
1344            }
1345            offset += err;
1346        }
1347    } else if (nwnames > P9_MAXWELEM) {
1348        err = -EINVAL;
1349        goto out_nofid;
1350    }
1351    fidp = get_fid(pdu, fid);
1352    if (fidp == NULL) {
1353        err = -ENOENT;
1354        goto out_nofid;
1355    }
1356
1357    v9fs_path_init(&dpath);
1358    v9fs_path_init(&path);
1359
1360    err = fid_to_qid(pdu, fidp, &qid);
1361    if (err < 0) {
1362        goto out;
1363    }
1364
1365    /*
1366     * Both dpath and path initially poin to fidp.
1367     * Needed to handle request with nwnames == 0
1368     */
1369    v9fs_path_copy(&dpath, &fidp->path);
1370    v9fs_path_copy(&path, &fidp->path);
1371    for (name_idx = 0; name_idx < nwnames; name_idx++) {
1372        if (not_same_qid(&pdu->s->root_qid, &qid) ||
1373            strcmp("..", wnames[name_idx].data)) {
1374            err = v9fs_co_name_to_path(pdu, &dpath, wnames[name_idx].data,
1375                                       &path);
1376            if (err < 0) {
1377                goto out;
1378            }
1379
1380            err = v9fs_co_lstat(pdu, &path, &stbuf);
1381            if (err < 0) {
1382                goto out;
1383            }
1384            stat_to_qid(&stbuf, &qid);
1385            v9fs_path_copy(&dpath, &path);
1386        }
1387        memcpy(&qids[name_idx], &qid, sizeof(qid));
1388    }
1389    if (fid == newfid) {
1390        if (fidp->fid_type != P9_FID_NONE) {
1391            err = -EINVAL;
1392            goto out;
1393        }
1394        v9fs_path_write_lock(s);
1395        v9fs_path_copy(&fidp->path, &path);
1396        v9fs_path_unlock(s);
1397    } else {
1398        newfidp = alloc_fid(s, newfid);
1399        if (newfidp == NULL) {
1400            err = -EINVAL;
1401            goto out;
1402        }
1403        newfidp->uid = fidp->uid;
1404        v9fs_path_copy(&newfidp->path, &path);
1405    }
1406    err = v9fs_walk_marshal(pdu, nwnames, qids);
1407    trace_v9fs_walk_return(pdu->tag, pdu->id, nwnames, qids);
1408out:
1409    put_fid(pdu, fidp);
1410    if (newfidp) {
1411        put_fid(pdu, newfidp);
1412    }
1413    v9fs_path_free(&dpath);
1414    v9fs_path_free(&path);
1415out_nofid:
1416    pdu_complete(pdu, err);
1417    if (nwnames && nwnames <= P9_MAXWELEM) {
1418        for (name_idx = 0; name_idx < nwnames; name_idx++) {
1419            v9fs_string_free(&wnames[name_idx]);
1420        }
1421        g_free(wnames);
1422        g_free(qids);
1423    }
1424}
1425
1426static int32_t coroutine_fn get_iounit(V9fsPDU *pdu, V9fsPath *path)
1427{
1428    struct statfs stbuf;
1429    int32_t iounit = 0;
1430    V9fsState *s = pdu->s;
1431
1432    /*
1433     * iounit should be multiples of f_bsize (host filesystem block size
1434     * and as well as less than (client msize - P9_IOHDRSZ))
1435     */
1436    if (!v9fs_co_statfs(pdu, path, &stbuf)) {
1437        iounit = stbuf.f_bsize;
1438        iounit *= (s->msize - P9_IOHDRSZ)/stbuf.f_bsize;
1439    }
1440    if (!iounit) {
1441        iounit = s->msize - P9_IOHDRSZ;
1442    }
1443    return iounit;
1444}
1445
1446static void coroutine_fn v9fs_open(void *opaque)
1447{
1448    int flags;
1449    int32_t fid;
1450    int32_t mode;
1451    V9fsQID qid;
1452    int iounit = 0;
1453    ssize_t err = 0;
1454    size_t offset = 7;
1455    struct stat stbuf;
1456    V9fsFidState *fidp;
1457    V9fsPDU *pdu = opaque;
1458    V9fsState *s = pdu->s;
1459
1460    if (s->proto_version == V9FS_PROTO_2000L) {
1461        err = pdu_unmarshal(pdu, offset, "dd", &fid, &mode);
1462    } else {
1463        uint8_t modebyte;
1464        err = pdu_unmarshal(pdu, offset, "db", &fid, &modebyte);
1465        mode = modebyte;
1466    }
1467    if (err < 0) {
1468        goto out_nofid;
1469    }
1470    trace_v9fs_open(pdu->tag, pdu->id, fid, mode);
1471
1472    fidp = get_fid(pdu, fid);
1473    if (fidp == NULL) {
1474        err = -ENOENT;
1475        goto out_nofid;
1476    }
1477    if (fidp->fid_type != P9_FID_NONE) {
1478        err = -EINVAL;
1479        goto out;
1480    }
1481
1482    err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1483    if (err < 0) {
1484        goto out;
1485    }
1486    stat_to_qid(&stbuf, &qid);
1487    if (S_ISDIR(stbuf.st_mode)) {
1488        err = v9fs_co_opendir(pdu, fidp);
1489        if (err < 0) {
1490            goto out;
1491        }
1492        fidp->fid_type = P9_FID_DIR;
1493        err = pdu_marshal(pdu, offset, "Qd", &qid, 0);
1494        if (err < 0) {
1495            goto out;
1496        }
1497        err += offset;
1498    } else {
1499        if (s->proto_version == V9FS_PROTO_2000L) {
1500            flags = get_dotl_openflags(s, mode);
1501        } else {
1502            flags = omode_to_uflags(mode);
1503        }
1504        if (is_ro_export(&s->ctx)) {
1505            if (mode & O_WRONLY || mode & O_RDWR ||
1506                mode & O_APPEND || mode & O_TRUNC) {
1507                err = -EROFS;
1508                goto out;
1509            }
1510        }
1511        err = v9fs_co_open(pdu, fidp, flags);
1512        if (err < 0) {
1513            goto out;
1514        }
1515        fidp->fid_type = P9_FID_FILE;
1516        fidp->open_flags = flags;
1517        if (flags & O_EXCL) {
1518            /*
1519             * We let the host file system do O_EXCL check
1520             * We should not reclaim such fd
1521             */
1522            fidp->flags |= FID_NON_RECLAIMABLE;
1523        }
1524        iounit = get_iounit(pdu, &fidp->path);
1525        err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1526        if (err < 0) {
1527            goto out;
1528        }
1529        err += offset;
1530    }
1531    trace_v9fs_open_return(pdu->tag, pdu->id,
1532                           qid.type, qid.version, qid.path, iounit);
1533out:
1534    put_fid(pdu, fidp);
1535out_nofid:
1536    pdu_complete(pdu, err);
1537}
1538
1539static void coroutine_fn v9fs_lcreate(void *opaque)
1540{
1541    int32_t dfid, flags, mode;
1542    gid_t gid;
1543    ssize_t err = 0;
1544    ssize_t offset = 7;
1545    V9fsString name;
1546    V9fsFidState *fidp;
1547    struct stat stbuf;
1548    V9fsQID qid;
1549    int32_t iounit;
1550    V9fsPDU *pdu = opaque;
1551
1552    v9fs_string_init(&name);
1553    err = pdu_unmarshal(pdu, offset, "dsddd", &dfid,
1554                        &name, &flags, &mode, &gid);
1555    if (err < 0) {
1556        goto out_nofid;
1557    }
1558    trace_v9fs_lcreate(pdu->tag, pdu->id, dfid, flags, mode, gid);
1559
1560    if (name_is_illegal(name.data)) {
1561        err = -ENOENT;
1562        goto out_nofid;
1563    }
1564
1565    if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
1566        err = -EEXIST;
1567        goto out_nofid;
1568    }
1569
1570    fidp = get_fid(pdu, dfid);
1571    if (fidp == NULL) {
1572        err = -ENOENT;
1573        goto out_nofid;
1574    }
1575    if (fidp->fid_type != P9_FID_NONE) {
1576        err = -EINVAL;
1577        goto out;
1578    }
1579
1580    flags = get_dotl_openflags(pdu->s, flags);
1581    err = v9fs_co_open2(pdu, fidp, &name, gid,
1582                        flags | O_CREAT, mode, &stbuf);
1583    if (err < 0) {
1584        goto out;
1585    }
1586    fidp->fid_type = P9_FID_FILE;
1587    fidp->open_flags = flags;
1588    if (flags & O_EXCL) {
1589        /*
1590         * We let the host file system do O_EXCL check
1591         * We should not reclaim such fd
1592         */
1593        fidp->flags |= FID_NON_RECLAIMABLE;
1594    }
1595    iounit =  get_iounit(pdu, &fidp->path);
1596    stat_to_qid(&stbuf, &qid);
1597    err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1598    if (err < 0) {
1599        goto out;
1600    }
1601    err += offset;
1602    trace_v9fs_lcreate_return(pdu->tag, pdu->id,
1603                              qid.type, qid.version, qid.path, iounit);
1604out:
1605    put_fid(pdu, fidp);
1606out_nofid:
1607    pdu_complete(pdu, err);
1608    v9fs_string_free(&name);
1609}
1610
1611static void coroutine_fn v9fs_fsync(void *opaque)
1612{
1613    int err;
1614    int32_t fid;
1615    int datasync;
1616    size_t offset = 7;
1617    V9fsFidState *fidp;
1618    V9fsPDU *pdu = opaque;
1619
1620    err = pdu_unmarshal(pdu, offset, "dd", &fid, &datasync);
1621    if (err < 0) {
1622        goto out_nofid;
1623    }
1624    trace_v9fs_fsync(pdu->tag, pdu->id, fid, datasync);
1625
1626    fidp = get_fid(pdu, fid);
1627    if (fidp == NULL) {
1628        err = -ENOENT;
1629        goto out_nofid;
1630    }
1631    err = v9fs_co_fsync(pdu, fidp, datasync);
1632    if (!err) {
1633        err = offset;
1634    }
1635    put_fid(pdu, fidp);
1636out_nofid:
1637    pdu_complete(pdu, err);
1638}
1639
1640static void coroutine_fn v9fs_clunk(void *opaque)
1641{
1642    int err;
1643    int32_t fid;
1644    size_t offset = 7;
1645    V9fsFidState *fidp;
1646    V9fsPDU *pdu = opaque;
1647    V9fsState *s = pdu->s;
1648
1649    err = pdu_unmarshal(pdu, offset, "d", &fid);
1650    if (err < 0) {
1651        goto out_nofid;
1652    }
1653    trace_v9fs_clunk(pdu->tag, pdu->id, fid);
1654
1655    fidp = clunk_fid(s, fid);
1656    if (fidp == NULL) {
1657        err = -ENOENT;
1658        goto out_nofid;
1659    }
1660    /*
1661     * Bump the ref so that put_fid will
1662     * free the fid.
1663     */
1664    fidp->ref++;
1665    err = put_fid(pdu, fidp);
1666    if (!err) {
1667        err = offset;
1668    }
1669out_nofid:
1670    pdu_complete(pdu, err);
1671}
1672
1673/*
1674 * Create a QEMUIOVector for a sub-region of PDU iovecs
1675 *
1676 * @qiov:       uninitialized QEMUIOVector
1677 * @skip:       number of bytes to skip from beginning of PDU
1678 * @size:       number of bytes to include
1679 * @is_write:   true - write, false - read
1680 *
1681 * The resulting QEMUIOVector has heap-allocated iovecs and must be cleaned up
1682 * with qemu_iovec_destroy().
1683 */
1684static void v9fs_init_qiov_from_pdu(QEMUIOVector *qiov, V9fsPDU *pdu,
1685                                    size_t skip, size_t size,
1686                                    bool is_write)
1687{
1688    QEMUIOVector elem;
1689    struct iovec *iov;
1690    unsigned int niov;
1691
1692    if (is_write) {
1693        pdu->s->transport->init_out_iov_from_pdu(pdu, &iov, &niov, size + skip);
1694    } else {
1695        pdu->s->transport->init_in_iov_from_pdu(pdu, &iov, &niov, size + skip);
1696    }
1697
1698    qemu_iovec_init_external(&elem, iov, niov);
1699    qemu_iovec_init(qiov, niov);
1700    qemu_iovec_concat(qiov, &elem, skip, size);
1701}
1702
1703static int v9fs_xattr_read(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
1704                           uint64_t off, uint32_t max_count)
1705{
1706    ssize_t err;
1707    size_t offset = 7;
1708    uint64_t read_count;
1709    QEMUIOVector qiov_full;
1710
1711    if (fidp->fs.xattr.len < off) {
1712        read_count = 0;
1713    } else {
1714        read_count = fidp->fs.xattr.len - off;
1715    }
1716    if (read_count > max_count) {
1717        read_count = max_count;
1718    }
1719    err = pdu_marshal(pdu, offset, "d", read_count);
1720    if (err < 0) {
1721        return err;
1722    }
1723    offset += err;
1724
1725    v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, read_count, false);
1726    err = v9fs_pack(qiov_full.iov, qiov_full.niov, 0,
1727                    ((char *)fidp->fs.xattr.value) + off,
1728                    read_count);
1729    qemu_iovec_destroy(&qiov_full);
1730    if (err < 0) {
1731        return err;
1732    }
1733    offset += err;
1734    return offset;
1735}
1736
1737static int coroutine_fn v9fs_do_readdir_with_stat(V9fsPDU *pdu,
1738                                                  V9fsFidState *fidp,
1739                                                  uint32_t max_count)
1740{
1741    V9fsPath path;
1742    V9fsStat v9stat;
1743    int len, err = 0;
1744    int32_t count = 0;
1745    struct stat stbuf;
1746    off_t saved_dir_pos;
1747    struct dirent *dent;
1748
1749    /* save the directory position */
1750    saved_dir_pos = v9fs_co_telldir(pdu, fidp);
1751    if (saved_dir_pos < 0) {
1752        return saved_dir_pos;
1753    }
1754
1755    while (1) {
1756        v9fs_path_init(&path);
1757
1758        v9fs_readdir_lock(&fidp->fs.dir);
1759
1760        err = v9fs_co_readdir(pdu, fidp, &dent);
1761        if (err || !dent) {
1762            break;
1763        }
1764        err = v9fs_co_name_to_path(pdu, &fidp->path, dent->d_name, &path);
1765        if (err < 0) {
1766            break;
1767        }
1768        err = v9fs_co_lstat(pdu, &path, &stbuf);
1769        if (err < 0) {
1770            break;
1771        }
1772        err = stat_to_v9stat(pdu, &path, dent->d_name, &stbuf, &v9stat);
1773        if (err < 0) {
1774            break;
1775        }
1776        if ((count + v9stat.size + 2) > max_count) {
1777            v9fs_readdir_unlock(&fidp->fs.dir);
1778
1779            /* Ran out of buffer. Set dir back to old position and return */
1780            v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
1781            v9fs_stat_free(&v9stat);
1782            v9fs_path_free(&path);
1783            return count;
1784        }
1785
1786        /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
1787        len = pdu_marshal(pdu, 11 + count, "S", &v9stat);
1788
1789        v9fs_readdir_unlock(&fidp->fs.dir);
1790
1791        if (len < 0) {
1792            v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
1793            v9fs_stat_free(&v9stat);
1794            v9fs_path_free(&path);
1795            return len;
1796        }
1797        count += len;
1798        v9fs_stat_free(&v9stat);
1799        v9fs_path_free(&path);
1800        saved_dir_pos = dent->d_off;
1801    }
1802
1803    v9fs_readdir_unlock(&fidp->fs.dir);
1804
1805    v9fs_path_free(&path);
1806    if (err < 0) {
1807        return err;
1808    }
1809    return count;
1810}
1811
1812static void coroutine_fn v9fs_read(void *opaque)
1813{
1814    int32_t fid;
1815    uint64_t off;
1816    ssize_t err = 0;
1817    int32_t count = 0;
1818    size_t offset = 7;
1819    uint32_t max_count;
1820    V9fsFidState *fidp;
1821    V9fsPDU *pdu = opaque;
1822    V9fsState *s = pdu->s;
1823
1824    err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &max_count);
1825    if (err < 0) {
1826        goto out_nofid;
1827    }
1828    trace_v9fs_read(pdu->tag, pdu->id, fid, off, max_count);
1829
1830    fidp = get_fid(pdu, fid);
1831    if (fidp == NULL) {
1832        err = -EINVAL;
1833        goto out_nofid;
1834    }
1835    if (fidp->fid_type == P9_FID_DIR) {
1836
1837        if (off == 0) {
1838            v9fs_co_rewinddir(pdu, fidp);
1839        }
1840        count = v9fs_do_readdir_with_stat(pdu, fidp, max_count);
1841        if (count < 0) {
1842            err = count;
1843            goto out;
1844        }
1845        err = pdu_marshal(pdu, offset, "d", count);
1846        if (err < 0) {
1847            goto out;
1848        }
1849        err += offset + count;
1850    } else if (fidp->fid_type == P9_FID_FILE) {
1851        QEMUIOVector qiov_full;
1852        QEMUIOVector qiov;
1853        int32_t len;
1854
1855        v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset + 4, max_count, false);
1856        qemu_iovec_init(&qiov, qiov_full.niov);
1857        do {
1858            qemu_iovec_reset(&qiov);
1859            qemu_iovec_concat(&qiov, &qiov_full, count, qiov_full.size - count);
1860            if (0) {
1861                print_sg(qiov.iov, qiov.niov);
1862            }
1863            /* Loop in case of EINTR */
1864            do {
1865                len = v9fs_co_preadv(pdu, fidp, qiov.iov, qiov.niov, off);
1866                if (len >= 0) {
1867                    off   += len;
1868                    count += len;
1869                }
1870            } while (len == -EINTR && !pdu->cancelled);
1871            if (len < 0) {
1872                /* IO error return the error */
1873                err = len;
1874                goto out_free_iovec;
1875            }
1876        } while (count < max_count && len > 0);
1877        err = pdu_marshal(pdu, offset, "d", count);
1878        if (err < 0) {
1879            goto out_free_iovec;
1880        }
1881        err += offset + count;
1882out_free_iovec:
1883        qemu_iovec_destroy(&qiov);
1884        qemu_iovec_destroy(&qiov_full);
1885    } else if (fidp->fid_type == P9_FID_XATTR) {
1886        err = v9fs_xattr_read(s, pdu, fidp, off, max_count);
1887    } else {
1888        err = -EINVAL;
1889    }
1890    trace_v9fs_read_return(pdu->tag, pdu->id, count, err);
1891out:
1892    put_fid(pdu, fidp);
1893out_nofid:
1894    pdu_complete(pdu, err);
1895}
1896
1897static size_t v9fs_readdir_data_size(V9fsString *name)
1898{
1899    /*
1900     * Size of each dirent on the wire: size of qid (13) + size of offset (8)
1901     * size of type (1) + size of name.size (2) + strlen(name.data)
1902     */
1903    return 24 + v9fs_string_size(name);
1904}
1905
1906static int coroutine_fn v9fs_do_readdir(V9fsPDU *pdu, V9fsFidState *fidp,
1907                                        int32_t max_count)
1908{
1909    size_t size;
1910    V9fsQID qid;
1911    V9fsString name;
1912    int len, err = 0;
1913    int32_t count = 0;
1914    off_t saved_dir_pos;
1915    struct dirent *dent;
1916
1917    /* save the directory position */
1918    saved_dir_pos = v9fs_co_telldir(pdu, fidp);
1919    if (saved_dir_pos < 0) {
1920        return saved_dir_pos;
1921    }
1922
1923    while (1) {
1924        v9fs_readdir_lock(&fidp->fs.dir);
1925
1926        err = v9fs_co_readdir(pdu, fidp, &dent);
1927        if (err || !dent) {
1928            break;
1929        }
1930        v9fs_string_init(&name);
1931        v9fs_string_sprintf(&name, "%s", dent->d_name);
1932        if ((count + v9fs_readdir_data_size(&name)) > max_count) {
1933            v9fs_readdir_unlock(&fidp->fs.dir);
1934
1935            /* Ran out of buffer. Set dir back to old position and return */
1936            v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
1937            v9fs_string_free(&name);
1938            return count;
1939        }
1940        /*
1941         * Fill up just the path field of qid because the client uses
1942         * only that. To fill the entire qid structure we will have
1943         * to stat each dirent found, which is expensive
1944         */
1945        size = MIN(sizeof(dent->d_ino), sizeof(qid.path));
1946        memcpy(&qid.path, &dent->d_ino, size);
1947        /* Fill the other fields with dummy values */
1948        qid.type = 0;
1949        qid.version = 0;
1950
1951        /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
1952        len = pdu_marshal(pdu, 11 + count, "Qqbs",
1953                          &qid, dent->d_off,
1954                          dent->d_type, &name);
1955
1956        v9fs_readdir_unlock(&fidp->fs.dir);
1957
1958        if (len < 0) {
1959            v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
1960            v9fs_string_free(&name);
1961            return len;
1962        }
1963        count += len;
1964        v9fs_string_free(&name);
1965        saved_dir_pos = dent->d_off;
1966    }
1967
1968    v9fs_readdir_unlock(&fidp->fs.dir);
1969
1970    if (err < 0) {
1971        return err;
1972    }
1973    return count;
1974}
1975
1976static void coroutine_fn v9fs_readdir(void *opaque)
1977{
1978    int32_t fid;
1979    V9fsFidState *fidp;
1980    ssize_t retval = 0;
1981    size_t offset = 7;
1982    uint64_t initial_offset;
1983    int32_t count;
1984    uint32_t max_count;
1985    V9fsPDU *pdu = opaque;
1986
1987    retval = pdu_unmarshal(pdu, offset, "dqd", &fid,
1988                           &initial_offset, &max_count);
1989    if (retval < 0) {
1990        goto out_nofid;
1991    }
1992    trace_v9fs_readdir(pdu->tag, pdu->id, fid, initial_offset, max_count);
1993
1994    fidp = get_fid(pdu, fid);
1995    if (fidp == NULL) {
1996        retval = -EINVAL;
1997        goto out_nofid;
1998    }
1999    if (!fidp->fs.dir.stream) {
2000        retval = -EINVAL;
2001        goto out;
2002    }
2003    if (initial_offset == 0) {
2004        v9fs_co_rewinddir(pdu, fidp);
2005    } else {
2006        v9fs_co_seekdir(pdu, fidp, initial_offset);
2007    }
2008    count = v9fs_do_readdir(pdu, fidp, max_count);
2009    if (count < 0) {
2010        retval = count;
2011        goto out;
2012    }
2013    retval = pdu_marshal(pdu, offset, "d", count);
2014    if (retval < 0) {
2015        goto out;
2016    }
2017    retval += count + offset;
2018    trace_v9fs_readdir_return(pdu->tag, pdu->id, count, retval);
2019out:
2020    put_fid(pdu, fidp);
2021out_nofid:
2022    pdu_complete(pdu, retval);
2023}
2024
2025static int v9fs_xattr_write(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
2026                            uint64_t off, uint32_t count,
2027                            struct iovec *sg, int cnt)
2028{
2029    int i, to_copy;
2030    ssize_t err = 0;
2031    uint64_t write_count;
2032    size_t offset = 7;
2033
2034
2035    if (fidp->fs.xattr.len < off) {
2036        err = -ENOSPC;
2037        goto out;
2038    }
2039    write_count = fidp->fs.xattr.len - off;
2040    if (write_count > count) {
2041        write_count = count;
2042    }
2043    err = pdu_marshal(pdu, offset, "d", write_count);
2044    if (err < 0) {
2045        return err;
2046    }
2047    err += offset;
2048    fidp->fs.xattr.copied_len += write_count;
2049    /*
2050     * Now copy the content from sg list
2051     */
2052    for (i = 0; i < cnt; i++) {
2053        if (write_count > sg[i].iov_len) {
2054            to_copy = sg[i].iov_len;
2055        } else {
2056            to_copy = write_count;
2057        }
2058        memcpy((char *)fidp->fs.xattr.value + off, sg[i].iov_base, to_copy);
2059        /* updating vs->off since we are not using below */
2060        off += to_copy;
2061        write_count -= to_copy;
2062    }
2063out:
2064    return err;
2065}
2066
2067static void coroutine_fn v9fs_write(void *opaque)
2068{
2069    ssize_t err;
2070    int32_t fid;
2071    uint64_t off;
2072    uint32_t count;
2073    int32_t len = 0;
2074    int32_t total = 0;
2075    size_t offset = 7;
2076    V9fsFidState *fidp;
2077    V9fsPDU *pdu = opaque;
2078    V9fsState *s = pdu->s;
2079    QEMUIOVector qiov_full;
2080    QEMUIOVector qiov;
2081
2082    err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &count);
2083    if (err < 0) {
2084        pdu_complete(pdu, err);
2085        return;
2086    }
2087    offset += err;
2088    v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, count, true);
2089    trace_v9fs_write(pdu->tag, pdu->id, fid, off, count, qiov_full.niov);
2090
2091    fidp = get_fid(pdu, fid);
2092    if (fidp == NULL) {
2093        err = -EINVAL;
2094        goto out_nofid;
2095    }
2096    if (fidp->fid_type == P9_FID_FILE) {
2097        if (fidp->fs.fd == -1) {
2098            err = -EINVAL;
2099            goto out;
2100        }
2101    } else if (fidp->fid_type == P9_FID_XATTR) {
2102        /*
2103         * setxattr operation
2104         */
2105        err = v9fs_xattr_write(s, pdu, fidp, off, count,
2106                               qiov_full.iov, qiov_full.niov);
2107        goto out;
2108    } else {
2109        err = -EINVAL;
2110        goto out;
2111    }
2112    qemu_iovec_init(&qiov, qiov_full.niov);
2113    do {
2114        qemu_iovec_reset(&qiov);
2115        qemu_iovec_concat(&qiov, &qiov_full, total, qiov_full.size - total);
2116        if (0) {
2117            print_sg(qiov.iov, qiov.niov);
2118        }
2119        /* Loop in case of EINTR */
2120        do {
2121            len = v9fs_co_pwritev(pdu, fidp, qiov.iov, qiov.niov, off);
2122            if (len >= 0) {
2123                off   += len;
2124                total += len;
2125            }
2126        } while (len == -EINTR && !pdu->cancelled);
2127        if (len < 0) {
2128            /* IO error return the error */
2129            err = len;
2130            goto out_qiov;
2131        }
2132    } while (total < count && len > 0);
2133
2134    offset = 7;
2135    err = pdu_marshal(pdu, offset, "d", total);
2136    if (err < 0) {
2137        goto out_qiov;
2138    }
2139    err += offset;
2140    trace_v9fs_write_return(pdu->tag, pdu->id, total, err);
2141out_qiov:
2142    qemu_iovec_destroy(&qiov);
2143out:
2144    put_fid(pdu, fidp);
2145out_nofid:
2146    qemu_iovec_destroy(&qiov_full);
2147    pdu_complete(pdu, err);
2148}
2149
2150static void coroutine_fn v9fs_create(void *opaque)
2151{
2152    int32_t fid;
2153    int err = 0;
2154    size_t offset = 7;
2155    V9fsFidState *fidp;
2156    V9fsQID qid;
2157    int32_t perm;
2158    int8_t mode;
2159    V9fsPath path;
2160    struct stat stbuf;
2161    V9fsString name;
2162    V9fsString extension;
2163    int iounit;
2164    V9fsPDU *pdu = opaque;
2165    V9fsState *s = pdu->s;
2166
2167    v9fs_path_init(&path);
2168    v9fs_string_init(&name);
2169    v9fs_string_init(&extension);
2170    err = pdu_unmarshal(pdu, offset, "dsdbs", &fid, &name,
2171                        &perm, &mode, &extension);
2172    if (err < 0) {
2173        goto out_nofid;
2174    }
2175    trace_v9fs_create(pdu->tag, pdu->id, fid, name.data, perm, mode);
2176
2177    if (name_is_illegal(name.data)) {
2178        err = -ENOENT;
2179        goto out_nofid;
2180    }
2181
2182    if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2183        err = -EEXIST;
2184        goto out_nofid;
2185    }
2186
2187    fidp = get_fid(pdu, fid);
2188    if (fidp == NULL) {
2189        err = -EINVAL;
2190        goto out_nofid;
2191    }
2192    if (fidp->fid_type != P9_FID_NONE) {
2193        err = -EINVAL;
2194        goto out;
2195    }
2196    if (perm & P9_STAT_MODE_DIR) {
2197        err = v9fs_co_mkdir(pdu, fidp, &name, perm & 0777,
2198                            fidp->uid, -1, &stbuf);
2199        if (err < 0) {
2200            goto out;
2201        }
2202        err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2203        if (err < 0) {
2204            goto out;
2205        }
2206        v9fs_path_write_lock(s);
2207        v9fs_path_copy(&fidp->path, &path);
2208        v9fs_path_unlock(s);
2209        err = v9fs_co_opendir(pdu, fidp);
2210        if (err < 0) {
2211            goto out;
2212        }
2213        fidp->fid_type = P9_FID_DIR;
2214    } else if (perm & P9_STAT_MODE_SYMLINK) {
2215        err = v9fs_co_symlink(pdu, fidp, &name,
2216                              extension.data, -1 , &stbuf);
2217        if (err < 0) {
2218            goto out;
2219        }
2220        err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2221        if (err < 0) {
2222            goto out;
2223        }
2224        v9fs_path_write_lock(s);
2225        v9fs_path_copy(&fidp->path, &path);
2226        v9fs_path_unlock(s);
2227    } else if (perm & P9_STAT_MODE_LINK) {
2228        int32_t ofid = atoi(extension.data);
2229        V9fsFidState *ofidp = get_fid(pdu, ofid);
2230        if (ofidp == NULL) {
2231            err = -EINVAL;
2232            goto out;
2233        }
2234        err = v9fs_co_link(pdu, ofidp, fidp, &name);
2235        put_fid(pdu, ofidp);
2236        if (err < 0) {
2237            goto out;
2238        }
2239        err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2240        if (err < 0) {
2241            fidp->fid_type = P9_FID_NONE;
2242            goto out;
2243        }
2244        v9fs_path_write_lock(s);
2245        v9fs_path_copy(&fidp->path, &path);
2246        v9fs_path_unlock(s);
2247        err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
2248        if (err < 0) {
2249            fidp->fid_type = P9_FID_NONE;
2250            goto out;
2251        }
2252    } else if (perm & P9_STAT_MODE_DEVICE) {
2253        char ctype;
2254        uint32_t major, minor;
2255        mode_t nmode = 0;
2256
2257        if (sscanf(extension.data, "%c %u %u", &ctype, &major, &minor) != 3) {
2258            err = -errno;
2259            goto out;
2260        }
2261
2262        switch (ctype) {
2263        case 'c':
2264            nmode = S_IFCHR;
2265            break;
2266        case 'b':
2267            nmode = S_IFBLK;
2268            break;
2269        default:
2270            err = -EIO;
2271            goto out;
2272        }
2273
2274        nmode |= perm & 0777;
2275        err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2276                            makedev(major, minor), nmode, &stbuf);
2277        if (err < 0) {
2278            goto out;
2279        }
2280        err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2281        if (err < 0) {
2282            goto out;
2283        }
2284        v9fs_path_write_lock(s);
2285        v9fs_path_copy(&fidp->path, &path);
2286        v9fs_path_unlock(s);
2287    } else if (perm & P9_STAT_MODE_NAMED_PIPE) {
2288        err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2289                            0, S_IFIFO | (perm & 0777), &stbuf);
2290        if (err < 0) {
2291            goto out;
2292        }
2293        err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2294        if (err < 0) {
2295            goto out;
2296        }
2297        v9fs_path_write_lock(s);
2298        v9fs_path_copy(&fidp->path, &path);
2299        v9fs_path_unlock(s);
2300    } else if (perm & P9_STAT_MODE_SOCKET) {
2301        err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2302                            0, S_IFSOCK | (perm & 0777), &stbuf);
2303        if (err < 0) {
2304            goto out;
2305        }
2306        err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2307        if (err < 0) {
2308            goto out;
2309        }
2310        v9fs_path_write_lock(s);
2311        v9fs_path_copy(&fidp->path, &path);
2312        v9fs_path_unlock(s);
2313    } else {
2314        err = v9fs_co_open2(pdu, fidp, &name, -1,
2315                            omode_to_uflags(mode)|O_CREAT, perm, &stbuf);
2316        if (err < 0) {
2317            goto out;
2318        }
2319        fidp->fid_type = P9_FID_FILE;
2320        fidp->open_flags = omode_to_uflags(mode);
2321        if (fidp->open_flags & O_EXCL) {
2322            /*
2323             * We let the host file system do O_EXCL check
2324             * We should not reclaim such fd
2325             */
2326            fidp->flags |= FID_NON_RECLAIMABLE;
2327        }
2328    }
2329    iounit = get_iounit(pdu, &fidp->path);
2330    stat_to_qid(&stbuf, &qid);
2331    err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
2332    if (err < 0) {
2333        goto out;
2334    }
2335    err += offset;
2336    trace_v9fs_create_return(pdu->tag, pdu->id,
2337                             qid.type, qid.version, qid.path, iounit);
2338out:
2339    put_fid(pdu, fidp);
2340out_nofid:
2341   pdu_complete(pdu, err);
2342   v9fs_string_free(&name);
2343   v9fs_string_free(&extension);
2344   v9fs_path_free(&path);
2345}
2346
2347static void coroutine_fn v9fs_symlink(void *opaque)
2348{
2349    V9fsPDU *pdu = opaque;
2350    V9fsString name;
2351    V9fsString symname;
2352    V9fsFidState *dfidp;
2353    V9fsQID qid;
2354    struct stat stbuf;
2355    int32_t dfid;
2356    int err = 0;
2357    gid_t gid;
2358    size_t offset = 7;
2359
2360    v9fs_string_init(&name);
2361    v9fs_string_init(&symname);
2362    err = pdu_unmarshal(pdu, offset, "dssd", &dfid, &name, &symname, &gid);
2363    if (err < 0) {
2364        goto out_nofid;
2365    }
2366    trace_v9fs_symlink(pdu->tag, pdu->id, dfid, name.data, symname.data, gid);
2367
2368    if (name_is_illegal(name.data)) {
2369        err = -ENOENT;
2370        goto out_nofid;
2371    }
2372
2373    if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2374        err = -EEXIST;
2375        goto out_nofid;
2376    }
2377
2378    dfidp = get_fid(pdu, dfid);
2379    if (dfidp == NULL) {
2380        err = -EINVAL;
2381        goto out_nofid;
2382    }
2383    err = v9fs_co_symlink(pdu, dfidp, &name, symname.data, gid, &stbuf);
2384    if (err < 0) {
2385        goto out;
2386    }
2387    stat_to_qid(&stbuf, &qid);
2388    err =  pdu_marshal(pdu, offset, "Q", &qid);
2389    if (err < 0) {
2390        goto out;
2391    }
2392    err += offset;
2393    trace_v9fs_symlink_return(pdu->tag, pdu->id,
2394                              qid.type, qid.version, qid.path);
2395out:
2396    put_fid(pdu, dfidp);
2397out_nofid:
2398    pdu_complete(pdu, err);
2399    v9fs_string_free(&name);
2400    v9fs_string_free(&symname);
2401}
2402
2403static void coroutine_fn v9fs_flush(void *opaque)
2404{
2405    ssize_t err;
2406    int16_t tag;
2407    size_t offset = 7;
2408    V9fsPDU *cancel_pdu = NULL;
2409    V9fsPDU *pdu = opaque;
2410    V9fsState *s = pdu->s;
2411
2412    err = pdu_unmarshal(pdu, offset, "w", &tag);
2413    if (err < 0) {
2414        pdu_complete(pdu, err);
2415        return;
2416    }
2417    trace_v9fs_flush(pdu->tag, pdu->id, tag);
2418
2419    if (pdu->tag == tag) {
2420        warn_report("the guest sent a self-referencing 9P flush request");
2421    } else {
2422        QLIST_FOREACH(cancel_pdu, &s->active_list, next) {
2423            if (cancel_pdu->tag == tag) {
2424                break;
2425            }
2426        }
2427    }
2428    if (cancel_pdu) {
2429        cancel_pdu->cancelled = 1;
2430        /*
2431         * Wait for pdu to complete.
2432         */
2433        qemu_co_queue_wait(&cancel_pdu->complete, NULL);
2434        if (!qemu_co_queue_next(&cancel_pdu->complete)) {
2435            cancel_pdu->cancelled = 0;
2436            pdu_free(cancel_pdu);
2437        }
2438    }
2439    pdu_complete(pdu, 7);
2440}
2441
2442static void coroutine_fn v9fs_link(void *opaque)
2443{
2444    V9fsPDU *pdu = opaque;
2445    int32_t dfid, oldfid;
2446    V9fsFidState *dfidp, *oldfidp;
2447    V9fsString name;
2448    size_t offset = 7;
2449    int err = 0;
2450
2451    v9fs_string_init(&name);
2452    err = pdu_unmarshal(pdu, offset, "dds", &dfid, &oldfid, &name);
2453    if (err < 0) {
2454        goto out_nofid;
2455    }
2456    trace_v9fs_link(pdu->tag, pdu->id, dfid, oldfid, name.data);
2457
2458    if (name_is_illegal(name.data)) {
2459        err = -ENOENT;
2460        goto out_nofid;
2461    }
2462
2463    if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2464        err = -EEXIST;
2465        goto out_nofid;
2466    }
2467
2468    dfidp = get_fid(pdu, dfid);
2469    if (dfidp == NULL) {
2470        err = -ENOENT;
2471        goto out_nofid;
2472    }
2473
2474    oldfidp = get_fid(pdu, oldfid);
2475    if (oldfidp == NULL) {
2476        err = -ENOENT;
2477        goto out;
2478    }
2479    err = v9fs_co_link(pdu, oldfidp, dfidp, &name);
2480    if (!err) {
2481        err = offset;
2482    }
2483    put_fid(pdu, oldfidp);
2484out:
2485    put_fid(pdu, dfidp);
2486out_nofid:
2487    v9fs_string_free(&name);
2488    pdu_complete(pdu, err);
2489}
2490
2491/* Only works with path name based fid */
2492static void coroutine_fn v9fs_remove(void *opaque)
2493{
2494    int32_t fid;
2495    int err = 0;
2496    size_t offset = 7;
2497    V9fsFidState *fidp;
2498    V9fsPDU *pdu = opaque;
2499
2500    err = pdu_unmarshal(pdu, offset, "d", &fid);
2501    if (err < 0) {
2502        goto out_nofid;
2503    }
2504    trace_v9fs_remove(pdu->tag, pdu->id, fid);
2505
2506    fidp = get_fid(pdu, fid);
2507    if (fidp == NULL) {
2508        err = -EINVAL;
2509        goto out_nofid;
2510    }
2511    /* if fs driver is not path based, return EOPNOTSUPP */
2512    if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
2513        err = -EOPNOTSUPP;
2514        goto out_err;
2515    }
2516    /*
2517     * IF the file is unlinked, we cannot reopen
2518     * the file later. So don't reclaim fd
2519     */
2520    err = v9fs_mark_fids_unreclaim(pdu, &fidp->path);
2521    if (err < 0) {
2522        goto out_err;
2523    }
2524    err = v9fs_co_remove(pdu, &fidp->path);
2525    if (!err) {
2526        err = offset;
2527    }
2528out_err:
2529    /* For TREMOVE we need to clunk the fid even on failed remove */
2530    clunk_fid(pdu->s, fidp->fid);
2531    put_fid(pdu, fidp);
2532out_nofid:
2533    pdu_complete(pdu, err);
2534}
2535
2536static void coroutine_fn v9fs_unlinkat(void *opaque)
2537{
2538    int err = 0;
2539    V9fsString name;
2540    int32_t dfid, flags, rflags = 0;
2541    size_t offset = 7;
2542    V9fsPath path;
2543    V9fsFidState *dfidp;
2544    V9fsPDU *pdu = opaque;
2545
2546    v9fs_string_init(&name);
2547    err = pdu_unmarshal(pdu, offset, "dsd", &dfid, &name, &flags);
2548    if (err < 0) {
2549        goto out_nofid;
2550    }
2551
2552    if (name_is_illegal(name.data)) {
2553        err = -ENOENT;
2554        goto out_nofid;
2555    }
2556
2557    if (!strcmp(".", name.data)) {
2558        err = -EINVAL;
2559        goto out_nofid;
2560    }
2561
2562    if (!strcmp("..", name.data)) {
2563        err = -ENOTEMPTY;
2564        goto out_nofid;
2565    }
2566
2567    if (flags & ~P9_DOTL_AT_REMOVEDIR) {
2568        err = -EINVAL;
2569        goto out_nofid;
2570    }
2571
2572    if (flags & P9_DOTL_AT_REMOVEDIR) {
2573        rflags |= AT_REMOVEDIR;
2574    }
2575
2576    dfidp = get_fid(pdu, dfid);
2577    if (dfidp == NULL) {
2578        err = -EINVAL;
2579        goto out_nofid;
2580    }
2581    /*
2582     * IF the file is unlinked, we cannot reopen
2583     * the file later. So don't reclaim fd
2584     */
2585    v9fs_path_init(&path);
2586    err = v9fs_co_name_to_path(pdu, &dfidp->path, name.data, &path);
2587    if (err < 0) {
2588        goto out_err;
2589    }
2590    err = v9fs_mark_fids_unreclaim(pdu, &path);
2591    if (err < 0) {
2592        goto out_err;
2593    }
2594    err = v9fs_co_unlinkat(pdu, &dfidp->path, &name, rflags);
2595    if (!err) {
2596        err = offset;
2597    }
2598out_err:
2599    put_fid(pdu, dfidp);
2600    v9fs_path_free(&path);
2601out_nofid:
2602    pdu_complete(pdu, err);
2603    v9fs_string_free(&name);
2604}
2605
2606
2607/* Only works with path name based fid */
2608static int coroutine_fn v9fs_complete_rename(V9fsPDU *pdu, V9fsFidState *fidp,
2609                                             int32_t newdirfid,
2610                                             V9fsString *name)
2611{
2612    int err = 0;
2613    V9fsPath new_path;
2614    V9fsFidState *tfidp;
2615    V9fsState *s = pdu->s;
2616    V9fsFidState *dirfidp = NULL;
2617
2618    v9fs_path_init(&new_path);
2619    if (newdirfid != -1) {
2620        dirfidp = get_fid(pdu, newdirfid);
2621        if (dirfidp == NULL) {
2622            err = -ENOENT;
2623            goto out_nofid;
2624        }
2625        if (fidp->fid_type != P9_FID_NONE) {
2626            err = -EINVAL;
2627            goto out;
2628        }
2629        err = v9fs_co_name_to_path(pdu, &dirfidp->path, name->data, &new_path);
2630        if (err < 0) {
2631            goto out;
2632        }
2633    } else {
2634        char *dir_name = g_path_get_dirname(fidp->path.data);
2635        V9fsPath dir_path;
2636
2637        v9fs_path_init(&dir_path);
2638        v9fs_path_sprintf(&dir_path, "%s", dir_name);
2639        g_free(dir_name);
2640
2641        err = v9fs_co_name_to_path(pdu, &dir_path, name->data, &new_path);
2642        v9fs_path_free(&dir_path);
2643        if (err < 0) {
2644            goto out;
2645        }
2646    }
2647    err = v9fs_co_rename(pdu, &fidp->path, &new_path);
2648    if (err < 0) {
2649        goto out;
2650    }
2651    /*
2652     * Fixup fid's pointing to the old name to
2653     * start pointing to the new name
2654     */
2655    for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
2656        if (v9fs_path_is_ancestor(&fidp->path, &tfidp->path)) {
2657            /* replace the name */
2658            v9fs_fix_path(&tfidp->path, &new_path, strlen(fidp->path.data));
2659        }
2660    }
2661out:
2662    if (dirfidp) {
2663        put_fid(pdu, dirfidp);
2664    }
2665    v9fs_path_free(&new_path);
2666out_nofid:
2667    return err;
2668}
2669
2670/* Only works with path name based fid */
2671static void coroutine_fn v9fs_rename(void *opaque)
2672{
2673    int32_t fid;
2674    ssize_t err = 0;
2675    size_t offset = 7;
2676    V9fsString name;
2677    int32_t newdirfid;
2678    V9fsFidState *fidp;
2679    V9fsPDU *pdu = opaque;
2680    V9fsState *s = pdu->s;
2681
2682    v9fs_string_init(&name);
2683    err = pdu_unmarshal(pdu, offset, "dds", &fid, &newdirfid, &name);
2684    if (err < 0) {
2685        goto out_nofid;
2686    }
2687
2688    if (name_is_illegal(name.data)) {
2689        err = -ENOENT;
2690        goto out_nofid;
2691    }
2692
2693    if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2694        err = -EISDIR;
2695        goto out_nofid;
2696    }
2697
2698    fidp = get_fid(pdu, fid);
2699    if (fidp == NULL) {
2700        err = -ENOENT;
2701        goto out_nofid;
2702    }
2703    if (fidp->fid_type != P9_FID_NONE) {
2704        err = -EINVAL;
2705        goto out;
2706    }
2707    /* if fs driver is not path based, return EOPNOTSUPP */
2708    if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
2709        err = -EOPNOTSUPP;
2710        goto out;
2711    }
2712    v9fs_path_write_lock(s);
2713    err = v9fs_complete_rename(pdu, fidp, newdirfid, &name);
2714    v9fs_path_unlock(s);
2715    if (!err) {
2716        err = offset;
2717    }
2718out:
2719    put_fid(pdu, fidp);
2720out_nofid:
2721    pdu_complete(pdu, err);
2722    v9fs_string_free(&name);
2723}
2724
2725static int coroutine_fn v9fs_fix_fid_paths(V9fsPDU *pdu, V9fsPath *olddir,
2726                                           V9fsString *old_name,
2727                                           V9fsPath *newdir,
2728                                           V9fsString *new_name)
2729{
2730    V9fsFidState *tfidp;
2731    V9fsPath oldpath, newpath;
2732    V9fsState *s = pdu->s;
2733    int err;
2734
2735    v9fs_path_init(&oldpath);
2736    v9fs_path_init(&newpath);
2737    err = v9fs_co_name_to_path(pdu, olddir, old_name->data, &oldpath);
2738    if (err < 0) {
2739        goto out;
2740    }
2741    err = v9fs_co_name_to_path(pdu, newdir, new_name->data, &newpath);
2742    if (err < 0) {
2743        goto out;
2744    }
2745
2746    /*
2747     * Fixup fid's pointing to the old name to
2748     * start pointing to the new name
2749     */
2750    for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
2751        if (v9fs_path_is_ancestor(&oldpath, &tfidp->path)) {
2752            /* replace the name */
2753            v9fs_fix_path(&tfidp->path, &newpath, strlen(oldpath.data));
2754        }
2755    }
2756out:
2757    v9fs_path_free(&oldpath);
2758    v9fs_path_free(&newpath);
2759    return err;
2760}
2761
2762static int coroutine_fn v9fs_complete_renameat(V9fsPDU *pdu, int32_t olddirfid,
2763                                               V9fsString *old_name,
2764                                               int32_t newdirfid,
2765                                               V9fsString *new_name)
2766{
2767    int err = 0;
2768    V9fsState *s = pdu->s;
2769    V9fsFidState *newdirfidp = NULL, *olddirfidp = NULL;
2770
2771    olddirfidp = get_fid(pdu, olddirfid);
2772    if (olddirfidp == NULL) {
2773        err = -ENOENT;
2774        goto out;
2775    }
2776    if (newdirfid != -1) {
2777        newdirfidp = get_fid(pdu, newdirfid);
2778        if (newdirfidp == NULL) {
2779            err = -ENOENT;
2780            goto out;
2781        }
2782    } else {
2783        newdirfidp = get_fid(pdu, olddirfid);
2784    }
2785
2786    err = v9fs_co_renameat(pdu, &olddirfidp->path, old_name,
2787                           &newdirfidp->path, new_name);
2788    if (err < 0) {
2789        goto out;
2790    }
2791    if (s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT) {
2792        /* Only for path based fid  we need to do the below fixup */
2793        err = v9fs_fix_fid_paths(pdu, &olddirfidp->path, old_name,
2794                                 &newdirfidp->path, new_name);
2795    }
2796out:
2797    if (olddirfidp) {
2798        put_fid(pdu, olddirfidp);
2799    }
2800    if (newdirfidp) {
2801        put_fid(pdu, newdirfidp);
2802    }
2803    return err;
2804}
2805
2806static void coroutine_fn v9fs_renameat(void *opaque)
2807{
2808    ssize_t err = 0;
2809    size_t offset = 7;
2810    V9fsPDU *pdu = opaque;
2811    V9fsState *s = pdu->s;
2812    int32_t olddirfid, newdirfid;
2813    V9fsString old_name, new_name;
2814
2815    v9fs_string_init(&old_name);
2816    v9fs_string_init(&new_name);
2817    err = pdu_unmarshal(pdu, offset, "dsds", &olddirfid,
2818                        &old_name, &newdirfid, &new_name);
2819    if (err < 0) {
2820        goto out_err;
2821    }
2822
2823    if (name_is_illegal(old_name.data) || name_is_illegal(new_name.data)) {
2824        err = -ENOENT;
2825        goto out_err;
2826    }
2827
2828    if (!strcmp(".", old_name.data) || !strcmp("..", old_name.data) ||
2829        !strcmp(".", new_name.data) || !strcmp("..", new_name.data)) {
2830        err = -EISDIR;
2831        goto out_err;
2832    }
2833
2834    v9fs_path_write_lock(s);
2835    err = v9fs_complete_renameat(pdu, olddirfid,
2836                                 &old_name, newdirfid, &new_name);
2837    v9fs_path_unlock(s);
2838    if (!err) {
2839        err = offset;
2840    }
2841
2842out_err:
2843    pdu_complete(pdu, err);
2844    v9fs_string_free(&old_name);
2845    v9fs_string_free(&new_name);
2846}
2847
2848static void coroutine_fn v9fs_wstat(void *opaque)
2849{
2850    int32_t fid;
2851    int err = 0;
2852    int16_t unused;
2853    V9fsStat v9stat;
2854    size_t offset = 7;
2855    struct stat stbuf;
2856    V9fsFidState *fidp;
2857    V9fsPDU *pdu = opaque;
2858    V9fsState *s = pdu->s;
2859
2860    v9fs_stat_init(&v9stat);
2861    err = pdu_unmarshal(pdu, offset, "dwS", &fid, &unused, &v9stat);
2862    if (err < 0) {
2863        goto out_nofid;
2864    }
2865    trace_v9fs_wstat(pdu->tag, pdu->id, fid,
2866                     v9stat.mode, v9stat.atime, v9stat.mtime);
2867
2868    fidp = get_fid(pdu, fid);
2869    if (fidp == NULL) {
2870        err = -EINVAL;
2871        goto out_nofid;
2872    }
2873    /* do we need to sync the file? */
2874    if (donttouch_stat(&v9stat)) {
2875        err = v9fs_co_fsync(pdu, fidp, 0);
2876        goto out;
2877    }
2878    if (v9stat.mode != -1) {
2879        uint32_t v9_mode;
2880        err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
2881        if (err < 0) {
2882            goto out;
2883        }
2884        v9_mode = stat_to_v9mode(&stbuf);
2885        if ((v9stat.mode & P9_STAT_MODE_TYPE_BITS) !=
2886            (v9_mode & P9_STAT_MODE_TYPE_BITS)) {
2887            /* Attempting to change the type */
2888            err = -EIO;
2889            goto out;
2890        }
2891        err = v9fs_co_chmod(pdu, &fidp->path,
2892                            v9mode_to_mode(v9stat.mode,
2893                                           &v9stat.extension));
2894        if (err < 0) {
2895            goto out;
2896        }
2897    }
2898    if (v9stat.mtime != -1 || v9stat.atime != -1) {
2899        struct timespec times[2];
2900        if (v9stat.atime != -1) {
2901            times[0].tv_sec = v9stat.atime;
2902            times[0].tv_nsec = 0;
2903        } else {
2904            times[0].tv_nsec = UTIME_OMIT;
2905        }
2906        if (v9stat.mtime != -1) {
2907            times[1].tv_sec = v9stat.mtime;
2908            times[1].tv_nsec = 0;
2909        } else {
2910            times[1].tv_nsec = UTIME_OMIT;
2911        }
2912        err = v9fs_co_utimensat(pdu, &fidp->path, times);
2913        if (err < 0) {
2914            goto out;
2915        }
2916    }
2917    if (v9stat.n_gid != -1 || v9stat.n_uid != -1) {
2918        err = v9fs_co_chown(pdu, &fidp->path, v9stat.n_uid, v9stat.n_gid);
2919        if (err < 0) {
2920            goto out;
2921        }
2922    }
2923    if (v9stat.name.size != 0) {
2924        v9fs_path_write_lock(s);
2925        err = v9fs_complete_rename(pdu, fidp, -1, &v9stat.name);
2926        v9fs_path_unlock(s);
2927        if (err < 0) {
2928            goto out;
2929        }
2930    }
2931    if (v9stat.length != -1) {
2932        err = v9fs_co_truncate(pdu, &fidp->path, v9stat.length);
2933        if (err < 0) {
2934            goto out;
2935        }
2936    }
2937    err = offset;
2938out:
2939    put_fid(pdu, fidp);
2940out_nofid:
2941    v9fs_stat_free(&v9stat);
2942    pdu_complete(pdu, err);
2943}
2944
2945static int v9fs_fill_statfs(V9fsState *s, V9fsPDU *pdu, struct statfs *stbuf)
2946{
2947    uint32_t f_type;
2948    uint32_t f_bsize;
2949    uint64_t f_blocks;
2950    uint64_t f_bfree;
2951    uint64_t f_bavail;
2952    uint64_t f_files;
2953    uint64_t f_ffree;
2954    uint64_t fsid_val;
2955    uint32_t f_namelen;
2956    size_t offset = 7;
2957    int32_t bsize_factor;
2958
2959    /*
2960     * compute bsize factor based on host file system block size
2961     * and client msize
2962     */
2963    bsize_factor = (s->msize - P9_IOHDRSZ)/stbuf->f_bsize;
2964    if (!bsize_factor) {
2965        bsize_factor = 1;
2966    }
2967    f_type  = stbuf->f_type;
2968    f_bsize = stbuf->f_bsize;
2969    f_bsize *= bsize_factor;
2970    /*
2971     * f_bsize is adjusted(multiplied) by bsize factor, so we need to
2972     * adjust(divide) the number of blocks, free blocks and available
2973     * blocks by bsize factor
2974     */
2975    f_blocks = stbuf->f_blocks/bsize_factor;
2976    f_bfree  = stbuf->f_bfree/bsize_factor;
2977    f_bavail = stbuf->f_bavail/bsize_factor;
2978    f_files  = stbuf->f_files;
2979    f_ffree  = stbuf->f_ffree;
2980    fsid_val = (unsigned int) stbuf->f_fsid.__val[0] |
2981               (unsigned long long)stbuf->f_fsid.__val[1] << 32;
2982    f_namelen = stbuf->f_namelen;
2983
2984    return pdu_marshal(pdu, offset, "ddqqqqqqd",
2985                       f_type, f_bsize, f_blocks, f_bfree,
2986                       f_bavail, f_files, f_ffree,
2987                       fsid_val, f_namelen);
2988}
2989
2990static void coroutine_fn v9fs_statfs(void *opaque)
2991{
2992    int32_t fid;
2993    ssize_t retval = 0;
2994    size_t offset = 7;
2995    V9fsFidState *fidp;
2996    struct statfs stbuf;
2997    V9fsPDU *pdu = opaque;
2998    V9fsState *s = pdu->s;
2999
3000    retval = pdu_unmarshal(pdu, offset, "d", &fid);
3001    if (retval < 0) {
3002        goto out_nofid;
3003    }
3004    fidp = get_fid(pdu, fid);
3005    if (fidp == NULL) {
3006        retval = -ENOENT;
3007        goto out_nofid;
3008    }
3009    retval = v9fs_co_statfs(pdu, &fidp->path, &stbuf);
3010    if (retval < 0) {
3011        goto out;
3012    }
3013    retval = v9fs_fill_statfs(s, pdu, &stbuf);
3014    if (retval < 0) {
3015        goto out;
3016    }
3017    retval += offset;
3018out:
3019    put_fid(pdu, fidp);
3020out_nofid:
3021    pdu_complete(pdu, retval);
3022}
3023
3024static void coroutine_fn v9fs_mknod(void *opaque)
3025{
3026
3027    int mode;
3028    gid_t gid;
3029    int32_t fid;
3030    V9fsQID qid;
3031    int err = 0;
3032    int major, minor;
3033    size_t offset = 7;
3034    V9fsString name;
3035    struct stat stbuf;
3036    V9fsFidState *fidp;
3037    V9fsPDU *pdu = opaque;
3038
3039    v9fs_string_init(&name);
3040    err = pdu_unmarshal(pdu, offset, "dsdddd", &fid, &name, &mode,
3041                        &major, &minor, &gid);
3042    if (err < 0) {
3043        goto out_nofid;
3044    }
3045    trace_v9fs_mknod(pdu->tag, pdu->id, fid, mode, major, minor);
3046
3047    if (name_is_illegal(name.data)) {
3048        err = -ENOENT;
3049        goto out_nofid;
3050    }
3051
3052    if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3053        err = -EEXIST;
3054        goto out_nofid;
3055    }
3056
3057    fidp = get_fid(pdu, fid);
3058    if (fidp == NULL) {
3059        err = -ENOENT;
3060        goto out_nofid;
3061    }
3062    err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, gid,
3063                        makedev(major, minor), mode, &stbuf);
3064    if (err < 0) {
3065        goto out;
3066    }
3067    stat_to_qid(&stbuf, &qid);
3068    err = pdu_marshal(pdu, offset, "Q", &qid);
3069    if (err < 0) {
3070        goto out;
3071    }
3072    err += offset;
3073    trace_v9fs_mknod_return(pdu->tag, pdu->id,
3074                            qid.type, qid.version, qid.path);
3075out:
3076    put_fid(pdu, fidp);
3077out_nofid:
3078    pdu_complete(pdu, err);
3079    v9fs_string_free(&name);
3080}
3081
3082/*
3083 * Implement posix byte range locking code
3084 * Server side handling of locking code is very simple, because 9p server in
3085 * QEMU can handle only one client. And most of the lock handling
3086 * (like conflict, merging) etc is done by the VFS layer itself, so no need to
3087 * do any thing in * qemu 9p server side lock code path.
3088 * So when a TLOCK request comes, always return success
3089 */
3090static void coroutine_fn v9fs_lock(void *opaque)
3091{
3092    V9fsFlock flock;
3093    size_t offset = 7;
3094    struct stat stbuf;
3095    V9fsFidState *fidp;
3096    int32_t fid, err = 0;
3097    V9fsPDU *pdu = opaque;
3098
3099    v9fs_string_init(&flock.client_id);
3100    err = pdu_unmarshal(pdu, offset, "dbdqqds", &fid, &flock.type,
3101                        &flock.flags, &flock.start, &flock.length,
3102                        &flock.proc_id, &flock.client_id);
3103    if (err < 0) {
3104        goto out_nofid;
3105    }
3106    trace_v9fs_lock(pdu->tag, pdu->id, fid,
3107                    flock.type, flock.start, flock.length);
3108
3109
3110    /* We support only block flag now (that too ignored currently) */
3111    if (flock.flags & ~P9_LOCK_FLAGS_BLOCK) {
3112        err = -EINVAL;
3113        goto out_nofid;
3114    }
3115    fidp = get_fid(pdu, fid);
3116    if (fidp == NULL) {
3117        err = -ENOENT;
3118        goto out_nofid;
3119    }
3120    err = v9fs_co_fstat(pdu, fidp, &stbuf);
3121    if (err < 0) {
3122        goto out;
3123    }
3124    err = pdu_marshal(pdu, offset, "b", P9_LOCK_SUCCESS);
3125    if (err < 0) {
3126        goto out;
3127    }
3128    err += offset;
3129    trace_v9fs_lock_return(pdu->tag, pdu->id, P9_LOCK_SUCCESS);
3130out:
3131    put_fid(pdu, fidp);
3132out_nofid:
3133    pdu_complete(pdu, err);
3134    v9fs_string_free(&flock.client_id);
3135}
3136
3137/*
3138 * When a TGETLOCK request comes, always return success because all lock
3139 * handling is done by client's VFS layer.
3140 */
3141static void coroutine_fn v9fs_getlock(void *opaque)
3142{
3143    size_t offset = 7;
3144    struct stat stbuf;
3145    V9fsFidState *fidp;
3146    V9fsGetlock glock;
3147    int32_t fid, err = 0;
3148    V9fsPDU *pdu = opaque;
3149
3150    v9fs_string_init(&glock.client_id);
3151    err = pdu_unmarshal(pdu, offset, "dbqqds", &fid, &glock.type,
3152                        &glock.start, &glock.length, &glock.proc_id,
3153                        &glock.client_id);
3154    if (err < 0) {
3155        goto out_nofid;
3156    }
3157    trace_v9fs_getlock(pdu->tag, pdu->id, fid,
3158                       glock.type, glock.start, glock.length);
3159
3160    fidp = get_fid(pdu, fid);
3161    if (fidp == NULL) {
3162        err = -ENOENT;
3163        goto out_nofid;
3164    }
3165    err = v9fs_co_fstat(pdu, fidp, &stbuf);
3166    if (err < 0) {
3167        goto out;
3168    }
3169    glock.type = P9_LOCK_TYPE_UNLCK;
3170    err = pdu_marshal(pdu, offset, "bqqds", glock.type,
3171                          glock.start, glock.length, glock.proc_id,
3172                          &glock.client_id);
3173    if (err < 0) {
3174        goto out;
3175    }
3176    err += offset;
3177    trace_v9fs_getlock_return(pdu->tag, pdu->id, glock.type, glock.start,
3178                              glock.length, glock.proc_id);
3179out:
3180    put_fid(pdu, fidp);
3181out_nofid:
3182    pdu_complete(pdu, err);
3183    v9fs_string_free(&glock.client_id);
3184}
3185
3186static void coroutine_fn v9fs_mkdir(void *opaque)
3187{
3188    V9fsPDU *pdu = opaque;
3189    size_t offset = 7;
3190    int32_t fid;
3191    struct stat stbuf;
3192    V9fsQID qid;
3193    V9fsString name;
3194    V9fsFidState *fidp;
3195    gid_t gid;
3196    int mode;
3197    int err = 0;
3198
3199    v9fs_string_init(&name);
3200    err = pdu_unmarshal(pdu, offset, "dsdd", &fid, &name, &mode, &gid);
3201    if (err < 0) {
3202        goto out_nofid;
3203    }
3204    trace_v9fs_mkdir(pdu->tag, pdu->id, fid, name.data, mode, gid);
3205
3206    if (name_is_illegal(name.data)) {
3207        err = -ENOENT;
3208        goto out_nofid;
3209    }
3210
3211    if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3212        err = -EEXIST;
3213        goto out_nofid;
3214    }
3215
3216    fidp = get_fid(pdu, fid);
3217    if (fidp == NULL) {
3218        err = -ENOENT;
3219        goto out_nofid;
3220    }
3221    err = v9fs_co_mkdir(pdu, fidp, &name, mode, fidp->uid, gid, &stbuf);
3222    if (err < 0) {
3223        goto out;
3224    }
3225    stat_to_qid(&stbuf, &qid);
3226    err = pdu_marshal(pdu, offset, "Q", &qid);
3227    if (err < 0) {
3228        goto out;
3229    }
3230    err += offset;
3231    trace_v9fs_mkdir_return(pdu->tag, pdu->id,
3232                            qid.type, qid.version, qid.path, err);
3233out:
3234    put_fid(pdu, fidp);
3235out_nofid:
3236    pdu_complete(pdu, err);
3237    v9fs_string_free(&name);
3238}
3239
3240static void coroutine_fn v9fs_xattrwalk(void *opaque)
3241{
3242    int64_t size;
3243    V9fsString name;
3244    ssize_t err = 0;
3245    size_t offset = 7;
3246    int32_t fid, newfid;
3247    V9fsFidState *file_fidp;
3248    V9fsFidState *xattr_fidp = NULL;
3249    V9fsPDU *pdu = opaque;
3250    V9fsState *s = pdu->s;
3251
3252    v9fs_string_init(&name);
3253    err = pdu_unmarshal(pdu, offset, "dds", &fid, &newfid, &name);
3254    if (err < 0) {
3255        goto out_nofid;
3256    }
3257    trace_v9fs_xattrwalk(pdu->tag, pdu->id, fid, newfid, name.data);
3258
3259    file_fidp = get_fid(pdu, fid);
3260    if (file_fidp == NULL) {
3261        err = -ENOENT;
3262        goto out_nofid;
3263    }
3264    xattr_fidp = alloc_fid(s, newfid);
3265    if (xattr_fidp == NULL) {
3266        err = -EINVAL;
3267        goto out;
3268    }
3269    v9fs_path_copy(&xattr_fidp->path, &file_fidp->path);
3270    if (!v9fs_string_size(&name)) {
3271        /*
3272         * listxattr request. Get the size first
3273         */
3274        size = v9fs_co_llistxattr(pdu, &xattr_fidp->path, NULL, 0);
3275        if (size < 0) {
3276            err = size;
3277            clunk_fid(s, xattr_fidp->fid);
3278            goto out;
3279        }
3280        /*
3281         * Read the xattr value
3282         */
3283        xattr_fidp->fs.xattr.len = size;
3284        xattr_fidp->fid_type = P9_FID_XATTR;
3285        xattr_fidp->fs.xattr.xattrwalk_fid = true;
3286        xattr_fidp->fs.xattr.value = g_malloc0(size);
3287        if (size) {
3288            err = v9fs_co_llistxattr(pdu, &xattr_fidp->path,
3289                                     xattr_fidp->fs.xattr.value,
3290                                     xattr_fidp->fs.xattr.len);
3291            if (err < 0) {
3292                clunk_fid(s, xattr_fidp->fid);
3293                goto out;
3294            }
3295        }
3296        err = pdu_marshal(pdu, offset, "q", size);
3297        if (err < 0) {
3298            goto out;
3299        }
3300        err += offset;
3301    } else {
3302        /*
3303         * specific xattr fid. We check for xattr
3304         * presence also collect the xattr size
3305         */
3306        size = v9fs_co_lgetxattr(pdu, &xattr_fidp->path,
3307                                 &name, NULL, 0);
3308        if (size < 0) {
3309            err = size;
3310            clunk_fid(s, xattr_fidp->fid);
3311            goto out;
3312        }
3313        /*
3314         * Read the xattr value
3315         */
3316        xattr_fidp->fs.xattr.len = size;
3317        xattr_fidp->fid_type = P9_FID_XATTR;
3318        xattr_fidp->fs.xattr.xattrwalk_fid = true;
3319        xattr_fidp->fs.xattr.value = g_malloc0(size);
3320        if (size) {
3321            err = v9fs_co_lgetxattr(pdu, &xattr_fidp->path,
3322                                    &name, xattr_fidp->fs.xattr.value,
3323                                    xattr_fidp->fs.xattr.len);
3324            if (err < 0) {
3325                clunk_fid(s, xattr_fidp->fid);
3326                goto out;
3327            }
3328        }
3329        err = pdu_marshal(pdu, offset, "q", size);
3330        if (err < 0) {
3331            goto out;
3332        }
3333        err += offset;
3334    }
3335    trace_v9fs_xattrwalk_return(pdu->tag, pdu->id, size);
3336out:
3337    put_fid(pdu, file_fidp);
3338    if (xattr_fidp) {
3339        put_fid(pdu, xattr_fidp);
3340    }
3341out_nofid:
3342    pdu_complete(pdu, err);
3343    v9fs_string_free(&name);
3344}
3345
3346static void coroutine_fn v9fs_xattrcreate(void *opaque)
3347{
3348    int flags, rflags = 0;
3349    int32_t fid;
3350    uint64_t size;
3351    ssize_t err = 0;
3352    V9fsString name;
3353    size_t offset = 7;
3354    V9fsFidState *file_fidp;
3355    V9fsFidState *xattr_fidp;
3356    V9fsPDU *pdu = opaque;
3357
3358    v9fs_string_init(&name);
3359    err = pdu_unmarshal(pdu, offset, "dsqd", &fid, &name, &size, &flags);
3360    if (err < 0) {
3361        goto out_nofid;
3362    }
3363    trace_v9fs_xattrcreate(pdu->tag, pdu->id, fid, name.data, size, flags);
3364
3365    if (flags & ~(P9_XATTR_CREATE | P9_XATTR_REPLACE)) {
3366        err = -EINVAL;
3367        goto out_nofid;
3368    }
3369
3370    if (flags & P9_XATTR_CREATE) {
3371        rflags |= XATTR_CREATE;
3372    }
3373
3374    if (flags & P9_XATTR_REPLACE) {
3375        rflags |= XATTR_REPLACE;
3376    }
3377
3378    if (size > XATTR_SIZE_MAX) {
3379        err = -E2BIG;
3380        goto out_nofid;
3381    }
3382
3383    file_fidp = get_fid(pdu, fid);
3384    if (file_fidp == NULL) {
3385        err = -EINVAL;
3386        goto out_nofid;
3387    }
3388    if (file_fidp->fid_type != P9_FID_NONE) {
3389        err = -EINVAL;
3390        goto out_put_fid;
3391    }
3392
3393    /* Make the file fid point to xattr */
3394    xattr_fidp = file_fidp;
3395    xattr_fidp->fid_type = P9_FID_XATTR;
3396    xattr_fidp->fs.xattr.copied_len = 0;
3397    xattr_fidp->fs.xattr.xattrwalk_fid = false;
3398    xattr_fidp->fs.xattr.len = size;
3399    xattr_fidp->fs.xattr.flags = rflags;
3400    v9fs_string_init(&xattr_fidp->fs.xattr.name);
3401    v9fs_string_copy(&xattr_fidp->fs.xattr.name, &name);
3402    xattr_fidp->fs.xattr.value = g_malloc0(size);
3403    err = offset;
3404out_put_fid:
3405    put_fid(pdu, file_fidp);
3406out_nofid:
3407    pdu_complete(pdu, err);
3408    v9fs_string_free(&name);
3409}
3410
3411static void coroutine_fn v9fs_readlink(void *opaque)
3412{
3413    V9fsPDU *pdu = opaque;
3414    size_t offset = 7;
3415    V9fsString target;
3416    int32_t fid;
3417    int err = 0;
3418    V9fsFidState *fidp;
3419
3420    err = pdu_unmarshal(pdu, offset, "d", &fid);
3421    if (err < 0) {
3422        goto out_nofid;
3423    }
3424    trace_v9fs_readlink(pdu->tag, pdu->id, fid);
3425    fidp = get_fid(pdu, fid);
3426    if (fidp == NULL) {
3427        err = -ENOENT;
3428        goto out_nofid;
3429    }
3430
3431    v9fs_string_init(&target);
3432    err = v9fs_co_readlink(pdu, &fidp->path, &target);
3433    if (err < 0) {
3434        goto out;
3435    }
3436    err = pdu_marshal(pdu, offset, "s", &target);
3437    if (err < 0) {
3438        v9fs_string_free(&target);
3439        goto out;
3440    }
3441    err += offset;
3442    trace_v9fs_readlink_return(pdu->tag, pdu->id, target.data);
3443    v9fs_string_free(&target);
3444out:
3445    put_fid(pdu, fidp);
3446out_nofid:
3447    pdu_complete(pdu, err);
3448}
3449
3450static CoroutineEntry *pdu_co_handlers[] = {
3451    [P9_TREADDIR] = v9fs_readdir,
3452    [P9_TSTATFS] = v9fs_statfs,
3453    [P9_TGETATTR] = v9fs_getattr,
3454    [P9_TSETATTR] = v9fs_setattr,
3455    [P9_TXATTRWALK] = v9fs_xattrwalk,
3456    [P9_TXATTRCREATE] = v9fs_xattrcreate,
3457    [P9_TMKNOD] = v9fs_mknod,
3458    [P9_TRENAME] = v9fs_rename,
3459    [P9_TLOCK] = v9fs_lock,
3460    [P9_TGETLOCK] = v9fs_getlock,
3461    [P9_TRENAMEAT] = v9fs_renameat,
3462    [P9_TREADLINK] = v9fs_readlink,
3463    [P9_TUNLINKAT] = v9fs_unlinkat,
3464    [P9_TMKDIR] = v9fs_mkdir,
3465    [P9_TVERSION] = v9fs_version,
3466    [P9_TLOPEN] = v9fs_open,
3467    [P9_TATTACH] = v9fs_attach,
3468    [P9_TSTAT] = v9fs_stat,
3469    [P9_TWALK] = v9fs_walk,
3470    [P9_TCLUNK] = v9fs_clunk,
3471    [P9_TFSYNC] = v9fs_fsync,
3472    [P9_TOPEN] = v9fs_open,
3473    [P9_TREAD] = v9fs_read,
3474#if 0
3475    [P9_TAUTH] = v9fs_auth,
3476#endif
3477    [P9_TFLUSH] = v9fs_flush,
3478    [P9_TLINK] = v9fs_link,
3479    [P9_TSYMLINK] = v9fs_symlink,
3480    [P9_TCREATE] = v9fs_create,
3481    [P9_TLCREATE] = v9fs_lcreate,
3482    [P9_TWRITE] = v9fs_write,
3483    [P9_TWSTAT] = v9fs_wstat,
3484    [P9_TREMOVE] = v9fs_remove,
3485};
3486
3487static void coroutine_fn v9fs_op_not_supp(void *opaque)
3488{
3489    V9fsPDU *pdu = opaque;
3490    pdu_complete(pdu, -EOPNOTSUPP);
3491}
3492
3493static void coroutine_fn v9fs_fs_ro(void *opaque)
3494{
3495    V9fsPDU *pdu = opaque;
3496    pdu_complete(pdu, -EROFS);
3497}
3498
3499static inline bool is_read_only_op(V9fsPDU *pdu)
3500{
3501    switch (pdu->id) {
3502    case P9_TREADDIR:
3503    case P9_TSTATFS:
3504    case P9_TGETATTR:
3505    case P9_TXATTRWALK:
3506    case P9_TLOCK:
3507    case P9_TGETLOCK:
3508    case P9_TREADLINK:
3509    case P9_TVERSION:
3510    case P9_TLOPEN:
3511    case P9_TATTACH:
3512    case P9_TSTAT:
3513    case P9_TWALK:
3514    case P9_TCLUNK:
3515    case P9_TFSYNC:
3516    case P9_TOPEN:
3517    case P9_TREAD:
3518    case P9_TAUTH:
3519    case P9_TFLUSH:
3520        return 1;
3521    default:
3522        return 0;
3523    }
3524}
3525
3526void pdu_submit(V9fsPDU *pdu, P9MsgHeader *hdr)
3527{
3528    Coroutine *co;
3529    CoroutineEntry *handler;
3530    V9fsState *s = pdu->s;
3531
3532    pdu->size = le32_to_cpu(hdr->size_le);
3533    pdu->id = hdr->id;
3534    pdu->tag = le16_to_cpu(hdr->tag_le);
3535
3536    if (pdu->id >= ARRAY_SIZE(pdu_co_handlers) ||
3537        (pdu_co_handlers[pdu->id] == NULL)) {
3538        handler = v9fs_op_not_supp;
3539    } else if (is_ro_export(&s->ctx) && !is_read_only_op(pdu)) {
3540        handler = v9fs_fs_ro;
3541    } else {
3542        handler = pdu_co_handlers[pdu->id];
3543    }
3544
3545    qemu_co_queue_init(&pdu->complete);
3546    co = qemu_coroutine_create(handler, pdu);
3547    qemu_coroutine_enter(co);
3548}
3549
3550/* Returns 0 on success, 1 on failure. */
3551int v9fs_device_realize_common(V9fsState *s, const V9fsTransport *t,
3552                               Error **errp)
3553{
3554    int i, len;
3555    struct stat stat;
3556    FsDriverEntry *fse;
3557    V9fsPath path;
3558    int rc = 1;
3559
3560    assert(!s->transport);
3561    s->transport = t;
3562
3563    /* initialize pdu allocator */
3564    QLIST_INIT(&s->free_list);
3565    QLIST_INIT(&s->active_list);
3566    for (i = 0; i < MAX_REQ; i++) {
3567        QLIST_INSERT_HEAD(&s->free_list, &s->pdus[i], next);
3568        s->pdus[i].s = s;
3569        s->pdus[i].idx = i;
3570    }
3571
3572    v9fs_path_init(&path);
3573
3574    fse = get_fsdev_fsentry(s->fsconf.fsdev_id);
3575
3576    if (!fse) {
3577        /* We don't have a fsdev identified by fsdev_id */
3578        error_setg(errp, "9pfs device couldn't find fsdev with the "
3579                   "id = %s",
3580                   s->fsconf.fsdev_id ? s->fsconf.fsdev_id : "NULL");
3581        goto out;
3582    }
3583
3584    if (!s->fsconf.tag) {
3585        /* we haven't specified a mount_tag */
3586        error_setg(errp, "fsdev with id %s needs mount_tag arguments",
3587                   s->fsconf.fsdev_id);
3588        goto out;
3589    }
3590
3591    s->ctx.export_flags = fse->export_flags;
3592    s->ctx.fs_root = g_strdup(fse->path);
3593    s->ctx.exops.get_st_gen = NULL;
3594    len = strlen(s->fsconf.tag);
3595    if (len > MAX_TAG_LEN - 1) {
3596        error_setg(errp, "mount tag '%s' (%d bytes) is longer than "
3597                   "maximum (%d bytes)", s->fsconf.tag, len, MAX_TAG_LEN - 1);
3598        goto out;
3599    }
3600
3601    s->tag = g_strdup(s->fsconf.tag);
3602    s->ctx.uid = -1;
3603
3604    s->ops = fse->ops;
3605
3606    s->ctx.fmode = fse->fmode;
3607    s->ctx.dmode = fse->dmode;
3608
3609    s->fid_list = NULL;
3610    qemu_co_rwlock_init(&s->rename_lock);
3611
3612    if (s->ops->init(&s->ctx, errp) < 0) {
3613        error_prepend(errp, "cannot initialize fsdev '%s': ",
3614                      s->fsconf.fsdev_id);
3615        goto out;
3616    }
3617
3618    /*
3619     * Check details of export path, We need to use fs driver
3620     * call back to do that. Since we are in the init path, we don't
3621     * use co-routines here.
3622     */
3623    if (s->ops->name_to_path(&s->ctx, NULL, "/", &path) < 0) {
3624        error_setg(errp,
3625                   "error in converting name to path %s", strerror(errno));
3626        goto out;
3627    }
3628    if (s->ops->lstat(&s->ctx, &path, &stat)) {
3629        error_setg(errp, "share path %s does not exist", fse->path);
3630        goto out;
3631    } else if (!S_ISDIR(stat.st_mode)) {
3632        error_setg(errp, "share path %s is not a directory", fse->path);
3633        goto out;
3634    }
3635
3636    s->ctx.fst = &fse->fst;
3637    fsdev_throttle_init(s->ctx.fst);
3638
3639    v9fs_path_free(&path);
3640
3641    rc = 0;
3642out:
3643    if (rc) {
3644        if (s->ops && s->ops->cleanup && s->ctx.private) {
3645            s->ops->cleanup(&s->ctx);
3646        }
3647        g_free(s->tag);
3648        g_free(s->ctx.fs_root);
3649        v9fs_path_free(&path);
3650    }
3651    return rc;
3652}
3653
3654void v9fs_device_unrealize_common(V9fsState *s, Error **errp)
3655{
3656    if (s->ops->cleanup) {
3657        s->ops->cleanup(&s->ctx);
3658    }
3659    fsdev_throttle_cleanup(s->ctx.fst);
3660    g_free(s->tag);
3661    g_free(s->ctx.fs_root);
3662}
3663
3664typedef struct VirtfsCoResetData {
3665    V9fsPDU pdu;
3666    bool done;
3667} VirtfsCoResetData;
3668
3669static void coroutine_fn virtfs_co_reset(void *opaque)
3670{
3671    VirtfsCoResetData *data = opaque;
3672
3673    virtfs_reset(&data->pdu);
3674    data->done = true;
3675}
3676
3677void v9fs_reset(V9fsState *s)
3678{
3679    VirtfsCoResetData data = { .pdu = { .s = s }, .done = false };
3680    Coroutine *co;
3681
3682    while (!QLIST_EMPTY(&s->active_list)) {
3683        aio_poll(qemu_get_aio_context(), true);
3684    }
3685
3686    co = qemu_coroutine_create(virtfs_co_reset, &data);
3687    qemu_coroutine_enter(co);
3688
3689    while (!data.done) {
3690        aio_poll(qemu_get_aio_context(), true);
3691    }
3692}
3693
3694static void __attribute__((__constructor__)) v9fs_set_fd_limit(void)
3695{
3696    struct rlimit rlim;
3697    if (getrlimit(RLIMIT_NOFILE, &rlim) < 0) {
3698        error_report("Failed to get the resource limit");
3699        exit(1);
3700    }
3701    open_fd_hw = rlim.rlim_cur - MIN(400, rlim.rlim_cur/3);
3702    open_fd_rc = rlim.rlim_cur/2;
3703}
3704