4 * Copyright IBM, Corp. 2010
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
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"
20 #include "qemu/sockets.h"
21 #include "virtio-9p.h"
22 #include "fsdev/qemu-fsdev.h"
26 #include "migration/blocker.h"
27 #include "sysemu/qtest.h"
31 static int open_fd_rc
;
45 static ssize_t
pdu_marshal(V9fsPDU
*pdu
, size_t offset
, const char *fmt
, ...)
51 ret
= pdu
->s
->transport
->pdu_vmarshal(pdu
, offset
, fmt
, ap
);
57 static ssize_t
pdu_unmarshal(V9fsPDU
*pdu
, size_t offset
, const char *fmt
, ...)
63 ret
= pdu
->s
->transport
->pdu_vunmarshal(pdu
, offset
, fmt
, ap
);
69 static int omode_to_uflags(int8_t mode
)
103 typedef struct DotlOpenflagMap
{
108 static int dotl_to_open_flags(int flags
)
112 * We have same bits for P9_DOTL_READONLY, P9_DOTL_WRONLY
113 * and P9_DOTL_NOACCESS
115 int oflags
= flags
& O_ACCMODE
;
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
},
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
;
143 void cred_init(FsCred
*credp
)
151 static int get_dotl_openflags(V9fsState
*s
, int oflags
)
155 * Filter the client open flags
157 flags
= dotl_to_open_flags(oflags
);
158 flags
&= ~(O_NOCTTY
| O_ASYNC
| O_CREAT
);
160 * Ignore direct disk access hint until the server supports it.
166 void v9fs_path_init(V9fsPath
*path
)
172 void v9fs_path_free(V9fsPath
*path
)
180 void GCC_FMT_ATTR(2, 3)
181 v9fs_path_sprintf(V9fsPath
*path
, const char *fmt
, ...)
185 v9fs_path_free(path
);
188 /* Bump the size for including terminating NULL */
189 path
->size
= g_vasprintf(&path
->data
, fmt
, ap
) + 1;
193 void v9fs_path_copy(V9fsPath
*lhs
, V9fsPath
*rhs
)
196 lhs
->data
= g_malloc(rhs
->size
);
197 memcpy(lhs
->data
, rhs
->data
, rhs
->size
);
198 lhs
->size
= rhs
->size
;
201 int v9fs_name_to_path(V9fsState
*s
, V9fsPath
*dirpath
,
202 const char *name
, V9fsPath
*path
)
205 err
= s
->ops
->name_to_path(&s
->ctx
, dirpath
, name
, path
);
213 * Return TRUE if s1 is an ancestor of s2.
215 * E.g. "a/b" is an ancestor of "a/b/c" but not of "a/bc/d".
216 * As a special case, We treat s1 as ancestor of s2 if they are same!
218 static int v9fs_path_is_ancestor(V9fsPath
*s1
, V9fsPath
*s2
)
220 if (!strncmp(s1
->data
, s2
->data
, s1
->size
- 1)) {
221 if (s2
->data
[s1
->size
- 1] == '\0' || s2
->data
[s1
->size
- 1] == '/') {
228 static size_t v9fs_string_size(V9fsString
*str
)
234 * returns 0 if fid got re-opened, 1 if not, < 0 on error */
235 static int coroutine_fn
v9fs_reopen_fid(V9fsPDU
*pdu
, V9fsFidState
*f
)
238 if (f
->fid_type
== P9_FID_FILE
) {
239 if (f
->fs
.fd
== -1) {
241 err
= v9fs_co_open(pdu
, f
, f
->open_flags
);
242 } while (err
== -EINTR
&& !pdu
->cancelled
);
244 } else if (f
->fid_type
== P9_FID_DIR
) {
245 if (f
->fs
.dir
.stream
== NULL
) {
247 err
= v9fs_co_opendir(pdu
, f
);
248 } while (err
== -EINTR
&& !pdu
->cancelled
);
254 static V9fsFidState
*coroutine_fn
get_fid(V9fsPDU
*pdu
, int32_t fid
)
258 V9fsState
*s
= pdu
->s
;
260 for (f
= s
->fid_list
; f
; f
= f
->next
) {
264 * Update the fid ref upfront so that
265 * we don't get reclaimed when we yield
270 * check whether we need to reopen the
271 * file. We might have closed the fd
272 * while trying to free up some file
275 err
= v9fs_reopen_fid(pdu
, f
);
281 * Mark the fid as referenced so that the LRU
282 * reclaim won't close the file descriptor
284 f
->flags
|= FID_REFERENCED
;
291 static V9fsFidState
*alloc_fid(V9fsState
*s
, int32_t fid
)
295 for (f
= s
->fid_list
; f
; f
= f
->next
) {
296 /* If fid is already there return NULL */
302 f
= g_malloc0(sizeof(V9fsFidState
));
304 f
->fid_type
= P9_FID_NONE
;
307 * Mark the fid as referenced so that the LRU
308 * reclaim won't close the file descriptor
310 f
->flags
|= FID_REFERENCED
;
311 f
->next
= s
->fid_list
;
314 v9fs_readdir_init(&f
->fs
.dir
);
315 v9fs_readdir_init(&f
->fs_reclaim
.dir
);
320 static int coroutine_fn
v9fs_xattr_fid_clunk(V9fsPDU
*pdu
, V9fsFidState
*fidp
)
324 if (fidp
->fs
.xattr
.xattrwalk_fid
) {
325 /* getxattr/listxattr fid */
329 * if this is fid for setxattr. clunk should
330 * result in setxattr localcall
332 if (fidp
->fs
.xattr
.len
!= fidp
->fs
.xattr
.copied_len
) {
333 /* clunk after partial write */
337 if (fidp
->fs
.xattr
.len
) {
338 retval
= v9fs_co_lsetxattr(pdu
, &fidp
->path
, &fidp
->fs
.xattr
.name
,
339 fidp
->fs
.xattr
.value
,
341 fidp
->fs
.xattr
.flags
);
343 retval
= v9fs_co_lremovexattr(pdu
, &fidp
->path
, &fidp
->fs
.xattr
.name
);
346 v9fs_string_free(&fidp
->fs
.xattr
.name
);
348 g_free(fidp
->fs
.xattr
.value
);
352 static int coroutine_fn
free_fid(V9fsPDU
*pdu
, V9fsFidState
*fidp
)
356 if (fidp
->fid_type
== P9_FID_FILE
) {
357 /* If we reclaimed the fd no need to close */
358 if (fidp
->fs
.fd
!= -1) {
359 retval
= v9fs_co_close(pdu
, &fidp
->fs
);
361 } else if (fidp
->fid_type
== P9_FID_DIR
) {
362 if (fidp
->fs
.dir
.stream
!= NULL
) {
363 retval
= v9fs_co_closedir(pdu
, &fidp
->fs
);
365 } else if (fidp
->fid_type
== P9_FID_XATTR
) {
366 retval
= v9fs_xattr_fid_clunk(pdu
, fidp
);
368 v9fs_path_free(&fidp
->path
);
373 static int coroutine_fn
put_fid(V9fsPDU
*pdu
, V9fsFidState
*fidp
)
378 * Don't free the fid if it is in reclaim list
380 if (!fidp
->ref
&& fidp
->clunked
) {
381 if (fidp
->fid
== pdu
->s
->root_fid
) {
383 * if the clunked fid is root fid then we
384 * have unmounted the fs on the client side.
385 * delete the migration blocker. Ideally, this
386 * should be hooked to transport close notification
388 if (pdu
->s
->migration_blocker
) {
389 migrate_del_blocker(pdu
->s
->migration_blocker
);
390 error_free(pdu
->s
->migration_blocker
);
391 pdu
->s
->migration_blocker
= NULL
;
394 return free_fid(pdu
, fidp
);
399 static V9fsFidState
*clunk_fid(V9fsState
*s
, int32_t fid
)
401 V9fsFidState
**fidpp
, *fidp
;
403 for (fidpp
= &s
->fid_list
; *fidpp
; fidpp
= &(*fidpp
)->next
) {
404 if ((*fidpp
)->fid
== fid
) {
408 if (*fidpp
== NULL
) {
417 void coroutine_fn
v9fs_reclaim_fd(V9fsPDU
*pdu
)
419 int reclaim_count
= 0;
420 V9fsState
*s
= pdu
->s
;
421 V9fsFidState
*f
, *reclaim_list
= NULL
;
423 for (f
= s
->fid_list
; f
; f
= f
->next
) {
425 * Unlink fids cannot be reclaimed. Check
426 * for them and skip them. Also skip fids
427 * currently being operated on.
429 if (f
->ref
|| f
->flags
& FID_NON_RECLAIMABLE
) {
433 * if it is a recently referenced fid
434 * we leave the fid untouched and clear the
435 * reference bit. We come back to it later
436 * in the next iteration. (a simple LRU without
437 * moving list elements around)
439 if (f
->flags
& FID_REFERENCED
) {
440 f
->flags
&= ~FID_REFERENCED
;
444 * Add fids to reclaim list.
446 if (f
->fid_type
== P9_FID_FILE
) {
447 if (f
->fs
.fd
!= -1) {
449 * Up the reference count so that
450 * a clunk request won't free this fid
453 f
->rclm_lst
= reclaim_list
;
455 f
->fs_reclaim
.fd
= f
->fs
.fd
;
459 } else if (f
->fid_type
== P9_FID_DIR
) {
460 if (f
->fs
.dir
.stream
!= NULL
) {
462 * Up the reference count so that
463 * a clunk request won't free this fid
466 f
->rclm_lst
= reclaim_list
;
468 f
->fs_reclaim
.dir
.stream
= f
->fs
.dir
.stream
;
469 f
->fs
.dir
.stream
= NULL
;
473 if (reclaim_count
>= open_fd_rc
) {
478 * Now close the fid in reclaim list. Free them if they
479 * are already clunked.
481 while (reclaim_list
) {
483 reclaim_list
= f
->rclm_lst
;
484 if (f
->fid_type
== P9_FID_FILE
) {
485 v9fs_co_close(pdu
, &f
->fs_reclaim
);
486 } else if (f
->fid_type
== P9_FID_DIR
) {
487 v9fs_co_closedir(pdu
, &f
->fs_reclaim
);
491 * Now drop the fid reference, free it
498 static int coroutine_fn
v9fs_mark_fids_unreclaim(V9fsPDU
*pdu
, V9fsPath
*path
)
501 V9fsState
*s
= pdu
->s
;
502 V9fsFidState
*fidp
, head_fid
;
504 head_fid
.next
= s
->fid_list
;
505 for (fidp
= s
->fid_list
; fidp
; fidp
= fidp
->next
) {
506 if (fidp
->path
.size
!= path
->size
) {
509 if (!memcmp(fidp
->path
.data
, path
->data
, path
->size
)) {
510 /* Mark the fid non reclaimable. */
511 fidp
->flags
|= FID_NON_RECLAIMABLE
;
513 /* reopen the file/dir if already closed */
514 err
= v9fs_reopen_fid(pdu
, fidp
);
519 * Go back to head of fid list because
520 * the list could have got updated when
521 * switched to the worker thread
531 static void coroutine_fn
virtfs_reset(V9fsPDU
*pdu
)
533 V9fsState
*s
= pdu
->s
;
537 while (s
->fid_list
) {
543 s
->fid_list
= fidp
->next
;
550 #define P9_QID_TYPE_DIR 0x80
551 #define P9_QID_TYPE_SYMLINK 0x02
553 #define P9_STAT_MODE_DIR 0x80000000
554 #define P9_STAT_MODE_APPEND 0x40000000
555 #define P9_STAT_MODE_EXCL 0x20000000
556 #define P9_STAT_MODE_MOUNT 0x10000000
557 #define P9_STAT_MODE_AUTH 0x08000000
558 #define P9_STAT_MODE_TMP 0x04000000
559 #define P9_STAT_MODE_SYMLINK 0x02000000
560 #define P9_STAT_MODE_LINK 0x01000000
561 #define P9_STAT_MODE_DEVICE 0x00800000
562 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
563 #define P9_STAT_MODE_SOCKET 0x00100000
564 #define P9_STAT_MODE_SETUID 0x00080000
565 #define P9_STAT_MODE_SETGID 0x00040000
566 #define P9_STAT_MODE_SETVTX 0x00010000
568 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR | \
569 P9_STAT_MODE_SYMLINK | \
570 P9_STAT_MODE_LINK | \
571 P9_STAT_MODE_DEVICE | \
572 P9_STAT_MODE_NAMED_PIPE | \
575 /* This is the algorithm from ufs in spfs */
576 static void stat_to_qid(const struct stat
*stbuf
, V9fsQID
*qidp
)
580 memset(&qidp
->path
, 0, sizeof(qidp
->path
));
581 size
= MIN(sizeof(stbuf
->st_ino
), sizeof(qidp
->path
));
582 memcpy(&qidp
->path
, &stbuf
->st_ino
, size
);
583 qidp
->version
= stbuf
->st_mtime
^ (stbuf
->st_size
<< 8);
585 if (S_ISDIR(stbuf
->st_mode
)) {
586 qidp
->type
|= P9_QID_TYPE_DIR
;
588 if (S_ISLNK(stbuf
->st_mode
)) {
589 qidp
->type
|= P9_QID_TYPE_SYMLINK
;
593 static int coroutine_fn
fid_to_qid(V9fsPDU
*pdu
, V9fsFidState
*fidp
,
599 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
603 stat_to_qid(&stbuf
, qidp
);
607 V9fsPDU
*pdu_alloc(V9fsState
*s
)
611 if (!QLIST_EMPTY(&s
->free_list
)) {
612 pdu
= QLIST_FIRST(&s
->free_list
);
613 QLIST_REMOVE(pdu
, next
);
614 QLIST_INSERT_HEAD(&s
->active_list
, pdu
, next
);
619 void pdu_free(V9fsPDU
*pdu
)
621 V9fsState
*s
= pdu
->s
;
623 g_assert(!pdu
->cancelled
);
624 QLIST_REMOVE(pdu
, next
);
625 QLIST_INSERT_HEAD(&s
->free_list
, pdu
, next
);
628 static void coroutine_fn
pdu_complete(V9fsPDU
*pdu
, ssize_t len
)
630 int8_t id
= pdu
->id
+ 1; /* Response */
631 V9fsState
*s
= pdu
->s
;
635 * The 9p spec requires that successfully cancelled pdus receive no reply.
636 * Sending a reply would confuse clients because they would
637 * assume that any EINTR is the actual result of the operation,
638 * rather than a consequence of the cancellation. However, if
639 * the operation completed (succesfully or with an error other
640 * than caused be cancellation), we do send out that reply, both
641 * for efficiency and to avoid confusing the rest of the state machine
642 * that assumes passing a non-error here will mean a successful
643 * transmission of the reply.
645 bool discard
= pdu
->cancelled
&& len
== -EINTR
;
647 trace_v9fs_rcancel(pdu
->tag
, pdu
->id
);
656 if (s
->proto_version
!= V9FS_PROTO_2000L
) {
659 str
.data
= strerror(err
);
660 str
.size
= strlen(str
.data
);
662 ret
= pdu_marshal(pdu
, len
, "s", &str
);
670 ret
= pdu_marshal(pdu
, len
, "d", err
);
676 if (s
->proto_version
== V9FS_PROTO_2000L
) {
679 trace_v9fs_rerror(pdu
->tag
, pdu
->id
, err
); /* Trace ERROR */
682 /* fill out the header */
683 if (pdu_marshal(pdu
, 0, "dbw", (int32_t)len
, id
, pdu
->tag
) < 0) {
687 /* keep these in sync */
692 pdu
->s
->transport
->push_and_notify(pdu
);
694 /* Now wakeup anybody waiting in flush for this request */
695 if (!qemu_co_queue_next(&pdu
->complete
)) {
700 static mode_t
v9mode_to_mode(uint32_t mode
, V9fsString
*extension
)
705 if (mode
& P9_STAT_MODE_DIR
) {
709 if (mode
& P9_STAT_MODE_SYMLINK
) {
712 if (mode
& P9_STAT_MODE_SOCKET
) {
715 if (mode
& P9_STAT_MODE_NAMED_PIPE
) {
718 if (mode
& P9_STAT_MODE_DEVICE
) {
719 if (extension
->size
&& extension
->data
[0] == 'c') {
730 if (mode
& P9_STAT_MODE_SETUID
) {
733 if (mode
& P9_STAT_MODE_SETGID
) {
736 if (mode
& P9_STAT_MODE_SETVTX
) {
743 static int donttouch_stat(V9fsStat
*stat
)
745 if (stat
->type
== -1 &&
747 stat
->qid
.type
== -1 &&
748 stat
->qid
.version
== -1 &&
749 stat
->qid
.path
== -1 &&
753 stat
->length
== -1 &&
760 stat
->n_muid
== -1) {
767 static void v9fs_stat_init(V9fsStat
*stat
)
769 v9fs_string_init(&stat
->name
);
770 v9fs_string_init(&stat
->uid
);
771 v9fs_string_init(&stat
->gid
);
772 v9fs_string_init(&stat
->muid
);
773 v9fs_string_init(&stat
->extension
);
776 static void v9fs_stat_free(V9fsStat
*stat
)
778 v9fs_string_free(&stat
->name
);
779 v9fs_string_free(&stat
->uid
);
780 v9fs_string_free(&stat
->gid
);
781 v9fs_string_free(&stat
->muid
);
782 v9fs_string_free(&stat
->extension
);
785 static uint32_t stat_to_v9mode(const struct stat
*stbuf
)
789 mode
= stbuf
->st_mode
& 0777;
790 if (S_ISDIR(stbuf
->st_mode
)) {
791 mode
|= P9_STAT_MODE_DIR
;
794 if (S_ISLNK(stbuf
->st_mode
)) {
795 mode
|= P9_STAT_MODE_SYMLINK
;
798 if (S_ISSOCK(stbuf
->st_mode
)) {
799 mode
|= P9_STAT_MODE_SOCKET
;
802 if (S_ISFIFO(stbuf
->st_mode
)) {
803 mode
|= P9_STAT_MODE_NAMED_PIPE
;
806 if (S_ISBLK(stbuf
->st_mode
) || S_ISCHR(stbuf
->st_mode
)) {
807 mode
|= P9_STAT_MODE_DEVICE
;
810 if (stbuf
->st_mode
& S_ISUID
) {
811 mode
|= P9_STAT_MODE_SETUID
;
814 if (stbuf
->st_mode
& S_ISGID
) {
815 mode
|= P9_STAT_MODE_SETGID
;
818 if (stbuf
->st_mode
& S_ISVTX
) {
819 mode
|= P9_STAT_MODE_SETVTX
;
825 static int coroutine_fn
stat_to_v9stat(V9fsPDU
*pdu
, V9fsPath
*path
,
826 const char *basename
,
827 const struct stat
*stbuf
,
832 memset(v9stat
, 0, sizeof(*v9stat
));
834 stat_to_qid(stbuf
, &v9stat
->qid
);
835 v9stat
->mode
= stat_to_v9mode(stbuf
);
836 v9stat
->atime
= stbuf
->st_atime
;
837 v9stat
->mtime
= stbuf
->st_mtime
;
838 v9stat
->length
= stbuf
->st_size
;
840 v9fs_string_free(&v9stat
->uid
);
841 v9fs_string_free(&v9stat
->gid
);
842 v9fs_string_free(&v9stat
->muid
);
844 v9stat
->n_uid
= stbuf
->st_uid
;
845 v9stat
->n_gid
= stbuf
->st_gid
;
848 v9fs_string_free(&v9stat
->extension
);
850 if (v9stat
->mode
& P9_STAT_MODE_SYMLINK
) {
851 err
= v9fs_co_readlink(pdu
, path
, &v9stat
->extension
);
855 } else if (v9stat
->mode
& P9_STAT_MODE_DEVICE
) {
856 v9fs_string_sprintf(&v9stat
->extension
, "%c %u %u",
857 S_ISCHR(stbuf
->st_mode
) ? 'c' : 'b',
858 major(stbuf
->st_rdev
), minor(stbuf
->st_rdev
));
859 } else if (S_ISDIR(stbuf
->st_mode
) || S_ISREG(stbuf
->st_mode
)) {
860 v9fs_string_sprintf(&v9stat
->extension
, "%s %lu",
861 "HARDLINKCOUNT", (unsigned long)stbuf
->st_nlink
);
864 v9fs_string_sprintf(&v9stat
->name
, "%s", basename
);
867 v9fs_string_size(&v9stat
->name
) +
868 v9fs_string_size(&v9stat
->uid
) +
869 v9fs_string_size(&v9stat
->gid
) +
870 v9fs_string_size(&v9stat
->muid
) +
871 v9fs_string_size(&v9stat
->extension
);
875 #define P9_STATS_MODE 0x00000001ULL
876 #define P9_STATS_NLINK 0x00000002ULL
877 #define P9_STATS_UID 0x00000004ULL
878 #define P9_STATS_GID 0x00000008ULL
879 #define P9_STATS_RDEV 0x00000010ULL
880 #define P9_STATS_ATIME 0x00000020ULL
881 #define P9_STATS_MTIME 0x00000040ULL
882 #define P9_STATS_CTIME 0x00000080ULL
883 #define P9_STATS_INO 0x00000100ULL
884 #define P9_STATS_SIZE 0x00000200ULL
885 #define P9_STATS_BLOCKS 0x00000400ULL
887 #define P9_STATS_BTIME 0x00000800ULL
888 #define P9_STATS_GEN 0x00001000ULL
889 #define P9_STATS_DATA_VERSION 0x00002000ULL
891 #define P9_STATS_BASIC 0x000007ffULL /* Mask for fields up to BLOCKS */
892 #define P9_STATS_ALL 0x00003fffULL /* Mask for All fields above */
895 static void stat_to_v9stat_dotl(V9fsState
*s
, const struct stat
*stbuf
,
896 V9fsStatDotl
*v9lstat
)
898 memset(v9lstat
, 0, sizeof(*v9lstat
));
900 v9lstat
->st_mode
= stbuf
->st_mode
;
901 v9lstat
->st_nlink
= stbuf
->st_nlink
;
902 v9lstat
->st_uid
= stbuf
->st_uid
;
903 v9lstat
->st_gid
= stbuf
->st_gid
;
904 v9lstat
->st_rdev
= stbuf
->st_rdev
;
905 v9lstat
->st_size
= stbuf
->st_size
;
906 v9lstat
->st_blksize
= stbuf
->st_blksize
;
907 v9lstat
->st_blocks
= stbuf
->st_blocks
;
908 v9lstat
->st_atime_sec
= stbuf
->st_atime
;
909 v9lstat
->st_atime_nsec
= stbuf
->st_atim
.tv_nsec
;
910 v9lstat
->st_mtime_sec
= stbuf
->st_mtime
;
911 v9lstat
->st_mtime_nsec
= stbuf
->st_mtim
.tv_nsec
;
912 v9lstat
->st_ctime_sec
= stbuf
->st_ctime
;
913 v9lstat
->st_ctime_nsec
= stbuf
->st_ctim
.tv_nsec
;
914 /* Currently we only support BASIC fields in stat */
915 v9lstat
->st_result_mask
= P9_STATS_BASIC
;
917 stat_to_qid(stbuf
, &v9lstat
->qid
);
920 static void print_sg(struct iovec
*sg
, int cnt
)
924 printf("sg[%d]: {", cnt
);
925 for (i
= 0; i
< cnt
; i
++) {
929 printf("(%p, %zd)", sg
[i
].iov_base
, sg
[i
].iov_len
);
934 /* Will call this only for path name based fid */
935 static void v9fs_fix_path(V9fsPath
*dst
, V9fsPath
*src
, int len
)
938 v9fs_path_init(&str
);
939 v9fs_path_copy(&str
, dst
);
940 v9fs_path_sprintf(dst
, "%s%s", src
->data
, str
.data
+ len
);
941 v9fs_path_free(&str
);
944 static inline bool is_ro_export(FsContext
*ctx
)
946 return ctx
->export_flags
& V9FS_RDONLY
;
949 static void coroutine_fn
v9fs_version(void *opaque
)
952 V9fsPDU
*pdu
= opaque
;
953 V9fsState
*s
= pdu
->s
;
957 v9fs_string_init(&version
);
958 err
= pdu_unmarshal(pdu
, offset
, "ds", &s
->msize
, &version
);
962 trace_v9fs_version(pdu
->tag
, pdu
->id
, s
->msize
, version
.data
);
966 if (!strcmp(version
.data
, "9P2000.u")) {
967 s
->proto_version
= V9FS_PROTO_2000U
;
968 } else if (!strcmp(version
.data
, "9P2000.L")) {
969 s
->proto_version
= V9FS_PROTO_2000L
;
971 v9fs_string_sprintf(&version
, "unknown");
974 err
= pdu_marshal(pdu
, offset
, "ds", s
->msize
, &version
);
979 trace_v9fs_version_return(pdu
->tag
, pdu
->id
, s
->msize
, version
.data
);
981 pdu_complete(pdu
, err
);
982 v9fs_string_free(&version
);
985 static void coroutine_fn
v9fs_attach(void *opaque
)
987 V9fsPDU
*pdu
= opaque
;
988 V9fsState
*s
= pdu
->s
;
989 int32_t fid
, afid
, n_uname
;
990 V9fsString uname
, aname
;
995 Error
*local_err
= NULL
;
997 v9fs_string_init(&uname
);
998 v9fs_string_init(&aname
);
999 err
= pdu_unmarshal(pdu
, offset
, "ddssd", &fid
,
1000 &afid
, &uname
, &aname
, &n_uname
);
1004 trace_v9fs_attach(pdu
->tag
, pdu
->id
, fid
, afid
, uname
.data
, aname
.data
);
1006 fidp
= alloc_fid(s
, fid
);
1011 fidp
->uid
= n_uname
;
1012 err
= v9fs_co_name_to_path(pdu
, NULL
, "/", &fidp
->path
);
1018 err
= fid_to_qid(pdu
, fidp
, &qid
);
1026 * disable migration if we haven't done already.
1027 * attach could get called multiple times for the same export.
1029 if (!s
->migration_blocker
) {
1030 error_setg(&s
->migration_blocker
,
1031 "Migration is disabled when VirtFS export path '%s' is mounted in the guest using mount_tag '%s'",
1032 s
->ctx
.fs_root
? s
->ctx
.fs_root
: "NULL", s
->tag
);
1033 err
= migrate_add_blocker(s
->migration_blocker
, &local_err
);
1035 error_free(local_err
);
1036 error_free(s
->migration_blocker
);
1037 s
->migration_blocker
= NULL
;
1044 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
1051 memcpy(&s
->root_qid
, &qid
, sizeof(qid
));
1052 trace_v9fs_attach_return(pdu
->tag
, pdu
->id
,
1053 qid
.type
, qid
.version
, qid
.path
);
1057 pdu_complete(pdu
, err
);
1058 v9fs_string_free(&uname
);
1059 v9fs_string_free(&aname
);
1062 static void coroutine_fn
v9fs_stat(void *opaque
)
1070 V9fsPDU
*pdu
= opaque
;
1073 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
1077 trace_v9fs_stat(pdu
->tag
, pdu
->id
, fid
);
1079 fidp
= get_fid(pdu
, fid
);
1084 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1088 basename
= g_path_get_basename(fidp
->path
.data
);
1089 err
= stat_to_v9stat(pdu
, &fidp
->path
, basename
, &stbuf
, &v9stat
);
1094 err
= pdu_marshal(pdu
, offset
, "wS", 0, &v9stat
);
1096 v9fs_stat_free(&v9stat
);
1099 trace_v9fs_stat_return(pdu
->tag
, pdu
->id
, v9stat
.mode
,
1100 v9stat
.atime
, v9stat
.mtime
, v9stat
.length
);
1102 v9fs_stat_free(&v9stat
);
1106 pdu_complete(pdu
, err
);
1109 static void coroutine_fn
v9fs_getattr(void *opaque
)
1116 uint64_t request_mask
;
1117 V9fsStatDotl v9stat_dotl
;
1118 V9fsPDU
*pdu
= opaque
;
1119 V9fsState
*s
= pdu
->s
;
1121 retval
= pdu_unmarshal(pdu
, offset
, "dq", &fid
, &request_mask
);
1125 trace_v9fs_getattr(pdu
->tag
, pdu
->id
, fid
, request_mask
);
1127 fidp
= get_fid(pdu
, fid
);
1133 * Currently we only support BASIC fields in stat, so there is no
1134 * need to look at request_mask.
1136 retval
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1140 stat_to_v9stat_dotl(s
, &stbuf
, &v9stat_dotl
);
1142 /* fill st_gen if requested and supported by underlying fs */
1143 if (request_mask
& P9_STATS_GEN
) {
1144 retval
= v9fs_co_st_gen(pdu
, &fidp
->path
, stbuf
.st_mode
, &v9stat_dotl
);
1147 /* we have valid st_gen: update result mask */
1148 v9stat_dotl
.st_result_mask
|= P9_STATS_GEN
;
1151 /* request cancelled, e.g. by Tflush */
1154 /* failed to get st_gen: not fatal, ignore */
1158 retval
= pdu_marshal(pdu
, offset
, "A", &v9stat_dotl
);
1163 trace_v9fs_getattr_return(pdu
->tag
, pdu
->id
, v9stat_dotl
.st_result_mask
,
1164 v9stat_dotl
.st_mode
, v9stat_dotl
.st_uid
,
1165 v9stat_dotl
.st_gid
);
1169 pdu_complete(pdu
, retval
);
1172 /* Attribute flags */
1173 #define P9_ATTR_MODE (1 << 0)
1174 #define P9_ATTR_UID (1 << 1)
1175 #define P9_ATTR_GID (1 << 2)
1176 #define P9_ATTR_SIZE (1 << 3)
1177 #define P9_ATTR_ATIME (1 << 4)
1178 #define P9_ATTR_MTIME (1 << 5)
1179 #define P9_ATTR_CTIME (1 << 6)
1180 #define P9_ATTR_ATIME_SET (1 << 7)
1181 #define P9_ATTR_MTIME_SET (1 << 8)
1183 #define P9_ATTR_MASK 127
1185 static void coroutine_fn
v9fs_setattr(void *opaque
)
1192 V9fsPDU
*pdu
= opaque
;
1194 err
= pdu_unmarshal(pdu
, offset
, "dI", &fid
, &v9iattr
);
1199 fidp
= get_fid(pdu
, fid
);
1204 if (v9iattr
.valid
& P9_ATTR_MODE
) {
1205 err
= v9fs_co_chmod(pdu
, &fidp
->path
, v9iattr
.mode
);
1210 if (v9iattr
.valid
& (P9_ATTR_ATIME
| P9_ATTR_MTIME
)) {
1211 struct timespec times
[2];
1212 if (v9iattr
.valid
& P9_ATTR_ATIME
) {
1213 if (v9iattr
.valid
& P9_ATTR_ATIME_SET
) {
1214 times
[0].tv_sec
= v9iattr
.atime_sec
;
1215 times
[0].tv_nsec
= v9iattr
.atime_nsec
;
1217 times
[0].tv_nsec
= UTIME_NOW
;
1220 times
[0].tv_nsec
= UTIME_OMIT
;
1222 if (v9iattr
.valid
& P9_ATTR_MTIME
) {
1223 if (v9iattr
.valid
& P9_ATTR_MTIME_SET
) {
1224 times
[1].tv_sec
= v9iattr
.mtime_sec
;
1225 times
[1].tv_nsec
= v9iattr
.mtime_nsec
;
1227 times
[1].tv_nsec
= UTIME_NOW
;
1230 times
[1].tv_nsec
= UTIME_OMIT
;
1232 err
= v9fs_co_utimensat(pdu
, &fidp
->path
, times
);
1238 * If the only valid entry in iattr is ctime we can call
1239 * chown(-1,-1) to update the ctime of the file
1241 if ((v9iattr
.valid
& (P9_ATTR_UID
| P9_ATTR_GID
)) ||
1242 ((v9iattr
.valid
& P9_ATTR_CTIME
)
1243 && !((v9iattr
.valid
& P9_ATTR_MASK
) & ~P9_ATTR_CTIME
))) {
1244 if (!(v9iattr
.valid
& P9_ATTR_UID
)) {
1247 if (!(v9iattr
.valid
& P9_ATTR_GID
)) {
1250 err
= v9fs_co_chown(pdu
, &fidp
->path
, v9iattr
.uid
,
1256 if (v9iattr
.valid
& (P9_ATTR_SIZE
)) {
1257 err
= v9fs_co_truncate(pdu
, &fidp
->path
, v9iattr
.size
);
1266 pdu_complete(pdu
, err
);
1269 static int v9fs_walk_marshal(V9fsPDU
*pdu
, uint16_t nwnames
, V9fsQID
*qids
)
1275 err
= pdu_marshal(pdu
, offset
, "w", nwnames
);
1280 for (i
= 0; i
< nwnames
; i
++) {
1281 err
= pdu_marshal(pdu
, offset
, "Q", &qids
[i
]);
1290 static bool name_is_illegal(const char *name
)
1292 return !*name
|| strchr(name
, '/') != NULL
;
1295 static bool not_same_qid(const V9fsQID
*qid1
, const V9fsQID
*qid2
)
1298 qid1
->type
!= qid2
->type
||
1299 qid1
->version
!= qid2
->version
||
1300 qid1
->path
!= qid2
->path
;
1303 static void coroutine_fn
v9fs_walk(void *opaque
)
1306 V9fsQID
*qids
= NULL
;
1308 V9fsPath dpath
, path
;
1312 int32_t fid
, newfid
;
1313 V9fsString
*wnames
= NULL
;
1315 V9fsFidState
*newfidp
= NULL
;
1316 V9fsPDU
*pdu
= opaque
;
1317 V9fsState
*s
= pdu
->s
;
1320 err
= pdu_unmarshal(pdu
, offset
, "ddw", &fid
, &newfid
, &nwnames
);
1322 pdu_complete(pdu
, err
);
1327 trace_v9fs_walk(pdu
->tag
, pdu
->id
, fid
, newfid
, nwnames
);
1329 if (nwnames
&& nwnames
<= P9_MAXWELEM
) {
1330 wnames
= g_malloc0(sizeof(wnames
[0]) * nwnames
);
1331 qids
= g_malloc0(sizeof(qids
[0]) * nwnames
);
1332 for (i
= 0; i
< nwnames
; i
++) {
1333 err
= pdu_unmarshal(pdu
, offset
, "s", &wnames
[i
]);
1337 if (name_is_illegal(wnames
[i
].data
)) {
1343 } else if (nwnames
> P9_MAXWELEM
) {
1347 fidp
= get_fid(pdu
, fid
);
1353 v9fs_path_init(&dpath
);
1354 v9fs_path_init(&path
);
1356 err
= fid_to_qid(pdu
, fidp
, &qid
);
1362 * Both dpath and path initially poin to fidp.
1363 * Needed to handle request with nwnames == 0
1365 v9fs_path_copy(&dpath
, &fidp
->path
);
1366 v9fs_path_copy(&path
, &fidp
->path
);
1367 for (name_idx
= 0; name_idx
< nwnames
; name_idx
++) {
1368 if (not_same_qid(&pdu
->s
->root_qid
, &qid
) ||
1369 strcmp("..", wnames
[name_idx
].data
)) {
1370 err
= v9fs_co_name_to_path(pdu
, &dpath
, wnames
[name_idx
].data
,
1376 err
= v9fs_co_lstat(pdu
, &path
, &stbuf
);
1380 stat_to_qid(&stbuf
, &qid
);
1381 v9fs_path_copy(&dpath
, &path
);
1383 memcpy(&qids
[name_idx
], &qid
, sizeof(qid
));
1385 if (fid
== newfid
) {
1386 if (fidp
->fid_type
!= P9_FID_NONE
) {
1390 v9fs_path_copy(&fidp
->path
, &path
);
1392 newfidp
= alloc_fid(s
, newfid
);
1393 if (newfidp
== NULL
) {
1397 newfidp
->uid
= fidp
->uid
;
1398 v9fs_path_copy(&newfidp
->path
, &path
);
1400 err
= v9fs_walk_marshal(pdu
, nwnames
, qids
);
1401 trace_v9fs_walk_return(pdu
->tag
, pdu
->id
, nwnames
, qids
);
1405 put_fid(pdu
, newfidp
);
1407 v9fs_path_free(&dpath
);
1408 v9fs_path_free(&path
);
1410 pdu_complete(pdu
, err
);
1411 if (nwnames
&& nwnames
<= P9_MAXWELEM
) {
1412 for (name_idx
= 0; name_idx
< nwnames
; name_idx
++) {
1413 v9fs_string_free(&wnames
[name_idx
]);
1420 static int32_t coroutine_fn
get_iounit(V9fsPDU
*pdu
, V9fsPath
*path
)
1422 struct statfs stbuf
;
1424 V9fsState
*s
= pdu
->s
;
1427 * iounit should be multiples of f_bsize (host filesystem block size
1428 * and as well as less than (client msize - P9_IOHDRSZ))
1430 if (!v9fs_co_statfs(pdu
, path
, &stbuf
)) {
1431 iounit
= stbuf
.f_bsize
;
1432 iounit
*= (s
->msize
- P9_IOHDRSZ
)/stbuf
.f_bsize
;
1435 iounit
= s
->msize
- P9_IOHDRSZ
;
1440 static void coroutine_fn
v9fs_open(void *opaque
)
1451 V9fsPDU
*pdu
= opaque
;
1452 V9fsState
*s
= pdu
->s
;
1454 if (s
->proto_version
== V9FS_PROTO_2000L
) {
1455 err
= pdu_unmarshal(pdu
, offset
, "dd", &fid
, &mode
);
1458 err
= pdu_unmarshal(pdu
, offset
, "db", &fid
, &modebyte
);
1464 trace_v9fs_open(pdu
->tag
, pdu
->id
, fid
, mode
);
1466 fidp
= get_fid(pdu
, fid
);
1471 if (fidp
->fid_type
!= P9_FID_NONE
) {
1476 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1480 stat_to_qid(&stbuf
, &qid
);
1481 if (S_ISDIR(stbuf
.st_mode
)) {
1482 err
= v9fs_co_opendir(pdu
, fidp
);
1486 fidp
->fid_type
= P9_FID_DIR
;
1487 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, 0);
1493 if (s
->proto_version
== V9FS_PROTO_2000L
) {
1494 flags
= get_dotl_openflags(s
, mode
);
1496 flags
= omode_to_uflags(mode
);
1498 if (is_ro_export(&s
->ctx
)) {
1499 if (mode
& O_WRONLY
|| mode
& O_RDWR
||
1500 mode
& O_APPEND
|| mode
& O_TRUNC
) {
1505 err
= v9fs_co_open(pdu
, fidp
, flags
);
1509 fidp
->fid_type
= P9_FID_FILE
;
1510 fidp
->open_flags
= flags
;
1511 if (flags
& O_EXCL
) {
1513 * We let the host file system do O_EXCL check
1514 * We should not reclaim such fd
1516 fidp
->flags
|= FID_NON_RECLAIMABLE
;
1518 iounit
= get_iounit(pdu
, &fidp
->path
);
1519 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
1525 trace_v9fs_open_return(pdu
->tag
, pdu
->id
,
1526 qid
.type
, qid
.version
, qid
.path
, iounit
);
1530 pdu_complete(pdu
, err
);
1533 static void coroutine_fn
v9fs_lcreate(void *opaque
)
1535 int32_t dfid
, flags
, mode
;
1544 V9fsPDU
*pdu
= opaque
;
1546 v9fs_string_init(&name
);
1547 err
= pdu_unmarshal(pdu
, offset
, "dsddd", &dfid
,
1548 &name
, &flags
, &mode
, &gid
);
1552 trace_v9fs_lcreate(pdu
->tag
, pdu
->id
, dfid
, flags
, mode
, gid
);
1554 if (name_is_illegal(name
.data
)) {
1559 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
1564 fidp
= get_fid(pdu
, dfid
);
1569 if (fidp
->fid_type
!= P9_FID_NONE
) {
1574 flags
= get_dotl_openflags(pdu
->s
, flags
);
1575 err
= v9fs_co_open2(pdu
, fidp
, &name
, gid
,
1576 flags
| O_CREAT
, mode
, &stbuf
);
1580 fidp
->fid_type
= P9_FID_FILE
;
1581 fidp
->open_flags
= flags
;
1582 if (flags
& O_EXCL
) {
1584 * We let the host file system do O_EXCL check
1585 * We should not reclaim such fd
1587 fidp
->flags
|= FID_NON_RECLAIMABLE
;
1589 iounit
= get_iounit(pdu
, &fidp
->path
);
1590 stat_to_qid(&stbuf
, &qid
);
1591 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
1596 trace_v9fs_lcreate_return(pdu
->tag
, pdu
->id
,
1597 qid
.type
, qid
.version
, qid
.path
, iounit
);
1601 pdu_complete(pdu
, err
);
1602 v9fs_string_free(&name
);
1605 static void coroutine_fn
v9fs_fsync(void *opaque
)
1612 V9fsPDU
*pdu
= opaque
;
1614 err
= pdu_unmarshal(pdu
, offset
, "dd", &fid
, &datasync
);
1618 trace_v9fs_fsync(pdu
->tag
, pdu
->id
, fid
, datasync
);
1620 fidp
= get_fid(pdu
, fid
);
1625 err
= v9fs_co_fsync(pdu
, fidp
, datasync
);
1631 pdu_complete(pdu
, err
);
1634 static void coroutine_fn
v9fs_clunk(void *opaque
)
1640 V9fsPDU
*pdu
= opaque
;
1641 V9fsState
*s
= pdu
->s
;
1643 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
1647 trace_v9fs_clunk(pdu
->tag
, pdu
->id
, fid
);
1649 fidp
= clunk_fid(s
, fid
);
1655 * Bump the ref so that put_fid will
1659 err
= put_fid(pdu
, fidp
);
1664 pdu_complete(pdu
, err
);
1668 * Create a QEMUIOVector for a sub-region of PDU iovecs
1670 * @qiov: uninitialized QEMUIOVector
1671 * @skip: number of bytes to skip from beginning of PDU
1672 * @size: number of bytes to include
1673 * @is_write: true - write, false - read
1675 * The resulting QEMUIOVector has heap-allocated iovecs and must be cleaned up
1676 * with qemu_iovec_destroy().
1678 static void v9fs_init_qiov_from_pdu(QEMUIOVector
*qiov
, V9fsPDU
*pdu
,
1679 size_t skip
, size_t size
,
1687 pdu
->s
->transport
->init_out_iov_from_pdu(pdu
, &iov
, &niov
, size
+ skip
);
1689 pdu
->s
->transport
->init_in_iov_from_pdu(pdu
, &iov
, &niov
, size
+ skip
);
1692 qemu_iovec_init_external(&elem
, iov
, niov
);
1693 qemu_iovec_init(qiov
, niov
);
1694 qemu_iovec_concat(qiov
, &elem
, skip
, size
);
1697 static int v9fs_xattr_read(V9fsState
*s
, V9fsPDU
*pdu
, V9fsFidState
*fidp
,
1698 uint64_t off
, uint32_t max_count
)
1702 uint64_t read_count
;
1703 QEMUIOVector qiov_full
;
1705 if (fidp
->fs
.xattr
.len
< off
) {
1708 read_count
= fidp
->fs
.xattr
.len
- off
;
1710 if (read_count
> max_count
) {
1711 read_count
= max_count
;
1713 err
= pdu_marshal(pdu
, offset
, "d", read_count
);
1719 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
, read_count
, false);
1720 err
= v9fs_pack(qiov_full
.iov
, qiov_full
.niov
, 0,
1721 ((char *)fidp
->fs
.xattr
.value
) + off
,
1723 qemu_iovec_destroy(&qiov_full
);
1731 static int coroutine_fn
v9fs_do_readdir_with_stat(V9fsPDU
*pdu
,
1740 off_t saved_dir_pos
;
1741 struct dirent
*dent
;
1743 /* save the directory position */
1744 saved_dir_pos
= v9fs_co_telldir(pdu
, fidp
);
1745 if (saved_dir_pos
< 0) {
1746 return saved_dir_pos
;
1750 v9fs_path_init(&path
);
1752 v9fs_readdir_lock(&fidp
->fs
.dir
);
1754 err
= v9fs_co_readdir(pdu
, fidp
, &dent
);
1758 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, dent
->d_name
, &path
);
1762 err
= v9fs_co_lstat(pdu
, &path
, &stbuf
);
1766 err
= stat_to_v9stat(pdu
, &path
, dent
->d_name
, &stbuf
, &v9stat
);
1770 if ((count
+ v9stat
.size
+ 2) > max_count
) {
1771 v9fs_readdir_unlock(&fidp
->fs
.dir
);
1773 /* Ran out of buffer. Set dir back to old position and return */
1774 v9fs_co_seekdir(pdu
, fidp
, saved_dir_pos
);
1775 v9fs_stat_free(&v9stat
);
1776 v9fs_path_free(&path
);
1780 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
1781 len
= pdu_marshal(pdu
, 11 + count
, "S", &v9stat
);
1783 v9fs_readdir_unlock(&fidp
->fs
.dir
);
1786 v9fs_co_seekdir(pdu
, fidp
, saved_dir_pos
);
1787 v9fs_stat_free(&v9stat
);
1788 v9fs_path_free(&path
);
1792 v9fs_stat_free(&v9stat
);
1793 v9fs_path_free(&path
);
1794 saved_dir_pos
= dent
->d_off
;
1797 v9fs_readdir_unlock(&fidp
->fs
.dir
);
1799 v9fs_path_free(&path
);
1806 static void coroutine_fn
v9fs_read(void *opaque
)
1815 V9fsPDU
*pdu
= opaque
;
1816 V9fsState
*s
= pdu
->s
;
1818 err
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
, &off
, &max_count
);
1822 trace_v9fs_read(pdu
->tag
, pdu
->id
, fid
, off
, max_count
);
1824 fidp
= get_fid(pdu
, fid
);
1829 if (fidp
->fid_type
== P9_FID_DIR
) {
1832 v9fs_co_rewinddir(pdu
, fidp
);
1834 count
= v9fs_do_readdir_with_stat(pdu
, fidp
, max_count
);
1839 err
= pdu_marshal(pdu
, offset
, "d", count
);
1843 err
+= offset
+ count
;
1844 } else if (fidp
->fid_type
== P9_FID_FILE
) {
1845 QEMUIOVector qiov_full
;
1849 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
+ 4, max_count
, false);
1850 qemu_iovec_init(&qiov
, qiov_full
.niov
);
1852 qemu_iovec_reset(&qiov
);
1853 qemu_iovec_concat(&qiov
, &qiov_full
, count
, qiov_full
.size
- count
);
1855 print_sg(qiov
.iov
, qiov
.niov
);
1857 /* Loop in case of EINTR */
1859 len
= v9fs_co_preadv(pdu
, fidp
, qiov
.iov
, qiov
.niov
, off
);
1864 } while (len
== -EINTR
&& !pdu
->cancelled
);
1866 /* IO error return the error */
1868 goto out_free_iovec
;
1870 } while (count
< max_count
&& len
> 0);
1871 err
= pdu_marshal(pdu
, offset
, "d", count
);
1873 goto out_free_iovec
;
1875 err
+= offset
+ count
;
1877 qemu_iovec_destroy(&qiov
);
1878 qemu_iovec_destroy(&qiov_full
);
1879 } else if (fidp
->fid_type
== P9_FID_XATTR
) {
1880 err
= v9fs_xattr_read(s
, pdu
, fidp
, off
, max_count
);
1884 trace_v9fs_read_return(pdu
->tag
, pdu
->id
, count
, err
);
1888 pdu_complete(pdu
, err
);
1891 static size_t v9fs_readdir_data_size(V9fsString
*name
)
1894 * Size of each dirent on the wire: size of qid (13) + size of offset (8)
1895 * size of type (1) + size of name.size (2) + strlen(name.data)
1897 return 24 + v9fs_string_size(name
);
1900 static int coroutine_fn
v9fs_do_readdir(V9fsPDU
*pdu
, V9fsFidState
*fidp
,
1908 off_t saved_dir_pos
;
1909 struct dirent
*dent
;
1911 /* save the directory position */
1912 saved_dir_pos
= v9fs_co_telldir(pdu
, fidp
);
1913 if (saved_dir_pos
< 0) {
1914 return saved_dir_pos
;
1918 v9fs_readdir_lock(&fidp
->fs
.dir
);
1920 err
= v9fs_co_readdir(pdu
, fidp
, &dent
);
1924 v9fs_string_init(&name
);
1925 v9fs_string_sprintf(&name
, "%s", dent
->d_name
);
1926 if ((count
+ v9fs_readdir_data_size(&name
)) > max_count
) {
1927 v9fs_readdir_unlock(&fidp
->fs
.dir
);
1929 /* Ran out of buffer. Set dir back to old position and return */
1930 v9fs_co_seekdir(pdu
, fidp
, saved_dir_pos
);
1931 v9fs_string_free(&name
);
1935 * Fill up just the path field of qid because the client uses
1936 * only that. To fill the entire qid structure we will have
1937 * to stat each dirent found, which is expensive
1939 size
= MIN(sizeof(dent
->d_ino
), sizeof(qid
.path
));
1940 memcpy(&qid
.path
, &dent
->d_ino
, size
);
1941 /* Fill the other fields with dummy values */
1945 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
1946 len
= pdu_marshal(pdu
, 11 + count
, "Qqbs",
1948 dent
->d_type
, &name
);
1950 v9fs_readdir_unlock(&fidp
->fs
.dir
);
1953 v9fs_co_seekdir(pdu
, fidp
, saved_dir_pos
);
1954 v9fs_string_free(&name
);
1958 v9fs_string_free(&name
);
1959 saved_dir_pos
= dent
->d_off
;
1962 v9fs_readdir_unlock(&fidp
->fs
.dir
);
1970 static void coroutine_fn
v9fs_readdir(void *opaque
)
1976 uint64_t initial_offset
;
1979 V9fsPDU
*pdu
= opaque
;
1981 retval
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
,
1982 &initial_offset
, &max_count
);
1986 trace_v9fs_readdir(pdu
->tag
, pdu
->id
, fid
, initial_offset
, max_count
);
1988 fidp
= get_fid(pdu
, fid
);
1993 if (!fidp
->fs
.dir
.stream
) {
1997 if (initial_offset
== 0) {
1998 v9fs_co_rewinddir(pdu
, fidp
);
2000 v9fs_co_seekdir(pdu
, fidp
, initial_offset
);
2002 count
= v9fs_do_readdir(pdu
, fidp
, max_count
);
2007 retval
= pdu_marshal(pdu
, offset
, "d", count
);
2011 retval
+= count
+ offset
;
2012 trace_v9fs_readdir_return(pdu
->tag
, pdu
->id
, count
, retval
);
2016 pdu_complete(pdu
, retval
);
2019 static int v9fs_xattr_write(V9fsState
*s
, V9fsPDU
*pdu
, V9fsFidState
*fidp
,
2020 uint64_t off
, uint32_t count
,
2021 struct iovec
*sg
, int cnt
)
2025 uint64_t write_count
;
2029 if (fidp
->fs
.xattr
.len
< off
) {
2033 write_count
= fidp
->fs
.xattr
.len
- off
;
2034 if (write_count
> count
) {
2035 write_count
= count
;
2037 err
= pdu_marshal(pdu
, offset
, "d", write_count
);
2042 fidp
->fs
.xattr
.copied_len
+= write_count
;
2044 * Now copy the content from sg list
2046 for (i
= 0; i
< cnt
; i
++) {
2047 if (write_count
> sg
[i
].iov_len
) {
2048 to_copy
= sg
[i
].iov_len
;
2050 to_copy
= write_count
;
2052 memcpy((char *)fidp
->fs
.xattr
.value
+ off
, sg
[i
].iov_base
, to_copy
);
2053 /* updating vs->off since we are not using below */
2055 write_count
-= to_copy
;
2061 static void coroutine_fn
v9fs_write(void *opaque
)
2071 V9fsPDU
*pdu
= opaque
;
2072 V9fsState
*s
= pdu
->s
;
2073 QEMUIOVector qiov_full
;
2076 err
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
, &off
, &count
);
2078 pdu_complete(pdu
, err
);
2082 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
, count
, true);
2083 trace_v9fs_write(pdu
->tag
, pdu
->id
, fid
, off
, count
, qiov_full
.niov
);
2085 fidp
= get_fid(pdu
, fid
);
2090 if (fidp
->fid_type
== P9_FID_FILE
) {
2091 if (fidp
->fs
.fd
== -1) {
2095 } else if (fidp
->fid_type
== P9_FID_XATTR
) {
2097 * setxattr operation
2099 err
= v9fs_xattr_write(s
, pdu
, fidp
, off
, count
,
2100 qiov_full
.iov
, qiov_full
.niov
);
2106 qemu_iovec_init(&qiov
, qiov_full
.niov
);
2108 qemu_iovec_reset(&qiov
);
2109 qemu_iovec_concat(&qiov
, &qiov_full
, total
, qiov_full
.size
- total
);
2111 print_sg(qiov
.iov
, qiov
.niov
);
2113 /* Loop in case of EINTR */
2115 len
= v9fs_co_pwritev(pdu
, fidp
, qiov
.iov
, qiov
.niov
, off
);
2120 } while (len
== -EINTR
&& !pdu
->cancelled
);
2122 /* IO error return the error */
2126 } while (total
< count
&& len
> 0);
2129 err
= pdu_marshal(pdu
, offset
, "d", total
);
2134 trace_v9fs_write_return(pdu
->tag
, pdu
->id
, total
, err
);
2136 qemu_iovec_destroy(&qiov
);
2140 qemu_iovec_destroy(&qiov_full
);
2141 pdu_complete(pdu
, err
);
2144 static void coroutine_fn
v9fs_create(void *opaque
)
2156 V9fsString extension
;
2158 V9fsPDU
*pdu
= opaque
;
2160 v9fs_path_init(&path
);
2161 v9fs_string_init(&name
);
2162 v9fs_string_init(&extension
);
2163 err
= pdu_unmarshal(pdu
, offset
, "dsdbs", &fid
, &name
,
2164 &perm
, &mode
, &extension
);
2168 trace_v9fs_create(pdu
->tag
, pdu
->id
, fid
, name
.data
, perm
, mode
);
2170 if (name_is_illegal(name
.data
)) {
2175 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2180 fidp
= get_fid(pdu
, fid
);
2185 if (fidp
->fid_type
!= P9_FID_NONE
) {
2189 if (perm
& P9_STAT_MODE_DIR
) {
2190 err
= v9fs_co_mkdir(pdu
, fidp
, &name
, perm
& 0777,
2191 fidp
->uid
, -1, &stbuf
);
2195 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2199 v9fs_path_copy(&fidp
->path
, &path
);
2200 err
= v9fs_co_opendir(pdu
, fidp
);
2204 fidp
->fid_type
= P9_FID_DIR
;
2205 } else if (perm
& P9_STAT_MODE_SYMLINK
) {
2206 err
= v9fs_co_symlink(pdu
, fidp
, &name
,
2207 extension
.data
, -1 , &stbuf
);
2211 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2215 v9fs_path_copy(&fidp
->path
, &path
);
2216 } else if (perm
& P9_STAT_MODE_LINK
) {
2217 int32_t ofid
= atoi(extension
.data
);
2218 V9fsFidState
*ofidp
= get_fid(pdu
, ofid
);
2219 if (ofidp
== NULL
) {
2223 err
= v9fs_co_link(pdu
, ofidp
, fidp
, &name
);
2224 put_fid(pdu
, ofidp
);
2228 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2230 fidp
->fid_type
= P9_FID_NONE
;
2233 v9fs_path_copy(&fidp
->path
, &path
);
2234 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
2236 fidp
->fid_type
= P9_FID_NONE
;
2239 } else if (perm
& P9_STAT_MODE_DEVICE
) {
2241 uint32_t major
, minor
;
2244 if (sscanf(extension
.data
, "%c %u %u", &ctype
, &major
, &minor
) != 3) {
2261 nmode
|= perm
& 0777;
2262 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2263 makedev(major
, minor
), nmode
, &stbuf
);
2267 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2271 v9fs_path_copy(&fidp
->path
, &path
);
2272 } else if (perm
& P9_STAT_MODE_NAMED_PIPE
) {
2273 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2274 0, S_IFIFO
| (perm
& 0777), &stbuf
);
2278 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2282 v9fs_path_copy(&fidp
->path
, &path
);
2283 } else if (perm
& P9_STAT_MODE_SOCKET
) {
2284 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2285 0, S_IFSOCK
| (perm
& 0777), &stbuf
);
2289 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2293 v9fs_path_copy(&fidp
->path
, &path
);
2295 err
= v9fs_co_open2(pdu
, fidp
, &name
, -1,
2296 omode_to_uflags(mode
)|O_CREAT
, perm
, &stbuf
);
2300 fidp
->fid_type
= P9_FID_FILE
;
2301 fidp
->open_flags
= omode_to_uflags(mode
);
2302 if (fidp
->open_flags
& O_EXCL
) {
2304 * We let the host file system do O_EXCL check
2305 * We should not reclaim such fd
2307 fidp
->flags
|= FID_NON_RECLAIMABLE
;
2310 iounit
= get_iounit(pdu
, &fidp
->path
);
2311 stat_to_qid(&stbuf
, &qid
);
2312 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
2317 trace_v9fs_create_return(pdu
->tag
, pdu
->id
,
2318 qid
.type
, qid
.version
, qid
.path
, iounit
);
2322 pdu_complete(pdu
, err
);
2323 v9fs_string_free(&name
);
2324 v9fs_string_free(&extension
);
2325 v9fs_path_free(&path
);
2328 static void coroutine_fn
v9fs_symlink(void *opaque
)
2330 V9fsPDU
*pdu
= opaque
;
2333 V9fsFidState
*dfidp
;
2341 v9fs_string_init(&name
);
2342 v9fs_string_init(&symname
);
2343 err
= pdu_unmarshal(pdu
, offset
, "dssd", &dfid
, &name
, &symname
, &gid
);
2347 trace_v9fs_symlink(pdu
->tag
, pdu
->id
, dfid
, name
.data
, symname
.data
, gid
);
2349 if (name_is_illegal(name
.data
)) {
2354 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2359 dfidp
= get_fid(pdu
, dfid
);
2360 if (dfidp
== NULL
) {
2364 err
= v9fs_co_symlink(pdu
, dfidp
, &name
, symname
.data
, gid
, &stbuf
);
2368 stat_to_qid(&stbuf
, &qid
);
2369 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
2374 trace_v9fs_symlink_return(pdu
->tag
, pdu
->id
,
2375 qid
.type
, qid
.version
, qid
.path
);
2377 put_fid(pdu
, dfidp
);
2379 pdu_complete(pdu
, err
);
2380 v9fs_string_free(&name
);
2381 v9fs_string_free(&symname
);
2384 static void coroutine_fn
v9fs_flush(void *opaque
)
2389 V9fsPDU
*cancel_pdu
= NULL
;
2390 V9fsPDU
*pdu
= opaque
;
2391 V9fsState
*s
= pdu
->s
;
2393 err
= pdu_unmarshal(pdu
, offset
, "w", &tag
);
2395 pdu_complete(pdu
, err
);
2398 trace_v9fs_flush(pdu
->tag
, pdu
->id
, tag
);
2400 if (pdu
->tag
== tag
) {
2401 warn_report("the guest sent a self-referencing 9P flush request");
2403 QLIST_FOREACH(cancel_pdu
, &s
->active_list
, next
) {
2404 if (cancel_pdu
->tag
== tag
) {
2410 cancel_pdu
->cancelled
= 1;
2412 * Wait for pdu to complete.
2414 qemu_co_queue_wait(&cancel_pdu
->complete
, NULL
);
2415 if (!qemu_co_queue_next(&cancel_pdu
->complete
)) {
2416 cancel_pdu
->cancelled
= 0;
2417 pdu_free(cancel_pdu
);
2420 pdu_complete(pdu
, 7);
2423 static void coroutine_fn
v9fs_link(void *opaque
)
2425 V9fsPDU
*pdu
= opaque
;
2426 int32_t dfid
, oldfid
;
2427 V9fsFidState
*dfidp
, *oldfidp
;
2432 v9fs_string_init(&name
);
2433 err
= pdu_unmarshal(pdu
, offset
, "dds", &dfid
, &oldfid
, &name
);
2437 trace_v9fs_link(pdu
->tag
, pdu
->id
, dfid
, oldfid
, name
.data
);
2439 if (name_is_illegal(name
.data
)) {
2444 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2449 dfidp
= get_fid(pdu
, dfid
);
2450 if (dfidp
== NULL
) {
2455 oldfidp
= get_fid(pdu
, oldfid
);
2456 if (oldfidp
== NULL
) {
2460 err
= v9fs_co_link(pdu
, oldfidp
, dfidp
, &name
);
2464 put_fid(pdu
, oldfidp
);
2466 put_fid(pdu
, dfidp
);
2468 v9fs_string_free(&name
);
2469 pdu_complete(pdu
, err
);
2472 /* Only works with path name based fid */
2473 static void coroutine_fn
v9fs_remove(void *opaque
)
2479 V9fsPDU
*pdu
= opaque
;
2481 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
2485 trace_v9fs_remove(pdu
->tag
, pdu
->id
, fid
);
2487 fidp
= get_fid(pdu
, fid
);
2492 /* if fs driver is not path based, return EOPNOTSUPP */
2493 if (!(pdu
->s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
)) {
2498 * IF the file is unlinked, we cannot reopen
2499 * the file later. So don't reclaim fd
2501 err
= v9fs_mark_fids_unreclaim(pdu
, &fidp
->path
);
2505 err
= v9fs_co_remove(pdu
, &fidp
->path
);
2510 /* For TREMOVE we need to clunk the fid even on failed remove */
2511 clunk_fid(pdu
->s
, fidp
->fid
);
2514 pdu_complete(pdu
, err
);
2517 static void coroutine_fn
v9fs_unlinkat(void *opaque
)
2521 int32_t dfid
, flags
;
2524 V9fsFidState
*dfidp
;
2525 V9fsPDU
*pdu
= opaque
;
2527 v9fs_string_init(&name
);
2528 err
= pdu_unmarshal(pdu
, offset
, "dsd", &dfid
, &name
, &flags
);
2533 if (name_is_illegal(name
.data
)) {
2538 if (!strcmp(".", name
.data
)) {
2543 if (!strcmp("..", name
.data
)) {
2548 dfidp
= get_fid(pdu
, dfid
);
2549 if (dfidp
== NULL
) {
2554 * IF the file is unlinked, we cannot reopen
2555 * the file later. So don't reclaim fd
2557 v9fs_path_init(&path
);
2558 err
= v9fs_co_name_to_path(pdu
, &dfidp
->path
, name
.data
, &path
);
2562 err
= v9fs_mark_fids_unreclaim(pdu
, &path
);
2566 err
= v9fs_co_unlinkat(pdu
, &dfidp
->path
, &name
, flags
);
2571 put_fid(pdu
, dfidp
);
2572 v9fs_path_free(&path
);
2574 pdu_complete(pdu
, err
);
2575 v9fs_string_free(&name
);
2579 /* Only works with path name based fid */
2580 static int coroutine_fn
v9fs_complete_rename(V9fsPDU
*pdu
, V9fsFidState
*fidp
,
2586 V9fsFidState
*tfidp
;
2587 V9fsState
*s
= pdu
->s
;
2588 V9fsFidState
*dirfidp
= NULL
;
2590 v9fs_path_init(&new_path
);
2591 if (newdirfid
!= -1) {
2592 dirfidp
= get_fid(pdu
, newdirfid
);
2593 if (dirfidp
== NULL
) {
2597 if (fidp
->fid_type
!= P9_FID_NONE
) {
2601 err
= v9fs_co_name_to_path(pdu
, &dirfidp
->path
, name
->data
, &new_path
);
2606 char *dir_name
= g_path_get_dirname(fidp
->path
.data
);
2609 v9fs_path_init(&dir_path
);
2610 v9fs_path_sprintf(&dir_path
, "%s", dir_name
);
2613 err
= v9fs_co_name_to_path(pdu
, &dir_path
, name
->data
, &new_path
);
2614 v9fs_path_free(&dir_path
);
2619 err
= v9fs_co_rename(pdu
, &fidp
->path
, &new_path
);
2624 * Fixup fid's pointing to the old name to
2625 * start pointing to the new name
2627 for (tfidp
= s
->fid_list
; tfidp
; tfidp
= tfidp
->next
) {
2628 if (v9fs_path_is_ancestor(&fidp
->path
, &tfidp
->path
)) {
2629 /* replace the name */
2630 v9fs_fix_path(&tfidp
->path
, &new_path
, strlen(fidp
->path
.data
));
2635 put_fid(pdu
, dirfidp
);
2637 v9fs_path_free(&new_path
);
2642 /* Only works with path name based fid */
2643 static void coroutine_fn
v9fs_rename(void *opaque
)
2651 V9fsPDU
*pdu
= opaque
;
2652 V9fsState
*s
= pdu
->s
;
2654 v9fs_string_init(&name
);
2655 err
= pdu_unmarshal(pdu
, offset
, "dds", &fid
, &newdirfid
, &name
);
2660 if (name_is_illegal(name
.data
)) {
2665 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2670 fidp
= get_fid(pdu
, fid
);
2675 if (fidp
->fid_type
!= P9_FID_NONE
) {
2679 /* if fs driver is not path based, return EOPNOTSUPP */
2680 if (!(pdu
->s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
)) {
2684 v9fs_path_write_lock(s
);
2685 err
= v9fs_complete_rename(pdu
, fidp
, newdirfid
, &name
);
2686 v9fs_path_unlock(s
);
2693 pdu_complete(pdu
, err
);
2694 v9fs_string_free(&name
);
2697 static int coroutine_fn
v9fs_fix_fid_paths(V9fsPDU
*pdu
, V9fsPath
*olddir
,
2698 V9fsString
*old_name
,
2700 V9fsString
*new_name
)
2702 V9fsFidState
*tfidp
;
2703 V9fsPath oldpath
, newpath
;
2704 V9fsState
*s
= pdu
->s
;
2707 v9fs_path_init(&oldpath
);
2708 v9fs_path_init(&newpath
);
2709 err
= v9fs_co_name_to_path(pdu
, olddir
, old_name
->data
, &oldpath
);
2713 err
= v9fs_co_name_to_path(pdu
, newdir
, new_name
->data
, &newpath
);
2719 * Fixup fid's pointing to the old name to
2720 * start pointing to the new name
2722 for (tfidp
= s
->fid_list
; tfidp
; tfidp
= tfidp
->next
) {
2723 if (v9fs_path_is_ancestor(&oldpath
, &tfidp
->path
)) {
2724 /* replace the name */
2725 v9fs_fix_path(&tfidp
->path
, &newpath
, strlen(oldpath
.data
));
2729 v9fs_path_free(&oldpath
);
2730 v9fs_path_free(&newpath
);
2734 static int coroutine_fn
v9fs_complete_renameat(V9fsPDU
*pdu
, int32_t olddirfid
,
2735 V9fsString
*old_name
,
2737 V9fsString
*new_name
)
2740 V9fsState
*s
= pdu
->s
;
2741 V9fsFidState
*newdirfidp
= NULL
, *olddirfidp
= NULL
;
2743 olddirfidp
= get_fid(pdu
, olddirfid
);
2744 if (olddirfidp
== NULL
) {
2748 if (newdirfid
!= -1) {
2749 newdirfidp
= get_fid(pdu
, newdirfid
);
2750 if (newdirfidp
== NULL
) {
2755 newdirfidp
= get_fid(pdu
, olddirfid
);
2758 err
= v9fs_co_renameat(pdu
, &olddirfidp
->path
, old_name
,
2759 &newdirfidp
->path
, new_name
);
2763 if (s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
) {
2764 /* Only for path based fid we need to do the below fixup */
2765 err
= v9fs_fix_fid_paths(pdu
, &olddirfidp
->path
, old_name
,
2766 &newdirfidp
->path
, new_name
);
2770 put_fid(pdu
, olddirfidp
);
2773 put_fid(pdu
, newdirfidp
);
2778 static void coroutine_fn
v9fs_renameat(void *opaque
)
2782 V9fsPDU
*pdu
= opaque
;
2783 V9fsState
*s
= pdu
->s
;
2784 int32_t olddirfid
, newdirfid
;
2785 V9fsString old_name
, new_name
;
2787 v9fs_string_init(&old_name
);
2788 v9fs_string_init(&new_name
);
2789 err
= pdu_unmarshal(pdu
, offset
, "dsds", &olddirfid
,
2790 &old_name
, &newdirfid
, &new_name
);
2795 if (name_is_illegal(old_name
.data
) || name_is_illegal(new_name
.data
)) {
2800 if (!strcmp(".", old_name
.data
) || !strcmp("..", old_name
.data
) ||
2801 !strcmp(".", new_name
.data
) || !strcmp("..", new_name
.data
)) {
2806 v9fs_path_write_lock(s
);
2807 err
= v9fs_complete_renameat(pdu
, olddirfid
,
2808 &old_name
, newdirfid
, &new_name
);
2809 v9fs_path_unlock(s
);
2815 pdu_complete(pdu
, err
);
2816 v9fs_string_free(&old_name
);
2817 v9fs_string_free(&new_name
);
2820 static void coroutine_fn
v9fs_wstat(void *opaque
)
2829 V9fsPDU
*pdu
= opaque
;
2831 v9fs_stat_init(&v9stat
);
2832 err
= pdu_unmarshal(pdu
, offset
, "dwS", &fid
, &unused
, &v9stat
);
2836 trace_v9fs_wstat(pdu
->tag
, pdu
->id
, fid
,
2837 v9stat
.mode
, v9stat
.atime
, v9stat
.mtime
);
2839 fidp
= get_fid(pdu
, fid
);
2844 /* do we need to sync the file? */
2845 if (donttouch_stat(&v9stat
)) {
2846 err
= v9fs_co_fsync(pdu
, fidp
, 0);
2849 if (v9stat
.mode
!= -1) {
2851 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
2855 v9_mode
= stat_to_v9mode(&stbuf
);
2856 if ((v9stat
.mode
& P9_STAT_MODE_TYPE_BITS
) !=
2857 (v9_mode
& P9_STAT_MODE_TYPE_BITS
)) {
2858 /* Attempting to change the type */
2862 err
= v9fs_co_chmod(pdu
, &fidp
->path
,
2863 v9mode_to_mode(v9stat
.mode
,
2864 &v9stat
.extension
));
2869 if (v9stat
.mtime
!= -1 || v9stat
.atime
!= -1) {
2870 struct timespec times
[2];
2871 if (v9stat
.atime
!= -1) {
2872 times
[0].tv_sec
= v9stat
.atime
;
2873 times
[0].tv_nsec
= 0;
2875 times
[0].tv_nsec
= UTIME_OMIT
;
2877 if (v9stat
.mtime
!= -1) {
2878 times
[1].tv_sec
= v9stat
.mtime
;
2879 times
[1].tv_nsec
= 0;
2881 times
[1].tv_nsec
= UTIME_OMIT
;
2883 err
= v9fs_co_utimensat(pdu
, &fidp
->path
, times
);
2888 if (v9stat
.n_gid
!= -1 || v9stat
.n_uid
!= -1) {
2889 err
= v9fs_co_chown(pdu
, &fidp
->path
, v9stat
.n_uid
, v9stat
.n_gid
);
2894 if (v9stat
.name
.size
!= 0) {
2895 err
= v9fs_complete_rename(pdu
, fidp
, -1, &v9stat
.name
);
2900 if (v9stat
.length
!= -1) {
2901 err
= v9fs_co_truncate(pdu
, &fidp
->path
, v9stat
.length
);
2910 v9fs_stat_free(&v9stat
);
2911 pdu_complete(pdu
, err
);
2914 static int v9fs_fill_statfs(V9fsState
*s
, V9fsPDU
*pdu
, struct statfs
*stbuf
)
2926 int32_t bsize_factor
;
2929 * compute bsize factor based on host file system block size
2932 bsize_factor
= (s
->msize
- P9_IOHDRSZ
)/stbuf
->f_bsize
;
2933 if (!bsize_factor
) {
2936 f_type
= stbuf
->f_type
;
2937 f_bsize
= stbuf
->f_bsize
;
2938 f_bsize
*= bsize_factor
;
2940 * f_bsize is adjusted(multiplied) by bsize factor, so we need to
2941 * adjust(divide) the number of blocks, free blocks and available
2942 * blocks by bsize factor
2944 f_blocks
= stbuf
->f_blocks
/bsize_factor
;
2945 f_bfree
= stbuf
->f_bfree
/bsize_factor
;
2946 f_bavail
= stbuf
->f_bavail
/bsize_factor
;
2947 f_files
= stbuf
->f_files
;
2948 f_ffree
= stbuf
->f_ffree
;
2949 fsid_val
= (unsigned int) stbuf
->f_fsid
.__val
[0] |
2950 (unsigned long long)stbuf
->f_fsid
.__val
[1] << 32;
2951 f_namelen
= stbuf
->f_namelen
;
2953 return pdu_marshal(pdu
, offset
, "ddqqqqqqd",
2954 f_type
, f_bsize
, f_blocks
, f_bfree
,
2955 f_bavail
, f_files
, f_ffree
,
2956 fsid_val
, f_namelen
);
2959 static void coroutine_fn
v9fs_statfs(void *opaque
)
2965 struct statfs stbuf
;
2966 V9fsPDU
*pdu
= opaque
;
2967 V9fsState
*s
= pdu
->s
;
2969 retval
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
2973 fidp
= get_fid(pdu
, fid
);
2978 retval
= v9fs_co_statfs(pdu
, &fidp
->path
, &stbuf
);
2982 retval
= v9fs_fill_statfs(s
, pdu
, &stbuf
);
2990 pdu_complete(pdu
, retval
);
2993 static void coroutine_fn
v9fs_mknod(void *opaque
)
3006 V9fsPDU
*pdu
= opaque
;
3008 v9fs_string_init(&name
);
3009 err
= pdu_unmarshal(pdu
, offset
, "dsdddd", &fid
, &name
, &mode
,
3010 &major
, &minor
, &gid
);
3014 trace_v9fs_mknod(pdu
->tag
, pdu
->id
, fid
, mode
, major
, minor
);
3016 if (name_is_illegal(name
.data
)) {
3021 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3026 fidp
= get_fid(pdu
, fid
);
3031 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, gid
,
3032 makedev(major
, minor
), mode
, &stbuf
);
3036 stat_to_qid(&stbuf
, &qid
);
3037 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
3042 trace_v9fs_mknod_return(pdu
->tag
, pdu
->id
,
3043 qid
.type
, qid
.version
, qid
.path
);
3047 pdu_complete(pdu
, err
);
3048 v9fs_string_free(&name
);
3052 * Implement posix byte range locking code
3053 * Server side handling of locking code is very simple, because 9p server in
3054 * QEMU can handle only one client. And most of the lock handling
3055 * (like conflict, merging) etc is done by the VFS layer itself, so no need to
3056 * do any thing in * qemu 9p server side lock code path.
3057 * So when a TLOCK request comes, always return success
3059 static void coroutine_fn
v9fs_lock(void *opaque
)
3065 int32_t fid
, err
= 0;
3066 V9fsPDU
*pdu
= opaque
;
3068 v9fs_string_init(&flock
.client_id
);
3069 err
= pdu_unmarshal(pdu
, offset
, "dbdqqds", &fid
, &flock
.type
,
3070 &flock
.flags
, &flock
.start
, &flock
.length
,
3071 &flock
.proc_id
, &flock
.client_id
);
3075 trace_v9fs_lock(pdu
->tag
, pdu
->id
, fid
,
3076 flock
.type
, flock
.start
, flock
.length
);
3079 /* We support only block flag now (that too ignored currently) */
3080 if (flock
.flags
& ~P9_LOCK_FLAGS_BLOCK
) {
3084 fidp
= get_fid(pdu
, fid
);
3089 err
= v9fs_co_fstat(pdu
, fidp
, &stbuf
);
3093 err
= pdu_marshal(pdu
, offset
, "b", P9_LOCK_SUCCESS
);
3098 trace_v9fs_lock_return(pdu
->tag
, pdu
->id
, P9_LOCK_SUCCESS
);
3102 pdu_complete(pdu
, err
);
3103 v9fs_string_free(&flock
.client_id
);
3107 * When a TGETLOCK request comes, always return success because all lock
3108 * handling is done by client's VFS layer.
3110 static void coroutine_fn
v9fs_getlock(void *opaque
)
3116 int32_t fid
, err
= 0;
3117 V9fsPDU
*pdu
= opaque
;
3119 v9fs_string_init(&glock
.client_id
);
3120 err
= pdu_unmarshal(pdu
, offset
, "dbqqds", &fid
, &glock
.type
,
3121 &glock
.start
, &glock
.length
, &glock
.proc_id
,
3126 trace_v9fs_getlock(pdu
->tag
, pdu
->id
, fid
,
3127 glock
.type
, glock
.start
, glock
.length
);
3129 fidp
= get_fid(pdu
, fid
);
3134 err
= v9fs_co_fstat(pdu
, fidp
, &stbuf
);
3138 glock
.type
= P9_LOCK_TYPE_UNLCK
;
3139 err
= pdu_marshal(pdu
, offset
, "bqqds", glock
.type
,
3140 glock
.start
, glock
.length
, glock
.proc_id
,
3146 trace_v9fs_getlock_return(pdu
->tag
, pdu
->id
, glock
.type
, glock
.start
,
3147 glock
.length
, glock
.proc_id
);
3151 pdu_complete(pdu
, err
);
3152 v9fs_string_free(&glock
.client_id
);
3155 static void coroutine_fn
v9fs_mkdir(void *opaque
)
3157 V9fsPDU
*pdu
= opaque
;
3168 v9fs_string_init(&name
);
3169 err
= pdu_unmarshal(pdu
, offset
, "dsdd", &fid
, &name
, &mode
, &gid
);
3173 trace_v9fs_mkdir(pdu
->tag
, pdu
->id
, fid
, name
.data
, mode
, gid
);
3175 if (name_is_illegal(name
.data
)) {
3180 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3185 fidp
= get_fid(pdu
, fid
);
3190 err
= v9fs_co_mkdir(pdu
, fidp
, &name
, mode
, fidp
->uid
, gid
, &stbuf
);
3194 stat_to_qid(&stbuf
, &qid
);
3195 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
3200 trace_v9fs_mkdir_return(pdu
->tag
, pdu
->id
,
3201 qid
.type
, qid
.version
, qid
.path
, err
);
3205 pdu_complete(pdu
, err
);
3206 v9fs_string_free(&name
);
3209 static void coroutine_fn
v9fs_xattrwalk(void *opaque
)
3215 int32_t fid
, newfid
;
3216 V9fsFidState
*file_fidp
;
3217 V9fsFidState
*xattr_fidp
= NULL
;
3218 V9fsPDU
*pdu
= opaque
;
3219 V9fsState
*s
= pdu
->s
;
3221 v9fs_string_init(&name
);
3222 err
= pdu_unmarshal(pdu
, offset
, "dds", &fid
, &newfid
, &name
);
3226 trace_v9fs_xattrwalk(pdu
->tag
, pdu
->id
, fid
, newfid
, name
.data
);
3228 file_fidp
= get_fid(pdu
, fid
);
3229 if (file_fidp
== NULL
) {
3233 xattr_fidp
= alloc_fid(s
, newfid
);
3234 if (xattr_fidp
== NULL
) {
3238 v9fs_path_copy(&xattr_fidp
->path
, &file_fidp
->path
);
3239 if (!v9fs_string_size(&name
)) {
3241 * listxattr request. Get the size first
3243 size
= v9fs_co_llistxattr(pdu
, &xattr_fidp
->path
, NULL
, 0);
3246 clunk_fid(s
, xattr_fidp
->fid
);
3250 * Read the xattr value
3252 xattr_fidp
->fs
.xattr
.len
= size
;
3253 xattr_fidp
->fid_type
= P9_FID_XATTR
;
3254 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= true;
3256 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
3257 err
= v9fs_co_llistxattr(pdu
, &xattr_fidp
->path
,
3258 xattr_fidp
->fs
.xattr
.value
,
3259 xattr_fidp
->fs
.xattr
.len
);
3261 clunk_fid(s
, xattr_fidp
->fid
);
3265 err
= pdu_marshal(pdu
, offset
, "q", size
);
3272 * specific xattr fid. We check for xattr
3273 * presence also collect the xattr size
3275 size
= v9fs_co_lgetxattr(pdu
, &xattr_fidp
->path
,
3279 clunk_fid(s
, xattr_fidp
->fid
);
3283 * Read the xattr value
3285 xattr_fidp
->fs
.xattr
.len
= size
;
3286 xattr_fidp
->fid_type
= P9_FID_XATTR
;
3287 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= true;
3289 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
3290 err
= v9fs_co_lgetxattr(pdu
, &xattr_fidp
->path
,
3291 &name
, xattr_fidp
->fs
.xattr
.value
,
3292 xattr_fidp
->fs
.xattr
.len
);
3294 clunk_fid(s
, xattr_fidp
->fid
);
3298 err
= pdu_marshal(pdu
, offset
, "q", size
);
3304 trace_v9fs_xattrwalk_return(pdu
->tag
, pdu
->id
, size
);
3306 put_fid(pdu
, file_fidp
);
3308 put_fid(pdu
, xattr_fidp
);
3311 pdu_complete(pdu
, err
);
3312 v9fs_string_free(&name
);
3315 static void coroutine_fn
v9fs_xattrcreate(void *opaque
)
3323 V9fsFidState
*file_fidp
;
3324 V9fsFidState
*xattr_fidp
;
3325 V9fsPDU
*pdu
= opaque
;
3327 v9fs_string_init(&name
);
3328 err
= pdu_unmarshal(pdu
, offset
, "dsqd", &fid
, &name
, &size
, &flags
);
3332 trace_v9fs_xattrcreate(pdu
->tag
, pdu
->id
, fid
, name
.data
, size
, flags
);
3334 if (size
> XATTR_SIZE_MAX
) {
3339 file_fidp
= get_fid(pdu
, fid
);
3340 if (file_fidp
== NULL
) {
3344 if (file_fidp
->fid_type
!= P9_FID_NONE
) {
3349 /* Make the file fid point to xattr */
3350 xattr_fidp
= file_fidp
;
3351 xattr_fidp
->fid_type
= P9_FID_XATTR
;
3352 xattr_fidp
->fs
.xattr
.copied_len
= 0;
3353 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= false;
3354 xattr_fidp
->fs
.xattr
.len
= size
;
3355 xattr_fidp
->fs
.xattr
.flags
= flags
;
3356 v9fs_string_init(&xattr_fidp
->fs
.xattr
.name
);
3357 v9fs_string_copy(&xattr_fidp
->fs
.xattr
.name
, &name
);
3358 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
3361 put_fid(pdu
, file_fidp
);
3363 pdu_complete(pdu
, err
);
3364 v9fs_string_free(&name
);
3367 static void coroutine_fn
v9fs_readlink(void *opaque
)
3369 V9fsPDU
*pdu
= opaque
;
3376 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
3380 trace_v9fs_readlink(pdu
->tag
, pdu
->id
, fid
);
3381 fidp
= get_fid(pdu
, fid
);
3387 v9fs_string_init(&target
);
3388 err
= v9fs_co_readlink(pdu
, &fidp
->path
, &target
);
3392 err
= pdu_marshal(pdu
, offset
, "s", &target
);
3394 v9fs_string_free(&target
);
3398 trace_v9fs_readlink_return(pdu
->tag
, pdu
->id
, target
.data
);
3399 v9fs_string_free(&target
);
3403 pdu_complete(pdu
, err
);
3406 static CoroutineEntry
*pdu_co_handlers
[] = {
3407 [P9_TREADDIR
] = v9fs_readdir
,
3408 [P9_TSTATFS
] = v9fs_statfs
,
3409 [P9_TGETATTR
] = v9fs_getattr
,
3410 [P9_TSETATTR
] = v9fs_setattr
,
3411 [P9_TXATTRWALK
] = v9fs_xattrwalk
,
3412 [P9_TXATTRCREATE
] = v9fs_xattrcreate
,
3413 [P9_TMKNOD
] = v9fs_mknod
,
3414 [P9_TRENAME
] = v9fs_rename
,
3415 [P9_TLOCK
] = v9fs_lock
,
3416 [P9_TGETLOCK
] = v9fs_getlock
,
3417 [P9_TRENAMEAT
] = v9fs_renameat
,
3418 [P9_TREADLINK
] = v9fs_readlink
,
3419 [P9_TUNLINKAT
] = v9fs_unlinkat
,
3420 [P9_TMKDIR
] = v9fs_mkdir
,
3421 [P9_TVERSION
] = v9fs_version
,
3422 [P9_TLOPEN
] = v9fs_open
,
3423 [P9_TATTACH
] = v9fs_attach
,
3424 [P9_TSTAT
] = v9fs_stat
,
3425 [P9_TWALK
] = v9fs_walk
,
3426 [P9_TCLUNK
] = v9fs_clunk
,
3427 [P9_TFSYNC
] = v9fs_fsync
,
3428 [P9_TOPEN
] = v9fs_open
,
3429 [P9_TREAD
] = v9fs_read
,
3431 [P9_TAUTH
] = v9fs_auth
,
3433 [P9_TFLUSH
] = v9fs_flush
,
3434 [P9_TLINK
] = v9fs_link
,
3435 [P9_TSYMLINK
] = v9fs_symlink
,
3436 [P9_TCREATE
] = v9fs_create
,
3437 [P9_TLCREATE
] = v9fs_lcreate
,
3438 [P9_TWRITE
] = v9fs_write
,
3439 [P9_TWSTAT
] = v9fs_wstat
,
3440 [P9_TREMOVE
] = v9fs_remove
,
3443 static void coroutine_fn
v9fs_op_not_supp(void *opaque
)
3445 V9fsPDU
*pdu
= opaque
;
3446 pdu_complete(pdu
, -EOPNOTSUPP
);
3449 static void coroutine_fn
v9fs_fs_ro(void *opaque
)
3451 V9fsPDU
*pdu
= opaque
;
3452 pdu_complete(pdu
, -EROFS
);
3455 static inline bool is_read_only_op(V9fsPDU
*pdu
)
3482 void pdu_submit(V9fsPDU
*pdu
, P9MsgHeader
*hdr
)
3485 CoroutineEntry
*handler
;
3486 V9fsState
*s
= pdu
->s
;
3488 pdu
->size
= le32_to_cpu(hdr
->size_le
);
3490 pdu
->tag
= le16_to_cpu(hdr
->tag_le
);
3492 if (pdu
->id
>= ARRAY_SIZE(pdu_co_handlers
) ||
3493 (pdu_co_handlers
[pdu
->id
] == NULL
)) {
3494 handler
= v9fs_op_not_supp
;
3495 } else if (is_ro_export(&s
->ctx
) && !is_read_only_op(pdu
)) {
3496 handler
= v9fs_fs_ro
;
3498 handler
= pdu_co_handlers
[pdu
->id
];
3501 qemu_co_queue_init(&pdu
->complete
);
3502 co
= qemu_coroutine_create(handler
, pdu
);
3503 qemu_coroutine_enter(co
);
3506 /* Returns 0 on success, 1 on failure. */
3507 int v9fs_device_realize_common(V9fsState
*s
, const V9fsTransport
*t
,
3516 assert(!s
->transport
);
3519 /* initialize pdu allocator */
3520 QLIST_INIT(&s
->free_list
);
3521 QLIST_INIT(&s
->active_list
);
3522 for (i
= 0; i
< MAX_REQ
; i
++) {
3523 QLIST_INSERT_HEAD(&s
->free_list
, &s
->pdus
[i
], next
);
3528 v9fs_path_init(&path
);
3530 fse
= get_fsdev_fsentry(s
->fsconf
.fsdev_id
);
3533 /* We don't have a fsdev identified by fsdev_id */
3534 error_setg(errp
, "9pfs device couldn't find fsdev with the "
3536 s
->fsconf
.fsdev_id
? s
->fsconf
.fsdev_id
: "NULL");
3540 if (!s
->fsconf
.tag
) {
3541 /* we haven't specified a mount_tag */
3542 error_setg(errp
, "fsdev with id %s needs mount_tag arguments",
3543 s
->fsconf
.fsdev_id
);
3547 s
->ctx
.export_flags
= fse
->export_flags
;
3548 s
->ctx
.fs_root
= g_strdup(fse
->path
);
3549 s
->ctx
.exops
.get_st_gen
= NULL
;
3550 len
= strlen(s
->fsconf
.tag
);
3551 if (len
> MAX_TAG_LEN
- 1) {
3552 error_setg(errp
, "mount tag '%s' (%d bytes) is longer than "
3553 "maximum (%d bytes)", s
->fsconf
.tag
, len
, MAX_TAG_LEN
- 1);
3557 s
->tag
= g_strdup(s
->fsconf
.tag
);
3562 s
->ctx
.fmode
= fse
->fmode
;
3563 s
->ctx
.dmode
= fse
->dmode
;
3566 qemu_co_rwlock_init(&s
->rename_lock
);
3568 if (s
->ops
->init(&s
->ctx
, errp
) < 0) {
3569 error_prepend(errp
, "cannot initialize fsdev '%s': ",
3570 s
->fsconf
.fsdev_id
);
3575 * Check details of export path, We need to use fs driver
3576 * call back to do that. Since we are in the init path, we don't
3577 * use co-routines here.
3579 if (s
->ops
->name_to_path(&s
->ctx
, NULL
, "/", &path
) < 0) {
3581 "error in converting name to path %s", strerror(errno
));
3584 if (s
->ops
->lstat(&s
->ctx
, &path
, &stat
)) {
3585 error_setg(errp
, "share path %s does not exist", fse
->path
);
3587 } else if (!S_ISDIR(stat
.st_mode
)) {
3588 error_setg(errp
, "share path %s is not a directory", fse
->path
);
3592 s
->ctx
.fst
= &fse
->fst
;
3593 fsdev_throttle_init(s
->ctx
.fst
);
3595 v9fs_path_free(&path
);
3600 if (s
->ops
&& s
->ops
->cleanup
&& s
->ctx
.private) {
3601 s
->ops
->cleanup(&s
->ctx
);
3604 g_free(s
->ctx
.fs_root
);
3605 v9fs_path_free(&path
);
3610 void v9fs_device_unrealize_common(V9fsState
*s
, Error
**errp
)
3612 if (s
->ops
->cleanup
) {
3613 s
->ops
->cleanup(&s
->ctx
);
3615 fsdev_throttle_cleanup(s
->ctx
.fst
);
3617 g_free(s
->ctx
.fs_root
);
3620 typedef struct VirtfsCoResetData
{
3623 } VirtfsCoResetData
;
3625 static void coroutine_fn
virtfs_co_reset(void *opaque
)
3627 VirtfsCoResetData
*data
= opaque
;
3629 virtfs_reset(&data
->pdu
);
3633 void v9fs_reset(V9fsState
*s
)
3635 VirtfsCoResetData data
= { .pdu
= { .s
= s
}, .done
= false };
3638 while (!QLIST_EMPTY(&s
->active_list
)) {
3639 aio_poll(qemu_get_aio_context(), true);
3642 co
= qemu_coroutine_create(virtfs_co_reset
, &data
);
3643 qemu_coroutine_enter(co
);
3645 while (!data
.done
) {
3646 aio_poll(qemu_get_aio_context(), true);
3650 static void __attribute__((__constructor__
)) v9fs_set_fd_limit(void)
3653 if (getrlimit(RLIMIT_NOFILE
, &rlim
) < 0) {
3654 error_report("Failed to get the resource limit");
3657 open_fd_hw
= rlim
.rlim_cur
- MIN(400, rlim
.rlim_cur
/3);
3658 open_fd_rc
= rlim
.rlim_cur
/2;