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.
15 * Not so fast! You might want to read the 9p developer docs first:
16 * https://wiki.qemu.org/Documentation/9p
19 #include "qemu/osdep.h"
21 #include <linux/limits.h>
23 #include <glib/gprintf.h>
24 #include "hw/virtio/virtio.h"
25 #include "qapi/error.h"
26 #include "qemu/error-report.h"
28 #include "qemu/main-loop.h"
29 #include "qemu/sockets.h"
30 #include "virtio-9p.h"
31 #include "fsdev/qemu-fsdev.h"
36 #include "migration/blocker.h"
37 #include "qemu/xxhash.h"
42 static int open_fd_rc
;
56 P9ARRAY_DEFINE_TYPE(V9fsPath
, v9fs_path_free
);
58 static ssize_t
pdu_marshal(V9fsPDU
*pdu
, size_t offset
, const char *fmt
, ...)
64 ret
= pdu
->s
->transport
->pdu_vmarshal(pdu
, offset
, fmt
, ap
);
70 static ssize_t
pdu_unmarshal(V9fsPDU
*pdu
, size_t offset
, const char *fmt
, ...)
76 ret
= pdu
->s
->transport
->pdu_vunmarshal(pdu
, offset
, fmt
, ap
);
82 static int omode_to_uflags(int8_t mode
)
105 if (mode
& Oappend
) {
116 typedef struct DotlOpenflagMap
{
121 static int dotl_to_open_flags(int flags
)
125 * We have same bits for P9_DOTL_READONLY, P9_DOTL_WRONLY
126 * and P9_DOTL_NOACCESS
128 int oflags
= flags
& O_ACCMODE
;
130 DotlOpenflagMap dotl_oflag_map
[] = {
131 { P9_DOTL_CREATE
, O_CREAT
},
132 { P9_DOTL_EXCL
, O_EXCL
},
133 { P9_DOTL_NOCTTY
, O_NOCTTY
},
134 { P9_DOTL_TRUNC
, O_TRUNC
},
135 { P9_DOTL_APPEND
, O_APPEND
},
136 { P9_DOTL_NONBLOCK
, O_NONBLOCK
} ,
137 { P9_DOTL_DSYNC
, O_DSYNC
},
138 { P9_DOTL_FASYNC
, FASYNC
},
139 #ifndef CONFIG_DARWIN
140 { P9_DOTL_NOATIME
, O_NOATIME
},
142 * On Darwin, we could map to F_NOCACHE, which is
143 * similar, but doesn't quite have the same
144 * semantics. However, we don't support O_DIRECT
145 * even on linux at the moment, so we just ignore
148 { P9_DOTL_DIRECT
, O_DIRECT
},
150 { P9_DOTL_LARGEFILE
, O_LARGEFILE
},
151 { P9_DOTL_DIRECTORY
, O_DIRECTORY
},
152 { P9_DOTL_NOFOLLOW
, O_NOFOLLOW
},
153 { P9_DOTL_SYNC
, O_SYNC
},
156 for (i
= 0; i
< ARRAY_SIZE(dotl_oflag_map
); i
++) {
157 if (flags
& dotl_oflag_map
[i
].dotl_flag
) {
158 oflags
|= dotl_oflag_map
[i
].open_flag
;
165 void cred_init(FsCred
*credp
)
173 static int get_dotl_openflags(V9fsState
*s
, int oflags
)
177 * Filter the client open flags
179 flags
= dotl_to_open_flags(oflags
);
180 flags
&= ~(O_NOCTTY
| O_ASYNC
| O_CREAT
);
181 #ifndef CONFIG_DARWIN
183 * Ignore direct disk access hint until the server supports it.
190 void v9fs_path_init(V9fsPath
*path
)
196 void v9fs_path_free(V9fsPath
*path
)
204 void G_GNUC_PRINTF(2, 3)
205 v9fs_path_sprintf(V9fsPath
*path
, const char *fmt
, ...)
209 v9fs_path_free(path
);
212 /* Bump the size for including terminating NULL */
213 path
->size
= g_vasprintf(&path
->data
, fmt
, ap
) + 1;
217 void v9fs_path_copy(V9fsPath
*dst
, const V9fsPath
*src
)
220 dst
->size
= src
->size
;
221 dst
->data
= g_memdup(src
->data
, src
->size
);
224 int v9fs_name_to_path(V9fsState
*s
, V9fsPath
*dirpath
,
225 const char *name
, V9fsPath
*path
)
228 err
= s
->ops
->name_to_path(&s
->ctx
, dirpath
, name
, path
);
236 * Return TRUE if s1 is an ancestor of s2.
238 * E.g. "a/b" is an ancestor of "a/b/c" but not of "a/bc/d".
239 * As a special case, We treat s1 as ancestor of s2 if they are same!
241 static int v9fs_path_is_ancestor(V9fsPath
*s1
, V9fsPath
*s2
)
243 if (!strncmp(s1
->data
, s2
->data
, s1
->size
- 1)) {
244 if (s2
->data
[s1
->size
- 1] == '\0' || s2
->data
[s1
->size
- 1] == '/') {
251 static size_t v9fs_string_size(V9fsString
*str
)
257 * returns 0 if fid got re-opened, 1 if not, < 0 on error
259 static int coroutine_fn
v9fs_reopen_fid(V9fsPDU
*pdu
, V9fsFidState
*f
)
262 if (f
->fid_type
== P9_FID_FILE
) {
263 if (f
->fs
.fd
== -1) {
265 err
= v9fs_co_open(pdu
, f
, f
->open_flags
);
266 } while (err
== -EINTR
&& !pdu
->cancelled
);
268 } else if (f
->fid_type
== P9_FID_DIR
) {
269 if (f
->fs
.dir
.stream
== NULL
) {
271 err
= v9fs_co_opendir(pdu
, f
);
272 } while (err
== -EINTR
&& !pdu
->cancelled
);
278 static V9fsFidState
*coroutine_fn
get_fid(V9fsPDU
*pdu
, int32_t fid
)
282 V9fsState
*s
= pdu
->s
;
284 f
= g_hash_table_lookup(s
->fids
, GINT_TO_POINTER(fid
));
288 * Update the fid ref upfront so that
289 * we don't get reclaimed when we yield
294 * check whether we need to reopen the
295 * file. We might have closed the fd
296 * while trying to free up some file
299 err
= v9fs_reopen_fid(pdu
, f
);
305 * Mark the fid as referenced so that the LRU
306 * reclaim won't close the file descriptor
308 f
->flags
|= FID_REFERENCED
;
314 static V9fsFidState
*alloc_fid(V9fsState
*s
, int32_t fid
)
318 f
= g_hash_table_lookup(s
->fids
, GINT_TO_POINTER(fid
));
320 /* If fid is already there return NULL */
324 f
= g_new0(V9fsFidState
, 1);
326 f
->fid_type
= P9_FID_NONE
;
329 * Mark the fid as referenced so that the LRU
330 * reclaim won't close the file descriptor
332 f
->flags
|= FID_REFERENCED
;
333 g_hash_table_insert(s
->fids
, GINT_TO_POINTER(fid
), f
);
335 v9fs_readdir_init(s
->proto_version
, &f
->fs
.dir
);
336 v9fs_readdir_init(s
->proto_version
, &f
->fs_reclaim
.dir
);
341 static int coroutine_fn
v9fs_xattr_fid_clunk(V9fsPDU
*pdu
, V9fsFidState
*fidp
)
345 if (fidp
->fs
.xattr
.xattrwalk_fid
) {
346 /* getxattr/listxattr fid */
350 * if this is fid for setxattr. clunk should
351 * result in setxattr localcall
353 if (fidp
->fs
.xattr
.len
!= fidp
->fs
.xattr
.copied_len
) {
354 /* clunk after partial write */
358 if (fidp
->fs
.xattr
.len
) {
359 retval
= v9fs_co_lsetxattr(pdu
, &fidp
->path
, &fidp
->fs
.xattr
.name
,
360 fidp
->fs
.xattr
.value
,
362 fidp
->fs
.xattr
.flags
);
364 retval
= v9fs_co_lremovexattr(pdu
, &fidp
->path
, &fidp
->fs
.xattr
.name
);
367 v9fs_string_free(&fidp
->fs
.xattr
.name
);
369 g_free(fidp
->fs
.xattr
.value
);
373 static int coroutine_fn
free_fid(V9fsPDU
*pdu
, V9fsFidState
*fidp
)
377 if (fidp
->fid_type
== P9_FID_FILE
) {
378 /* If we reclaimed the fd no need to close */
379 if (fidp
->fs
.fd
!= -1) {
380 retval
= v9fs_co_close(pdu
, &fidp
->fs
);
382 } else if (fidp
->fid_type
== P9_FID_DIR
) {
383 if (fidp
->fs
.dir
.stream
!= NULL
) {
384 retval
= v9fs_co_closedir(pdu
, &fidp
->fs
);
386 } else if (fidp
->fid_type
== P9_FID_XATTR
) {
387 retval
= v9fs_xattr_fid_clunk(pdu
, fidp
);
389 v9fs_path_free(&fidp
->path
);
394 static int coroutine_fn
put_fid(V9fsPDU
*pdu
, V9fsFidState
*fidp
)
399 * Don't free the fid if it is in reclaim list
401 if (!fidp
->ref
&& fidp
->clunked
) {
402 if (fidp
->fid
== pdu
->s
->root_fid
) {
404 * if the clunked fid is root fid then we
405 * have unmounted the fs on the client side.
406 * delete the migration blocker. Ideally, this
407 * should be hooked to transport close notification
409 if (pdu
->s
->migration_blocker
) {
410 migrate_del_blocker(pdu
->s
->migration_blocker
);
411 error_free(pdu
->s
->migration_blocker
);
412 pdu
->s
->migration_blocker
= NULL
;
415 return free_fid(pdu
, fidp
);
420 static V9fsFidState
*clunk_fid(V9fsState
*s
, int32_t fid
)
424 /* TODO: Use g_hash_table_steal_extended() instead? */
425 fidp
= g_hash_table_lookup(s
->fids
, GINT_TO_POINTER(fid
));
427 g_hash_table_remove(s
->fids
, GINT_TO_POINTER(fid
));
428 fidp
->clunked
= true;
434 void coroutine_fn
v9fs_reclaim_fd(V9fsPDU
*pdu
)
436 int reclaim_count
= 0;
437 V9fsState
*s
= pdu
->s
;
442 g_hash_table_iter_init(&iter
, s
->fids
);
444 QSLIST_HEAD(, V9fsFidState
) reclaim_list
=
445 QSLIST_HEAD_INITIALIZER(reclaim_list
);
447 while (g_hash_table_iter_next(&iter
, &fid
, (gpointer
*) &f
)) {
449 * Unlink fids cannot be reclaimed. Check
450 * for them and skip them. Also skip fids
451 * currently being operated on.
453 if (f
->ref
|| f
->flags
& FID_NON_RECLAIMABLE
) {
457 * if it is a recently referenced fid
458 * we leave the fid untouched and clear the
459 * reference bit. We come back to it later
460 * in the next iteration. (a simple LRU without
461 * moving list elements around)
463 if (f
->flags
& FID_REFERENCED
) {
464 f
->flags
&= ~FID_REFERENCED
;
468 * Add fids to reclaim list.
470 if (f
->fid_type
== P9_FID_FILE
) {
471 if (f
->fs
.fd
!= -1) {
473 * Up the reference count so that
474 * a clunk request won't free this fid
477 QSLIST_INSERT_HEAD(&reclaim_list
, f
, reclaim_next
);
478 f
->fs_reclaim
.fd
= f
->fs
.fd
;
482 } else if (f
->fid_type
== P9_FID_DIR
) {
483 if (f
->fs
.dir
.stream
!= NULL
) {
485 * Up the reference count so that
486 * a clunk request won't free this fid
489 QSLIST_INSERT_HEAD(&reclaim_list
, f
, reclaim_next
);
490 f
->fs_reclaim
.dir
.stream
= f
->fs
.dir
.stream
;
491 f
->fs
.dir
.stream
= NULL
;
495 if (reclaim_count
>= open_fd_rc
) {
500 * Now close the fid in reclaim list. Free them if they
501 * are already clunked.
503 while (!QSLIST_EMPTY(&reclaim_list
)) {
504 f
= QSLIST_FIRST(&reclaim_list
);
505 QSLIST_REMOVE(&reclaim_list
, f
, V9fsFidState
, reclaim_next
);
506 if (f
->fid_type
== P9_FID_FILE
) {
507 v9fs_co_close(pdu
, &f
->fs_reclaim
);
508 } else if (f
->fid_type
== P9_FID_DIR
) {
509 v9fs_co_closedir(pdu
, &f
->fs_reclaim
);
512 * Now drop the fid reference, free it
520 * This is used when a path is removed from the directory tree. Any
521 * fids that still reference it must not be closed from then on, since
522 * they cannot be reopened.
524 static int coroutine_fn
v9fs_mark_fids_unreclaim(V9fsPDU
*pdu
, V9fsPath
*path
)
527 V9fsState
*s
= pdu
->s
;
532 * The most common case is probably that we have exactly one
533 * fid for the given path, so preallocate exactly one.
535 g_autoptr(GArray
) to_reopen
= g_array_sized_new(FALSE
, FALSE
,
536 sizeof(V9fsFidState
*), 1);
539 g_hash_table_iter_init(&iter
, s
->fids
);
542 * We iterate over the fid table looking for the entries we need
543 * to reopen, and store them in to_reopen. This is because
544 * v9fs_reopen_fid() and put_fid() yield. This allows the fid table
545 * to be modified in the meantime, invalidating our iterator.
547 while (g_hash_table_iter_next(&iter
, &fid
, (gpointer
*) &fidp
)) {
548 if (fidp
->path
.size
== path
->size
&&
549 !memcmp(fidp
->path
.data
, path
->data
, path
->size
)) {
551 * Ensure the fid survives a potential clunk request during
552 * v9fs_reopen_fid or put_fid.
555 fidp
->flags
|= FID_NON_RECLAIMABLE
;
556 g_array_append_val(to_reopen
, fidp
);
560 for (i
= 0; i
< to_reopen
->len
; i
++) {
561 fidp
= g_array_index(to_reopen
, V9fsFidState
*, i
);
562 /* reopen the file/dir if already closed */
563 err
= v9fs_reopen_fid(pdu
, fidp
);
569 for (i
= 0; i
< to_reopen
->len
; i
++) {
570 put_fid(pdu
, g_array_index(to_reopen
, V9fsFidState
*, i
));
575 static void coroutine_fn
virtfs_reset(V9fsPDU
*pdu
)
577 V9fsState
*s
= pdu
->s
;
581 * Get a list of all the values (fid states) in the table, which
584 g_autoptr(GList
) fids
= g_hash_table_get_values(s
->fids
);
586 /* ... remove from the table, taking over ownership. */
587 g_hash_table_steal_all(s
->fids
);
590 * This allows us to release our references to them asynchronously without
591 * iterating over the hash table and risking iterator invalidation
592 * through concurrent modifications.
594 for (freeing
= fids
; freeing
; freeing
= freeing
->next
) {
595 fidp
= freeing
->data
;
597 fidp
->clunked
= true;
602 #define P9_QID_TYPE_DIR 0x80
603 #define P9_QID_TYPE_SYMLINK 0x02
605 #define P9_STAT_MODE_DIR 0x80000000
606 #define P9_STAT_MODE_APPEND 0x40000000
607 #define P9_STAT_MODE_EXCL 0x20000000
608 #define P9_STAT_MODE_MOUNT 0x10000000
609 #define P9_STAT_MODE_AUTH 0x08000000
610 #define P9_STAT_MODE_TMP 0x04000000
611 #define P9_STAT_MODE_SYMLINK 0x02000000
612 #define P9_STAT_MODE_LINK 0x01000000
613 #define P9_STAT_MODE_DEVICE 0x00800000
614 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
615 #define P9_STAT_MODE_SOCKET 0x00100000
616 #define P9_STAT_MODE_SETUID 0x00080000
617 #define P9_STAT_MODE_SETGID 0x00040000
618 #define P9_STAT_MODE_SETVTX 0x00010000
620 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR | \
621 P9_STAT_MODE_SYMLINK | \
622 P9_STAT_MODE_LINK | \
623 P9_STAT_MODE_DEVICE | \
624 P9_STAT_MODE_NAMED_PIPE | \
627 /* Mirrors all bits of a byte. So e.g. binary 10100000 would become 00000101. */
628 static inline uint8_t mirror8bit(uint8_t byte
)
630 return (byte
* 0x0202020202ULL
& 0x010884422010ULL
) % 1023;
633 /* Same as mirror8bit() just for a 64 bit data type instead for a byte. */
634 static inline uint64_t mirror64bit(uint64_t value
)
636 return ((uint64_t)mirror8bit(value
& 0xff) << 56) |
637 ((uint64_t)mirror8bit((value
>> 8) & 0xff) << 48) |
638 ((uint64_t)mirror8bit((value
>> 16) & 0xff) << 40) |
639 ((uint64_t)mirror8bit((value
>> 24) & 0xff) << 32) |
640 ((uint64_t)mirror8bit((value
>> 32) & 0xff) << 24) |
641 ((uint64_t)mirror8bit((value
>> 40) & 0xff) << 16) |
642 ((uint64_t)mirror8bit((value
>> 48) & 0xff) << 8) |
643 ((uint64_t)mirror8bit((value
>> 56) & 0xff));
647 * Parameter k for the Exponential Golomb algorihm to be used.
649 * The smaller this value, the smaller the minimum bit count for the Exp.
650 * Golomb generated affixes will be (at lowest index) however for the
651 * price of having higher maximum bit count of generated affixes (at highest
652 * index). Likewise increasing this parameter yields in smaller maximum bit
653 * count for the price of having higher minimum bit count.
655 * In practice that means: a good value for k depends on the expected amount
656 * of devices to be exposed by one export. For a small amount of devices k
657 * should be small, for a large amount of devices k might be increased
658 * instead. The default of k=0 should be fine for most users though.
660 * IMPORTANT: In case this ever becomes a runtime parameter; the value of
661 * k should not change as long as guest is still running! Because that would
662 * cause completely different inode numbers to be generated on guest.
664 #define EXP_GOLOMB_K 0
667 * expGolombEncode() - Exponential Golomb algorithm for arbitrary k
670 * @n: natural number (or index) of the prefix to be generated
672 * @k: parameter k of Exp. Golomb algorithm to be used
673 * (see comment on EXP_GOLOMB_K macro for details about k)
674 * Return: prefix for given @n and @k
676 * The Exponential Golomb algorithm generates prefixes (NOT suffixes!)
677 * with growing length and with the mathematical property of being
678 * "prefix-free". The latter means the generated prefixes can be prepended
679 * in front of arbitrary numbers and the resulting concatenated numbers are
680 * guaranteed to be always unique.
682 * This is a minor adjustment to the original Exp. Golomb algorithm in the
683 * sense that lowest allowed index (@n) starts with 1, not with zero.
685 static VariLenAffix
expGolombEncode(uint64_t n
, int k
)
687 const uint64_t value
= n
+ (1 << k
) - 1;
688 const int bits
= (int) log2(value
) + 1;
689 return (VariLenAffix
) {
690 .type
= AffixType_Prefix
,
692 .bits
= bits
+ MAX((bits
- 1 - k
), 0)
697 * invertAffix() - Converts a suffix into a prefix, or a prefix into a suffix.
698 * @affix: either suffix or prefix to be inverted
699 * Return: inversion of passed @affix
701 * Simply mirror all bits of the affix value, for the purpose to preserve
702 * respectively the mathematical "prefix-free" or "suffix-free" property
703 * after the conversion.
705 * If a passed prefix is suitable to create unique numbers, then the
706 * returned suffix is suitable to create unique numbers as well (and vice
709 static VariLenAffix
invertAffix(const VariLenAffix
*affix
)
711 return (VariLenAffix
) {
713 (affix
->type
== AffixType_Suffix
) ?
714 AffixType_Prefix
: AffixType_Suffix
,
716 mirror64bit(affix
->value
) >>
717 ((sizeof(affix
->value
) * 8) - affix
->bits
),
723 * affixForIndex() - Generates suffix numbers with "suffix-free" property.
724 * @index: natural number (or index) of the suffix to be generated
726 * Return: Suffix suitable to assemble unique number.
728 * This is just a wrapper function on top of the Exp. Golomb algorithm.
730 * Since the Exp. Golomb algorithm generates prefixes, but we need suffixes,
731 * this function converts the Exp. Golomb prefixes into appropriate suffixes
732 * which are still suitable for generating unique numbers.
734 static VariLenAffix
affixForIndex(uint64_t index
)
737 prefix
= expGolombEncode(index
, EXP_GOLOMB_K
);
738 return invertAffix(&prefix
); /* convert prefix to suffix */
741 /* creative abuse of tb_hash_func7, which is based on xxhash */
742 static uint32_t qpp_hash(QppEntry e
)
744 return qemu_xxhash7(e
.ino_prefix
, e
.dev
, 0, 0, 0);
747 static uint32_t qpf_hash(QpfEntry e
)
749 return qemu_xxhash7(e
.ino
, e
.dev
, 0, 0, 0);
752 static bool qpd_cmp_func(const void *obj
, const void *userp
)
754 const QpdEntry
*e1
= obj
, *e2
= userp
;
755 return e1
->dev
== e2
->dev
;
758 static bool qpp_cmp_func(const void *obj
, const void *userp
)
760 const QppEntry
*e1
= obj
, *e2
= userp
;
761 return e1
->dev
== e2
->dev
&& e1
->ino_prefix
== e2
->ino_prefix
;
764 static bool qpf_cmp_func(const void *obj
, const void *userp
)
766 const QpfEntry
*e1
= obj
, *e2
= userp
;
767 return e1
->dev
== e2
->dev
&& e1
->ino
== e2
->ino
;
770 static void qp_table_remove(void *p
, uint32_t h
, void *up
)
775 static void qp_table_destroy(struct qht
*ht
)
777 if (!ht
|| !ht
->map
) {
780 qht_iter(ht
, qp_table_remove
, NULL
);
784 static void qpd_table_init(struct qht
*ht
)
786 qht_init(ht
, qpd_cmp_func
, 1, QHT_MODE_AUTO_RESIZE
);
789 static void qpp_table_init(struct qht
*ht
)
791 qht_init(ht
, qpp_cmp_func
, 1, QHT_MODE_AUTO_RESIZE
);
794 static void qpf_table_init(struct qht
*ht
)
796 qht_init(ht
, qpf_cmp_func
, 1 << 16, QHT_MODE_AUTO_RESIZE
);
800 * Returns how many (high end) bits of inode numbers of the passed fs
801 * device shall be used (in combination with the device number) to
802 * generate hash values for qpp_table entries.
804 * This function is required if variable length suffixes are used for inode
805 * number mapping on guest level. Since a device may end up having multiple
806 * entries in qpp_table, each entry most probably with a different suffix
807 * length, we thus need this function in conjunction with qpd_table to
808 * "agree" about a fix amount of bits (per device) to be always used for
809 * generating hash values for the purpose of accessing qpp_table in order
810 * get consistent behaviour when accessing qpp_table.
812 static int qid_inode_prefix_hash_bits(V9fsPDU
*pdu
, dev_t dev
)
820 val
= qht_lookup(&pdu
->s
->qpd_table
, &lookup
, hash
);
822 val
= g_new0(QpdEntry
, 1);
824 affix
= affixForIndex(pdu
->s
->qp_affix_next
);
825 val
->prefix_bits
= affix
.bits
;
826 qht_insert(&pdu
->s
->qpd_table
, val
, hash
, NULL
);
827 pdu
->s
->qp_ndevices
++;
829 return val
->prefix_bits
;
833 * Slow / full mapping host inode nr -> guest inode nr.
835 * This function performs a slower and much more costly remapping of an
836 * original file inode number on host to an appropriate different inode
837 * number on guest. For every (dev, inode) combination on host a new
838 * sequential number is generated, cached and exposed as inode number on
841 * This is just a "last resort" fallback solution if the much faster/cheaper
842 * qid_path_suffixmap() failed. In practice this slow / full mapping is not
843 * expected ever to be used at all though.
845 * See qid_path_suffixmap() for details
848 static int qid_path_fullmap(V9fsPDU
*pdu
, const struct stat
*stbuf
,
852 .dev
= stbuf
->st_dev
,
855 uint32_t hash
= qpf_hash(lookup
);
858 val
= qht_lookup(&pdu
->s
->qpf_table
, &lookup
, hash
);
861 if (pdu
->s
->qp_fullpath_next
== 0) {
862 /* no more files can be mapped :'( */
864 "9p: No more prefixes available for remapping inodes from "
870 val
= g_new0(QpfEntry
, 1);
873 /* new unique inode and device combo */
874 affix
= affixForIndex(
875 1ULL << (sizeof(pdu
->s
->qp_affix_next
) * 8)
877 val
->path
= (pdu
->s
->qp_fullpath_next
++ << affix
.bits
) | affix
.value
;
878 pdu
->s
->qp_fullpath_next
&= ((1ULL << (64 - affix
.bits
)) - 1);
879 qht_insert(&pdu
->s
->qpf_table
, val
, hash
, NULL
);
887 * Quick mapping host inode nr -> guest inode nr.
889 * This function performs quick remapping of an original file inode number
890 * on host to an appropriate different inode number on guest. This remapping
891 * of inodes is required to avoid inode nr collisions on guest which would
892 * happen if the 9p export contains more than 1 exported file system (or
893 * more than 1 file system data set), because unlike on host level where the
894 * files would have different device nrs, all files exported by 9p would
895 * share the same device nr on guest (the device nr of the virtual 9p device
898 * Inode remapping is performed by chopping off high end bits of the original
899 * inode number from host, shifting the result upwards and then assigning a
900 * generated suffix number for the low end bits, where the same suffix number
901 * will be shared by all inodes with the same device id AND the same high end
902 * bits that have been chopped off. That approach utilizes the fact that inode
903 * numbers very likely share the same high end bits (i.e. due to their common
904 * sequential generation by file systems) and hence we only have to generate
905 * and track a very limited amount of suffixes in practice due to that.
907 * We generate variable size suffixes for that purpose. The 1st generated
908 * suffix will only have 1 bit and hence we only need to chop off 1 bit from
909 * the original inode number. The subsequent suffixes being generated will
910 * grow in (bit) size subsequently, i.e. the 2nd and 3rd suffix being
911 * generated will have 3 bits and hence we have to chop off 3 bits from their
912 * original inodes, and so on. That approach of using variable length suffixes
913 * (i.e. over fixed size ones) utilizes the fact that in practice only a very
914 * limited amount of devices are shared by the same export (e.g. typically
915 * less than 2 dozen devices per 9p export), so in practice we need to chop
916 * off less bits than with fixed size prefixes and yet are flexible to add
917 * new devices at runtime below host's export directory at any time without
918 * having to reboot guest nor requiring to reconfigure guest for that. And due
919 * to the very limited amount of original high end bits that we chop off that
920 * way, the total amount of suffixes we need to generate is less than by using
921 * fixed size prefixes and hence it also improves performance of the inode
922 * remapping algorithm, and finally has the nice side effect that the inode
923 * numbers on guest will be much smaller & human friendly. ;-)
925 static int qid_path_suffixmap(V9fsPDU
*pdu
, const struct stat
*stbuf
,
928 const int ino_hash_bits
= qid_inode_prefix_hash_bits(pdu
, stbuf
->st_dev
);
930 .dev
= stbuf
->st_dev
,
931 .ino_prefix
= (uint16_t) (stbuf
->st_ino
>> (64 - ino_hash_bits
))
933 uint32_t hash
= qpp_hash(lookup
);
935 val
= qht_lookup(&pdu
->s
->qpp_table
, &lookup
, hash
);
938 if (pdu
->s
->qp_affix_next
== 0) {
939 /* we ran out of affixes */
941 "9p: Potential degraded performance of inode remapping"
946 val
= g_new0(QppEntry
, 1);
949 /* new unique inode affix and device combo */
950 val
->qp_affix_index
= pdu
->s
->qp_affix_next
++;
951 val
->qp_affix
= affixForIndex(val
->qp_affix_index
);
952 qht_insert(&pdu
->s
->qpp_table
, val
, hash
, NULL
);
954 /* assuming generated affix to be suffix type, not prefix */
955 *path
= (stbuf
->st_ino
<< val
->qp_affix
.bits
) | val
->qp_affix
.value
;
959 static int stat_to_qid(V9fsPDU
*pdu
, const struct stat
*stbuf
, V9fsQID
*qidp
)
964 if (pdu
->s
->ctx
.export_flags
& V9FS_REMAP_INODES
) {
965 /* map inode+device to qid path (fast path) */
966 err
= qid_path_suffixmap(pdu
, stbuf
, &qidp
->path
);
967 if (err
== -ENFILE
) {
968 /* fast path didn't work, fall back to full map */
969 err
= qid_path_fullmap(pdu
, stbuf
, &qidp
->path
);
975 if (pdu
->s
->dev_id
!= stbuf
->st_dev
) {
976 if (pdu
->s
->ctx
.export_flags
& V9FS_FORBID_MULTIDEVS
) {
978 "9p: Multiple devices detected in same VirtFS export. "
979 "Access of guest to additional devices is (partly) "
980 "denied due to virtfs option 'multidevs=forbid' being "
986 "9p: Multiple devices detected in same VirtFS export, "
987 "which might lead to file ID collisions and severe "
988 "misbehaviours on guest! You should either use a "
989 "separate export for each device shared from host or "
990 "use virtfs option 'multidevs=remap'!"
994 memset(&qidp
->path
, 0, sizeof(qidp
->path
));
995 size
= MIN(sizeof(stbuf
->st_ino
), sizeof(qidp
->path
));
996 memcpy(&qidp
->path
, &stbuf
->st_ino
, size
);
999 qidp
->version
= stbuf
->st_mtime
^ (stbuf
->st_size
<< 8);
1001 if (S_ISDIR(stbuf
->st_mode
)) {
1002 qidp
->type
|= P9_QID_TYPE_DIR
;
1004 if (S_ISLNK(stbuf
->st_mode
)) {
1005 qidp
->type
|= P9_QID_TYPE_SYMLINK
;
1011 V9fsPDU
*pdu_alloc(V9fsState
*s
)
1013 V9fsPDU
*pdu
= NULL
;
1015 if (!QLIST_EMPTY(&s
->free_list
)) {
1016 pdu
= QLIST_FIRST(&s
->free_list
);
1017 QLIST_REMOVE(pdu
, next
);
1018 QLIST_INSERT_HEAD(&s
->active_list
, pdu
, next
);
1023 void pdu_free(V9fsPDU
*pdu
)
1025 V9fsState
*s
= pdu
->s
;
1027 g_assert(!pdu
->cancelled
);
1028 QLIST_REMOVE(pdu
, next
);
1029 QLIST_INSERT_HEAD(&s
->free_list
, pdu
, next
);
1032 static void coroutine_fn
pdu_complete(V9fsPDU
*pdu
, ssize_t len
)
1034 int8_t id
= pdu
->id
+ 1; /* Response */
1035 V9fsState
*s
= pdu
->s
;
1039 * The 9p spec requires that successfully cancelled pdus receive no reply.
1040 * Sending a reply would confuse clients because they would
1041 * assume that any EINTR is the actual result of the operation,
1042 * rather than a consequence of the cancellation. However, if
1043 * the operation completed (succesfully or with an error other
1044 * than caused be cancellation), we do send out that reply, both
1045 * for efficiency and to avoid confusing the rest of the state machine
1046 * that assumes passing a non-error here will mean a successful
1047 * transmission of the reply.
1049 bool discard
= pdu
->cancelled
&& len
== -EINTR
;
1051 trace_v9fs_rcancel(pdu
->tag
, pdu
->id
);
1060 if (s
->proto_version
!= V9FS_PROTO_2000L
) {
1063 str
.data
= strerror(err
);
1064 str
.size
= strlen(str
.data
);
1066 ret
= pdu_marshal(pdu
, len
, "s", &str
);
1073 err
= errno_to_dotl(err
);
1076 ret
= pdu_marshal(pdu
, len
, "d", err
);
1082 if (s
->proto_version
== V9FS_PROTO_2000L
) {
1085 trace_v9fs_rerror(pdu
->tag
, pdu
->id
, err
); /* Trace ERROR */
1088 /* fill out the header */
1089 if (pdu_marshal(pdu
, 0, "dbw", (int32_t)len
, id
, pdu
->tag
) < 0) {
1093 /* keep these in sync */
1098 pdu
->s
->transport
->push_and_notify(pdu
);
1100 /* Now wakeup anybody waiting in flush for this request */
1101 if (!qemu_co_queue_next(&pdu
->complete
)) {
1106 static mode_t
v9mode_to_mode(uint32_t mode
, V9fsString
*extension
)
1111 if (mode
& P9_STAT_MODE_DIR
) {
1115 if (mode
& P9_STAT_MODE_SYMLINK
) {
1118 if (mode
& P9_STAT_MODE_SOCKET
) {
1121 if (mode
& P9_STAT_MODE_NAMED_PIPE
) {
1124 if (mode
& P9_STAT_MODE_DEVICE
) {
1125 if (extension
->size
&& extension
->data
[0] == 'c') {
1132 if (!(ret
& ~0777)) {
1136 if (mode
& P9_STAT_MODE_SETUID
) {
1139 if (mode
& P9_STAT_MODE_SETGID
) {
1142 if (mode
& P9_STAT_MODE_SETVTX
) {
1149 static int donttouch_stat(V9fsStat
*stat
)
1151 if (stat
->type
== -1 &&
1153 stat
->qid
.type
== 0xff &&
1154 stat
->qid
.version
== (uint32_t) -1 &&
1155 stat
->qid
.path
== (uint64_t) -1 &&
1157 stat
->atime
== -1 &&
1158 stat
->mtime
== -1 &&
1159 stat
->length
== -1 &&
1164 stat
->n_uid
== -1 &&
1165 stat
->n_gid
== -1 &&
1166 stat
->n_muid
== -1) {
1173 static void v9fs_stat_init(V9fsStat
*stat
)
1175 v9fs_string_init(&stat
->name
);
1176 v9fs_string_init(&stat
->uid
);
1177 v9fs_string_init(&stat
->gid
);
1178 v9fs_string_init(&stat
->muid
);
1179 v9fs_string_init(&stat
->extension
);
1182 static void v9fs_stat_free(V9fsStat
*stat
)
1184 v9fs_string_free(&stat
->name
);
1185 v9fs_string_free(&stat
->uid
);
1186 v9fs_string_free(&stat
->gid
);
1187 v9fs_string_free(&stat
->muid
);
1188 v9fs_string_free(&stat
->extension
);
1191 static uint32_t stat_to_v9mode(const struct stat
*stbuf
)
1195 mode
= stbuf
->st_mode
& 0777;
1196 if (S_ISDIR(stbuf
->st_mode
)) {
1197 mode
|= P9_STAT_MODE_DIR
;
1200 if (S_ISLNK(stbuf
->st_mode
)) {
1201 mode
|= P9_STAT_MODE_SYMLINK
;
1204 if (S_ISSOCK(stbuf
->st_mode
)) {
1205 mode
|= P9_STAT_MODE_SOCKET
;
1208 if (S_ISFIFO(stbuf
->st_mode
)) {
1209 mode
|= P9_STAT_MODE_NAMED_PIPE
;
1212 if (S_ISBLK(stbuf
->st_mode
) || S_ISCHR(stbuf
->st_mode
)) {
1213 mode
|= P9_STAT_MODE_DEVICE
;
1216 if (stbuf
->st_mode
& S_ISUID
) {
1217 mode
|= P9_STAT_MODE_SETUID
;
1220 if (stbuf
->st_mode
& S_ISGID
) {
1221 mode
|= P9_STAT_MODE_SETGID
;
1224 if (stbuf
->st_mode
& S_ISVTX
) {
1225 mode
|= P9_STAT_MODE_SETVTX
;
1231 static int coroutine_fn
stat_to_v9stat(V9fsPDU
*pdu
, V9fsPath
*path
,
1232 const char *basename
,
1233 const struct stat
*stbuf
,
1238 memset(v9stat
, 0, sizeof(*v9stat
));
1240 err
= stat_to_qid(pdu
, stbuf
, &v9stat
->qid
);
1244 v9stat
->mode
= stat_to_v9mode(stbuf
);
1245 v9stat
->atime
= stbuf
->st_atime
;
1246 v9stat
->mtime
= stbuf
->st_mtime
;
1247 v9stat
->length
= stbuf
->st_size
;
1249 v9fs_string_free(&v9stat
->uid
);
1250 v9fs_string_free(&v9stat
->gid
);
1251 v9fs_string_free(&v9stat
->muid
);
1253 v9stat
->n_uid
= stbuf
->st_uid
;
1254 v9stat
->n_gid
= stbuf
->st_gid
;
1257 v9fs_string_free(&v9stat
->extension
);
1259 if (v9stat
->mode
& P9_STAT_MODE_SYMLINK
) {
1260 err
= v9fs_co_readlink(pdu
, path
, &v9stat
->extension
);
1264 } else if (v9stat
->mode
& P9_STAT_MODE_DEVICE
) {
1265 v9fs_string_sprintf(&v9stat
->extension
, "%c %u %u",
1266 S_ISCHR(stbuf
->st_mode
) ? 'c' : 'b',
1267 major(stbuf
->st_rdev
), minor(stbuf
->st_rdev
));
1268 } else if (S_ISDIR(stbuf
->st_mode
) || S_ISREG(stbuf
->st_mode
)) {
1269 v9fs_string_sprintf(&v9stat
->extension
, "%s %lu",
1270 "HARDLINKCOUNT", (unsigned long)stbuf
->st_nlink
);
1273 v9fs_string_sprintf(&v9stat
->name
, "%s", basename
);
1276 v9fs_string_size(&v9stat
->name
) +
1277 v9fs_string_size(&v9stat
->uid
) +
1278 v9fs_string_size(&v9stat
->gid
) +
1279 v9fs_string_size(&v9stat
->muid
) +
1280 v9fs_string_size(&v9stat
->extension
);
1284 #define P9_STATS_MODE 0x00000001ULL
1285 #define P9_STATS_NLINK 0x00000002ULL
1286 #define P9_STATS_UID 0x00000004ULL
1287 #define P9_STATS_GID 0x00000008ULL
1288 #define P9_STATS_RDEV 0x00000010ULL
1289 #define P9_STATS_ATIME 0x00000020ULL
1290 #define P9_STATS_MTIME 0x00000040ULL
1291 #define P9_STATS_CTIME 0x00000080ULL
1292 #define P9_STATS_INO 0x00000100ULL
1293 #define P9_STATS_SIZE 0x00000200ULL
1294 #define P9_STATS_BLOCKS 0x00000400ULL
1296 #define P9_STATS_BTIME 0x00000800ULL
1297 #define P9_STATS_GEN 0x00001000ULL
1298 #define P9_STATS_DATA_VERSION 0x00002000ULL
1300 #define P9_STATS_BASIC 0x000007ffULL /* Mask for fields up to BLOCKS */
1301 #define P9_STATS_ALL 0x00003fffULL /* Mask for All fields above */
1305 * blksize_to_iounit() - Block size exposed to 9p client.
1306 * Return: block size
1308 * @pdu: 9p client request
1309 * @blksize: host filesystem's block size
1311 * Convert host filesystem's block size into an appropriate block size for
1312 * 9p client (guest OS side). The value returned suggests an "optimum" block
1313 * size for 9p I/O, i.e. to maximize performance.
1315 static int32_t blksize_to_iounit(const V9fsPDU
*pdu
, int32_t blksize
)
1318 V9fsState
*s
= pdu
->s
;
1321 * iounit should be multiples of blksize (host filesystem block size)
1322 * as well as less than (client msize - P9_IOHDRSZ)
1325 iounit
= QEMU_ALIGN_DOWN(s
->msize
- P9_IOHDRSZ
, blksize
);
1328 iounit
= s
->msize
- P9_IOHDRSZ
;
1333 static int32_t stat_to_iounit(const V9fsPDU
*pdu
, const struct stat
*stbuf
)
1335 return blksize_to_iounit(pdu
, stbuf
->st_blksize
);
1338 static int stat_to_v9stat_dotl(V9fsPDU
*pdu
, const struct stat
*stbuf
,
1339 V9fsStatDotl
*v9lstat
)
1341 memset(v9lstat
, 0, sizeof(*v9lstat
));
1343 v9lstat
->st_mode
= stbuf
->st_mode
;
1344 v9lstat
->st_nlink
= stbuf
->st_nlink
;
1345 v9lstat
->st_uid
= stbuf
->st_uid
;
1346 v9lstat
->st_gid
= stbuf
->st_gid
;
1347 v9lstat
->st_rdev
= host_dev_to_dotl_dev(stbuf
->st_rdev
);
1348 v9lstat
->st_size
= stbuf
->st_size
;
1349 v9lstat
->st_blksize
= stat_to_iounit(pdu
, stbuf
);
1350 v9lstat
->st_blocks
= stbuf
->st_blocks
;
1351 v9lstat
->st_atime_sec
= stbuf
->st_atime
;
1352 v9lstat
->st_mtime_sec
= stbuf
->st_mtime
;
1353 v9lstat
->st_ctime_sec
= stbuf
->st_ctime
;
1354 #ifdef CONFIG_DARWIN
1355 v9lstat
->st_atime_nsec
= stbuf
->st_atimespec
.tv_nsec
;
1356 v9lstat
->st_mtime_nsec
= stbuf
->st_mtimespec
.tv_nsec
;
1357 v9lstat
->st_ctime_nsec
= stbuf
->st_ctimespec
.tv_nsec
;
1359 v9lstat
->st_atime_nsec
= stbuf
->st_atim
.tv_nsec
;
1360 v9lstat
->st_mtime_nsec
= stbuf
->st_mtim
.tv_nsec
;
1361 v9lstat
->st_ctime_nsec
= stbuf
->st_ctim
.tv_nsec
;
1363 /* Currently we only support BASIC fields in stat */
1364 v9lstat
->st_result_mask
= P9_STATS_BASIC
;
1366 return stat_to_qid(pdu
, stbuf
, &v9lstat
->qid
);
1369 static void print_sg(struct iovec
*sg
, int cnt
)
1373 printf("sg[%d]: {", cnt
);
1374 for (i
= 0; i
< cnt
; i
++) {
1378 printf("(%p, %zd)", sg
[i
].iov_base
, sg
[i
].iov_len
);
1383 /* Will call this only for path name based fid */
1384 static void v9fs_fix_path(V9fsPath
*dst
, V9fsPath
*src
, int len
)
1387 v9fs_path_init(&str
);
1388 v9fs_path_copy(&str
, dst
);
1389 v9fs_path_sprintf(dst
, "%s%s", src
->data
, str
.data
+ len
);
1390 v9fs_path_free(&str
);
1393 static inline bool is_ro_export(FsContext
*ctx
)
1395 return ctx
->export_flags
& V9FS_RDONLY
;
1398 static void coroutine_fn
v9fs_version(void *opaque
)
1401 V9fsPDU
*pdu
= opaque
;
1402 V9fsState
*s
= pdu
->s
;
1406 v9fs_string_init(&version
);
1407 err
= pdu_unmarshal(pdu
, offset
, "ds", &s
->msize
, &version
);
1411 trace_v9fs_version(pdu
->tag
, pdu
->id
, s
->msize
, version
.data
);
1415 if (!strcmp(version
.data
, "9P2000.u")) {
1416 s
->proto_version
= V9FS_PROTO_2000U
;
1417 } else if (!strcmp(version
.data
, "9P2000.L")) {
1418 s
->proto_version
= V9FS_PROTO_2000L
;
1420 v9fs_string_sprintf(&version
, "unknown");
1421 /* skip min. msize check, reporting invalid version has priority */
1425 if (s
->msize
< P9_MIN_MSIZE
) {
1428 "9pfs: Client requested msize < minimum msize ("
1429 stringify(P9_MIN_MSIZE
) ") supported by this server."
1434 /* 8192 is the default msize of Linux clients */
1435 if (s
->msize
<= 8192 && !(s
->ctx
.export_flags
& V9FS_NO_PERF_WARN
)) {
1437 "9p: degraded performance: a reasonable high msize should be "
1438 "chosen on client/guest side (chosen msize is <= 8192). See "
1439 "https://wiki.qemu.org/Documentation/9psetup#msize for details."
1444 err
= pdu_marshal(pdu
, offset
, "ds", s
->msize
, &version
);
1449 trace_v9fs_version_return(pdu
->tag
, pdu
->id
, s
->msize
, version
.data
);
1451 pdu_complete(pdu
, err
);
1452 v9fs_string_free(&version
);
1455 static void coroutine_fn
v9fs_attach(void *opaque
)
1457 V9fsPDU
*pdu
= opaque
;
1458 V9fsState
*s
= pdu
->s
;
1459 int32_t fid
, afid
, n_uname
;
1460 V9fsString uname
, aname
;
1467 v9fs_string_init(&uname
);
1468 v9fs_string_init(&aname
);
1469 err
= pdu_unmarshal(pdu
, offset
, "ddssd", &fid
,
1470 &afid
, &uname
, &aname
, &n_uname
);
1474 trace_v9fs_attach(pdu
->tag
, pdu
->id
, fid
, afid
, uname
.data
, aname
.data
);
1476 fidp
= alloc_fid(s
, fid
);
1481 fidp
->uid
= n_uname
;
1482 err
= v9fs_co_name_to_path(pdu
, NULL
, "/", &fidp
->path
);
1488 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1494 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
1502 * disable migration if we haven't done already.
1503 * attach could get called multiple times for the same export.
1505 if (!s
->migration_blocker
) {
1506 error_setg(&s
->migration_blocker
,
1507 "Migration is disabled when VirtFS export path '%s' is mounted in the guest using mount_tag '%s'",
1508 s
->ctx
.fs_root
? s
->ctx
.fs_root
: "NULL", s
->tag
);
1509 err
= migrate_add_blocker(s
->migration_blocker
, NULL
);
1511 error_free(s
->migration_blocker
);
1512 s
->migration_blocker
= NULL
;
1519 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
1526 memcpy(&s
->root_st
, &stbuf
, sizeof(stbuf
));
1527 trace_v9fs_attach_return(pdu
->tag
, pdu
->id
,
1528 qid
.type
, qid
.version
, qid
.path
);
1532 pdu_complete(pdu
, err
);
1533 v9fs_string_free(&uname
);
1534 v9fs_string_free(&aname
);
1537 static void coroutine_fn
v9fs_stat(void *opaque
)
1545 V9fsPDU
*pdu
= opaque
;
1548 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
1552 trace_v9fs_stat(pdu
->tag
, pdu
->id
, fid
);
1554 fidp
= get_fid(pdu
, fid
);
1559 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1563 basename
= g_path_get_basename(fidp
->path
.data
);
1564 err
= stat_to_v9stat(pdu
, &fidp
->path
, basename
, &stbuf
, &v9stat
);
1569 err
= pdu_marshal(pdu
, offset
, "wS", 0, &v9stat
);
1571 v9fs_stat_free(&v9stat
);
1574 trace_v9fs_stat_return(pdu
->tag
, pdu
->id
, v9stat
.mode
,
1575 v9stat
.atime
, v9stat
.mtime
, v9stat
.length
);
1577 v9fs_stat_free(&v9stat
);
1581 pdu_complete(pdu
, err
);
1584 static void coroutine_fn
v9fs_getattr(void *opaque
)
1591 uint64_t request_mask
;
1592 V9fsStatDotl v9stat_dotl
;
1593 V9fsPDU
*pdu
= opaque
;
1595 retval
= pdu_unmarshal(pdu
, offset
, "dq", &fid
, &request_mask
);
1599 trace_v9fs_getattr(pdu
->tag
, pdu
->id
, fid
, request_mask
);
1601 fidp
= get_fid(pdu
, fid
);
1607 * Currently we only support BASIC fields in stat, so there is no
1608 * need to look at request_mask.
1610 retval
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1614 retval
= stat_to_v9stat_dotl(pdu
, &stbuf
, &v9stat_dotl
);
1619 /* fill st_gen if requested and supported by underlying fs */
1620 if (request_mask
& P9_STATS_GEN
) {
1621 retval
= v9fs_co_st_gen(pdu
, &fidp
->path
, stbuf
.st_mode
, &v9stat_dotl
);
1624 /* we have valid st_gen: update result mask */
1625 v9stat_dotl
.st_result_mask
|= P9_STATS_GEN
;
1628 /* request cancelled, e.g. by Tflush */
1631 /* failed to get st_gen: not fatal, ignore */
1635 retval
= pdu_marshal(pdu
, offset
, "A", &v9stat_dotl
);
1640 trace_v9fs_getattr_return(pdu
->tag
, pdu
->id
, v9stat_dotl
.st_result_mask
,
1641 v9stat_dotl
.st_mode
, v9stat_dotl
.st_uid
,
1642 v9stat_dotl
.st_gid
);
1646 pdu_complete(pdu
, retval
);
1649 /* Attribute flags */
1650 #define P9_ATTR_MODE (1 << 0)
1651 #define P9_ATTR_UID (1 << 1)
1652 #define P9_ATTR_GID (1 << 2)
1653 #define P9_ATTR_SIZE (1 << 3)
1654 #define P9_ATTR_ATIME (1 << 4)
1655 #define P9_ATTR_MTIME (1 << 5)
1656 #define P9_ATTR_CTIME (1 << 6)
1657 #define P9_ATTR_ATIME_SET (1 << 7)
1658 #define P9_ATTR_MTIME_SET (1 << 8)
1660 #define P9_ATTR_MASK 127
1662 static void coroutine_fn
v9fs_setattr(void *opaque
)
1669 V9fsPDU
*pdu
= opaque
;
1671 err
= pdu_unmarshal(pdu
, offset
, "dI", &fid
, &v9iattr
);
1676 trace_v9fs_setattr(pdu
->tag
, pdu
->id
, fid
,
1677 v9iattr
.valid
, v9iattr
.mode
, v9iattr
.uid
, v9iattr
.gid
,
1678 v9iattr
.size
, v9iattr
.atime_sec
, v9iattr
.mtime_sec
);
1680 fidp
= get_fid(pdu
, fid
);
1685 if (v9iattr
.valid
& P9_ATTR_MODE
) {
1686 err
= v9fs_co_chmod(pdu
, &fidp
->path
, v9iattr
.mode
);
1691 if (v9iattr
.valid
& (P9_ATTR_ATIME
| P9_ATTR_MTIME
)) {
1692 struct timespec times
[2];
1693 if (v9iattr
.valid
& P9_ATTR_ATIME
) {
1694 if (v9iattr
.valid
& P9_ATTR_ATIME_SET
) {
1695 times
[0].tv_sec
= v9iattr
.atime_sec
;
1696 times
[0].tv_nsec
= v9iattr
.atime_nsec
;
1698 times
[0].tv_nsec
= UTIME_NOW
;
1701 times
[0].tv_nsec
= UTIME_OMIT
;
1703 if (v9iattr
.valid
& P9_ATTR_MTIME
) {
1704 if (v9iattr
.valid
& P9_ATTR_MTIME_SET
) {
1705 times
[1].tv_sec
= v9iattr
.mtime_sec
;
1706 times
[1].tv_nsec
= v9iattr
.mtime_nsec
;
1708 times
[1].tv_nsec
= UTIME_NOW
;
1711 times
[1].tv_nsec
= UTIME_OMIT
;
1713 err
= v9fs_co_utimensat(pdu
, &fidp
->path
, times
);
1719 * If the only valid entry in iattr is ctime we can call
1720 * chown(-1,-1) to update the ctime of the file
1722 if ((v9iattr
.valid
& (P9_ATTR_UID
| P9_ATTR_GID
)) ||
1723 ((v9iattr
.valid
& P9_ATTR_CTIME
)
1724 && !((v9iattr
.valid
& P9_ATTR_MASK
) & ~P9_ATTR_CTIME
))) {
1725 if (!(v9iattr
.valid
& P9_ATTR_UID
)) {
1728 if (!(v9iattr
.valid
& P9_ATTR_GID
)) {
1731 err
= v9fs_co_chown(pdu
, &fidp
->path
, v9iattr
.uid
,
1737 if (v9iattr
.valid
& (P9_ATTR_SIZE
)) {
1738 err
= v9fs_co_truncate(pdu
, &fidp
->path
, v9iattr
.size
);
1744 trace_v9fs_setattr_return(pdu
->tag
, pdu
->id
);
1748 pdu_complete(pdu
, err
);
1751 static int v9fs_walk_marshal(V9fsPDU
*pdu
, uint16_t nwnames
, V9fsQID
*qids
)
1757 err
= pdu_marshal(pdu
, offset
, "w", nwnames
);
1762 for (i
= 0; i
< nwnames
; i
++) {
1763 err
= pdu_marshal(pdu
, offset
, "Q", &qids
[i
]);
1772 static bool name_is_illegal(const char *name
)
1774 return !*name
|| strchr(name
, '/') != NULL
;
1777 static bool same_stat_id(const struct stat
*a
, const struct stat
*b
)
1779 return a
->st_dev
== b
->st_dev
&& a
->st_ino
== b
->st_ino
;
1782 static void coroutine_fn
v9fs_walk(void *opaque
)
1784 int name_idx
, nwalked
;
1785 g_autofree V9fsQID
*qids
= NULL
;
1786 int i
, err
= 0, any_err
= 0;
1787 V9fsPath dpath
, path
;
1788 P9ARRAY_REF(V9fsPath
) pathes
= NULL
;
1790 struct stat stbuf
, fidst
;
1791 g_autofree
struct stat
*stbufs
= NULL
;
1793 int32_t fid
, newfid
;
1794 P9ARRAY_REF(V9fsString
) wnames
= NULL
;
1796 V9fsFidState
*newfidp
= NULL
;
1797 V9fsPDU
*pdu
= opaque
;
1798 V9fsState
*s
= pdu
->s
;
1801 err
= pdu_unmarshal(pdu
, offset
, "ddw", &fid
, &newfid
, &nwnames
);
1803 pdu_complete(pdu
, err
);
1808 trace_v9fs_walk(pdu
->tag
, pdu
->id
, fid
, newfid
, nwnames
);
1810 if (nwnames
> P9_MAXWELEM
) {
1815 P9ARRAY_NEW(V9fsString
, wnames
, nwnames
);
1816 qids
= g_new0(V9fsQID
, nwnames
);
1817 stbufs
= g_new0(struct stat
, nwnames
);
1818 P9ARRAY_NEW(V9fsPath
, pathes
, nwnames
);
1819 for (i
= 0; i
< nwnames
; i
++) {
1820 err
= pdu_unmarshal(pdu
, offset
, "s", &wnames
[i
]);
1824 if (name_is_illegal(wnames
[i
].data
)) {
1831 fidp
= get_fid(pdu
, fid
);
1837 v9fs_path_init(&dpath
);
1838 v9fs_path_init(&path
);
1840 * Both dpath and path initially point to fidp.
1841 * Needed to handle request with nwnames == 0
1843 v9fs_path_copy(&dpath
, &fidp
->path
);
1844 v9fs_path_copy(&path
, &fidp
->path
);
1847 * To keep latency (i.e. overall execution time for processing this
1848 * Twalk client request) as small as possible, run all the required fs
1849 * driver code altogether inside the following block.
1851 v9fs_co_run_in_worker({
1853 if (v9fs_request_cancelled(pdu
)) {
1854 any_err
|= err
= -EINTR
;
1857 err
= s
->ops
->lstat(&s
->ctx
, &dpath
, &fidst
);
1859 any_err
|= err
= -errno
;
1863 for (; nwalked
< nwnames
; nwalked
++) {
1864 if (v9fs_request_cancelled(pdu
)) {
1865 any_err
|= err
= -EINTR
;
1868 if (!same_stat_id(&pdu
->s
->root_st
, &stbuf
) ||
1869 strcmp("..", wnames
[nwalked
].data
))
1871 err
= s
->ops
->name_to_path(&s
->ctx
, &dpath
,
1872 wnames
[nwalked
].data
,
1875 any_err
|= err
= -errno
;
1878 if (v9fs_request_cancelled(pdu
)) {
1879 any_err
|= err
= -EINTR
;
1882 err
= s
->ops
->lstat(&s
->ctx
, &pathes
[nwalked
], &stbuf
);
1884 any_err
|= err
= -errno
;
1887 stbufs
[nwalked
] = stbuf
;
1888 v9fs_path_copy(&dpath
, &pathes
[nwalked
]);
1893 * Handle all the rest of this Twalk request on main thread ...
1895 * NOTE: -EINTR is an exception where we deviate from the protocol spec
1896 * and simply send a (R)Lerror response instead of bothering to assemble
1897 * a (deducted) Rwalk response; because -EINTR is always the result of a
1898 * Tflush request, so client would no longer wait for a response in this
1901 if ((err
< 0 && !nwalked
) || err
== -EINTR
) {
1905 any_err
|= err
= stat_to_qid(pdu
, &fidst
, &qid
);
1906 if (err
< 0 && !nwalked
) {
1911 /* reset dpath and path */
1912 v9fs_path_copy(&dpath
, &fidp
->path
);
1913 v9fs_path_copy(&path
, &fidp
->path
);
1915 for (name_idx
= 0; name_idx
< nwalked
; name_idx
++) {
1916 if (!same_stat_id(&pdu
->s
->root_st
, &stbuf
) ||
1917 strcmp("..", wnames
[name_idx
].data
))
1919 stbuf
= stbufs
[name_idx
];
1920 any_err
|= err
= stat_to_qid(pdu
, &stbuf
, &qid
);
1924 v9fs_path_copy(&path
, &pathes
[name_idx
]);
1925 v9fs_path_copy(&dpath
, &path
);
1927 memcpy(&qids
[name_idx
], &qid
, sizeof(qid
));
1931 /* don't send any QIDs, send Rlerror instead */
1934 /* send QIDs (not Rlerror), but fid MUST remain unaffected */
1938 if (fid
== newfid
) {
1939 if (fidp
->fid_type
!= P9_FID_NONE
) {
1943 v9fs_path_write_lock(s
);
1944 v9fs_path_copy(&fidp
->path
, &path
);
1945 v9fs_path_unlock(s
);
1947 newfidp
= alloc_fid(s
, newfid
);
1948 if (newfidp
== NULL
) {
1952 newfidp
->uid
= fidp
->uid
;
1953 v9fs_path_copy(&newfidp
->path
, &path
);
1956 err
= v9fs_walk_marshal(pdu
, name_idx
, qids
);
1957 trace_v9fs_walk_return(pdu
->tag
, pdu
->id
, name_idx
, qids
);
1961 put_fid(pdu
, newfidp
);
1963 v9fs_path_free(&dpath
);
1964 v9fs_path_free(&path
);
1966 pdu_complete(pdu
, err
);
1969 static int32_t coroutine_fn
get_iounit(V9fsPDU
*pdu
, V9fsPath
*path
)
1971 struct statfs stbuf
;
1972 int err
= v9fs_co_statfs(pdu
, path
, &stbuf
);
1974 return blksize_to_iounit(pdu
, (err
>= 0) ? stbuf
.f_bsize
: 0);
1977 static void coroutine_fn
v9fs_open(void *opaque
)
1988 V9fsPDU
*pdu
= opaque
;
1989 V9fsState
*s
= pdu
->s
;
1991 if (s
->proto_version
== V9FS_PROTO_2000L
) {
1992 err
= pdu_unmarshal(pdu
, offset
, "dd", &fid
, &mode
);
1995 err
= pdu_unmarshal(pdu
, offset
, "db", &fid
, &modebyte
);
2001 trace_v9fs_open(pdu
->tag
, pdu
->id
, fid
, mode
);
2003 fidp
= get_fid(pdu
, fid
);
2008 if (fidp
->fid_type
!= P9_FID_NONE
) {
2013 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
2017 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
2021 if (S_ISDIR(stbuf
.st_mode
)) {
2022 err
= v9fs_co_opendir(pdu
, fidp
);
2026 fidp
->fid_type
= P9_FID_DIR
;
2027 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, 0);
2033 if (s
->proto_version
== V9FS_PROTO_2000L
) {
2034 flags
= get_dotl_openflags(s
, mode
);
2036 flags
= omode_to_uflags(mode
);
2038 if (is_ro_export(&s
->ctx
)) {
2039 if (mode
& O_WRONLY
|| mode
& O_RDWR
||
2040 mode
& O_APPEND
|| mode
& O_TRUNC
) {
2045 err
= v9fs_co_open(pdu
, fidp
, flags
);
2049 fidp
->fid_type
= P9_FID_FILE
;
2050 fidp
->open_flags
= flags
;
2051 if (flags
& O_EXCL
) {
2053 * We let the host file system do O_EXCL check
2054 * We should not reclaim such fd
2056 fidp
->flags
|= FID_NON_RECLAIMABLE
;
2058 iounit
= get_iounit(pdu
, &fidp
->path
);
2059 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
2065 trace_v9fs_open_return(pdu
->tag
, pdu
->id
,
2066 qid
.type
, qid
.version
, qid
.path
, iounit
);
2070 pdu_complete(pdu
, err
);
2073 static void coroutine_fn
v9fs_lcreate(void *opaque
)
2075 int32_t dfid
, flags
, mode
;
2084 V9fsPDU
*pdu
= opaque
;
2086 v9fs_string_init(&name
);
2087 err
= pdu_unmarshal(pdu
, offset
, "dsddd", &dfid
,
2088 &name
, &flags
, &mode
, &gid
);
2092 trace_v9fs_lcreate(pdu
->tag
, pdu
->id
, dfid
, flags
, mode
, gid
);
2094 if (name_is_illegal(name
.data
)) {
2099 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2104 fidp
= get_fid(pdu
, dfid
);
2109 if (fidp
->fid_type
!= P9_FID_NONE
) {
2114 flags
= get_dotl_openflags(pdu
->s
, flags
);
2115 err
= v9fs_co_open2(pdu
, fidp
, &name
, gid
,
2116 flags
| O_CREAT
, mode
, &stbuf
);
2120 fidp
->fid_type
= P9_FID_FILE
;
2121 fidp
->open_flags
= flags
;
2122 if (flags
& O_EXCL
) {
2124 * We let the host file system do O_EXCL check
2125 * We should not reclaim such fd
2127 fidp
->flags
|= FID_NON_RECLAIMABLE
;
2129 iounit
= get_iounit(pdu
, &fidp
->path
);
2130 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
2134 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
2139 trace_v9fs_lcreate_return(pdu
->tag
, pdu
->id
,
2140 qid
.type
, qid
.version
, qid
.path
, iounit
);
2144 pdu_complete(pdu
, err
);
2145 v9fs_string_free(&name
);
2148 static void coroutine_fn
v9fs_fsync(void *opaque
)
2155 V9fsPDU
*pdu
= opaque
;
2157 err
= pdu_unmarshal(pdu
, offset
, "dd", &fid
, &datasync
);
2161 trace_v9fs_fsync(pdu
->tag
, pdu
->id
, fid
, datasync
);
2163 fidp
= get_fid(pdu
, fid
);
2168 err
= v9fs_co_fsync(pdu
, fidp
, datasync
);
2174 pdu_complete(pdu
, err
);
2177 static void coroutine_fn
v9fs_clunk(void *opaque
)
2183 V9fsPDU
*pdu
= opaque
;
2184 V9fsState
*s
= pdu
->s
;
2186 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
2190 trace_v9fs_clunk(pdu
->tag
, pdu
->id
, fid
);
2192 fidp
= clunk_fid(s
, fid
);
2198 * Bump the ref so that put_fid will
2202 err
= put_fid(pdu
, fidp
);
2207 pdu_complete(pdu
, err
);
2211 * Create a QEMUIOVector for a sub-region of PDU iovecs
2213 * @qiov: uninitialized QEMUIOVector
2214 * @skip: number of bytes to skip from beginning of PDU
2215 * @size: number of bytes to include
2216 * @is_write: true - write, false - read
2218 * The resulting QEMUIOVector has heap-allocated iovecs and must be cleaned up
2219 * with qemu_iovec_destroy().
2221 static void v9fs_init_qiov_from_pdu(QEMUIOVector
*qiov
, V9fsPDU
*pdu
,
2222 size_t skip
, size_t size
,
2230 pdu
->s
->transport
->init_out_iov_from_pdu(pdu
, &iov
, &niov
, size
+ skip
);
2232 pdu
->s
->transport
->init_in_iov_from_pdu(pdu
, &iov
, &niov
, size
+ skip
);
2235 qemu_iovec_init_external(&elem
, iov
, niov
);
2236 qemu_iovec_init(qiov
, niov
);
2237 qemu_iovec_concat(qiov
, &elem
, skip
, size
);
2240 static int v9fs_xattr_read(V9fsState
*s
, V9fsPDU
*pdu
, V9fsFidState
*fidp
,
2241 uint64_t off
, uint32_t max_count
)
2245 uint64_t read_count
;
2246 QEMUIOVector qiov_full
;
2248 if (fidp
->fs
.xattr
.len
< off
) {
2251 read_count
= fidp
->fs
.xattr
.len
- off
;
2253 if (read_count
> max_count
) {
2254 read_count
= max_count
;
2256 err
= pdu_marshal(pdu
, offset
, "d", read_count
);
2262 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
, read_count
, false);
2263 err
= v9fs_pack(qiov_full
.iov
, qiov_full
.niov
, 0,
2264 ((char *)fidp
->fs
.xattr
.value
) + off
,
2266 qemu_iovec_destroy(&qiov_full
);
2274 static int coroutine_fn
v9fs_do_readdir_with_stat(V9fsPDU
*pdu
,
2283 off_t saved_dir_pos
;
2284 struct dirent
*dent
;
2286 /* save the directory position */
2287 saved_dir_pos
= v9fs_co_telldir(pdu
, fidp
);
2288 if (saved_dir_pos
< 0) {
2289 return saved_dir_pos
;
2293 v9fs_path_init(&path
);
2295 v9fs_readdir_lock(&fidp
->fs
.dir
);
2297 err
= v9fs_co_readdir(pdu
, fidp
, &dent
);
2301 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, dent
->d_name
, &path
);
2305 err
= v9fs_co_lstat(pdu
, &path
, &stbuf
);
2309 err
= stat_to_v9stat(pdu
, &path
, dent
->d_name
, &stbuf
, &v9stat
);
2313 if ((count
+ v9stat
.size
+ 2) > max_count
) {
2314 v9fs_readdir_unlock(&fidp
->fs
.dir
);
2316 /* Ran out of buffer. Set dir back to old position and return */
2317 v9fs_co_seekdir(pdu
, fidp
, saved_dir_pos
);
2318 v9fs_stat_free(&v9stat
);
2319 v9fs_path_free(&path
);
2323 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2324 len
= pdu_marshal(pdu
, 11 + count
, "S", &v9stat
);
2326 v9fs_readdir_unlock(&fidp
->fs
.dir
);
2329 v9fs_co_seekdir(pdu
, fidp
, saved_dir_pos
);
2330 v9fs_stat_free(&v9stat
);
2331 v9fs_path_free(&path
);
2335 v9fs_stat_free(&v9stat
);
2336 v9fs_path_free(&path
);
2337 saved_dir_pos
= qemu_dirent_off(dent
);
2340 v9fs_readdir_unlock(&fidp
->fs
.dir
);
2342 v9fs_path_free(&path
);
2349 static void coroutine_fn
v9fs_read(void *opaque
)
2358 V9fsPDU
*pdu
= opaque
;
2359 V9fsState
*s
= pdu
->s
;
2361 err
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
, &off
, &max_count
);
2365 trace_v9fs_read(pdu
->tag
, pdu
->id
, fid
, off
, max_count
);
2367 fidp
= get_fid(pdu
, fid
);
2372 if (fidp
->fid_type
== P9_FID_DIR
) {
2373 if (s
->proto_version
!= V9FS_PROTO_2000U
) {
2375 "9p: bad client: T_read request on directory only expected "
2376 "with 9P2000.u protocol version"
2382 v9fs_co_rewinddir(pdu
, fidp
);
2384 count
= v9fs_do_readdir_with_stat(pdu
, fidp
, max_count
);
2389 err
= pdu_marshal(pdu
, offset
, "d", count
);
2393 err
+= offset
+ count
;
2394 } else if (fidp
->fid_type
== P9_FID_FILE
) {
2395 QEMUIOVector qiov_full
;
2399 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
+ 4, max_count
, false);
2400 qemu_iovec_init(&qiov
, qiov_full
.niov
);
2402 qemu_iovec_reset(&qiov
);
2403 qemu_iovec_concat(&qiov
, &qiov_full
, count
, qiov_full
.size
- count
);
2405 print_sg(qiov
.iov
, qiov
.niov
);
2407 /* Loop in case of EINTR */
2409 len
= v9fs_co_preadv(pdu
, fidp
, qiov
.iov
, qiov
.niov
, off
);
2414 } while (len
== -EINTR
&& !pdu
->cancelled
);
2416 /* IO error return the error */
2418 goto out_free_iovec
;
2420 } while (count
< max_count
&& len
> 0);
2421 err
= pdu_marshal(pdu
, offset
, "d", count
);
2423 goto out_free_iovec
;
2425 err
+= offset
+ count
;
2427 qemu_iovec_destroy(&qiov
);
2428 qemu_iovec_destroy(&qiov_full
);
2429 } else if (fidp
->fid_type
== P9_FID_XATTR
) {
2430 err
= v9fs_xattr_read(s
, pdu
, fidp
, off
, max_count
);
2434 trace_v9fs_read_return(pdu
->tag
, pdu
->id
, count
, err
);
2438 pdu_complete(pdu
, err
);
2442 * v9fs_readdir_response_size() - Returns size required in Rreaddir response
2443 * for the passed dirent @name.
2445 * @name: directory entry's name (i.e. file name, directory name)
2446 * Return: required size in bytes
2448 size_t v9fs_readdir_response_size(V9fsString
*name
)
2451 * Size of each dirent on the wire: size of qid (13) + size of offset (8)
2452 * size of type (1) + size of name.size (2) + strlen(name.data)
2454 return 24 + v9fs_string_size(name
);
2457 static void v9fs_free_dirents(struct V9fsDirEnt
*e
)
2459 struct V9fsDirEnt
*next
= NULL
;
2461 for (; e
; e
= next
) {
2469 static int coroutine_fn
v9fs_do_readdir(V9fsPDU
*pdu
, V9fsFidState
*fidp
,
2470 off_t offset
, int32_t max_count
)
2478 struct dirent
*dent
;
2480 struct V9fsDirEnt
*entries
= NULL
;
2483 * inode remapping requires the device id, which in turn might be
2484 * different for different directory entries, so if inode remapping is
2485 * enabled we have to make a full stat for each directory entry
2487 const bool dostat
= pdu
->s
->ctx
.export_flags
& V9FS_REMAP_INODES
;
2490 * Fetch all required directory entries altogether on a background IO
2491 * thread from fs driver. We don't want to do that for each entry
2492 * individually, because hopping between threads (this main IO thread
2493 * and background IO driver thread) would sum up to huge latencies.
2495 count
= v9fs_co_readdir_many(pdu
, fidp
, &entries
, offset
, max_count
,
2504 for (struct V9fsDirEnt
*e
= entries
; e
; e
= e
->next
) {
2507 if (pdu
->s
->ctx
.export_flags
& V9FS_REMAP_INODES
) {
2509 /* e->st should never be NULL, but just to be sure */
2516 err
= stat_to_qid(pdu
, st
, &qid
);
2522 * Fill up just the path field of qid because the client uses
2523 * only that. To fill the entire qid structure we will have
2524 * to stat each dirent found, which is expensive. For the
2525 * latter reason we don't call stat_to_qid() here. Only drawback
2526 * is that no multi-device export detection of stat_to_qid()
2527 * would be done and provided as error to the user here. But
2528 * user would get that error anyway when accessing those
2529 * files/dirs through other ways.
2531 size
= MIN(sizeof(dent
->d_ino
), sizeof(qid
.path
));
2532 memcpy(&qid
.path
, &dent
->d_ino
, size
);
2533 /* Fill the other fields with dummy values */
2538 off
= qemu_dirent_off(dent
);
2539 v9fs_string_init(&name
);
2540 v9fs_string_sprintf(&name
, "%s", dent
->d_name
);
2542 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2543 len
= pdu_marshal(pdu
, 11 + count
, "Qqbs",
2545 dent
->d_type
, &name
);
2547 v9fs_string_free(&name
);
2558 v9fs_free_dirents(entries
);
2565 static void coroutine_fn
v9fs_readdir(void *opaque
)
2571 uint64_t initial_offset
;
2574 V9fsPDU
*pdu
= opaque
;
2575 V9fsState
*s
= pdu
->s
;
2577 retval
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
,
2578 &initial_offset
, &max_count
);
2582 trace_v9fs_readdir(pdu
->tag
, pdu
->id
, fid
, initial_offset
, max_count
);
2584 /* Enough space for a R_readdir header: size[4] Rreaddir tag[2] count[4] */
2585 if (max_count
> s
->msize
- 11) {
2586 max_count
= s
->msize
- 11;
2588 "9p: bad client: T_readdir with count > msize - 11"
2592 fidp
= get_fid(pdu
, fid
);
2597 if (!fidp
->fs
.dir
.stream
) {
2601 if (s
->proto_version
!= V9FS_PROTO_2000L
) {
2603 "9p: bad client: T_readdir request only expected with 9P2000.L "
2606 retval
= -EOPNOTSUPP
;
2609 count
= v9fs_do_readdir(pdu
, fidp
, (off_t
) initial_offset
, max_count
);
2614 retval
= pdu_marshal(pdu
, offset
, "d", count
);
2618 retval
+= count
+ offset
;
2619 trace_v9fs_readdir_return(pdu
->tag
, pdu
->id
, count
, retval
);
2623 pdu_complete(pdu
, retval
);
2626 static int v9fs_xattr_write(V9fsState
*s
, V9fsPDU
*pdu
, V9fsFidState
*fidp
,
2627 uint64_t off
, uint32_t count
,
2628 struct iovec
*sg
, int cnt
)
2632 uint64_t write_count
;
2636 if (fidp
->fs
.xattr
.len
< off
) {
2639 write_count
= fidp
->fs
.xattr
.len
- off
;
2640 if (write_count
> count
) {
2641 write_count
= count
;
2643 err
= pdu_marshal(pdu
, offset
, "d", write_count
);
2648 fidp
->fs
.xattr
.copied_len
+= write_count
;
2650 * Now copy the content from sg list
2652 for (i
= 0; i
< cnt
; i
++) {
2653 if (write_count
> sg
[i
].iov_len
) {
2654 to_copy
= sg
[i
].iov_len
;
2656 to_copy
= write_count
;
2658 memcpy((char *)fidp
->fs
.xattr
.value
+ off
, sg
[i
].iov_base
, to_copy
);
2659 /* updating vs->off since we are not using below */
2661 write_count
-= to_copy
;
2667 static void coroutine_fn
v9fs_write(void *opaque
)
2677 V9fsPDU
*pdu
= opaque
;
2678 V9fsState
*s
= pdu
->s
;
2679 QEMUIOVector qiov_full
;
2682 err
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
, &off
, &count
);
2684 pdu_complete(pdu
, err
);
2688 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
, count
, true);
2689 trace_v9fs_write(pdu
->tag
, pdu
->id
, fid
, off
, count
, qiov_full
.niov
);
2691 fidp
= get_fid(pdu
, fid
);
2696 if (fidp
->fid_type
== P9_FID_FILE
) {
2697 if (fidp
->fs
.fd
== -1) {
2701 } else if (fidp
->fid_type
== P9_FID_XATTR
) {
2703 * setxattr operation
2705 err
= v9fs_xattr_write(s
, pdu
, fidp
, off
, count
,
2706 qiov_full
.iov
, qiov_full
.niov
);
2712 qemu_iovec_init(&qiov
, qiov_full
.niov
);
2714 qemu_iovec_reset(&qiov
);
2715 qemu_iovec_concat(&qiov
, &qiov_full
, total
, qiov_full
.size
- total
);
2717 print_sg(qiov
.iov
, qiov
.niov
);
2719 /* Loop in case of EINTR */
2721 len
= v9fs_co_pwritev(pdu
, fidp
, qiov
.iov
, qiov
.niov
, off
);
2726 } while (len
== -EINTR
&& !pdu
->cancelled
);
2728 /* IO error return the error */
2732 } while (total
< count
&& len
> 0);
2735 err
= pdu_marshal(pdu
, offset
, "d", total
);
2740 trace_v9fs_write_return(pdu
->tag
, pdu
->id
, total
, err
);
2742 qemu_iovec_destroy(&qiov
);
2746 qemu_iovec_destroy(&qiov_full
);
2747 pdu_complete(pdu
, err
);
2750 static void coroutine_fn
v9fs_create(void *opaque
)
2762 V9fsString extension
;
2764 V9fsPDU
*pdu
= opaque
;
2765 V9fsState
*s
= pdu
->s
;
2767 v9fs_path_init(&path
);
2768 v9fs_string_init(&name
);
2769 v9fs_string_init(&extension
);
2770 err
= pdu_unmarshal(pdu
, offset
, "dsdbs", &fid
, &name
,
2771 &perm
, &mode
, &extension
);
2775 trace_v9fs_create(pdu
->tag
, pdu
->id
, fid
, name
.data
, perm
, mode
);
2777 if (name_is_illegal(name
.data
)) {
2782 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2787 fidp
= get_fid(pdu
, fid
);
2792 if (fidp
->fid_type
!= P9_FID_NONE
) {
2796 if (perm
& P9_STAT_MODE_DIR
) {
2797 err
= v9fs_co_mkdir(pdu
, fidp
, &name
, perm
& 0777,
2798 fidp
->uid
, -1, &stbuf
);
2802 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2806 v9fs_path_write_lock(s
);
2807 v9fs_path_copy(&fidp
->path
, &path
);
2808 v9fs_path_unlock(s
);
2809 err
= v9fs_co_opendir(pdu
, fidp
);
2813 fidp
->fid_type
= P9_FID_DIR
;
2814 } else if (perm
& P9_STAT_MODE_SYMLINK
) {
2815 err
= v9fs_co_symlink(pdu
, fidp
, &name
,
2816 extension
.data
, -1 , &stbuf
);
2820 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2824 v9fs_path_write_lock(s
);
2825 v9fs_path_copy(&fidp
->path
, &path
);
2826 v9fs_path_unlock(s
);
2827 } else if (perm
& P9_STAT_MODE_LINK
) {
2828 int32_t ofid
= atoi(extension
.data
);
2829 V9fsFidState
*ofidp
= get_fid(pdu
, ofid
);
2830 if (ofidp
== NULL
) {
2834 err
= v9fs_co_link(pdu
, ofidp
, fidp
, &name
);
2835 put_fid(pdu
, ofidp
);
2839 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2841 fidp
->fid_type
= P9_FID_NONE
;
2844 v9fs_path_write_lock(s
);
2845 v9fs_path_copy(&fidp
->path
, &path
);
2846 v9fs_path_unlock(s
);
2847 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
2849 fidp
->fid_type
= P9_FID_NONE
;
2852 } else if (perm
& P9_STAT_MODE_DEVICE
) {
2854 uint32_t major
, minor
;
2857 if (sscanf(extension
.data
, "%c %u %u", &ctype
, &major
, &minor
) != 3) {
2874 nmode
|= perm
& 0777;
2875 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2876 makedev(major
, minor
), nmode
, &stbuf
);
2880 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2884 v9fs_path_write_lock(s
);
2885 v9fs_path_copy(&fidp
->path
, &path
);
2886 v9fs_path_unlock(s
);
2887 } else if (perm
& P9_STAT_MODE_NAMED_PIPE
) {
2888 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2889 0, S_IFIFO
| (perm
& 0777), &stbuf
);
2893 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2897 v9fs_path_write_lock(s
);
2898 v9fs_path_copy(&fidp
->path
, &path
);
2899 v9fs_path_unlock(s
);
2900 } else if (perm
& P9_STAT_MODE_SOCKET
) {
2901 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2902 0, S_IFSOCK
| (perm
& 0777), &stbuf
);
2906 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2910 v9fs_path_write_lock(s
);
2911 v9fs_path_copy(&fidp
->path
, &path
);
2912 v9fs_path_unlock(s
);
2914 err
= v9fs_co_open2(pdu
, fidp
, &name
, -1,
2915 omode_to_uflags(mode
) | O_CREAT
, perm
, &stbuf
);
2919 fidp
->fid_type
= P9_FID_FILE
;
2920 fidp
->open_flags
= omode_to_uflags(mode
);
2921 if (fidp
->open_flags
& O_EXCL
) {
2923 * We let the host file system do O_EXCL check
2924 * We should not reclaim such fd
2926 fidp
->flags
|= FID_NON_RECLAIMABLE
;
2929 iounit
= get_iounit(pdu
, &fidp
->path
);
2930 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
2934 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
2939 trace_v9fs_create_return(pdu
->tag
, pdu
->id
,
2940 qid
.type
, qid
.version
, qid
.path
, iounit
);
2944 pdu_complete(pdu
, err
);
2945 v9fs_string_free(&name
);
2946 v9fs_string_free(&extension
);
2947 v9fs_path_free(&path
);
2950 static void coroutine_fn
v9fs_symlink(void *opaque
)
2952 V9fsPDU
*pdu
= opaque
;
2955 V9fsFidState
*dfidp
;
2963 v9fs_string_init(&name
);
2964 v9fs_string_init(&symname
);
2965 err
= pdu_unmarshal(pdu
, offset
, "dssd", &dfid
, &name
, &symname
, &gid
);
2969 trace_v9fs_symlink(pdu
->tag
, pdu
->id
, dfid
, name
.data
, symname
.data
, gid
);
2971 if (name_is_illegal(name
.data
)) {
2976 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2981 dfidp
= get_fid(pdu
, dfid
);
2982 if (dfidp
== NULL
) {
2986 err
= v9fs_co_symlink(pdu
, dfidp
, &name
, symname
.data
, gid
, &stbuf
);
2990 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
2994 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
2999 trace_v9fs_symlink_return(pdu
->tag
, pdu
->id
,
3000 qid
.type
, qid
.version
, qid
.path
);
3002 put_fid(pdu
, dfidp
);
3004 pdu_complete(pdu
, err
);
3005 v9fs_string_free(&name
);
3006 v9fs_string_free(&symname
);
3009 static void coroutine_fn
v9fs_flush(void *opaque
)
3014 V9fsPDU
*cancel_pdu
= NULL
;
3015 V9fsPDU
*pdu
= opaque
;
3016 V9fsState
*s
= pdu
->s
;
3018 err
= pdu_unmarshal(pdu
, offset
, "w", &tag
);
3020 pdu_complete(pdu
, err
);
3023 trace_v9fs_flush(pdu
->tag
, pdu
->id
, tag
);
3025 if (pdu
->tag
== tag
) {
3026 warn_report("the guest sent a self-referencing 9P flush request");
3028 QLIST_FOREACH(cancel_pdu
, &s
->active_list
, next
) {
3029 if (cancel_pdu
->tag
== tag
) {
3035 cancel_pdu
->cancelled
= 1;
3037 * Wait for pdu to complete.
3039 qemu_co_queue_wait(&cancel_pdu
->complete
, NULL
);
3040 if (!qemu_co_queue_next(&cancel_pdu
->complete
)) {
3041 cancel_pdu
->cancelled
= 0;
3042 pdu_free(cancel_pdu
);
3045 pdu_complete(pdu
, 7);
3048 static void coroutine_fn
v9fs_link(void *opaque
)
3050 V9fsPDU
*pdu
= opaque
;
3051 int32_t dfid
, oldfid
;
3052 V9fsFidState
*dfidp
, *oldfidp
;
3057 v9fs_string_init(&name
);
3058 err
= pdu_unmarshal(pdu
, offset
, "dds", &dfid
, &oldfid
, &name
);
3062 trace_v9fs_link(pdu
->tag
, pdu
->id
, dfid
, oldfid
, name
.data
);
3064 if (name_is_illegal(name
.data
)) {
3069 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3074 dfidp
= get_fid(pdu
, dfid
);
3075 if (dfidp
== NULL
) {
3080 oldfidp
= get_fid(pdu
, oldfid
);
3081 if (oldfidp
== NULL
) {
3085 err
= v9fs_co_link(pdu
, oldfidp
, dfidp
, &name
);
3089 put_fid(pdu
, oldfidp
);
3091 put_fid(pdu
, dfidp
);
3093 v9fs_string_free(&name
);
3094 pdu_complete(pdu
, err
);
3097 /* Only works with path name based fid */
3098 static void coroutine_fn
v9fs_remove(void *opaque
)
3104 V9fsPDU
*pdu
= opaque
;
3106 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
3110 trace_v9fs_remove(pdu
->tag
, pdu
->id
, fid
);
3112 fidp
= get_fid(pdu
, fid
);
3117 /* if fs driver is not path based, return EOPNOTSUPP */
3118 if (!(pdu
->s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
)) {
3123 * IF the file is unlinked, we cannot reopen
3124 * the file later. So don't reclaim fd
3126 err
= v9fs_mark_fids_unreclaim(pdu
, &fidp
->path
);
3130 err
= v9fs_co_remove(pdu
, &fidp
->path
);
3135 /* For TREMOVE we need to clunk the fid even on failed remove */
3136 clunk_fid(pdu
->s
, fidp
->fid
);
3139 pdu_complete(pdu
, err
);
3142 static void coroutine_fn
v9fs_unlinkat(void *opaque
)
3146 int32_t dfid
, flags
, rflags
= 0;
3149 V9fsFidState
*dfidp
;
3150 V9fsPDU
*pdu
= opaque
;
3152 v9fs_string_init(&name
);
3153 err
= pdu_unmarshal(pdu
, offset
, "dsd", &dfid
, &name
, &flags
);
3158 if (name_is_illegal(name
.data
)) {
3163 if (!strcmp(".", name
.data
)) {
3168 if (!strcmp("..", name
.data
)) {
3173 if (flags
& ~P9_DOTL_AT_REMOVEDIR
) {
3178 if (flags
& P9_DOTL_AT_REMOVEDIR
) {
3179 rflags
|= AT_REMOVEDIR
;
3182 dfidp
= get_fid(pdu
, dfid
);
3183 if (dfidp
== NULL
) {
3188 * IF the file is unlinked, we cannot reopen
3189 * the file later. So don't reclaim fd
3191 v9fs_path_init(&path
);
3192 err
= v9fs_co_name_to_path(pdu
, &dfidp
->path
, name
.data
, &path
);
3196 err
= v9fs_mark_fids_unreclaim(pdu
, &path
);
3200 err
= v9fs_co_unlinkat(pdu
, &dfidp
->path
, &name
, rflags
);
3205 put_fid(pdu
, dfidp
);
3206 v9fs_path_free(&path
);
3208 pdu_complete(pdu
, err
);
3209 v9fs_string_free(&name
);
3213 /* Only works with path name based fid */
3214 static int coroutine_fn
v9fs_complete_rename(V9fsPDU
*pdu
, V9fsFidState
*fidp
,
3220 V9fsFidState
*tfidp
;
3221 V9fsState
*s
= pdu
->s
;
3222 V9fsFidState
*dirfidp
= NULL
;
3223 GHashTableIter iter
;
3226 v9fs_path_init(&new_path
);
3227 if (newdirfid
!= -1) {
3228 dirfidp
= get_fid(pdu
, newdirfid
);
3229 if (dirfidp
== NULL
) {
3232 if (fidp
->fid_type
!= P9_FID_NONE
) {
3236 err
= v9fs_co_name_to_path(pdu
, &dirfidp
->path
, name
->data
, &new_path
);
3241 char *dir_name
= g_path_get_dirname(fidp
->path
.data
);
3244 v9fs_path_init(&dir_path
);
3245 v9fs_path_sprintf(&dir_path
, "%s", dir_name
);
3248 err
= v9fs_co_name_to_path(pdu
, &dir_path
, name
->data
, &new_path
);
3249 v9fs_path_free(&dir_path
);
3254 err
= v9fs_co_rename(pdu
, &fidp
->path
, &new_path
);
3260 * Fixup fid's pointing to the old name to
3261 * start pointing to the new name
3263 g_hash_table_iter_init(&iter
, s
->fids
);
3264 while (g_hash_table_iter_next(&iter
, &fid
, (gpointer
*) &tfidp
)) {
3265 if (v9fs_path_is_ancestor(&fidp
->path
, &tfidp
->path
)) {
3266 /* replace the name */
3267 v9fs_fix_path(&tfidp
->path
, &new_path
, strlen(fidp
->path
.data
));
3272 put_fid(pdu
, dirfidp
);
3274 v9fs_path_free(&new_path
);
3278 /* Only works with path name based fid */
3279 static void coroutine_fn
v9fs_rename(void *opaque
)
3287 V9fsPDU
*pdu
= opaque
;
3288 V9fsState
*s
= pdu
->s
;
3290 v9fs_string_init(&name
);
3291 err
= pdu_unmarshal(pdu
, offset
, "dds", &fid
, &newdirfid
, &name
);
3296 if (name_is_illegal(name
.data
)) {
3301 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3306 fidp
= get_fid(pdu
, fid
);
3311 if (fidp
->fid_type
!= P9_FID_NONE
) {
3315 /* if fs driver is not path based, return EOPNOTSUPP */
3316 if (!(pdu
->s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
)) {
3320 v9fs_path_write_lock(s
);
3321 err
= v9fs_complete_rename(pdu
, fidp
, newdirfid
, &name
);
3322 v9fs_path_unlock(s
);
3329 pdu_complete(pdu
, err
);
3330 v9fs_string_free(&name
);
3333 static int coroutine_fn
v9fs_fix_fid_paths(V9fsPDU
*pdu
, V9fsPath
*olddir
,
3334 V9fsString
*old_name
,
3336 V9fsString
*new_name
)
3338 V9fsFidState
*tfidp
;
3339 V9fsPath oldpath
, newpath
;
3340 V9fsState
*s
= pdu
->s
;
3342 GHashTableIter iter
;
3345 v9fs_path_init(&oldpath
);
3346 v9fs_path_init(&newpath
);
3347 err
= v9fs_co_name_to_path(pdu
, olddir
, old_name
->data
, &oldpath
);
3351 err
= v9fs_co_name_to_path(pdu
, newdir
, new_name
->data
, &newpath
);
3357 * Fixup fid's pointing to the old name to
3358 * start pointing to the new name
3360 g_hash_table_iter_init(&iter
, s
->fids
);
3361 while (g_hash_table_iter_next(&iter
, &fid
, (gpointer
*) &tfidp
)) {
3362 if (v9fs_path_is_ancestor(&oldpath
, &tfidp
->path
)) {
3363 /* replace the name */
3364 v9fs_fix_path(&tfidp
->path
, &newpath
, strlen(oldpath
.data
));
3368 v9fs_path_free(&oldpath
);
3369 v9fs_path_free(&newpath
);
3373 static int coroutine_fn
v9fs_complete_renameat(V9fsPDU
*pdu
, int32_t olddirfid
,
3374 V9fsString
*old_name
,
3376 V9fsString
*new_name
)
3379 V9fsState
*s
= pdu
->s
;
3380 V9fsFidState
*newdirfidp
= NULL
, *olddirfidp
= NULL
;
3382 olddirfidp
= get_fid(pdu
, olddirfid
);
3383 if (olddirfidp
== NULL
) {
3387 if (newdirfid
!= -1) {
3388 newdirfidp
= get_fid(pdu
, newdirfid
);
3389 if (newdirfidp
== NULL
) {
3394 newdirfidp
= get_fid(pdu
, olddirfid
);
3397 err
= v9fs_co_renameat(pdu
, &olddirfidp
->path
, old_name
,
3398 &newdirfidp
->path
, new_name
);
3402 if (s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
) {
3403 /* Only for path based fid we need to do the below fixup */
3404 err
= v9fs_fix_fid_paths(pdu
, &olddirfidp
->path
, old_name
,
3405 &newdirfidp
->path
, new_name
);
3409 put_fid(pdu
, olddirfidp
);
3412 put_fid(pdu
, newdirfidp
);
3417 static void coroutine_fn
v9fs_renameat(void *opaque
)
3421 V9fsPDU
*pdu
= opaque
;
3422 V9fsState
*s
= pdu
->s
;
3423 int32_t olddirfid
, newdirfid
;
3424 V9fsString old_name
, new_name
;
3426 v9fs_string_init(&old_name
);
3427 v9fs_string_init(&new_name
);
3428 err
= pdu_unmarshal(pdu
, offset
, "dsds", &olddirfid
,
3429 &old_name
, &newdirfid
, &new_name
);
3434 if (name_is_illegal(old_name
.data
) || name_is_illegal(new_name
.data
)) {
3439 if (!strcmp(".", old_name
.data
) || !strcmp("..", old_name
.data
) ||
3440 !strcmp(".", new_name
.data
) || !strcmp("..", new_name
.data
)) {
3445 v9fs_path_write_lock(s
);
3446 err
= v9fs_complete_renameat(pdu
, olddirfid
,
3447 &old_name
, newdirfid
, &new_name
);
3448 v9fs_path_unlock(s
);
3454 pdu_complete(pdu
, err
);
3455 v9fs_string_free(&old_name
);
3456 v9fs_string_free(&new_name
);
3459 static void coroutine_fn
v9fs_wstat(void *opaque
)
3468 V9fsPDU
*pdu
= opaque
;
3469 V9fsState
*s
= pdu
->s
;
3471 v9fs_stat_init(&v9stat
);
3472 err
= pdu_unmarshal(pdu
, offset
, "dwS", &fid
, &unused
, &v9stat
);
3476 trace_v9fs_wstat(pdu
->tag
, pdu
->id
, fid
,
3477 v9stat
.mode
, v9stat
.atime
, v9stat
.mtime
);
3479 fidp
= get_fid(pdu
, fid
);
3484 /* do we need to sync the file? */
3485 if (donttouch_stat(&v9stat
)) {
3486 err
= v9fs_co_fsync(pdu
, fidp
, 0);
3489 if (v9stat
.mode
!= -1) {
3491 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
3495 v9_mode
= stat_to_v9mode(&stbuf
);
3496 if ((v9stat
.mode
& P9_STAT_MODE_TYPE_BITS
) !=
3497 (v9_mode
& P9_STAT_MODE_TYPE_BITS
)) {
3498 /* Attempting to change the type */
3502 err
= v9fs_co_chmod(pdu
, &fidp
->path
,
3503 v9mode_to_mode(v9stat
.mode
,
3504 &v9stat
.extension
));
3509 if (v9stat
.mtime
!= -1 || v9stat
.atime
!= -1) {
3510 struct timespec times
[2];
3511 if (v9stat
.atime
!= -1) {
3512 times
[0].tv_sec
= v9stat
.atime
;
3513 times
[0].tv_nsec
= 0;
3515 times
[0].tv_nsec
= UTIME_OMIT
;
3517 if (v9stat
.mtime
!= -1) {
3518 times
[1].tv_sec
= v9stat
.mtime
;
3519 times
[1].tv_nsec
= 0;
3521 times
[1].tv_nsec
= UTIME_OMIT
;
3523 err
= v9fs_co_utimensat(pdu
, &fidp
->path
, times
);
3528 if (v9stat
.n_gid
!= -1 || v9stat
.n_uid
!= -1) {
3529 err
= v9fs_co_chown(pdu
, &fidp
->path
, v9stat
.n_uid
, v9stat
.n_gid
);
3534 if (v9stat
.name
.size
!= 0) {
3535 v9fs_path_write_lock(s
);
3536 err
= v9fs_complete_rename(pdu
, fidp
, -1, &v9stat
.name
);
3537 v9fs_path_unlock(s
);
3542 if (v9stat
.length
!= -1) {
3543 err
= v9fs_co_truncate(pdu
, &fidp
->path
, v9stat
.length
);
3552 v9fs_stat_free(&v9stat
);
3553 pdu_complete(pdu
, err
);
3556 static int v9fs_fill_statfs(V9fsState
*s
, V9fsPDU
*pdu
, struct statfs
*stbuf
)
3568 int32_t bsize_factor
;
3571 * compute bsize factor based on host file system block size
3574 bsize_factor
= (s
->msize
- P9_IOHDRSZ
) / stbuf
->f_bsize
;
3575 if (!bsize_factor
) {
3578 f_type
= stbuf
->f_type
;
3579 f_bsize
= stbuf
->f_bsize
;
3580 f_bsize
*= bsize_factor
;
3582 * f_bsize is adjusted(multiplied) by bsize factor, so we need to
3583 * adjust(divide) the number of blocks, free blocks and available
3584 * blocks by bsize factor
3586 f_blocks
= stbuf
->f_blocks
/ bsize_factor
;
3587 f_bfree
= stbuf
->f_bfree
/ bsize_factor
;
3588 f_bavail
= stbuf
->f_bavail
/ bsize_factor
;
3589 f_files
= stbuf
->f_files
;
3590 f_ffree
= stbuf
->f_ffree
;
3591 #ifdef CONFIG_DARWIN
3592 fsid_val
= (unsigned int)stbuf
->f_fsid
.val
[0] |
3593 (unsigned long long)stbuf
->f_fsid
.val
[1] << 32;
3594 f_namelen
= NAME_MAX
;
3596 fsid_val
= (unsigned int) stbuf
->f_fsid
.__val
[0] |
3597 (unsigned long long)stbuf
->f_fsid
.__val
[1] << 32;
3598 f_namelen
= stbuf
->f_namelen
;
3601 return pdu_marshal(pdu
, offset
, "ddqqqqqqd",
3602 f_type
, f_bsize
, f_blocks
, f_bfree
,
3603 f_bavail
, f_files
, f_ffree
,
3604 fsid_val
, f_namelen
);
3607 static void coroutine_fn
v9fs_statfs(void *opaque
)
3613 struct statfs stbuf
;
3614 V9fsPDU
*pdu
= opaque
;
3615 V9fsState
*s
= pdu
->s
;
3617 retval
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
3621 fidp
= get_fid(pdu
, fid
);
3626 retval
= v9fs_co_statfs(pdu
, &fidp
->path
, &stbuf
);
3630 retval
= v9fs_fill_statfs(s
, pdu
, &stbuf
);
3638 pdu_complete(pdu
, retval
);
3641 static void coroutine_fn
v9fs_mknod(void *opaque
)
3654 V9fsPDU
*pdu
= opaque
;
3656 v9fs_string_init(&name
);
3657 err
= pdu_unmarshal(pdu
, offset
, "dsdddd", &fid
, &name
, &mode
,
3658 &major
, &minor
, &gid
);
3662 trace_v9fs_mknod(pdu
->tag
, pdu
->id
, fid
, mode
, major
, minor
);
3664 if (name_is_illegal(name
.data
)) {
3669 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3674 fidp
= get_fid(pdu
, fid
);
3679 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, gid
,
3680 makedev(major
, minor
), mode
, &stbuf
);
3684 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
3688 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
3693 trace_v9fs_mknod_return(pdu
->tag
, pdu
->id
,
3694 qid
.type
, qid
.version
, qid
.path
);
3698 pdu_complete(pdu
, err
);
3699 v9fs_string_free(&name
);
3703 * Implement posix byte range locking code
3704 * Server side handling of locking code is very simple, because 9p server in
3705 * QEMU can handle only one client. And most of the lock handling
3706 * (like conflict, merging) etc is done by the VFS layer itself, so no need to
3707 * do any thing in * qemu 9p server side lock code path.
3708 * So when a TLOCK request comes, always return success
3710 static void coroutine_fn
v9fs_lock(void *opaque
)
3716 int32_t fid
, err
= 0;
3717 V9fsPDU
*pdu
= opaque
;
3719 v9fs_string_init(&flock
.client_id
);
3720 err
= pdu_unmarshal(pdu
, offset
, "dbdqqds", &fid
, &flock
.type
,
3721 &flock
.flags
, &flock
.start
, &flock
.length
,
3722 &flock
.proc_id
, &flock
.client_id
);
3726 trace_v9fs_lock(pdu
->tag
, pdu
->id
, fid
,
3727 flock
.type
, flock
.start
, flock
.length
);
3730 /* We support only block flag now (that too ignored currently) */
3731 if (flock
.flags
& ~P9_LOCK_FLAGS_BLOCK
) {
3735 fidp
= get_fid(pdu
, fid
);
3740 err
= v9fs_co_fstat(pdu
, fidp
, &stbuf
);
3744 err
= pdu_marshal(pdu
, offset
, "b", P9_LOCK_SUCCESS
);
3749 trace_v9fs_lock_return(pdu
->tag
, pdu
->id
, P9_LOCK_SUCCESS
);
3753 pdu_complete(pdu
, err
);
3754 v9fs_string_free(&flock
.client_id
);
3758 * When a TGETLOCK request comes, always return success because all lock
3759 * handling is done by client's VFS layer.
3761 static void coroutine_fn
v9fs_getlock(void *opaque
)
3767 int32_t fid
, err
= 0;
3768 V9fsPDU
*pdu
= opaque
;
3770 v9fs_string_init(&glock
.client_id
);
3771 err
= pdu_unmarshal(pdu
, offset
, "dbqqds", &fid
, &glock
.type
,
3772 &glock
.start
, &glock
.length
, &glock
.proc_id
,
3777 trace_v9fs_getlock(pdu
->tag
, pdu
->id
, fid
,
3778 glock
.type
, glock
.start
, glock
.length
);
3780 fidp
= get_fid(pdu
, fid
);
3785 err
= v9fs_co_fstat(pdu
, fidp
, &stbuf
);
3789 glock
.type
= P9_LOCK_TYPE_UNLCK
;
3790 err
= pdu_marshal(pdu
, offset
, "bqqds", glock
.type
,
3791 glock
.start
, glock
.length
, glock
.proc_id
,
3797 trace_v9fs_getlock_return(pdu
->tag
, pdu
->id
, glock
.type
, glock
.start
,
3798 glock
.length
, glock
.proc_id
);
3802 pdu_complete(pdu
, err
);
3803 v9fs_string_free(&glock
.client_id
);
3806 static void coroutine_fn
v9fs_mkdir(void *opaque
)
3808 V9fsPDU
*pdu
= opaque
;
3819 v9fs_string_init(&name
);
3820 err
= pdu_unmarshal(pdu
, offset
, "dsdd", &fid
, &name
, &mode
, &gid
);
3824 trace_v9fs_mkdir(pdu
->tag
, pdu
->id
, fid
, name
.data
, mode
, gid
);
3826 if (name_is_illegal(name
.data
)) {
3831 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3836 fidp
= get_fid(pdu
, fid
);
3841 err
= v9fs_co_mkdir(pdu
, fidp
, &name
, mode
, fidp
->uid
, gid
, &stbuf
);
3845 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
3849 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
3854 trace_v9fs_mkdir_return(pdu
->tag
, pdu
->id
,
3855 qid
.type
, qid
.version
, qid
.path
, err
);
3859 pdu_complete(pdu
, err
);
3860 v9fs_string_free(&name
);
3863 static void coroutine_fn
v9fs_xattrwalk(void *opaque
)
3869 int32_t fid
, newfid
;
3870 V9fsFidState
*file_fidp
;
3871 V9fsFidState
*xattr_fidp
= NULL
;
3872 V9fsPDU
*pdu
= opaque
;
3873 V9fsState
*s
= pdu
->s
;
3875 v9fs_string_init(&name
);
3876 err
= pdu_unmarshal(pdu
, offset
, "dds", &fid
, &newfid
, &name
);
3880 trace_v9fs_xattrwalk(pdu
->tag
, pdu
->id
, fid
, newfid
, name
.data
);
3882 file_fidp
= get_fid(pdu
, fid
);
3883 if (file_fidp
== NULL
) {
3887 xattr_fidp
= alloc_fid(s
, newfid
);
3888 if (xattr_fidp
== NULL
) {
3892 v9fs_path_copy(&xattr_fidp
->path
, &file_fidp
->path
);
3893 if (!v9fs_string_size(&name
)) {
3895 * listxattr request. Get the size first
3897 size
= v9fs_co_llistxattr(pdu
, &xattr_fidp
->path
, NULL
, 0);
3900 clunk_fid(s
, xattr_fidp
->fid
);
3904 * Read the xattr value
3906 xattr_fidp
->fs
.xattr
.len
= size
;
3907 xattr_fidp
->fid_type
= P9_FID_XATTR
;
3908 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= true;
3909 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
3911 err
= v9fs_co_llistxattr(pdu
, &xattr_fidp
->path
,
3912 xattr_fidp
->fs
.xattr
.value
,
3913 xattr_fidp
->fs
.xattr
.len
);
3915 clunk_fid(s
, xattr_fidp
->fid
);
3919 err
= pdu_marshal(pdu
, offset
, "q", size
);
3926 * specific xattr fid. We check for xattr
3927 * presence also collect the xattr size
3929 size
= v9fs_co_lgetxattr(pdu
, &xattr_fidp
->path
,
3933 clunk_fid(s
, xattr_fidp
->fid
);
3937 * Read the xattr value
3939 xattr_fidp
->fs
.xattr
.len
= size
;
3940 xattr_fidp
->fid_type
= P9_FID_XATTR
;
3941 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= true;
3942 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
3944 err
= v9fs_co_lgetxattr(pdu
, &xattr_fidp
->path
,
3945 &name
, xattr_fidp
->fs
.xattr
.value
,
3946 xattr_fidp
->fs
.xattr
.len
);
3948 clunk_fid(s
, xattr_fidp
->fid
);
3952 err
= pdu_marshal(pdu
, offset
, "q", size
);
3958 trace_v9fs_xattrwalk_return(pdu
->tag
, pdu
->id
, size
);
3960 put_fid(pdu
, file_fidp
);
3962 put_fid(pdu
, xattr_fidp
);
3965 pdu_complete(pdu
, err
);
3966 v9fs_string_free(&name
);
3969 #if defined(CONFIG_LINUX)
3970 /* Currently, only Linux has XATTR_SIZE_MAX */
3971 #define P9_XATTR_SIZE_MAX XATTR_SIZE_MAX
3972 #elif defined(CONFIG_DARWIN)
3974 * Darwin doesn't seem to define a maximum xattr size in its user
3975 * space header, so manually configure it across platforms as 64k.
3977 * Having no limit at all can lead to QEMU crashing during large g_malloc()
3978 * calls. Because QEMU does not currently support macOS guests, the below
3979 * preliminary solution only works due to its being a reflection of the limit of
3982 #define P9_XATTR_SIZE_MAX 65536
3984 #error Missing definition for P9_XATTR_SIZE_MAX for this host system
3987 static void coroutine_fn
v9fs_xattrcreate(void *opaque
)
3989 int flags
, rflags
= 0;
3995 V9fsFidState
*file_fidp
;
3996 V9fsFidState
*xattr_fidp
;
3997 V9fsPDU
*pdu
= opaque
;
3999 v9fs_string_init(&name
);
4000 err
= pdu_unmarshal(pdu
, offset
, "dsqd", &fid
, &name
, &size
, &flags
);
4004 trace_v9fs_xattrcreate(pdu
->tag
, pdu
->id
, fid
, name
.data
, size
, flags
);
4006 if (flags
& ~(P9_XATTR_CREATE
| P9_XATTR_REPLACE
)) {
4011 if (flags
& P9_XATTR_CREATE
) {
4012 rflags
|= XATTR_CREATE
;
4015 if (flags
& P9_XATTR_REPLACE
) {
4016 rflags
|= XATTR_REPLACE
;
4019 if (size
> P9_XATTR_SIZE_MAX
) {
4024 file_fidp
= get_fid(pdu
, fid
);
4025 if (file_fidp
== NULL
) {
4029 if (file_fidp
->fid_type
!= P9_FID_NONE
) {
4034 /* Make the file fid point to xattr */
4035 xattr_fidp
= file_fidp
;
4036 xattr_fidp
->fid_type
= P9_FID_XATTR
;
4037 xattr_fidp
->fs
.xattr
.copied_len
= 0;
4038 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= false;
4039 xattr_fidp
->fs
.xattr
.len
= size
;
4040 xattr_fidp
->fs
.xattr
.flags
= rflags
;
4041 v9fs_string_init(&xattr_fidp
->fs
.xattr
.name
);
4042 v9fs_string_copy(&xattr_fidp
->fs
.xattr
.name
, &name
);
4043 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
4046 put_fid(pdu
, file_fidp
);
4048 pdu_complete(pdu
, err
);
4049 v9fs_string_free(&name
);
4052 static void coroutine_fn
v9fs_readlink(void *opaque
)
4054 V9fsPDU
*pdu
= opaque
;
4061 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
4065 trace_v9fs_readlink(pdu
->tag
, pdu
->id
, fid
);
4066 fidp
= get_fid(pdu
, fid
);
4072 v9fs_string_init(&target
);
4073 err
= v9fs_co_readlink(pdu
, &fidp
->path
, &target
);
4077 err
= pdu_marshal(pdu
, offset
, "s", &target
);
4079 v9fs_string_free(&target
);
4083 trace_v9fs_readlink_return(pdu
->tag
, pdu
->id
, target
.data
);
4084 v9fs_string_free(&target
);
4088 pdu_complete(pdu
, err
);
4091 static CoroutineEntry
*pdu_co_handlers
[] = {
4092 [P9_TREADDIR
] = v9fs_readdir
,
4093 [P9_TSTATFS
] = v9fs_statfs
,
4094 [P9_TGETATTR
] = v9fs_getattr
,
4095 [P9_TSETATTR
] = v9fs_setattr
,
4096 [P9_TXATTRWALK
] = v9fs_xattrwalk
,
4097 [P9_TXATTRCREATE
] = v9fs_xattrcreate
,
4098 [P9_TMKNOD
] = v9fs_mknod
,
4099 [P9_TRENAME
] = v9fs_rename
,
4100 [P9_TLOCK
] = v9fs_lock
,
4101 [P9_TGETLOCK
] = v9fs_getlock
,
4102 [P9_TRENAMEAT
] = v9fs_renameat
,
4103 [P9_TREADLINK
] = v9fs_readlink
,
4104 [P9_TUNLINKAT
] = v9fs_unlinkat
,
4105 [P9_TMKDIR
] = v9fs_mkdir
,
4106 [P9_TVERSION
] = v9fs_version
,
4107 [P9_TLOPEN
] = v9fs_open
,
4108 [P9_TATTACH
] = v9fs_attach
,
4109 [P9_TSTAT
] = v9fs_stat
,
4110 [P9_TWALK
] = v9fs_walk
,
4111 [P9_TCLUNK
] = v9fs_clunk
,
4112 [P9_TFSYNC
] = v9fs_fsync
,
4113 [P9_TOPEN
] = v9fs_open
,
4114 [P9_TREAD
] = v9fs_read
,
4116 [P9_TAUTH
] = v9fs_auth
,
4118 [P9_TFLUSH
] = v9fs_flush
,
4119 [P9_TLINK
] = v9fs_link
,
4120 [P9_TSYMLINK
] = v9fs_symlink
,
4121 [P9_TCREATE
] = v9fs_create
,
4122 [P9_TLCREATE
] = v9fs_lcreate
,
4123 [P9_TWRITE
] = v9fs_write
,
4124 [P9_TWSTAT
] = v9fs_wstat
,
4125 [P9_TREMOVE
] = v9fs_remove
,
4128 static void coroutine_fn
v9fs_op_not_supp(void *opaque
)
4130 V9fsPDU
*pdu
= opaque
;
4131 pdu_complete(pdu
, -EOPNOTSUPP
);
4134 static void coroutine_fn
v9fs_fs_ro(void *opaque
)
4136 V9fsPDU
*pdu
= opaque
;
4137 pdu_complete(pdu
, -EROFS
);
4140 static inline bool is_read_only_op(V9fsPDU
*pdu
)
4167 void pdu_submit(V9fsPDU
*pdu
, P9MsgHeader
*hdr
)
4170 CoroutineEntry
*handler
;
4171 V9fsState
*s
= pdu
->s
;
4173 pdu
->size
= le32_to_cpu(hdr
->size_le
);
4175 pdu
->tag
= le16_to_cpu(hdr
->tag_le
);
4177 if (pdu
->id
>= ARRAY_SIZE(pdu_co_handlers
) ||
4178 (pdu_co_handlers
[pdu
->id
] == NULL
)) {
4179 handler
= v9fs_op_not_supp
;
4180 } else if (is_ro_export(&s
->ctx
) && !is_read_only_op(pdu
)) {
4181 handler
= v9fs_fs_ro
;
4183 handler
= pdu_co_handlers
[pdu
->id
];
4186 qemu_co_queue_init(&pdu
->complete
);
4187 co
= qemu_coroutine_create(handler
, pdu
);
4188 qemu_coroutine_enter(co
);
4191 /* Returns 0 on success, 1 on failure. */
4192 int v9fs_device_realize_common(V9fsState
*s
, const V9fsTransport
*t
,
4202 assert(!s
->transport
);
4205 /* initialize pdu allocator */
4206 QLIST_INIT(&s
->free_list
);
4207 QLIST_INIT(&s
->active_list
);
4208 for (i
= 0; i
< MAX_REQ
; i
++) {
4209 QLIST_INSERT_HEAD(&s
->free_list
, &s
->pdus
[i
], next
);
4214 v9fs_path_init(&path
);
4216 fse
= get_fsdev_fsentry(s
->fsconf
.fsdev_id
);
4219 /* We don't have a fsdev identified by fsdev_id */
4220 error_setg(errp
, "9pfs device couldn't find fsdev with the "
4222 s
->fsconf
.fsdev_id
? s
->fsconf
.fsdev_id
: "NULL");
4226 if (!s
->fsconf
.tag
) {
4227 /* we haven't specified a mount_tag */
4228 error_setg(errp
, "fsdev with id %s needs mount_tag arguments",
4229 s
->fsconf
.fsdev_id
);
4233 s
->ctx
.export_flags
= fse
->export_flags
;
4234 s
->ctx
.fs_root
= g_strdup(fse
->path
);
4235 s
->ctx
.exops
.get_st_gen
= NULL
;
4236 len
= strlen(s
->fsconf
.tag
);
4237 if (len
> MAX_TAG_LEN
- 1) {
4238 error_setg(errp
, "mount tag '%s' (%d bytes) is longer than "
4239 "maximum (%d bytes)", s
->fsconf
.tag
, len
, MAX_TAG_LEN
- 1);
4243 s
->tag
= g_strdup(s
->fsconf
.tag
);
4248 s
->ctx
.fmode
= fse
->fmode
;
4249 s
->ctx
.dmode
= fse
->dmode
;
4251 s
->fids
= g_hash_table_new(NULL
, NULL
);
4252 qemu_co_rwlock_init(&s
->rename_lock
);
4254 if (s
->ops
->init(&s
->ctx
, errp
) < 0) {
4255 error_prepend(errp
, "cannot initialize fsdev '%s': ",
4256 s
->fsconf
.fsdev_id
);
4261 * Check details of export path, We need to use fs driver
4262 * call back to do that. Since we are in the init path, we don't
4263 * use co-routines here.
4265 if (s
->ops
->name_to_path(&s
->ctx
, NULL
, "/", &path
) < 0) {
4267 "error in converting name to path %s", strerror(errno
));
4270 if (s
->ops
->lstat(&s
->ctx
, &path
, &stat
)) {
4271 error_setg(errp
, "share path %s does not exist", fse
->path
);
4273 } else if (!S_ISDIR(stat
.st_mode
)) {
4274 error_setg(errp
, "share path %s is not a directory", fse
->path
);
4278 s
->dev_id
= stat
.st_dev
;
4280 /* init inode remapping : */
4281 /* hash table for variable length inode suffixes */
4282 qpd_table_init(&s
->qpd_table
);
4283 /* hash table for slow/full inode remapping (most users won't need it) */
4284 qpf_table_init(&s
->qpf_table
);
4285 /* hash table for quick inode remapping */
4286 qpp_table_init(&s
->qpp_table
);
4288 s
->qp_affix_next
= 1; /* reserve 0 to detect overflow */
4289 s
->qp_fullpath_next
= 1;
4291 s
->ctx
.fst
= &fse
->fst
;
4292 fsdev_throttle_init(s
->ctx
.fst
);
4297 v9fs_device_unrealize_common(s
);
4299 v9fs_path_free(&path
);
4303 void v9fs_device_unrealize_common(V9fsState
*s
)
4305 if (s
->ops
&& s
->ops
->cleanup
) {
4306 s
->ops
->cleanup(&s
->ctx
);
4309 fsdev_throttle_cleanup(s
->ctx
.fst
);
4312 g_hash_table_destroy(s
->fids
);
4316 qp_table_destroy(&s
->qpd_table
);
4317 qp_table_destroy(&s
->qpp_table
);
4318 qp_table_destroy(&s
->qpf_table
);
4319 g_free(s
->ctx
.fs_root
);
4322 typedef struct VirtfsCoResetData
{
4325 } VirtfsCoResetData
;
4327 static void coroutine_fn
virtfs_co_reset(void *opaque
)
4329 VirtfsCoResetData
*data
= opaque
;
4331 virtfs_reset(&data
->pdu
);
4335 void v9fs_reset(V9fsState
*s
)
4337 VirtfsCoResetData data
= { .pdu
= { .s
= s
}, .done
= false };
4340 while (!QLIST_EMPTY(&s
->active_list
)) {
4341 aio_poll(qemu_get_aio_context(), true);
4344 co
= qemu_coroutine_create(virtfs_co_reset
, &data
);
4345 qemu_coroutine_enter(co
);
4347 while (!data
.done
) {
4348 aio_poll(qemu_get_aio_context(), true);
4352 static void __attribute__((__constructor__
)) v9fs_set_fd_limit(void)
4355 if (getrlimit(RLIMIT_NOFILE
, &rlim
) < 0) {
4356 error_report("Failed to get the resource limit");
4359 open_fd_hw
= rlim
.rlim_cur
- MIN(400, rlim
.rlim_cur
/ 3);
4360 open_fd_rc
= rlim
.rlim_cur
/ 2;