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 migrate_del_blocker(&pdu
->s
->migration_blocker
);
411 return free_fid(pdu
, fidp
);
416 static V9fsFidState
*clunk_fid(V9fsState
*s
, int32_t fid
)
420 /* TODO: Use g_hash_table_steal_extended() instead? */
421 fidp
= g_hash_table_lookup(s
->fids
, GINT_TO_POINTER(fid
));
423 g_hash_table_remove(s
->fids
, GINT_TO_POINTER(fid
));
424 fidp
->clunked
= true;
430 void coroutine_fn
v9fs_reclaim_fd(V9fsPDU
*pdu
)
432 int reclaim_count
= 0;
433 V9fsState
*s
= pdu
->s
;
438 g_hash_table_iter_init(&iter
, s
->fids
);
440 QSLIST_HEAD(, V9fsFidState
) reclaim_list
=
441 QSLIST_HEAD_INITIALIZER(reclaim_list
);
443 while (g_hash_table_iter_next(&iter
, &fid
, (gpointer
*) &f
)) {
445 * Unlink fids cannot be reclaimed. Check
446 * for them and skip them. Also skip fids
447 * currently being operated on.
449 if (f
->ref
|| f
->flags
& FID_NON_RECLAIMABLE
) {
453 * if it is a recently referenced fid
454 * we leave the fid untouched and clear the
455 * reference bit. We come back to it later
456 * in the next iteration. (a simple LRU without
457 * moving list elements around)
459 if (f
->flags
& FID_REFERENCED
) {
460 f
->flags
&= ~FID_REFERENCED
;
464 * Add fids to reclaim list.
466 if (f
->fid_type
== P9_FID_FILE
) {
467 if (f
->fs
.fd
!= -1) {
469 * Up the reference count so that
470 * a clunk request won't free this fid
473 QSLIST_INSERT_HEAD(&reclaim_list
, f
, reclaim_next
);
474 f
->fs_reclaim
.fd
= f
->fs
.fd
;
478 } else if (f
->fid_type
== P9_FID_DIR
) {
479 if (f
->fs
.dir
.stream
!= NULL
) {
481 * Up the reference count so that
482 * a clunk request won't free this fid
485 QSLIST_INSERT_HEAD(&reclaim_list
, f
, reclaim_next
);
486 f
->fs_reclaim
.dir
.stream
= f
->fs
.dir
.stream
;
487 f
->fs
.dir
.stream
= NULL
;
491 if (reclaim_count
>= open_fd_rc
) {
496 * Now close the fid in reclaim list. Free them if they
497 * are already clunked.
499 while (!QSLIST_EMPTY(&reclaim_list
)) {
500 f
= QSLIST_FIRST(&reclaim_list
);
501 QSLIST_REMOVE(&reclaim_list
, f
, V9fsFidState
, reclaim_next
);
502 if (f
->fid_type
== P9_FID_FILE
) {
503 v9fs_co_close(pdu
, &f
->fs_reclaim
);
504 } else if (f
->fid_type
== P9_FID_DIR
) {
505 v9fs_co_closedir(pdu
, &f
->fs_reclaim
);
508 * Now drop the fid reference, free it
516 * This is used when a path is removed from the directory tree. Any
517 * fids that still reference it must not be closed from then on, since
518 * they cannot be reopened.
520 static int coroutine_fn
v9fs_mark_fids_unreclaim(V9fsPDU
*pdu
, V9fsPath
*path
)
523 V9fsState
*s
= pdu
->s
;
528 * The most common case is probably that we have exactly one
529 * fid for the given path, so preallocate exactly one.
531 g_autoptr(GArray
) to_reopen
= g_array_sized_new(FALSE
, FALSE
,
532 sizeof(V9fsFidState
*), 1);
535 g_hash_table_iter_init(&iter
, s
->fids
);
538 * We iterate over the fid table looking for the entries we need
539 * to reopen, and store them in to_reopen. This is because
540 * v9fs_reopen_fid() and put_fid() yield. This allows the fid table
541 * to be modified in the meantime, invalidating our iterator.
543 while (g_hash_table_iter_next(&iter
, &fid
, (gpointer
*) &fidp
)) {
544 if (fidp
->path
.size
== path
->size
&&
545 !memcmp(fidp
->path
.data
, path
->data
, path
->size
)) {
547 * Ensure the fid survives a potential clunk request during
548 * v9fs_reopen_fid or put_fid.
551 fidp
->flags
|= FID_NON_RECLAIMABLE
;
552 g_array_append_val(to_reopen
, fidp
);
556 for (i
= 0; i
< to_reopen
->len
; i
++) {
557 fidp
= g_array_index(to_reopen
, V9fsFidState
*, i
);
558 /* reopen the file/dir if already closed */
559 err
= v9fs_reopen_fid(pdu
, fidp
);
565 for (i
= 0; i
< to_reopen
->len
; i
++) {
566 put_fid(pdu
, g_array_index(to_reopen
, V9fsFidState
*, i
));
571 static void coroutine_fn
virtfs_reset(V9fsPDU
*pdu
)
573 V9fsState
*s
= pdu
->s
;
577 * Get a list of all the values (fid states) in the table, which
580 g_autoptr(GList
) fids
= g_hash_table_get_values(s
->fids
);
582 /* ... remove from the table, taking over ownership. */
583 g_hash_table_steal_all(s
->fids
);
586 * This allows us to release our references to them asynchronously without
587 * iterating over the hash table and risking iterator invalidation
588 * through concurrent modifications.
590 for (freeing
= fids
; freeing
; freeing
= freeing
->next
) {
591 fidp
= freeing
->data
;
593 fidp
->clunked
= true;
598 #define P9_QID_TYPE_DIR 0x80
599 #define P9_QID_TYPE_SYMLINK 0x02
601 #define P9_STAT_MODE_DIR 0x80000000
602 #define P9_STAT_MODE_APPEND 0x40000000
603 #define P9_STAT_MODE_EXCL 0x20000000
604 #define P9_STAT_MODE_MOUNT 0x10000000
605 #define P9_STAT_MODE_AUTH 0x08000000
606 #define P9_STAT_MODE_TMP 0x04000000
607 #define P9_STAT_MODE_SYMLINK 0x02000000
608 #define P9_STAT_MODE_LINK 0x01000000
609 #define P9_STAT_MODE_DEVICE 0x00800000
610 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
611 #define P9_STAT_MODE_SOCKET 0x00100000
612 #define P9_STAT_MODE_SETUID 0x00080000
613 #define P9_STAT_MODE_SETGID 0x00040000
614 #define P9_STAT_MODE_SETVTX 0x00010000
616 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR | \
617 P9_STAT_MODE_SYMLINK | \
618 P9_STAT_MODE_LINK | \
619 P9_STAT_MODE_DEVICE | \
620 P9_STAT_MODE_NAMED_PIPE | \
623 /* Mirrors all bits of a byte. So e.g. binary 10100000 would become 00000101. */
624 static inline uint8_t mirror8bit(uint8_t byte
)
626 return (byte
* 0x0202020202ULL
& 0x010884422010ULL
) % 1023;
629 /* Same as mirror8bit() just for a 64 bit data type instead for a byte. */
630 static inline uint64_t mirror64bit(uint64_t value
)
632 return ((uint64_t)mirror8bit(value
& 0xff) << 56) |
633 ((uint64_t)mirror8bit((value
>> 8) & 0xff) << 48) |
634 ((uint64_t)mirror8bit((value
>> 16) & 0xff) << 40) |
635 ((uint64_t)mirror8bit((value
>> 24) & 0xff) << 32) |
636 ((uint64_t)mirror8bit((value
>> 32) & 0xff) << 24) |
637 ((uint64_t)mirror8bit((value
>> 40) & 0xff) << 16) |
638 ((uint64_t)mirror8bit((value
>> 48) & 0xff) << 8) |
639 ((uint64_t)mirror8bit((value
>> 56) & 0xff));
643 * Parameter k for the Exponential Golomb algorithm to be used.
645 * The smaller this value, the smaller the minimum bit count for the Exp.
646 * Golomb generated affixes will be (at lowest index) however for the
647 * price of having higher maximum bit count of generated affixes (at highest
648 * index). Likewise increasing this parameter yields in smaller maximum bit
649 * count for the price of having higher minimum bit count.
651 * In practice that means: a good value for k depends on the expected amount
652 * of devices to be exposed by one export. For a small amount of devices k
653 * should be small, for a large amount of devices k might be increased
654 * instead. The default of k=0 should be fine for most users though.
656 * IMPORTANT: In case this ever becomes a runtime parameter; the value of
657 * k should not change as long as guest is still running! Because that would
658 * cause completely different inode numbers to be generated on guest.
660 #define EXP_GOLOMB_K 0
663 * expGolombEncode() - Exponential Golomb algorithm for arbitrary k
666 * @n: natural number (or index) of the prefix to be generated
668 * @k: parameter k of Exp. Golomb algorithm to be used
669 * (see comment on EXP_GOLOMB_K macro for details about k)
670 * Return: prefix for given @n and @k
672 * The Exponential Golomb algorithm generates prefixes (NOT suffixes!)
673 * with growing length and with the mathematical property of being
674 * "prefix-free". The latter means the generated prefixes can be prepended
675 * in front of arbitrary numbers and the resulting concatenated numbers are
676 * guaranteed to be always unique.
678 * This is a minor adjustment to the original Exp. Golomb algorithm in the
679 * sense that lowest allowed index (@n) starts with 1, not with zero.
681 static VariLenAffix
expGolombEncode(uint64_t n
, int k
)
683 const uint64_t value
= n
+ (1 << k
) - 1;
684 const int bits
= (int) log2(value
) + 1;
685 return (VariLenAffix
) {
686 .type
= AffixType_Prefix
,
688 .bits
= bits
+ MAX((bits
- 1 - k
), 0)
693 * invertAffix() - Converts a suffix into a prefix, or a prefix into a suffix.
694 * @affix: either suffix or prefix to be inverted
695 * Return: inversion of passed @affix
697 * Simply mirror all bits of the affix value, for the purpose to preserve
698 * respectively the mathematical "prefix-free" or "suffix-free" property
699 * after the conversion.
701 * If a passed prefix is suitable to create unique numbers, then the
702 * returned suffix is suitable to create unique numbers as well (and vice
705 static VariLenAffix
invertAffix(const VariLenAffix
*affix
)
707 return (VariLenAffix
) {
709 (affix
->type
== AffixType_Suffix
) ?
710 AffixType_Prefix
: AffixType_Suffix
,
712 mirror64bit(affix
->value
) >>
713 ((sizeof(affix
->value
) * 8) - affix
->bits
),
719 * affixForIndex() - Generates suffix numbers with "suffix-free" property.
720 * @index: natural number (or index) of the suffix to be generated
722 * Return: Suffix suitable to assemble unique number.
724 * This is just a wrapper function on top of the Exp. Golomb algorithm.
726 * Since the Exp. Golomb algorithm generates prefixes, but we need suffixes,
727 * this function converts the Exp. Golomb prefixes into appropriate suffixes
728 * which are still suitable for generating unique numbers.
730 static VariLenAffix
affixForIndex(uint64_t index
)
733 prefix
= expGolombEncode(index
, EXP_GOLOMB_K
);
734 return invertAffix(&prefix
); /* convert prefix to suffix */
737 static uint32_t qpp_hash(QppEntry e
)
739 return qemu_xxhash4(e
.ino_prefix
, e
.dev
);
742 static uint32_t qpf_hash(QpfEntry e
)
744 return qemu_xxhash4(e
.ino
, e
.dev
);
747 static bool qpd_cmp_func(const void *obj
, const void *userp
)
749 const QpdEntry
*e1
= obj
, *e2
= userp
;
750 return e1
->dev
== e2
->dev
;
753 static bool qpp_cmp_func(const void *obj
, const void *userp
)
755 const QppEntry
*e1
= obj
, *e2
= userp
;
756 return e1
->dev
== e2
->dev
&& e1
->ino_prefix
== e2
->ino_prefix
;
759 static bool qpf_cmp_func(const void *obj
, const void *userp
)
761 const QpfEntry
*e1
= obj
, *e2
= userp
;
762 return e1
->dev
== e2
->dev
&& e1
->ino
== e2
->ino
;
765 static void qp_table_remove(void *p
, uint32_t h
, void *up
)
770 static void qp_table_destroy(struct qht
*ht
)
772 if (!ht
|| !ht
->map
) {
775 qht_iter(ht
, qp_table_remove
, NULL
);
779 static void qpd_table_init(struct qht
*ht
)
781 qht_init(ht
, qpd_cmp_func
, 1, QHT_MODE_AUTO_RESIZE
);
784 static void qpp_table_init(struct qht
*ht
)
786 qht_init(ht
, qpp_cmp_func
, 1, QHT_MODE_AUTO_RESIZE
);
789 static void qpf_table_init(struct qht
*ht
)
791 qht_init(ht
, qpf_cmp_func
, 1 << 16, QHT_MODE_AUTO_RESIZE
);
795 * Returns how many (high end) bits of inode numbers of the passed fs
796 * device shall be used (in combination with the device number) to
797 * generate hash values for qpp_table entries.
799 * This function is required if variable length suffixes are used for inode
800 * number mapping on guest level. Since a device may end up having multiple
801 * entries in qpp_table, each entry most probably with a different suffix
802 * length, we thus need this function in conjunction with qpd_table to
803 * "agree" about a fix amount of bits (per device) to be always used for
804 * generating hash values for the purpose of accessing qpp_table in order
805 * get consistent behaviour when accessing qpp_table.
807 static int qid_inode_prefix_hash_bits(V9fsPDU
*pdu
, dev_t dev
)
815 val
= qht_lookup(&pdu
->s
->qpd_table
, &lookup
, hash
);
817 val
= g_new0(QpdEntry
, 1);
819 affix
= affixForIndex(pdu
->s
->qp_affix_next
);
820 val
->prefix_bits
= affix
.bits
;
821 qht_insert(&pdu
->s
->qpd_table
, val
, hash
, NULL
);
822 pdu
->s
->qp_ndevices
++;
824 return val
->prefix_bits
;
828 * Slow / full mapping host inode nr -> guest inode nr.
830 * This function performs a slower and much more costly remapping of an
831 * original file inode number on host to an appropriate different inode
832 * number on guest. For every (dev, inode) combination on host a new
833 * sequential number is generated, cached and exposed as inode number on
836 * This is just a "last resort" fallback solution if the much faster/cheaper
837 * qid_path_suffixmap() failed. In practice this slow / full mapping is not
838 * expected ever to be used at all though.
840 * See qid_path_suffixmap() for details
843 static int qid_path_fullmap(V9fsPDU
*pdu
, const struct stat
*stbuf
,
847 .dev
= stbuf
->st_dev
,
850 uint32_t hash
= qpf_hash(lookup
);
853 val
= qht_lookup(&pdu
->s
->qpf_table
, &lookup
, hash
);
856 if (pdu
->s
->qp_fullpath_next
== 0) {
857 /* no more files can be mapped :'( */
859 "9p: No more prefixes available for remapping inodes from "
865 val
= g_new0(QpfEntry
, 1);
868 /* new unique inode and device combo */
869 affix
= affixForIndex(
870 1ULL << (sizeof(pdu
->s
->qp_affix_next
) * 8)
872 val
->path
= (pdu
->s
->qp_fullpath_next
++ << affix
.bits
) | affix
.value
;
873 pdu
->s
->qp_fullpath_next
&= ((1ULL << (64 - affix
.bits
)) - 1);
874 qht_insert(&pdu
->s
->qpf_table
, val
, hash
, NULL
);
882 * Quick mapping host inode nr -> guest inode nr.
884 * This function performs quick remapping of an original file inode number
885 * on host to an appropriate different inode number on guest. This remapping
886 * of inodes is required to avoid inode nr collisions on guest which would
887 * happen if the 9p export contains more than 1 exported file system (or
888 * more than 1 file system data set), because unlike on host level where the
889 * files would have different device nrs, all files exported by 9p would
890 * share the same device nr on guest (the device nr of the virtual 9p device
893 * Inode remapping is performed by chopping off high end bits of the original
894 * inode number from host, shifting the result upwards and then assigning a
895 * generated suffix number for the low end bits, where the same suffix number
896 * will be shared by all inodes with the same device id AND the same high end
897 * bits that have been chopped off. That approach utilizes the fact that inode
898 * numbers very likely share the same high end bits (i.e. due to their common
899 * sequential generation by file systems) and hence we only have to generate
900 * and track a very limited amount of suffixes in practice due to that.
902 * We generate variable size suffixes for that purpose. The 1st generated
903 * suffix will only have 1 bit and hence we only need to chop off 1 bit from
904 * the original inode number. The subsequent suffixes being generated will
905 * grow in (bit) size subsequently, i.e. the 2nd and 3rd suffix being
906 * generated will have 3 bits and hence we have to chop off 3 bits from their
907 * original inodes, and so on. That approach of using variable length suffixes
908 * (i.e. over fixed size ones) utilizes the fact that in practice only a very
909 * limited amount of devices are shared by the same export (e.g. typically
910 * less than 2 dozen devices per 9p export), so in practice we need to chop
911 * off less bits than with fixed size prefixes and yet are flexible to add
912 * new devices at runtime below host's export directory at any time without
913 * having to reboot guest nor requiring to reconfigure guest for that. And due
914 * to the very limited amount of original high end bits that we chop off that
915 * way, the total amount of suffixes we need to generate is less than by using
916 * fixed size prefixes and hence it also improves performance of the inode
917 * remapping algorithm, and finally has the nice side effect that the inode
918 * numbers on guest will be much smaller & human friendly. ;-)
920 static int qid_path_suffixmap(V9fsPDU
*pdu
, const struct stat
*stbuf
,
923 const int ino_hash_bits
= qid_inode_prefix_hash_bits(pdu
, stbuf
->st_dev
);
925 .dev
= stbuf
->st_dev
,
926 .ino_prefix
= (uint16_t) (stbuf
->st_ino
>> (64 - ino_hash_bits
))
928 uint32_t hash
= qpp_hash(lookup
);
930 val
= qht_lookup(&pdu
->s
->qpp_table
, &lookup
, hash
);
933 if (pdu
->s
->qp_affix_next
== 0) {
934 /* we ran out of affixes */
936 "9p: Potential degraded performance of inode remapping"
941 val
= g_new0(QppEntry
, 1);
944 /* new unique inode affix and device combo */
945 val
->qp_affix_index
= pdu
->s
->qp_affix_next
++;
946 val
->qp_affix
= affixForIndex(val
->qp_affix_index
);
947 qht_insert(&pdu
->s
->qpp_table
, val
, hash
, NULL
);
949 /* assuming generated affix to be suffix type, not prefix */
950 *path
= (stbuf
->st_ino
<< val
->qp_affix
.bits
) | val
->qp_affix
.value
;
954 static int stat_to_qid(V9fsPDU
*pdu
, const struct stat
*stbuf
, V9fsQID
*qidp
)
959 if (pdu
->s
->ctx
.export_flags
& V9FS_REMAP_INODES
) {
960 /* map inode+device to qid path (fast path) */
961 err
= qid_path_suffixmap(pdu
, stbuf
, &qidp
->path
);
962 if (err
== -ENFILE
) {
963 /* fast path didn't work, fall back to full map */
964 err
= qid_path_fullmap(pdu
, stbuf
, &qidp
->path
);
970 if (pdu
->s
->dev_id
!= stbuf
->st_dev
) {
971 if (pdu
->s
->ctx
.export_flags
& V9FS_FORBID_MULTIDEVS
) {
973 "9p: Multiple devices detected in same VirtFS export. "
974 "Access of guest to additional devices is (partly) "
975 "denied due to virtfs option 'multidevs=forbid' being "
981 "9p: Multiple devices detected in same VirtFS export, "
982 "which might lead to file ID collisions and severe "
983 "misbehaviours on guest! You should either use a "
984 "separate export for each device shared from host or "
985 "use virtfs option 'multidevs=remap'!"
989 memset(&qidp
->path
, 0, sizeof(qidp
->path
));
990 size
= MIN(sizeof(stbuf
->st_ino
), sizeof(qidp
->path
));
991 memcpy(&qidp
->path
, &stbuf
->st_ino
, size
);
994 qidp
->version
= stbuf
->st_mtime
^ (stbuf
->st_size
<< 8);
996 if (S_ISDIR(stbuf
->st_mode
)) {
997 qidp
->type
|= P9_QID_TYPE_DIR
;
999 if (S_ISLNK(stbuf
->st_mode
)) {
1000 qidp
->type
|= P9_QID_TYPE_SYMLINK
;
1006 V9fsPDU
*pdu_alloc(V9fsState
*s
)
1008 V9fsPDU
*pdu
= NULL
;
1010 if (!QLIST_EMPTY(&s
->free_list
)) {
1011 pdu
= QLIST_FIRST(&s
->free_list
);
1012 QLIST_REMOVE(pdu
, next
);
1013 QLIST_INSERT_HEAD(&s
->active_list
, pdu
, next
);
1018 void pdu_free(V9fsPDU
*pdu
)
1020 V9fsState
*s
= pdu
->s
;
1022 g_assert(!pdu
->cancelled
);
1023 QLIST_REMOVE(pdu
, next
);
1024 QLIST_INSERT_HEAD(&s
->free_list
, pdu
, next
);
1027 static void coroutine_fn
pdu_complete(V9fsPDU
*pdu
, ssize_t len
)
1029 int8_t id
= pdu
->id
+ 1; /* Response */
1030 V9fsState
*s
= pdu
->s
;
1034 * The 9p spec requires that successfully cancelled pdus receive no reply.
1035 * Sending a reply would confuse clients because they would
1036 * assume that any EINTR is the actual result of the operation,
1037 * rather than a consequence of the cancellation. However, if
1038 * the operation completed (successfully or with an error other
1039 * than caused be cancellation), we do send out that reply, both
1040 * for efficiency and to avoid confusing the rest of the state machine
1041 * that assumes passing a non-error here will mean a successful
1042 * transmission of the reply.
1044 bool discard
= pdu
->cancelled
&& len
== -EINTR
;
1046 trace_v9fs_rcancel(pdu
->tag
, pdu
->id
);
1055 if (s
->proto_version
!= V9FS_PROTO_2000L
) {
1058 str
.data
= strerror(err
);
1059 str
.size
= strlen(str
.data
);
1061 ret
= pdu_marshal(pdu
, len
, "s", &str
);
1068 err
= errno_to_dotl(err
);
1071 ret
= pdu_marshal(pdu
, len
, "d", err
);
1077 if (s
->proto_version
== V9FS_PROTO_2000L
) {
1080 trace_v9fs_rerror(pdu
->tag
, pdu
->id
, err
); /* Trace ERROR */
1083 /* fill out the header */
1084 if (pdu_marshal(pdu
, 0, "dbw", (int32_t)len
, id
, pdu
->tag
) < 0) {
1088 /* keep these in sync */
1093 pdu
->s
->transport
->push_and_notify(pdu
);
1095 /* Now wakeup anybody waiting in flush for this request */
1096 if (!qemu_co_queue_next(&pdu
->complete
)) {
1101 static mode_t
v9mode_to_mode(uint32_t mode
, V9fsString
*extension
)
1106 if (mode
& P9_STAT_MODE_DIR
) {
1110 if (mode
& P9_STAT_MODE_SYMLINK
) {
1113 if (mode
& P9_STAT_MODE_SOCKET
) {
1116 if (mode
& P9_STAT_MODE_NAMED_PIPE
) {
1119 if (mode
& P9_STAT_MODE_DEVICE
) {
1120 if (extension
->size
&& extension
->data
[0] == 'c') {
1127 if (!(ret
& ~0777)) {
1131 if (mode
& P9_STAT_MODE_SETUID
) {
1134 if (mode
& P9_STAT_MODE_SETGID
) {
1137 if (mode
& P9_STAT_MODE_SETVTX
) {
1144 static int donttouch_stat(V9fsStat
*stat
)
1146 if (stat
->type
== -1 &&
1148 stat
->qid
.type
== 0xff &&
1149 stat
->qid
.version
== (uint32_t) -1 &&
1150 stat
->qid
.path
== (uint64_t) -1 &&
1152 stat
->atime
== -1 &&
1153 stat
->mtime
== -1 &&
1154 stat
->length
== -1 &&
1159 stat
->n_uid
== -1 &&
1160 stat
->n_gid
== -1 &&
1161 stat
->n_muid
== -1) {
1168 static void v9fs_stat_init(V9fsStat
*stat
)
1170 v9fs_string_init(&stat
->name
);
1171 v9fs_string_init(&stat
->uid
);
1172 v9fs_string_init(&stat
->gid
);
1173 v9fs_string_init(&stat
->muid
);
1174 v9fs_string_init(&stat
->extension
);
1177 static void v9fs_stat_free(V9fsStat
*stat
)
1179 v9fs_string_free(&stat
->name
);
1180 v9fs_string_free(&stat
->uid
);
1181 v9fs_string_free(&stat
->gid
);
1182 v9fs_string_free(&stat
->muid
);
1183 v9fs_string_free(&stat
->extension
);
1186 static uint32_t stat_to_v9mode(const struct stat
*stbuf
)
1190 mode
= stbuf
->st_mode
& 0777;
1191 if (S_ISDIR(stbuf
->st_mode
)) {
1192 mode
|= P9_STAT_MODE_DIR
;
1195 if (S_ISLNK(stbuf
->st_mode
)) {
1196 mode
|= P9_STAT_MODE_SYMLINK
;
1199 if (S_ISSOCK(stbuf
->st_mode
)) {
1200 mode
|= P9_STAT_MODE_SOCKET
;
1203 if (S_ISFIFO(stbuf
->st_mode
)) {
1204 mode
|= P9_STAT_MODE_NAMED_PIPE
;
1207 if (S_ISBLK(stbuf
->st_mode
) || S_ISCHR(stbuf
->st_mode
)) {
1208 mode
|= P9_STAT_MODE_DEVICE
;
1211 if (stbuf
->st_mode
& S_ISUID
) {
1212 mode
|= P9_STAT_MODE_SETUID
;
1215 if (stbuf
->st_mode
& S_ISGID
) {
1216 mode
|= P9_STAT_MODE_SETGID
;
1219 if (stbuf
->st_mode
& S_ISVTX
) {
1220 mode
|= P9_STAT_MODE_SETVTX
;
1226 static int coroutine_fn
stat_to_v9stat(V9fsPDU
*pdu
, V9fsPath
*path
,
1227 const char *basename
,
1228 const struct stat
*stbuf
,
1233 memset(v9stat
, 0, sizeof(*v9stat
));
1235 err
= stat_to_qid(pdu
, stbuf
, &v9stat
->qid
);
1239 v9stat
->mode
= stat_to_v9mode(stbuf
);
1240 v9stat
->atime
= stbuf
->st_atime
;
1241 v9stat
->mtime
= stbuf
->st_mtime
;
1242 v9stat
->length
= stbuf
->st_size
;
1244 v9fs_string_free(&v9stat
->uid
);
1245 v9fs_string_free(&v9stat
->gid
);
1246 v9fs_string_free(&v9stat
->muid
);
1248 v9stat
->n_uid
= stbuf
->st_uid
;
1249 v9stat
->n_gid
= stbuf
->st_gid
;
1252 v9fs_string_free(&v9stat
->extension
);
1254 if (v9stat
->mode
& P9_STAT_MODE_SYMLINK
) {
1255 err
= v9fs_co_readlink(pdu
, path
, &v9stat
->extension
);
1259 } else if (v9stat
->mode
& P9_STAT_MODE_DEVICE
) {
1260 v9fs_string_sprintf(&v9stat
->extension
, "%c %u %u",
1261 S_ISCHR(stbuf
->st_mode
) ? 'c' : 'b',
1262 major(stbuf
->st_rdev
), minor(stbuf
->st_rdev
));
1263 } else if (S_ISDIR(stbuf
->st_mode
) || S_ISREG(stbuf
->st_mode
)) {
1264 v9fs_string_sprintf(&v9stat
->extension
, "%s %lu",
1265 "HARDLINKCOUNT", (unsigned long)stbuf
->st_nlink
);
1268 v9fs_string_sprintf(&v9stat
->name
, "%s", basename
);
1271 v9fs_string_size(&v9stat
->name
) +
1272 v9fs_string_size(&v9stat
->uid
) +
1273 v9fs_string_size(&v9stat
->gid
) +
1274 v9fs_string_size(&v9stat
->muid
) +
1275 v9fs_string_size(&v9stat
->extension
);
1279 #define P9_STATS_MODE 0x00000001ULL
1280 #define P9_STATS_NLINK 0x00000002ULL
1281 #define P9_STATS_UID 0x00000004ULL
1282 #define P9_STATS_GID 0x00000008ULL
1283 #define P9_STATS_RDEV 0x00000010ULL
1284 #define P9_STATS_ATIME 0x00000020ULL
1285 #define P9_STATS_MTIME 0x00000040ULL
1286 #define P9_STATS_CTIME 0x00000080ULL
1287 #define P9_STATS_INO 0x00000100ULL
1288 #define P9_STATS_SIZE 0x00000200ULL
1289 #define P9_STATS_BLOCKS 0x00000400ULL
1291 #define P9_STATS_BTIME 0x00000800ULL
1292 #define P9_STATS_GEN 0x00001000ULL
1293 #define P9_STATS_DATA_VERSION 0x00002000ULL
1295 #define P9_STATS_BASIC 0x000007ffULL /* Mask for fields up to BLOCKS */
1296 #define P9_STATS_ALL 0x00003fffULL /* Mask for All fields above */
1300 * blksize_to_iounit() - Block size exposed to 9p client.
1301 * Return: block size
1303 * @pdu: 9p client request
1304 * @blksize: host filesystem's block size
1306 * Convert host filesystem's block size into an appropriate block size for
1307 * 9p client (guest OS side). The value returned suggests an "optimum" block
1308 * size for 9p I/O, i.e. to maximize performance.
1310 static int32_t blksize_to_iounit(const V9fsPDU
*pdu
, int32_t blksize
)
1313 V9fsState
*s
= pdu
->s
;
1316 * iounit should be multiples of blksize (host filesystem block size)
1317 * as well as less than (client msize - P9_IOHDRSZ)
1320 iounit
= QEMU_ALIGN_DOWN(s
->msize
- P9_IOHDRSZ
, blksize
);
1323 iounit
= s
->msize
- P9_IOHDRSZ
;
1328 static int32_t stat_to_iounit(const V9fsPDU
*pdu
, const struct stat
*stbuf
)
1330 return blksize_to_iounit(pdu
, stbuf
->st_blksize
);
1333 static int stat_to_v9stat_dotl(V9fsPDU
*pdu
, const struct stat
*stbuf
,
1334 V9fsStatDotl
*v9lstat
)
1336 memset(v9lstat
, 0, sizeof(*v9lstat
));
1338 v9lstat
->st_mode
= stbuf
->st_mode
;
1339 v9lstat
->st_nlink
= stbuf
->st_nlink
;
1340 v9lstat
->st_uid
= stbuf
->st_uid
;
1341 v9lstat
->st_gid
= stbuf
->st_gid
;
1342 v9lstat
->st_rdev
= host_dev_to_dotl_dev(stbuf
->st_rdev
);
1343 v9lstat
->st_size
= stbuf
->st_size
;
1344 v9lstat
->st_blksize
= stat_to_iounit(pdu
, stbuf
);
1345 v9lstat
->st_blocks
= stbuf
->st_blocks
;
1346 v9lstat
->st_atime_sec
= stbuf
->st_atime
;
1347 v9lstat
->st_mtime_sec
= stbuf
->st_mtime
;
1348 v9lstat
->st_ctime_sec
= stbuf
->st_ctime
;
1349 #ifdef CONFIG_DARWIN
1350 v9lstat
->st_atime_nsec
= stbuf
->st_atimespec
.tv_nsec
;
1351 v9lstat
->st_mtime_nsec
= stbuf
->st_mtimespec
.tv_nsec
;
1352 v9lstat
->st_ctime_nsec
= stbuf
->st_ctimespec
.tv_nsec
;
1354 v9lstat
->st_atime_nsec
= stbuf
->st_atim
.tv_nsec
;
1355 v9lstat
->st_mtime_nsec
= stbuf
->st_mtim
.tv_nsec
;
1356 v9lstat
->st_ctime_nsec
= stbuf
->st_ctim
.tv_nsec
;
1358 /* Currently we only support BASIC fields in stat */
1359 v9lstat
->st_result_mask
= P9_STATS_BASIC
;
1361 return stat_to_qid(pdu
, stbuf
, &v9lstat
->qid
);
1364 static void print_sg(struct iovec
*sg
, int cnt
)
1368 printf("sg[%d]: {", cnt
);
1369 for (i
= 0; i
< cnt
; i
++) {
1373 printf("(%p, %zd)", sg
[i
].iov_base
, sg
[i
].iov_len
);
1378 /* Will call this only for path name based fid */
1379 static void v9fs_fix_path(V9fsPath
*dst
, V9fsPath
*src
, int len
)
1382 v9fs_path_init(&str
);
1383 v9fs_path_copy(&str
, dst
);
1384 v9fs_path_sprintf(dst
, "%s%s", src
->data
, str
.data
+ len
);
1385 v9fs_path_free(&str
);
1388 static inline bool is_ro_export(FsContext
*ctx
)
1390 return ctx
->export_flags
& V9FS_RDONLY
;
1393 static void coroutine_fn
v9fs_version(void *opaque
)
1396 V9fsPDU
*pdu
= opaque
;
1397 V9fsState
*s
= pdu
->s
;
1401 v9fs_string_init(&version
);
1402 err
= pdu_unmarshal(pdu
, offset
, "ds", &s
->msize
, &version
);
1406 trace_v9fs_version(pdu
->tag
, pdu
->id
, s
->msize
, version
.data
);
1410 if (!strcmp(version
.data
, "9P2000.u")) {
1411 s
->proto_version
= V9FS_PROTO_2000U
;
1412 } else if (!strcmp(version
.data
, "9P2000.L")) {
1413 s
->proto_version
= V9FS_PROTO_2000L
;
1415 v9fs_string_sprintf(&version
, "unknown");
1416 /* skip min. msize check, reporting invalid version has priority */
1420 if (s
->msize
< P9_MIN_MSIZE
) {
1423 "9pfs: Client requested msize < minimum msize ("
1424 stringify(P9_MIN_MSIZE
) ") supported by this server."
1429 /* 8192 is the default msize of Linux clients */
1430 if (s
->msize
<= 8192 && !(s
->ctx
.export_flags
& V9FS_NO_PERF_WARN
)) {
1432 "9p: degraded performance: a reasonable high msize should be "
1433 "chosen on client/guest side (chosen msize is <= 8192). See "
1434 "https://wiki.qemu.org/Documentation/9psetup#msize for details."
1439 err
= pdu_marshal(pdu
, offset
, "ds", s
->msize
, &version
);
1444 trace_v9fs_version_return(pdu
->tag
, pdu
->id
, s
->msize
, version
.data
);
1446 pdu_complete(pdu
, err
);
1447 v9fs_string_free(&version
);
1450 static void coroutine_fn
v9fs_attach(void *opaque
)
1452 V9fsPDU
*pdu
= opaque
;
1453 V9fsState
*s
= pdu
->s
;
1454 int32_t fid
, afid
, n_uname
;
1455 V9fsString uname
, aname
;
1462 v9fs_string_init(&uname
);
1463 v9fs_string_init(&aname
);
1464 err
= pdu_unmarshal(pdu
, offset
, "ddssd", &fid
,
1465 &afid
, &uname
, &aname
, &n_uname
);
1469 trace_v9fs_attach(pdu
->tag
, pdu
->id
, fid
, afid
, uname
.data
, aname
.data
);
1471 fidp
= alloc_fid(s
, fid
);
1476 fidp
->uid
= n_uname
;
1477 err
= v9fs_co_name_to_path(pdu
, NULL
, "/", &fidp
->path
);
1483 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1489 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
1497 * disable migration if we haven't done already.
1498 * attach could get called multiple times for the same export.
1500 if (!s
->migration_blocker
) {
1501 error_setg(&s
->migration_blocker
,
1502 "Migration is disabled when VirtFS export path '%s' is mounted in the guest using mount_tag '%s'",
1503 s
->ctx
.fs_root
? s
->ctx
.fs_root
: "NULL", s
->tag
);
1504 err
= migrate_add_blocker(&s
->migration_blocker
, NULL
);
1512 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
1519 memcpy(&s
->root_st
, &stbuf
, sizeof(stbuf
));
1520 trace_v9fs_attach_return(pdu
->tag
, pdu
->id
,
1521 qid
.type
, qid
.version
, qid
.path
);
1525 pdu_complete(pdu
, err
);
1526 v9fs_string_free(&uname
);
1527 v9fs_string_free(&aname
);
1530 static void coroutine_fn
v9fs_stat(void *opaque
)
1538 V9fsPDU
*pdu
= opaque
;
1541 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
1545 trace_v9fs_stat(pdu
->tag
, pdu
->id
, fid
);
1547 fidp
= get_fid(pdu
, fid
);
1552 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1556 basename
= g_path_get_basename(fidp
->path
.data
);
1557 err
= stat_to_v9stat(pdu
, &fidp
->path
, basename
, &stbuf
, &v9stat
);
1562 err
= pdu_marshal(pdu
, offset
, "wS", 0, &v9stat
);
1564 v9fs_stat_free(&v9stat
);
1567 trace_v9fs_stat_return(pdu
->tag
, pdu
->id
, v9stat
.mode
,
1568 v9stat
.atime
, v9stat
.mtime
, v9stat
.length
);
1570 v9fs_stat_free(&v9stat
);
1574 pdu_complete(pdu
, err
);
1577 static void coroutine_fn
v9fs_getattr(void *opaque
)
1584 uint64_t request_mask
;
1585 V9fsStatDotl v9stat_dotl
;
1586 V9fsPDU
*pdu
= opaque
;
1588 retval
= pdu_unmarshal(pdu
, offset
, "dq", &fid
, &request_mask
);
1592 trace_v9fs_getattr(pdu
->tag
, pdu
->id
, fid
, request_mask
);
1594 fidp
= get_fid(pdu
, fid
);
1600 * Currently we only support BASIC fields in stat, so there is no
1601 * need to look at request_mask.
1603 retval
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
1607 retval
= stat_to_v9stat_dotl(pdu
, &stbuf
, &v9stat_dotl
);
1612 /* fill st_gen if requested and supported by underlying fs */
1613 if (request_mask
& P9_STATS_GEN
) {
1614 retval
= v9fs_co_st_gen(pdu
, &fidp
->path
, stbuf
.st_mode
, &v9stat_dotl
);
1617 /* we have valid st_gen: update result mask */
1618 v9stat_dotl
.st_result_mask
|= P9_STATS_GEN
;
1621 /* request cancelled, e.g. by Tflush */
1624 /* failed to get st_gen: not fatal, ignore */
1628 retval
= pdu_marshal(pdu
, offset
, "A", &v9stat_dotl
);
1633 trace_v9fs_getattr_return(pdu
->tag
, pdu
->id
, v9stat_dotl
.st_result_mask
,
1634 v9stat_dotl
.st_mode
, v9stat_dotl
.st_uid
,
1635 v9stat_dotl
.st_gid
);
1639 pdu_complete(pdu
, retval
);
1642 /* Attribute flags */
1643 #define P9_ATTR_MODE (1 << 0)
1644 #define P9_ATTR_UID (1 << 1)
1645 #define P9_ATTR_GID (1 << 2)
1646 #define P9_ATTR_SIZE (1 << 3)
1647 #define P9_ATTR_ATIME (1 << 4)
1648 #define P9_ATTR_MTIME (1 << 5)
1649 #define P9_ATTR_CTIME (1 << 6)
1650 #define P9_ATTR_ATIME_SET (1 << 7)
1651 #define P9_ATTR_MTIME_SET (1 << 8)
1653 #define P9_ATTR_MASK 127
1655 static void coroutine_fn
v9fs_setattr(void *opaque
)
1662 V9fsPDU
*pdu
= opaque
;
1664 err
= pdu_unmarshal(pdu
, offset
, "dI", &fid
, &v9iattr
);
1669 trace_v9fs_setattr(pdu
->tag
, pdu
->id
, fid
,
1670 v9iattr
.valid
, v9iattr
.mode
, v9iattr
.uid
, v9iattr
.gid
,
1671 v9iattr
.size
, v9iattr
.atime_sec
, v9iattr
.mtime_sec
);
1673 fidp
= get_fid(pdu
, fid
);
1678 if (v9iattr
.valid
& P9_ATTR_MODE
) {
1679 err
= v9fs_co_chmod(pdu
, &fidp
->path
, v9iattr
.mode
);
1684 if (v9iattr
.valid
& (P9_ATTR_ATIME
| P9_ATTR_MTIME
)) {
1685 struct timespec times
[2];
1686 if (v9iattr
.valid
& P9_ATTR_ATIME
) {
1687 if (v9iattr
.valid
& P9_ATTR_ATIME_SET
) {
1688 times
[0].tv_sec
= v9iattr
.atime_sec
;
1689 times
[0].tv_nsec
= v9iattr
.atime_nsec
;
1691 times
[0].tv_nsec
= UTIME_NOW
;
1694 times
[0].tv_nsec
= UTIME_OMIT
;
1696 if (v9iattr
.valid
& P9_ATTR_MTIME
) {
1697 if (v9iattr
.valid
& P9_ATTR_MTIME_SET
) {
1698 times
[1].tv_sec
= v9iattr
.mtime_sec
;
1699 times
[1].tv_nsec
= v9iattr
.mtime_nsec
;
1701 times
[1].tv_nsec
= UTIME_NOW
;
1704 times
[1].tv_nsec
= UTIME_OMIT
;
1706 err
= v9fs_co_utimensat(pdu
, &fidp
->path
, times
);
1712 * If the only valid entry in iattr is ctime we can call
1713 * chown(-1,-1) to update the ctime of the file
1715 if ((v9iattr
.valid
& (P9_ATTR_UID
| P9_ATTR_GID
)) ||
1716 ((v9iattr
.valid
& P9_ATTR_CTIME
)
1717 && !((v9iattr
.valid
& P9_ATTR_MASK
) & ~P9_ATTR_CTIME
))) {
1718 if (!(v9iattr
.valid
& P9_ATTR_UID
)) {
1721 if (!(v9iattr
.valid
& P9_ATTR_GID
)) {
1724 err
= v9fs_co_chown(pdu
, &fidp
->path
, v9iattr
.uid
,
1730 if (v9iattr
.valid
& (P9_ATTR_SIZE
)) {
1731 err
= v9fs_co_truncate(pdu
, &fidp
->path
, v9iattr
.size
);
1737 trace_v9fs_setattr_return(pdu
->tag
, pdu
->id
);
1741 pdu_complete(pdu
, err
);
1744 static int v9fs_walk_marshal(V9fsPDU
*pdu
, uint16_t nwnames
, V9fsQID
*qids
)
1750 err
= pdu_marshal(pdu
, offset
, "w", nwnames
);
1755 for (i
= 0; i
< nwnames
; i
++) {
1756 err
= pdu_marshal(pdu
, offset
, "Q", &qids
[i
]);
1765 static bool name_is_illegal(const char *name
)
1767 return !*name
|| strchr(name
, '/') != NULL
;
1770 static bool same_stat_id(const struct stat
*a
, const struct stat
*b
)
1772 return a
->st_dev
== b
->st_dev
&& a
->st_ino
== b
->st_ino
;
1775 static void coroutine_fn
v9fs_walk(void *opaque
)
1777 int name_idx
, nwalked
;
1778 g_autofree V9fsQID
*qids
= NULL
;
1779 int i
, err
= 0, any_err
= 0;
1780 V9fsPath dpath
, path
;
1781 P9ARRAY_REF(V9fsPath
) pathes
= NULL
;
1783 struct stat stbuf
, fidst
;
1784 g_autofree
struct stat
*stbufs
= NULL
;
1786 int32_t fid
, newfid
;
1787 P9ARRAY_REF(V9fsString
) wnames
= NULL
;
1789 V9fsFidState
*newfidp
= NULL
;
1790 V9fsPDU
*pdu
= opaque
;
1791 V9fsState
*s
= pdu
->s
;
1794 err
= pdu_unmarshal(pdu
, offset
, "ddw", &fid
, &newfid
, &nwnames
);
1796 pdu_complete(pdu
, err
);
1801 trace_v9fs_walk(pdu
->tag
, pdu
->id
, fid
, newfid
, nwnames
);
1803 if (nwnames
> P9_MAXWELEM
) {
1808 P9ARRAY_NEW(V9fsString
, wnames
, nwnames
);
1809 qids
= g_new0(V9fsQID
, nwnames
);
1810 stbufs
= g_new0(struct stat
, nwnames
);
1811 P9ARRAY_NEW(V9fsPath
, pathes
, nwnames
);
1812 for (i
= 0; i
< nwnames
; i
++) {
1813 err
= pdu_unmarshal(pdu
, offset
, "s", &wnames
[i
]);
1817 if (name_is_illegal(wnames
[i
].data
)) {
1824 fidp
= get_fid(pdu
, fid
);
1830 v9fs_path_init(&dpath
);
1831 v9fs_path_init(&path
);
1833 * Both dpath and path initially point to fidp.
1834 * Needed to handle request with nwnames == 0
1836 v9fs_path_copy(&dpath
, &fidp
->path
);
1837 v9fs_path_copy(&path
, &fidp
->path
);
1840 * To keep latency (i.e. overall execution time for processing this
1841 * Twalk client request) as small as possible, run all the required fs
1842 * driver code altogether inside the following block.
1844 v9fs_co_run_in_worker({
1846 if (v9fs_request_cancelled(pdu
)) {
1847 any_err
|= err
= -EINTR
;
1850 err
= s
->ops
->lstat(&s
->ctx
, &dpath
, &fidst
);
1852 any_err
|= err
= -errno
;
1856 for (; nwalked
< nwnames
; nwalked
++) {
1857 if (v9fs_request_cancelled(pdu
)) {
1858 any_err
|= err
= -EINTR
;
1861 if (!same_stat_id(&pdu
->s
->root_st
, &stbuf
) ||
1862 strcmp("..", wnames
[nwalked
].data
))
1864 err
= s
->ops
->name_to_path(&s
->ctx
, &dpath
,
1865 wnames
[nwalked
].data
,
1868 any_err
|= err
= -errno
;
1871 if (v9fs_request_cancelled(pdu
)) {
1872 any_err
|= err
= -EINTR
;
1875 err
= s
->ops
->lstat(&s
->ctx
, &pathes
[nwalked
], &stbuf
);
1877 any_err
|= err
= -errno
;
1880 stbufs
[nwalked
] = stbuf
;
1881 v9fs_path_copy(&dpath
, &pathes
[nwalked
]);
1886 * Handle all the rest of this Twalk request on main thread ...
1888 * NOTE: -EINTR is an exception where we deviate from the protocol spec
1889 * and simply send a (R)Lerror response instead of bothering to assemble
1890 * a (deducted) Rwalk response; because -EINTR is always the result of a
1891 * Tflush request, so client would no longer wait for a response in this
1894 if ((err
< 0 && !nwalked
) || err
== -EINTR
) {
1898 any_err
|= err
= stat_to_qid(pdu
, &fidst
, &qid
);
1899 if (err
< 0 && !nwalked
) {
1904 /* reset dpath and path */
1905 v9fs_path_copy(&dpath
, &fidp
->path
);
1906 v9fs_path_copy(&path
, &fidp
->path
);
1908 for (name_idx
= 0; name_idx
< nwalked
; name_idx
++) {
1909 if (!same_stat_id(&pdu
->s
->root_st
, &stbuf
) ||
1910 strcmp("..", wnames
[name_idx
].data
))
1912 stbuf
= stbufs
[name_idx
];
1913 any_err
|= err
= stat_to_qid(pdu
, &stbuf
, &qid
);
1917 v9fs_path_copy(&path
, &pathes
[name_idx
]);
1918 v9fs_path_copy(&dpath
, &path
);
1920 memcpy(&qids
[name_idx
], &qid
, sizeof(qid
));
1924 /* don't send any QIDs, send Rlerror instead */
1927 /* send QIDs (not Rlerror), but fid MUST remain unaffected */
1931 if (fid
== newfid
) {
1932 if (fidp
->fid_type
!= P9_FID_NONE
) {
1936 v9fs_path_write_lock(s
);
1937 v9fs_path_copy(&fidp
->path
, &path
);
1938 v9fs_path_unlock(s
);
1940 newfidp
= alloc_fid(s
, newfid
);
1941 if (newfidp
== NULL
) {
1945 newfidp
->uid
= fidp
->uid
;
1946 v9fs_path_copy(&newfidp
->path
, &path
);
1949 err
= v9fs_walk_marshal(pdu
, name_idx
, qids
);
1950 trace_v9fs_walk_return(pdu
->tag
, pdu
->id
, name_idx
, qids
);
1954 put_fid(pdu
, newfidp
);
1956 v9fs_path_free(&dpath
);
1957 v9fs_path_free(&path
);
1959 pdu_complete(pdu
, err
);
1962 static int32_t coroutine_fn
get_iounit(V9fsPDU
*pdu
, V9fsPath
*path
)
1964 struct statfs stbuf
;
1965 int err
= v9fs_co_statfs(pdu
, path
, &stbuf
);
1967 return blksize_to_iounit(pdu
, (err
>= 0) ? stbuf
.f_bsize
: 0);
1970 static void coroutine_fn
v9fs_open(void *opaque
)
1981 V9fsPDU
*pdu
= opaque
;
1982 V9fsState
*s
= pdu
->s
;
1984 if (s
->proto_version
== V9FS_PROTO_2000L
) {
1985 err
= pdu_unmarshal(pdu
, offset
, "dd", &fid
, &mode
);
1988 err
= pdu_unmarshal(pdu
, offset
, "db", &fid
, &modebyte
);
1994 trace_v9fs_open(pdu
->tag
, pdu
->id
, fid
, mode
);
1996 fidp
= get_fid(pdu
, fid
);
2001 if (fidp
->fid_type
!= P9_FID_NONE
) {
2006 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
2010 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
2014 if (S_ISDIR(stbuf
.st_mode
)) {
2015 err
= v9fs_co_opendir(pdu
, fidp
);
2019 fidp
->fid_type
= P9_FID_DIR
;
2020 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, 0);
2026 if (s
->proto_version
== V9FS_PROTO_2000L
) {
2027 flags
= get_dotl_openflags(s
, mode
);
2029 flags
= omode_to_uflags(mode
);
2031 if (is_ro_export(&s
->ctx
)) {
2032 if (mode
& O_WRONLY
|| mode
& O_RDWR
||
2033 mode
& O_APPEND
|| mode
& O_TRUNC
) {
2038 err
= v9fs_co_open(pdu
, fidp
, flags
);
2042 fidp
->fid_type
= P9_FID_FILE
;
2043 fidp
->open_flags
= flags
;
2044 if (flags
& O_EXCL
) {
2046 * We let the host file system do O_EXCL check
2047 * We should not reclaim such fd
2049 fidp
->flags
|= FID_NON_RECLAIMABLE
;
2051 iounit
= get_iounit(pdu
, &fidp
->path
);
2052 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
2058 trace_v9fs_open_return(pdu
->tag
, pdu
->id
,
2059 qid
.type
, qid
.version
, qid
.path
, iounit
);
2063 pdu_complete(pdu
, err
);
2066 static void coroutine_fn
v9fs_lcreate(void *opaque
)
2068 int32_t dfid
, flags
, mode
;
2077 V9fsPDU
*pdu
= opaque
;
2079 v9fs_string_init(&name
);
2080 err
= pdu_unmarshal(pdu
, offset
, "dsddd", &dfid
,
2081 &name
, &flags
, &mode
, &gid
);
2085 trace_v9fs_lcreate(pdu
->tag
, pdu
->id
, dfid
, flags
, mode
, gid
);
2087 if (name_is_illegal(name
.data
)) {
2092 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2097 fidp
= get_fid(pdu
, dfid
);
2102 if (fidp
->fid_type
!= P9_FID_NONE
) {
2107 flags
= get_dotl_openflags(pdu
->s
, flags
);
2108 err
= v9fs_co_open2(pdu
, fidp
, &name
, gid
,
2109 flags
| O_CREAT
, mode
, &stbuf
);
2113 fidp
->fid_type
= P9_FID_FILE
;
2114 fidp
->open_flags
= flags
;
2115 if (flags
& O_EXCL
) {
2117 * We let the host file system do O_EXCL check
2118 * We should not reclaim such fd
2120 fidp
->flags
|= FID_NON_RECLAIMABLE
;
2122 iounit
= get_iounit(pdu
, &fidp
->path
);
2123 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
2127 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
2132 trace_v9fs_lcreate_return(pdu
->tag
, pdu
->id
,
2133 qid
.type
, qid
.version
, qid
.path
, iounit
);
2137 pdu_complete(pdu
, err
);
2138 v9fs_string_free(&name
);
2141 static void coroutine_fn
v9fs_fsync(void *opaque
)
2148 V9fsPDU
*pdu
= opaque
;
2150 err
= pdu_unmarshal(pdu
, offset
, "dd", &fid
, &datasync
);
2154 trace_v9fs_fsync(pdu
->tag
, pdu
->id
, fid
, datasync
);
2156 fidp
= get_fid(pdu
, fid
);
2161 err
= v9fs_co_fsync(pdu
, fidp
, datasync
);
2167 pdu_complete(pdu
, err
);
2170 static void coroutine_fn
v9fs_clunk(void *opaque
)
2176 V9fsPDU
*pdu
= opaque
;
2177 V9fsState
*s
= pdu
->s
;
2179 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
2183 trace_v9fs_clunk(pdu
->tag
, pdu
->id
, fid
);
2185 fidp
= clunk_fid(s
, fid
);
2191 * Bump the ref so that put_fid will
2195 err
= put_fid(pdu
, fidp
);
2200 pdu_complete(pdu
, err
);
2204 * Create a QEMUIOVector for a sub-region of PDU iovecs
2206 * @qiov: uninitialized QEMUIOVector
2207 * @skip: number of bytes to skip from beginning of PDU
2208 * @size: number of bytes to include
2209 * @is_write: true - write, false - read
2211 * The resulting QEMUIOVector has heap-allocated iovecs and must be cleaned up
2212 * with qemu_iovec_destroy().
2214 static void v9fs_init_qiov_from_pdu(QEMUIOVector
*qiov
, V9fsPDU
*pdu
,
2215 size_t skip
, size_t size
,
2223 pdu
->s
->transport
->init_out_iov_from_pdu(pdu
, &iov
, &niov
, size
+ skip
);
2225 pdu
->s
->transport
->init_in_iov_from_pdu(pdu
, &iov
, &niov
, size
+ skip
);
2228 qemu_iovec_init_external(&elem
, iov
, niov
);
2229 qemu_iovec_init(qiov
, niov
);
2230 qemu_iovec_concat(qiov
, &elem
, skip
, size
);
2233 static int v9fs_xattr_read(V9fsState
*s
, V9fsPDU
*pdu
, V9fsFidState
*fidp
,
2234 uint64_t off
, uint32_t max_count
)
2238 uint64_t read_count
;
2239 QEMUIOVector qiov_full
;
2241 if (fidp
->fs
.xattr
.len
< off
) {
2244 read_count
= fidp
->fs
.xattr
.len
- off
;
2246 if (read_count
> max_count
) {
2247 read_count
= max_count
;
2249 err
= pdu_marshal(pdu
, offset
, "d", read_count
);
2255 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
, read_count
, false);
2256 err
= v9fs_pack(qiov_full
.iov
, qiov_full
.niov
, 0,
2257 ((char *)fidp
->fs
.xattr
.value
) + off
,
2259 qemu_iovec_destroy(&qiov_full
);
2267 static int coroutine_fn
v9fs_do_readdir_with_stat(V9fsPDU
*pdu
,
2276 off_t saved_dir_pos
;
2277 struct dirent
*dent
;
2279 /* save the directory position */
2280 saved_dir_pos
= v9fs_co_telldir(pdu
, fidp
);
2281 if (saved_dir_pos
< 0) {
2282 return saved_dir_pos
;
2286 v9fs_path_init(&path
);
2288 v9fs_readdir_lock(&fidp
->fs
.dir
);
2290 err
= v9fs_co_readdir(pdu
, fidp
, &dent
);
2294 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, dent
->d_name
, &path
);
2298 err
= v9fs_co_lstat(pdu
, &path
, &stbuf
);
2302 err
= stat_to_v9stat(pdu
, &path
, dent
->d_name
, &stbuf
, &v9stat
);
2306 if ((count
+ v9stat
.size
+ 2) > max_count
) {
2307 v9fs_readdir_unlock(&fidp
->fs
.dir
);
2309 /* Ran out of buffer. Set dir back to old position and return */
2310 v9fs_co_seekdir(pdu
, fidp
, saved_dir_pos
);
2311 v9fs_stat_free(&v9stat
);
2312 v9fs_path_free(&path
);
2316 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2317 len
= pdu_marshal(pdu
, 11 + count
, "S", &v9stat
);
2319 v9fs_readdir_unlock(&fidp
->fs
.dir
);
2322 v9fs_co_seekdir(pdu
, fidp
, saved_dir_pos
);
2323 v9fs_stat_free(&v9stat
);
2324 v9fs_path_free(&path
);
2328 v9fs_stat_free(&v9stat
);
2329 v9fs_path_free(&path
);
2330 saved_dir_pos
= qemu_dirent_off(dent
);
2333 v9fs_readdir_unlock(&fidp
->fs
.dir
);
2335 v9fs_path_free(&path
);
2342 static void coroutine_fn
v9fs_read(void *opaque
)
2351 V9fsPDU
*pdu
= opaque
;
2352 V9fsState
*s
= pdu
->s
;
2354 err
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
, &off
, &max_count
);
2358 trace_v9fs_read(pdu
->tag
, pdu
->id
, fid
, off
, max_count
);
2360 fidp
= get_fid(pdu
, fid
);
2365 if (fidp
->fid_type
== P9_FID_DIR
) {
2366 if (s
->proto_version
!= V9FS_PROTO_2000U
) {
2368 "9p: bad client: T_read request on directory only expected "
2369 "with 9P2000.u protocol version"
2375 v9fs_co_rewinddir(pdu
, fidp
);
2377 count
= v9fs_do_readdir_with_stat(pdu
, fidp
, max_count
);
2382 err
= pdu_marshal(pdu
, offset
, "d", count
);
2386 err
+= offset
+ count
;
2387 } else if (fidp
->fid_type
== P9_FID_FILE
) {
2388 QEMUIOVector qiov_full
;
2392 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
+ 4, max_count
, false);
2393 qemu_iovec_init(&qiov
, qiov_full
.niov
);
2395 qemu_iovec_reset(&qiov
);
2396 qemu_iovec_concat(&qiov
, &qiov_full
, count
, qiov_full
.size
- count
);
2398 print_sg(qiov
.iov
, qiov
.niov
);
2400 /* Loop in case of EINTR */
2402 len
= v9fs_co_preadv(pdu
, fidp
, qiov
.iov
, qiov
.niov
, off
);
2407 } while (len
== -EINTR
&& !pdu
->cancelled
);
2409 /* IO error return the error */
2411 goto out_free_iovec
;
2413 } while (count
< max_count
&& len
> 0);
2414 err
= pdu_marshal(pdu
, offset
, "d", count
);
2416 goto out_free_iovec
;
2418 err
+= offset
+ count
;
2420 qemu_iovec_destroy(&qiov
);
2421 qemu_iovec_destroy(&qiov_full
);
2422 } else if (fidp
->fid_type
== P9_FID_XATTR
) {
2423 err
= v9fs_xattr_read(s
, pdu
, fidp
, off
, max_count
);
2427 trace_v9fs_read_return(pdu
->tag
, pdu
->id
, count
, err
);
2431 pdu_complete(pdu
, err
);
2435 * v9fs_readdir_response_size() - Returns size required in Rreaddir response
2436 * for the passed dirent @name.
2438 * @name: directory entry's name (i.e. file name, directory name)
2439 * Return: required size in bytes
2441 size_t v9fs_readdir_response_size(V9fsString
*name
)
2444 * Size of each dirent on the wire: size of qid (13) + size of offset (8)
2445 * size of type (1) + size of name.size (2) + strlen(name.data)
2447 return 24 + v9fs_string_size(name
);
2450 static void v9fs_free_dirents(struct V9fsDirEnt
*e
)
2452 struct V9fsDirEnt
*next
= NULL
;
2454 for (; e
; e
= next
) {
2462 static int coroutine_fn
v9fs_do_readdir(V9fsPDU
*pdu
, V9fsFidState
*fidp
,
2463 off_t offset
, int32_t max_count
)
2471 struct dirent
*dent
;
2473 struct V9fsDirEnt
*entries
= NULL
;
2476 * inode remapping requires the device id, which in turn might be
2477 * different for different directory entries, so if inode remapping is
2478 * enabled we have to make a full stat for each directory entry
2480 const bool dostat
= pdu
->s
->ctx
.export_flags
& V9FS_REMAP_INODES
;
2483 * Fetch all required directory entries altogether on a background IO
2484 * thread from fs driver. We don't want to do that for each entry
2485 * individually, because hopping between threads (this main IO thread
2486 * and background IO driver thread) would sum up to huge latencies.
2488 count
= v9fs_co_readdir_many(pdu
, fidp
, &entries
, offset
, max_count
,
2497 for (struct V9fsDirEnt
*e
= entries
; e
; e
= e
->next
) {
2500 if (pdu
->s
->ctx
.export_flags
& V9FS_REMAP_INODES
) {
2502 /* e->st should never be NULL, but just to be sure */
2509 err
= stat_to_qid(pdu
, st
, &qid
);
2515 * Fill up just the path field of qid because the client uses
2516 * only that. To fill the entire qid structure we will have
2517 * to stat each dirent found, which is expensive. For the
2518 * latter reason we don't call stat_to_qid() here. Only drawback
2519 * is that no multi-device export detection of stat_to_qid()
2520 * would be done and provided as error to the user here. But
2521 * user would get that error anyway when accessing those
2522 * files/dirs through other ways.
2524 size
= MIN(sizeof(dent
->d_ino
), sizeof(qid
.path
));
2525 memcpy(&qid
.path
, &dent
->d_ino
, size
);
2526 /* Fill the other fields with dummy values */
2531 off
= qemu_dirent_off(dent
);
2532 v9fs_string_init(&name
);
2533 v9fs_string_sprintf(&name
, "%s", dent
->d_name
);
2535 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2536 len
= pdu_marshal(pdu
, 11 + count
, "Qqbs",
2538 dent
->d_type
, &name
);
2540 v9fs_string_free(&name
);
2551 v9fs_free_dirents(entries
);
2558 static void coroutine_fn
v9fs_readdir(void *opaque
)
2564 uint64_t initial_offset
;
2567 V9fsPDU
*pdu
= opaque
;
2568 V9fsState
*s
= pdu
->s
;
2570 retval
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
,
2571 &initial_offset
, &max_count
);
2575 trace_v9fs_readdir(pdu
->tag
, pdu
->id
, fid
, initial_offset
, max_count
);
2577 /* Enough space for a R_readdir header: size[4] Rreaddir tag[2] count[4] */
2578 if (max_count
> s
->msize
- 11) {
2579 max_count
= s
->msize
- 11;
2581 "9p: bad client: T_readdir with count > msize - 11"
2585 fidp
= get_fid(pdu
, fid
);
2590 if (!fidp
->fs
.dir
.stream
) {
2594 if (s
->proto_version
!= V9FS_PROTO_2000L
) {
2596 "9p: bad client: T_readdir request only expected with 9P2000.L "
2599 retval
= -EOPNOTSUPP
;
2602 count
= v9fs_do_readdir(pdu
, fidp
, (off_t
) initial_offset
, max_count
);
2607 retval
= pdu_marshal(pdu
, offset
, "d", count
);
2611 retval
+= count
+ offset
;
2612 trace_v9fs_readdir_return(pdu
->tag
, pdu
->id
, count
, retval
);
2616 pdu_complete(pdu
, retval
);
2619 static int v9fs_xattr_write(V9fsState
*s
, V9fsPDU
*pdu
, V9fsFidState
*fidp
,
2620 uint64_t off
, uint32_t count
,
2621 struct iovec
*sg
, int cnt
)
2625 uint64_t write_count
;
2629 if (fidp
->fs
.xattr
.len
< off
) {
2632 write_count
= fidp
->fs
.xattr
.len
- off
;
2633 if (write_count
> count
) {
2634 write_count
= count
;
2636 err
= pdu_marshal(pdu
, offset
, "d", write_count
);
2641 fidp
->fs
.xattr
.copied_len
+= write_count
;
2643 * Now copy the content from sg list
2645 for (i
= 0; i
< cnt
; i
++) {
2646 if (write_count
> sg
[i
].iov_len
) {
2647 to_copy
= sg
[i
].iov_len
;
2649 to_copy
= write_count
;
2651 memcpy((char *)fidp
->fs
.xattr
.value
+ off
, sg
[i
].iov_base
, to_copy
);
2652 /* updating vs->off since we are not using below */
2654 write_count
-= to_copy
;
2660 static void coroutine_fn
v9fs_write(void *opaque
)
2670 V9fsPDU
*pdu
= opaque
;
2671 V9fsState
*s
= pdu
->s
;
2672 QEMUIOVector qiov_full
;
2675 err
= pdu_unmarshal(pdu
, offset
, "dqd", &fid
, &off
, &count
);
2677 pdu_complete(pdu
, err
);
2681 v9fs_init_qiov_from_pdu(&qiov_full
, pdu
, offset
, count
, true);
2682 trace_v9fs_write(pdu
->tag
, pdu
->id
, fid
, off
, count
, qiov_full
.niov
);
2684 fidp
= get_fid(pdu
, fid
);
2689 if (fidp
->fid_type
== P9_FID_FILE
) {
2690 if (fidp
->fs
.fd
== -1) {
2694 } else if (fidp
->fid_type
== P9_FID_XATTR
) {
2696 * setxattr operation
2698 err
= v9fs_xattr_write(s
, pdu
, fidp
, off
, count
,
2699 qiov_full
.iov
, qiov_full
.niov
);
2705 qemu_iovec_init(&qiov
, qiov_full
.niov
);
2707 qemu_iovec_reset(&qiov
);
2708 qemu_iovec_concat(&qiov
, &qiov_full
, total
, qiov_full
.size
- total
);
2710 print_sg(qiov
.iov
, qiov
.niov
);
2712 /* Loop in case of EINTR */
2714 len
= v9fs_co_pwritev(pdu
, fidp
, qiov
.iov
, qiov
.niov
, off
);
2719 } while (len
== -EINTR
&& !pdu
->cancelled
);
2721 /* IO error return the error */
2725 } while (total
< count
&& len
> 0);
2728 err
= pdu_marshal(pdu
, offset
, "d", total
);
2733 trace_v9fs_write_return(pdu
->tag
, pdu
->id
, total
, err
);
2735 qemu_iovec_destroy(&qiov
);
2739 qemu_iovec_destroy(&qiov_full
);
2740 pdu_complete(pdu
, err
);
2743 static void coroutine_fn
v9fs_create(void *opaque
)
2755 V9fsString extension
;
2757 V9fsPDU
*pdu
= opaque
;
2758 V9fsState
*s
= pdu
->s
;
2760 v9fs_path_init(&path
);
2761 v9fs_string_init(&name
);
2762 v9fs_string_init(&extension
);
2763 err
= pdu_unmarshal(pdu
, offset
, "dsdbs", &fid
, &name
,
2764 &perm
, &mode
, &extension
);
2768 trace_v9fs_create(pdu
->tag
, pdu
->id
, fid
, name
.data
, perm
, mode
);
2770 if (name_is_illegal(name
.data
)) {
2775 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2780 fidp
= get_fid(pdu
, fid
);
2785 if (fidp
->fid_type
!= P9_FID_NONE
) {
2789 if (perm
& P9_STAT_MODE_DIR
) {
2790 err
= v9fs_co_mkdir(pdu
, fidp
, &name
, perm
& 0777,
2791 fidp
->uid
, -1, &stbuf
);
2795 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2799 v9fs_path_write_lock(s
);
2800 v9fs_path_copy(&fidp
->path
, &path
);
2801 v9fs_path_unlock(s
);
2802 err
= v9fs_co_opendir(pdu
, fidp
);
2806 fidp
->fid_type
= P9_FID_DIR
;
2807 } else if (perm
& P9_STAT_MODE_SYMLINK
) {
2808 err
= v9fs_co_symlink(pdu
, fidp
, &name
,
2809 extension
.data
, -1 , &stbuf
);
2813 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2817 v9fs_path_write_lock(s
);
2818 v9fs_path_copy(&fidp
->path
, &path
);
2819 v9fs_path_unlock(s
);
2820 } else if (perm
& P9_STAT_MODE_LINK
) {
2821 int32_t ofid
= atoi(extension
.data
);
2822 V9fsFidState
*ofidp
= get_fid(pdu
, ofid
);
2823 if (ofidp
== NULL
) {
2827 err
= v9fs_co_link(pdu
, ofidp
, fidp
, &name
);
2828 put_fid(pdu
, ofidp
);
2832 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2834 fidp
->fid_type
= P9_FID_NONE
;
2837 v9fs_path_write_lock(s
);
2838 v9fs_path_copy(&fidp
->path
, &path
);
2839 v9fs_path_unlock(s
);
2840 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
2842 fidp
->fid_type
= P9_FID_NONE
;
2845 } else if (perm
& P9_STAT_MODE_DEVICE
) {
2847 uint32_t major
, minor
;
2850 if (sscanf(extension
.data
, "%c %u %u", &ctype
, &major
, &minor
) != 3) {
2867 nmode
|= perm
& 0777;
2868 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2869 makedev(major
, minor
), nmode
, &stbuf
);
2873 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2877 v9fs_path_write_lock(s
);
2878 v9fs_path_copy(&fidp
->path
, &path
);
2879 v9fs_path_unlock(s
);
2880 } else if (perm
& P9_STAT_MODE_NAMED_PIPE
) {
2881 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2882 0, S_IFIFO
| (perm
& 0777), &stbuf
);
2886 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2890 v9fs_path_write_lock(s
);
2891 v9fs_path_copy(&fidp
->path
, &path
);
2892 v9fs_path_unlock(s
);
2893 } else if (perm
& P9_STAT_MODE_SOCKET
) {
2894 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, -1,
2895 0, S_IFSOCK
| (perm
& 0777), &stbuf
);
2899 err
= v9fs_co_name_to_path(pdu
, &fidp
->path
, name
.data
, &path
);
2903 v9fs_path_write_lock(s
);
2904 v9fs_path_copy(&fidp
->path
, &path
);
2905 v9fs_path_unlock(s
);
2907 err
= v9fs_co_open2(pdu
, fidp
, &name
, -1,
2908 omode_to_uflags(mode
) | O_CREAT
, perm
, &stbuf
);
2912 fidp
->fid_type
= P9_FID_FILE
;
2913 fidp
->open_flags
= omode_to_uflags(mode
);
2914 if (fidp
->open_flags
& O_EXCL
) {
2916 * We let the host file system do O_EXCL check
2917 * We should not reclaim such fd
2919 fidp
->flags
|= FID_NON_RECLAIMABLE
;
2922 iounit
= get_iounit(pdu
, &fidp
->path
);
2923 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
2927 err
= pdu_marshal(pdu
, offset
, "Qd", &qid
, iounit
);
2932 trace_v9fs_create_return(pdu
->tag
, pdu
->id
,
2933 qid
.type
, qid
.version
, qid
.path
, iounit
);
2937 pdu_complete(pdu
, err
);
2938 v9fs_string_free(&name
);
2939 v9fs_string_free(&extension
);
2940 v9fs_path_free(&path
);
2943 static void coroutine_fn
v9fs_symlink(void *opaque
)
2945 V9fsPDU
*pdu
= opaque
;
2948 V9fsFidState
*dfidp
;
2956 v9fs_string_init(&name
);
2957 v9fs_string_init(&symname
);
2958 err
= pdu_unmarshal(pdu
, offset
, "dssd", &dfid
, &name
, &symname
, &gid
);
2962 trace_v9fs_symlink(pdu
->tag
, pdu
->id
, dfid
, name
.data
, symname
.data
, gid
);
2964 if (name_is_illegal(name
.data
)) {
2969 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
2974 dfidp
= get_fid(pdu
, dfid
);
2975 if (dfidp
== NULL
) {
2979 err
= v9fs_co_symlink(pdu
, dfidp
, &name
, symname
.data
, gid
, &stbuf
);
2983 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
2987 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
2992 trace_v9fs_symlink_return(pdu
->tag
, pdu
->id
,
2993 qid
.type
, qid
.version
, qid
.path
);
2995 put_fid(pdu
, dfidp
);
2997 pdu_complete(pdu
, err
);
2998 v9fs_string_free(&name
);
2999 v9fs_string_free(&symname
);
3002 static void coroutine_fn
v9fs_flush(void *opaque
)
3007 V9fsPDU
*cancel_pdu
= NULL
;
3008 V9fsPDU
*pdu
= opaque
;
3009 V9fsState
*s
= pdu
->s
;
3011 err
= pdu_unmarshal(pdu
, offset
, "w", &tag
);
3013 pdu_complete(pdu
, err
);
3016 trace_v9fs_flush(pdu
->tag
, pdu
->id
, tag
);
3018 if (pdu
->tag
== tag
) {
3019 warn_report("the guest sent a self-referencing 9P flush request");
3021 QLIST_FOREACH(cancel_pdu
, &s
->active_list
, next
) {
3022 if (cancel_pdu
->tag
== tag
) {
3028 cancel_pdu
->cancelled
= 1;
3030 * Wait for pdu to complete.
3032 qemu_co_queue_wait(&cancel_pdu
->complete
, NULL
);
3033 if (!qemu_co_queue_next(&cancel_pdu
->complete
)) {
3034 cancel_pdu
->cancelled
= 0;
3035 pdu_free(cancel_pdu
);
3038 pdu_complete(pdu
, 7);
3041 static void coroutine_fn
v9fs_link(void *opaque
)
3043 V9fsPDU
*pdu
= opaque
;
3044 int32_t dfid
, oldfid
;
3045 V9fsFidState
*dfidp
, *oldfidp
;
3050 v9fs_string_init(&name
);
3051 err
= pdu_unmarshal(pdu
, offset
, "dds", &dfid
, &oldfid
, &name
);
3055 trace_v9fs_link(pdu
->tag
, pdu
->id
, dfid
, oldfid
, name
.data
);
3057 if (name_is_illegal(name
.data
)) {
3062 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3067 dfidp
= get_fid(pdu
, dfid
);
3068 if (dfidp
== NULL
) {
3073 oldfidp
= get_fid(pdu
, oldfid
);
3074 if (oldfidp
== NULL
) {
3078 err
= v9fs_co_link(pdu
, oldfidp
, dfidp
, &name
);
3082 put_fid(pdu
, oldfidp
);
3084 put_fid(pdu
, dfidp
);
3086 v9fs_string_free(&name
);
3087 pdu_complete(pdu
, err
);
3090 /* Only works with path name based fid */
3091 static void coroutine_fn
v9fs_remove(void *opaque
)
3097 V9fsPDU
*pdu
= opaque
;
3099 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
3103 trace_v9fs_remove(pdu
->tag
, pdu
->id
, fid
);
3105 fidp
= get_fid(pdu
, fid
);
3110 /* if fs driver is not path based, return EOPNOTSUPP */
3111 if (!(pdu
->s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
)) {
3116 * IF the file is unlinked, we cannot reopen
3117 * the file later. So don't reclaim fd
3119 err
= v9fs_mark_fids_unreclaim(pdu
, &fidp
->path
);
3123 err
= v9fs_co_remove(pdu
, &fidp
->path
);
3128 /* For TREMOVE we need to clunk the fid even on failed remove */
3129 clunk_fid(pdu
->s
, fidp
->fid
);
3132 pdu_complete(pdu
, err
);
3135 static void coroutine_fn
v9fs_unlinkat(void *opaque
)
3139 int32_t dfid
, flags
, rflags
= 0;
3142 V9fsFidState
*dfidp
;
3143 V9fsPDU
*pdu
= opaque
;
3145 v9fs_string_init(&name
);
3146 err
= pdu_unmarshal(pdu
, offset
, "dsd", &dfid
, &name
, &flags
);
3151 if (name_is_illegal(name
.data
)) {
3156 if (!strcmp(".", name
.data
)) {
3161 if (!strcmp("..", name
.data
)) {
3166 if (flags
& ~P9_DOTL_AT_REMOVEDIR
) {
3171 if (flags
& P9_DOTL_AT_REMOVEDIR
) {
3172 rflags
|= AT_REMOVEDIR
;
3175 dfidp
= get_fid(pdu
, dfid
);
3176 if (dfidp
== NULL
) {
3181 * IF the file is unlinked, we cannot reopen
3182 * the file later. So don't reclaim fd
3184 v9fs_path_init(&path
);
3185 err
= v9fs_co_name_to_path(pdu
, &dfidp
->path
, name
.data
, &path
);
3189 err
= v9fs_mark_fids_unreclaim(pdu
, &path
);
3193 err
= v9fs_co_unlinkat(pdu
, &dfidp
->path
, &name
, rflags
);
3198 put_fid(pdu
, dfidp
);
3199 v9fs_path_free(&path
);
3201 pdu_complete(pdu
, err
);
3202 v9fs_string_free(&name
);
3206 /* Only works with path name based fid */
3207 static int coroutine_fn
v9fs_complete_rename(V9fsPDU
*pdu
, V9fsFidState
*fidp
,
3213 V9fsFidState
*tfidp
;
3214 V9fsState
*s
= pdu
->s
;
3215 V9fsFidState
*dirfidp
= NULL
;
3216 GHashTableIter iter
;
3219 v9fs_path_init(&new_path
);
3220 if (newdirfid
!= -1) {
3221 dirfidp
= get_fid(pdu
, newdirfid
);
3222 if (dirfidp
== NULL
) {
3225 if (fidp
->fid_type
!= P9_FID_NONE
) {
3229 err
= v9fs_co_name_to_path(pdu
, &dirfidp
->path
, name
->data
, &new_path
);
3234 char *dir_name
= g_path_get_dirname(fidp
->path
.data
);
3237 v9fs_path_init(&dir_path
);
3238 v9fs_path_sprintf(&dir_path
, "%s", dir_name
);
3241 err
= v9fs_co_name_to_path(pdu
, &dir_path
, name
->data
, &new_path
);
3242 v9fs_path_free(&dir_path
);
3247 err
= v9fs_co_rename(pdu
, &fidp
->path
, &new_path
);
3253 * Fixup fid's pointing to the old name to
3254 * start pointing to the new name
3256 g_hash_table_iter_init(&iter
, s
->fids
);
3257 while (g_hash_table_iter_next(&iter
, &fid
, (gpointer
*) &tfidp
)) {
3258 if (v9fs_path_is_ancestor(&fidp
->path
, &tfidp
->path
)) {
3259 /* replace the name */
3260 v9fs_fix_path(&tfidp
->path
, &new_path
, strlen(fidp
->path
.data
));
3265 put_fid(pdu
, dirfidp
);
3267 v9fs_path_free(&new_path
);
3271 /* Only works with path name based fid */
3272 static void coroutine_fn
v9fs_rename(void *opaque
)
3280 V9fsPDU
*pdu
= opaque
;
3281 V9fsState
*s
= pdu
->s
;
3283 v9fs_string_init(&name
);
3284 err
= pdu_unmarshal(pdu
, offset
, "dds", &fid
, &newdirfid
, &name
);
3289 if (name_is_illegal(name
.data
)) {
3294 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3299 fidp
= get_fid(pdu
, fid
);
3304 if (fidp
->fid_type
!= P9_FID_NONE
) {
3308 /* if fs driver is not path based, return EOPNOTSUPP */
3309 if (!(pdu
->s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
)) {
3313 v9fs_path_write_lock(s
);
3314 err
= v9fs_complete_rename(pdu
, fidp
, newdirfid
, &name
);
3315 v9fs_path_unlock(s
);
3322 pdu_complete(pdu
, err
);
3323 v9fs_string_free(&name
);
3326 static int coroutine_fn
v9fs_fix_fid_paths(V9fsPDU
*pdu
, V9fsPath
*olddir
,
3327 V9fsString
*old_name
,
3329 V9fsString
*new_name
)
3331 V9fsFidState
*tfidp
;
3332 V9fsPath oldpath
, newpath
;
3333 V9fsState
*s
= pdu
->s
;
3335 GHashTableIter iter
;
3338 v9fs_path_init(&oldpath
);
3339 v9fs_path_init(&newpath
);
3340 err
= v9fs_co_name_to_path(pdu
, olddir
, old_name
->data
, &oldpath
);
3344 err
= v9fs_co_name_to_path(pdu
, newdir
, new_name
->data
, &newpath
);
3350 * Fixup fid's pointing to the old name to
3351 * start pointing to the new name
3353 g_hash_table_iter_init(&iter
, s
->fids
);
3354 while (g_hash_table_iter_next(&iter
, &fid
, (gpointer
*) &tfidp
)) {
3355 if (v9fs_path_is_ancestor(&oldpath
, &tfidp
->path
)) {
3356 /* replace the name */
3357 v9fs_fix_path(&tfidp
->path
, &newpath
, strlen(oldpath
.data
));
3361 v9fs_path_free(&oldpath
);
3362 v9fs_path_free(&newpath
);
3366 static int coroutine_fn
v9fs_complete_renameat(V9fsPDU
*pdu
, int32_t olddirfid
,
3367 V9fsString
*old_name
,
3369 V9fsString
*new_name
)
3372 V9fsState
*s
= pdu
->s
;
3373 V9fsFidState
*newdirfidp
= NULL
, *olddirfidp
= NULL
;
3375 olddirfidp
= get_fid(pdu
, olddirfid
);
3376 if (olddirfidp
== NULL
) {
3380 if (newdirfid
!= -1) {
3381 newdirfidp
= get_fid(pdu
, newdirfid
);
3382 if (newdirfidp
== NULL
) {
3387 newdirfidp
= get_fid(pdu
, olddirfid
);
3390 err
= v9fs_co_renameat(pdu
, &olddirfidp
->path
, old_name
,
3391 &newdirfidp
->path
, new_name
);
3395 if (s
->ctx
.export_flags
& V9FS_PATHNAME_FSCONTEXT
) {
3396 /* Only for path based fid we need to do the below fixup */
3397 err
= v9fs_fix_fid_paths(pdu
, &olddirfidp
->path
, old_name
,
3398 &newdirfidp
->path
, new_name
);
3402 put_fid(pdu
, olddirfidp
);
3405 put_fid(pdu
, newdirfidp
);
3410 static void coroutine_fn
v9fs_renameat(void *opaque
)
3414 V9fsPDU
*pdu
= opaque
;
3415 V9fsState
*s
= pdu
->s
;
3416 int32_t olddirfid
, newdirfid
;
3417 V9fsString old_name
, new_name
;
3419 v9fs_string_init(&old_name
);
3420 v9fs_string_init(&new_name
);
3421 err
= pdu_unmarshal(pdu
, offset
, "dsds", &olddirfid
,
3422 &old_name
, &newdirfid
, &new_name
);
3427 if (name_is_illegal(old_name
.data
) || name_is_illegal(new_name
.data
)) {
3432 if (!strcmp(".", old_name
.data
) || !strcmp("..", old_name
.data
) ||
3433 !strcmp(".", new_name
.data
) || !strcmp("..", new_name
.data
)) {
3438 v9fs_path_write_lock(s
);
3439 err
= v9fs_complete_renameat(pdu
, olddirfid
,
3440 &old_name
, newdirfid
, &new_name
);
3441 v9fs_path_unlock(s
);
3447 pdu_complete(pdu
, err
);
3448 v9fs_string_free(&old_name
);
3449 v9fs_string_free(&new_name
);
3452 static void coroutine_fn
v9fs_wstat(void *opaque
)
3461 V9fsPDU
*pdu
= opaque
;
3462 V9fsState
*s
= pdu
->s
;
3464 v9fs_stat_init(&v9stat
);
3465 err
= pdu_unmarshal(pdu
, offset
, "dwS", &fid
, &unused
, &v9stat
);
3469 trace_v9fs_wstat(pdu
->tag
, pdu
->id
, fid
,
3470 v9stat
.mode
, v9stat
.atime
, v9stat
.mtime
);
3472 fidp
= get_fid(pdu
, fid
);
3477 /* do we need to sync the file? */
3478 if (donttouch_stat(&v9stat
)) {
3479 err
= v9fs_co_fsync(pdu
, fidp
, 0);
3482 if (v9stat
.mode
!= -1) {
3484 err
= v9fs_co_lstat(pdu
, &fidp
->path
, &stbuf
);
3488 v9_mode
= stat_to_v9mode(&stbuf
);
3489 if ((v9stat
.mode
& P9_STAT_MODE_TYPE_BITS
) !=
3490 (v9_mode
& P9_STAT_MODE_TYPE_BITS
)) {
3491 /* Attempting to change the type */
3495 err
= v9fs_co_chmod(pdu
, &fidp
->path
,
3496 v9mode_to_mode(v9stat
.mode
,
3497 &v9stat
.extension
));
3502 if (v9stat
.mtime
!= -1 || v9stat
.atime
!= -1) {
3503 struct timespec times
[2];
3504 if (v9stat
.atime
!= -1) {
3505 times
[0].tv_sec
= v9stat
.atime
;
3506 times
[0].tv_nsec
= 0;
3508 times
[0].tv_nsec
= UTIME_OMIT
;
3510 if (v9stat
.mtime
!= -1) {
3511 times
[1].tv_sec
= v9stat
.mtime
;
3512 times
[1].tv_nsec
= 0;
3514 times
[1].tv_nsec
= UTIME_OMIT
;
3516 err
= v9fs_co_utimensat(pdu
, &fidp
->path
, times
);
3521 if (v9stat
.n_gid
!= -1 || v9stat
.n_uid
!= -1) {
3522 err
= v9fs_co_chown(pdu
, &fidp
->path
, v9stat
.n_uid
, v9stat
.n_gid
);
3527 if (v9stat
.name
.size
!= 0) {
3528 v9fs_path_write_lock(s
);
3529 err
= v9fs_complete_rename(pdu
, fidp
, -1, &v9stat
.name
);
3530 v9fs_path_unlock(s
);
3535 if (v9stat
.length
!= -1) {
3536 err
= v9fs_co_truncate(pdu
, &fidp
->path
, v9stat
.length
);
3545 v9fs_stat_free(&v9stat
);
3546 pdu_complete(pdu
, err
);
3549 static int v9fs_fill_statfs(V9fsState
*s
, V9fsPDU
*pdu
, struct statfs
*stbuf
)
3561 int32_t bsize_factor
;
3564 * compute bsize factor based on host file system block size
3567 bsize_factor
= (s
->msize
- P9_IOHDRSZ
) / stbuf
->f_bsize
;
3568 if (!bsize_factor
) {
3571 f_type
= stbuf
->f_type
;
3572 f_bsize
= stbuf
->f_bsize
;
3573 f_bsize
*= bsize_factor
;
3575 * f_bsize is adjusted(multiplied) by bsize factor, so we need to
3576 * adjust(divide) the number of blocks, free blocks and available
3577 * blocks by bsize factor
3579 f_blocks
= stbuf
->f_blocks
/ bsize_factor
;
3580 f_bfree
= stbuf
->f_bfree
/ bsize_factor
;
3581 f_bavail
= stbuf
->f_bavail
/ bsize_factor
;
3582 f_files
= stbuf
->f_files
;
3583 f_ffree
= stbuf
->f_ffree
;
3584 #ifdef CONFIG_DARWIN
3585 fsid_val
= (unsigned int)stbuf
->f_fsid
.val
[0] |
3586 (unsigned long long)stbuf
->f_fsid
.val
[1] << 32;
3587 f_namelen
= NAME_MAX
;
3589 fsid_val
= (unsigned int) stbuf
->f_fsid
.__val
[0] |
3590 (unsigned long long)stbuf
->f_fsid
.__val
[1] << 32;
3591 f_namelen
= stbuf
->f_namelen
;
3594 return pdu_marshal(pdu
, offset
, "ddqqqqqqd",
3595 f_type
, f_bsize
, f_blocks
, f_bfree
,
3596 f_bavail
, f_files
, f_ffree
,
3597 fsid_val
, f_namelen
);
3600 static void coroutine_fn
v9fs_statfs(void *opaque
)
3606 struct statfs stbuf
;
3607 V9fsPDU
*pdu
= opaque
;
3608 V9fsState
*s
= pdu
->s
;
3610 retval
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
3614 fidp
= get_fid(pdu
, fid
);
3619 retval
= v9fs_co_statfs(pdu
, &fidp
->path
, &stbuf
);
3623 retval
= v9fs_fill_statfs(s
, pdu
, &stbuf
);
3631 pdu_complete(pdu
, retval
);
3634 static void coroutine_fn
v9fs_mknod(void *opaque
)
3647 V9fsPDU
*pdu
= opaque
;
3649 v9fs_string_init(&name
);
3650 err
= pdu_unmarshal(pdu
, offset
, "dsdddd", &fid
, &name
, &mode
,
3651 &major
, &minor
, &gid
);
3655 trace_v9fs_mknod(pdu
->tag
, pdu
->id
, fid
, mode
, major
, minor
);
3657 if (name_is_illegal(name
.data
)) {
3662 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3667 fidp
= get_fid(pdu
, fid
);
3672 err
= v9fs_co_mknod(pdu
, fidp
, &name
, fidp
->uid
, gid
,
3673 makedev(major
, minor
), mode
, &stbuf
);
3677 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
3681 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
3686 trace_v9fs_mknod_return(pdu
->tag
, pdu
->id
,
3687 qid
.type
, qid
.version
, qid
.path
);
3691 pdu_complete(pdu
, err
);
3692 v9fs_string_free(&name
);
3696 * Implement posix byte range locking code
3697 * Server side handling of locking code is very simple, because 9p server in
3698 * QEMU can handle only one client. And most of the lock handling
3699 * (like conflict, merging) etc is done by the VFS layer itself, so no need to
3700 * do any thing in * qemu 9p server side lock code path.
3701 * So when a TLOCK request comes, always return success
3703 static void coroutine_fn
v9fs_lock(void *opaque
)
3709 int32_t fid
, err
= 0;
3710 V9fsPDU
*pdu
= opaque
;
3712 v9fs_string_init(&flock
.client_id
);
3713 err
= pdu_unmarshal(pdu
, offset
, "dbdqqds", &fid
, &flock
.type
,
3714 &flock
.flags
, &flock
.start
, &flock
.length
,
3715 &flock
.proc_id
, &flock
.client_id
);
3719 trace_v9fs_lock(pdu
->tag
, pdu
->id
, fid
,
3720 flock
.type
, flock
.start
, flock
.length
);
3723 /* We support only block flag now (that too ignored currently) */
3724 if (flock
.flags
& ~P9_LOCK_FLAGS_BLOCK
) {
3728 fidp
= get_fid(pdu
, fid
);
3733 err
= v9fs_co_fstat(pdu
, fidp
, &stbuf
);
3737 err
= pdu_marshal(pdu
, offset
, "b", P9_LOCK_SUCCESS
);
3742 trace_v9fs_lock_return(pdu
->tag
, pdu
->id
, P9_LOCK_SUCCESS
);
3746 pdu_complete(pdu
, err
);
3747 v9fs_string_free(&flock
.client_id
);
3751 * When a TGETLOCK request comes, always return success because all lock
3752 * handling is done by client's VFS layer.
3754 static void coroutine_fn
v9fs_getlock(void *opaque
)
3760 int32_t fid
, err
= 0;
3761 V9fsPDU
*pdu
= opaque
;
3763 v9fs_string_init(&glock
.client_id
);
3764 err
= pdu_unmarshal(pdu
, offset
, "dbqqds", &fid
, &glock
.type
,
3765 &glock
.start
, &glock
.length
, &glock
.proc_id
,
3770 trace_v9fs_getlock(pdu
->tag
, pdu
->id
, fid
,
3771 glock
.type
, glock
.start
, glock
.length
);
3773 fidp
= get_fid(pdu
, fid
);
3778 err
= v9fs_co_fstat(pdu
, fidp
, &stbuf
);
3782 glock
.type
= P9_LOCK_TYPE_UNLCK
;
3783 err
= pdu_marshal(pdu
, offset
, "bqqds", glock
.type
,
3784 glock
.start
, glock
.length
, glock
.proc_id
,
3790 trace_v9fs_getlock_return(pdu
->tag
, pdu
->id
, glock
.type
, glock
.start
,
3791 glock
.length
, glock
.proc_id
);
3795 pdu_complete(pdu
, err
);
3796 v9fs_string_free(&glock
.client_id
);
3799 static void coroutine_fn
v9fs_mkdir(void *opaque
)
3801 V9fsPDU
*pdu
= opaque
;
3812 v9fs_string_init(&name
);
3813 err
= pdu_unmarshal(pdu
, offset
, "dsdd", &fid
, &name
, &mode
, &gid
);
3817 trace_v9fs_mkdir(pdu
->tag
, pdu
->id
, fid
, name
.data
, mode
, gid
);
3819 if (name_is_illegal(name
.data
)) {
3824 if (!strcmp(".", name
.data
) || !strcmp("..", name
.data
)) {
3829 fidp
= get_fid(pdu
, fid
);
3834 err
= v9fs_co_mkdir(pdu
, fidp
, &name
, mode
, fidp
->uid
, gid
, &stbuf
);
3838 err
= stat_to_qid(pdu
, &stbuf
, &qid
);
3842 err
= pdu_marshal(pdu
, offset
, "Q", &qid
);
3847 trace_v9fs_mkdir_return(pdu
->tag
, pdu
->id
,
3848 qid
.type
, qid
.version
, qid
.path
, err
);
3852 pdu_complete(pdu
, err
);
3853 v9fs_string_free(&name
);
3856 static void coroutine_fn
v9fs_xattrwalk(void *opaque
)
3862 int32_t fid
, newfid
;
3863 V9fsFidState
*file_fidp
;
3864 V9fsFidState
*xattr_fidp
= NULL
;
3865 V9fsPDU
*pdu
= opaque
;
3866 V9fsState
*s
= pdu
->s
;
3868 v9fs_string_init(&name
);
3869 err
= pdu_unmarshal(pdu
, offset
, "dds", &fid
, &newfid
, &name
);
3873 trace_v9fs_xattrwalk(pdu
->tag
, pdu
->id
, fid
, newfid
, name
.data
);
3875 file_fidp
= get_fid(pdu
, fid
);
3876 if (file_fidp
== NULL
) {
3880 xattr_fidp
= alloc_fid(s
, newfid
);
3881 if (xattr_fidp
== NULL
) {
3885 v9fs_path_copy(&xattr_fidp
->path
, &file_fidp
->path
);
3886 if (!v9fs_string_size(&name
)) {
3888 * listxattr request. Get the size first
3890 size
= v9fs_co_llistxattr(pdu
, &xattr_fidp
->path
, NULL
, 0);
3893 clunk_fid(s
, xattr_fidp
->fid
);
3897 * Read the xattr value
3899 xattr_fidp
->fs
.xattr
.len
= size
;
3900 xattr_fidp
->fid_type
= P9_FID_XATTR
;
3901 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= true;
3902 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
3904 err
= v9fs_co_llistxattr(pdu
, &xattr_fidp
->path
,
3905 xattr_fidp
->fs
.xattr
.value
,
3906 xattr_fidp
->fs
.xattr
.len
);
3908 clunk_fid(s
, xattr_fidp
->fid
);
3912 err
= pdu_marshal(pdu
, offset
, "q", size
);
3919 * specific xattr fid. We check for xattr
3920 * presence also collect the xattr size
3922 size
= v9fs_co_lgetxattr(pdu
, &xattr_fidp
->path
,
3926 clunk_fid(s
, xattr_fidp
->fid
);
3930 * Read the xattr value
3932 xattr_fidp
->fs
.xattr
.len
= size
;
3933 xattr_fidp
->fid_type
= P9_FID_XATTR
;
3934 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= true;
3935 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
3937 err
= v9fs_co_lgetxattr(pdu
, &xattr_fidp
->path
,
3938 &name
, xattr_fidp
->fs
.xattr
.value
,
3939 xattr_fidp
->fs
.xattr
.len
);
3941 clunk_fid(s
, xattr_fidp
->fid
);
3945 err
= pdu_marshal(pdu
, offset
, "q", size
);
3951 trace_v9fs_xattrwalk_return(pdu
->tag
, pdu
->id
, size
);
3953 put_fid(pdu
, file_fidp
);
3955 put_fid(pdu
, xattr_fidp
);
3958 pdu_complete(pdu
, err
);
3959 v9fs_string_free(&name
);
3962 #if defined(CONFIG_LINUX)
3963 /* Currently, only Linux has XATTR_SIZE_MAX */
3964 #define P9_XATTR_SIZE_MAX XATTR_SIZE_MAX
3965 #elif defined(CONFIG_DARWIN)
3967 * Darwin doesn't seem to define a maximum xattr size in its user
3968 * space header, so manually configure it across platforms as 64k.
3970 * Having no limit at all can lead to QEMU crashing during large g_malloc()
3971 * calls. Because QEMU does not currently support macOS guests, the below
3972 * preliminary solution only works due to its being a reflection of the limit of
3975 #define P9_XATTR_SIZE_MAX 65536
3977 #error Missing definition for P9_XATTR_SIZE_MAX for this host system
3980 static void coroutine_fn
v9fs_xattrcreate(void *opaque
)
3982 int flags
, rflags
= 0;
3988 V9fsFidState
*file_fidp
;
3989 V9fsFidState
*xattr_fidp
;
3990 V9fsPDU
*pdu
= opaque
;
3992 v9fs_string_init(&name
);
3993 err
= pdu_unmarshal(pdu
, offset
, "dsqd", &fid
, &name
, &size
, &flags
);
3997 trace_v9fs_xattrcreate(pdu
->tag
, pdu
->id
, fid
, name
.data
, size
, flags
);
3999 if (flags
& ~(P9_XATTR_CREATE
| P9_XATTR_REPLACE
)) {
4004 if (flags
& P9_XATTR_CREATE
) {
4005 rflags
|= XATTR_CREATE
;
4008 if (flags
& P9_XATTR_REPLACE
) {
4009 rflags
|= XATTR_REPLACE
;
4012 if (size
> P9_XATTR_SIZE_MAX
) {
4017 file_fidp
= get_fid(pdu
, fid
);
4018 if (file_fidp
== NULL
) {
4022 if (file_fidp
->fid_type
!= P9_FID_NONE
) {
4027 /* Make the file fid point to xattr */
4028 xattr_fidp
= file_fidp
;
4029 xattr_fidp
->fid_type
= P9_FID_XATTR
;
4030 xattr_fidp
->fs
.xattr
.copied_len
= 0;
4031 xattr_fidp
->fs
.xattr
.xattrwalk_fid
= false;
4032 xattr_fidp
->fs
.xattr
.len
= size
;
4033 xattr_fidp
->fs
.xattr
.flags
= rflags
;
4034 v9fs_string_init(&xattr_fidp
->fs
.xattr
.name
);
4035 v9fs_string_copy(&xattr_fidp
->fs
.xattr
.name
, &name
);
4036 xattr_fidp
->fs
.xattr
.value
= g_malloc0(size
);
4039 put_fid(pdu
, file_fidp
);
4041 pdu_complete(pdu
, err
);
4042 v9fs_string_free(&name
);
4045 static void coroutine_fn
v9fs_readlink(void *opaque
)
4047 V9fsPDU
*pdu
= opaque
;
4054 err
= pdu_unmarshal(pdu
, offset
, "d", &fid
);
4058 trace_v9fs_readlink(pdu
->tag
, pdu
->id
, fid
);
4059 fidp
= get_fid(pdu
, fid
);
4065 v9fs_string_init(&target
);
4066 err
= v9fs_co_readlink(pdu
, &fidp
->path
, &target
);
4070 err
= pdu_marshal(pdu
, offset
, "s", &target
);
4072 v9fs_string_free(&target
);
4076 trace_v9fs_readlink_return(pdu
->tag
, pdu
->id
, target
.data
);
4077 v9fs_string_free(&target
);
4081 pdu_complete(pdu
, err
);
4084 static CoroutineEntry
*pdu_co_handlers
[] = {
4085 [P9_TREADDIR
] = v9fs_readdir
,
4086 [P9_TSTATFS
] = v9fs_statfs
,
4087 [P9_TGETATTR
] = v9fs_getattr
,
4088 [P9_TSETATTR
] = v9fs_setattr
,
4089 [P9_TXATTRWALK
] = v9fs_xattrwalk
,
4090 [P9_TXATTRCREATE
] = v9fs_xattrcreate
,
4091 [P9_TMKNOD
] = v9fs_mknod
,
4092 [P9_TRENAME
] = v9fs_rename
,
4093 [P9_TLOCK
] = v9fs_lock
,
4094 [P9_TGETLOCK
] = v9fs_getlock
,
4095 [P9_TRENAMEAT
] = v9fs_renameat
,
4096 [P9_TREADLINK
] = v9fs_readlink
,
4097 [P9_TUNLINKAT
] = v9fs_unlinkat
,
4098 [P9_TMKDIR
] = v9fs_mkdir
,
4099 [P9_TVERSION
] = v9fs_version
,
4100 [P9_TLOPEN
] = v9fs_open
,
4101 [P9_TATTACH
] = v9fs_attach
,
4102 [P9_TSTAT
] = v9fs_stat
,
4103 [P9_TWALK
] = v9fs_walk
,
4104 [P9_TCLUNK
] = v9fs_clunk
,
4105 [P9_TFSYNC
] = v9fs_fsync
,
4106 [P9_TOPEN
] = v9fs_open
,
4107 [P9_TREAD
] = v9fs_read
,
4109 [P9_TAUTH
] = v9fs_auth
,
4111 [P9_TFLUSH
] = v9fs_flush
,
4112 [P9_TLINK
] = v9fs_link
,
4113 [P9_TSYMLINK
] = v9fs_symlink
,
4114 [P9_TCREATE
] = v9fs_create
,
4115 [P9_TLCREATE
] = v9fs_lcreate
,
4116 [P9_TWRITE
] = v9fs_write
,
4117 [P9_TWSTAT
] = v9fs_wstat
,
4118 [P9_TREMOVE
] = v9fs_remove
,
4121 static void coroutine_fn
v9fs_op_not_supp(void *opaque
)
4123 V9fsPDU
*pdu
= opaque
;
4124 pdu_complete(pdu
, -EOPNOTSUPP
);
4127 static void coroutine_fn
v9fs_fs_ro(void *opaque
)
4129 V9fsPDU
*pdu
= opaque
;
4130 pdu_complete(pdu
, -EROFS
);
4133 static inline bool is_read_only_op(V9fsPDU
*pdu
)
4160 void pdu_submit(V9fsPDU
*pdu
, P9MsgHeader
*hdr
)
4163 CoroutineEntry
*handler
;
4164 V9fsState
*s
= pdu
->s
;
4166 pdu
->size
= le32_to_cpu(hdr
->size_le
);
4168 pdu
->tag
= le16_to_cpu(hdr
->tag_le
);
4170 if (pdu
->id
>= ARRAY_SIZE(pdu_co_handlers
) ||
4171 (pdu_co_handlers
[pdu
->id
] == NULL
)) {
4172 handler
= v9fs_op_not_supp
;
4173 } else if (is_ro_export(&s
->ctx
) && !is_read_only_op(pdu
)) {
4174 handler
= v9fs_fs_ro
;
4176 handler
= pdu_co_handlers
[pdu
->id
];
4179 qemu_co_queue_init(&pdu
->complete
);
4180 co
= qemu_coroutine_create(handler
, pdu
);
4181 qemu_coroutine_enter(co
);
4184 /* Returns 0 on success, 1 on failure. */
4185 int v9fs_device_realize_common(V9fsState
*s
, const V9fsTransport
*t
,
4195 assert(!s
->transport
);
4198 /* initialize pdu allocator */
4199 QLIST_INIT(&s
->free_list
);
4200 QLIST_INIT(&s
->active_list
);
4201 for (i
= 0; i
< MAX_REQ
; i
++) {
4202 QLIST_INSERT_HEAD(&s
->free_list
, &s
->pdus
[i
], next
);
4207 v9fs_path_init(&path
);
4209 fse
= get_fsdev_fsentry(s
->fsconf
.fsdev_id
);
4212 /* We don't have a fsdev identified by fsdev_id */
4213 error_setg(errp
, "9pfs device couldn't find fsdev with the "
4215 s
->fsconf
.fsdev_id
? s
->fsconf
.fsdev_id
: "NULL");
4219 if (!s
->fsconf
.tag
) {
4220 /* we haven't specified a mount_tag */
4221 error_setg(errp
, "fsdev with id %s needs mount_tag arguments",
4222 s
->fsconf
.fsdev_id
);
4226 s
->ctx
.export_flags
= fse
->export_flags
;
4227 s
->ctx
.fs_root
= g_strdup(fse
->path
);
4228 s
->ctx
.exops
.get_st_gen
= NULL
;
4229 len
= strlen(s
->fsconf
.tag
);
4230 if (len
> MAX_TAG_LEN
- 1) {
4231 error_setg(errp
, "mount tag '%s' (%d bytes) is longer than "
4232 "maximum (%d bytes)", s
->fsconf
.tag
, len
, MAX_TAG_LEN
- 1);
4236 s
->tag
= g_strdup(s
->fsconf
.tag
);
4241 s
->ctx
.fmode
= fse
->fmode
;
4242 s
->ctx
.dmode
= fse
->dmode
;
4244 s
->fids
= g_hash_table_new(NULL
, NULL
);
4245 qemu_co_rwlock_init(&s
->rename_lock
);
4247 if (s
->ops
->init(&s
->ctx
, errp
) < 0) {
4248 error_prepend(errp
, "cannot initialize fsdev '%s': ",
4249 s
->fsconf
.fsdev_id
);
4254 * Check details of export path, We need to use fs driver
4255 * call back to do that. Since we are in the init path, we don't
4256 * use co-routines here.
4258 if (s
->ops
->name_to_path(&s
->ctx
, NULL
, "/", &path
) < 0) {
4260 "error in converting name to path %s", strerror(errno
));
4263 if (s
->ops
->lstat(&s
->ctx
, &path
, &stat
)) {
4264 error_setg(errp
, "share path %s does not exist", fse
->path
);
4266 } else if (!S_ISDIR(stat
.st_mode
)) {
4267 error_setg(errp
, "share path %s is not a directory", fse
->path
);
4271 s
->dev_id
= stat
.st_dev
;
4273 /* init inode remapping : */
4274 /* hash table for variable length inode suffixes */
4275 qpd_table_init(&s
->qpd_table
);
4276 /* hash table for slow/full inode remapping (most users won't need it) */
4277 qpf_table_init(&s
->qpf_table
);
4278 /* hash table for quick inode remapping */
4279 qpp_table_init(&s
->qpp_table
);
4281 s
->qp_affix_next
= 1; /* reserve 0 to detect overflow */
4282 s
->qp_fullpath_next
= 1;
4284 s
->ctx
.fst
= &fse
->fst
;
4285 fsdev_throttle_init(s
->ctx
.fst
);
4290 v9fs_device_unrealize_common(s
);
4292 v9fs_path_free(&path
);
4296 void v9fs_device_unrealize_common(V9fsState
*s
)
4298 if (s
->ops
&& s
->ops
->cleanup
) {
4299 s
->ops
->cleanup(&s
->ctx
);
4302 fsdev_throttle_cleanup(s
->ctx
.fst
);
4305 g_hash_table_destroy(s
->fids
);
4309 qp_table_destroy(&s
->qpd_table
);
4310 qp_table_destroy(&s
->qpp_table
);
4311 qp_table_destroy(&s
->qpf_table
);
4312 g_free(s
->ctx
.fs_root
);
4315 typedef struct VirtfsCoResetData
{
4318 } VirtfsCoResetData
;
4320 static void coroutine_fn
virtfs_co_reset(void *opaque
)
4322 VirtfsCoResetData
*data
= opaque
;
4324 virtfs_reset(&data
->pdu
);
4328 void v9fs_reset(V9fsState
*s
)
4330 VirtfsCoResetData data
= { .pdu
= { .s
= s
}, .done
= false };
4333 while (!QLIST_EMPTY(&s
->active_list
)) {
4334 aio_poll(qemu_get_aio_context(), true);
4337 co
= qemu_coroutine_create(virtfs_co_reset
, &data
);
4338 qemu_coroutine_enter(co
);
4340 while (!data
.done
) {
4341 aio_poll(qemu_get_aio_context(), true);
4345 static void __attribute__((__constructor__
)) v9fs_set_fd_limit(void)
4348 if (getrlimit(RLIMIT_NOFILE
, &rlim
) < 0) {
4349 error_report("Failed to get the resource limit");
4352 open_fd_hw
= rlim
.rlim_cur
- MIN(400, rlim
.rlim_cur
/ 3);
4353 open_fd_rc
= rlim
.rlim_cur
/ 2;