4 * Copyright IBM, Corp. 2010
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
14 #include "qemu/osdep.h"
15 #include <glib/gprintf.h>
16 #include "hw/virtio/virtio.h"
17 #include "qapi/error.h"
18 #include "qemu/error-report.h"
20 #include "qemu/main-loop.h"
21 #include "qemu/sockets.h"
22 #include "virtio-9p.h"
23 #include "fsdev/qemu-fsdev.h"
27 #include "migration/blocker.h"
28 #include "sysemu/qtest.h"
29 #include "qemu/xxhash.h"
31 #include <linux/limits.h>
35 static int open_fd_rc;
49 static ssize_t pdu_marshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
55 ret = pdu->s->transport->pdu_vmarshal(pdu, offset, fmt, ap);
61 static ssize_t pdu_unmarshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
67 ret = pdu->s->transport->pdu_vunmarshal(pdu, offset, fmt, ap);
73 static int omode_to_uflags(int8_t mode)
107 typedef struct DotlOpenflagMap {
112 static int dotl_to_open_flags(int flags)
116 * We have same bits for P9_DOTL_READONLY, P9_DOTL_WRONLY
117 * and P9_DOTL_NOACCESS
119 int oflags = flags & O_ACCMODE;
121 DotlOpenflagMap dotl_oflag_map[] = {
122 { P9_DOTL_CREATE, O_CREAT },
123 { P9_DOTL_EXCL, O_EXCL },
124 { P9_DOTL_NOCTTY , O_NOCTTY },
125 { P9_DOTL_TRUNC, O_TRUNC },
126 { P9_DOTL_APPEND, O_APPEND },
127 { P9_DOTL_NONBLOCK, O_NONBLOCK } ,
128 { P9_DOTL_DSYNC, O_DSYNC },
129 { P9_DOTL_FASYNC, FASYNC },
130 { P9_DOTL_DIRECT, O_DIRECT },
131 { P9_DOTL_LARGEFILE, O_LARGEFILE },
132 { P9_DOTL_DIRECTORY, O_DIRECTORY },
133 { P9_DOTL_NOFOLLOW, O_NOFOLLOW },
134 { P9_DOTL_NOATIME, O_NOATIME },
135 { P9_DOTL_SYNC, O_SYNC },
138 for (i = 0; i < ARRAY_SIZE(dotl_oflag_map); i++) {
139 if (flags & dotl_oflag_map[i].dotl_flag) {
140 oflags |= dotl_oflag_map[i].open_flag;
147 void cred_init(FsCred *credp)
155 static int get_dotl_openflags(V9fsState *s, int oflags)
159 * Filter the client open flags
161 flags = dotl_to_open_flags(oflags);
162 flags &= ~(O_NOCTTY | O_ASYNC | O_CREAT);
164 * Ignore direct disk access hint until the server supports it.
170 void v9fs_path_init(V9fsPath *path)
176 void v9fs_path_free(V9fsPath *path)
184 void GCC_FMT_ATTR(2, 3)
185 v9fs_path_sprintf(V9fsPath *path, const char *fmt, ...)
189 v9fs_path_free(path);
192 /* Bump the size for including terminating NULL */
193 path->size = g_vasprintf(&path->data, fmt, ap) + 1;
197 void v9fs_path_copy(V9fsPath *dst, const V9fsPath *src)
200 dst->size = src->size;
201 dst->data = g_memdup(src->data, src->size);
204 int v9fs_name_to_path(V9fsState *s, V9fsPath *dirpath,
205 const char *name, V9fsPath *path)
208 err = s->ops->name_to_path(&s->ctx, dirpath, name, path);
216 * Return TRUE if s1 is an ancestor of s2.
218 * E.g. "a/b" is an ancestor of "a/b/c" but not of "a/bc/d".
219 * As a special case, We treat s1 as ancestor of s2 if they are same!
221 static int v9fs_path_is_ancestor(V9fsPath *s1, V9fsPath *s2)
223 if (!strncmp(s1->data, s2->data, s1->size - 1)) {
224 if (s2->data[s1->size - 1] == '\0' || s2->data[s1->size - 1] == '/') {
231 static size_t v9fs_string_size(V9fsString *str)
237 * returns 0 if fid got re-opened, 1 if not, < 0 on error */
238 static int coroutine_fn v9fs_reopen_fid(V9fsPDU *pdu, V9fsFidState *f)
241 if (f->fid_type == P9_FID_FILE) {
242 if (f->fs.fd == -1) {
244 err = v9fs_co_open(pdu, f, f->open_flags);
245 } while (err == -EINTR && !pdu->cancelled);
247 } else if (f->fid_type == P9_FID_DIR) {
248 if (f->fs.dir.stream == NULL) {
250 err = v9fs_co_opendir(pdu, f);
251 } while (err == -EINTR && !pdu->cancelled);
257 static V9fsFidState *coroutine_fn get_fid(V9fsPDU *pdu, int32_t fid)
261 V9fsState *s = pdu->s;
263 for (f = s->fid_list; f; f = f->next) {
267 * Update the fid ref upfront so that
268 * we don't get reclaimed when we yield
273 * check whether we need to reopen the
274 * file. We might have closed the fd
275 * while trying to free up some file
278 err = v9fs_reopen_fid(pdu, f);
284 * Mark the fid as referenced so that the LRU
285 * reclaim won't close the file descriptor
287 f->flags |= FID_REFERENCED;
294 static V9fsFidState *alloc_fid(V9fsState *s, int32_t fid)
298 for (f = s->fid_list; f; f = f->next) {
299 /* If fid is already there return NULL */
305 f = g_malloc0(sizeof(V9fsFidState));
307 f->fid_type = P9_FID_NONE;
310 * Mark the fid as referenced so that the LRU
311 * reclaim won't close the file descriptor
313 f->flags |= FID_REFERENCED;
314 f->next = s->fid_list;
317 v9fs_readdir_init(s->proto_version, &f->fs.dir);
318 v9fs_readdir_init(s->proto_version, &f->fs_reclaim.dir);
323 static int coroutine_fn v9fs_xattr_fid_clunk(V9fsPDU *pdu, V9fsFidState *fidp)
327 if (fidp->fs.xattr.xattrwalk_fid) {
328 /* getxattr/listxattr fid */
332 * if this is fid for setxattr. clunk should
333 * result in setxattr localcall
335 if (fidp->fs.xattr.len != fidp->fs.xattr.copied_len) {
336 /* clunk after partial write */
340 if (fidp->fs.xattr.len) {
341 retval = v9fs_co_lsetxattr(pdu, &fidp->path, &fidp->fs.xattr.name,
342 fidp->fs.xattr.value,
344 fidp->fs.xattr.flags);
346 retval = v9fs_co_lremovexattr(pdu, &fidp->path, &fidp->fs.xattr.name);
349 v9fs_string_free(&fidp->fs.xattr.name);
351 g_free(fidp->fs.xattr.value);
355 static int coroutine_fn free_fid(V9fsPDU *pdu, V9fsFidState *fidp)
359 if (fidp->fid_type == P9_FID_FILE) {
360 /* If we reclaimed the fd no need to close */
361 if (fidp->fs.fd != -1) {
362 retval = v9fs_co_close(pdu, &fidp->fs);
364 } else if (fidp->fid_type == P9_FID_DIR) {
365 if (fidp->fs.dir.stream != NULL) {
366 retval = v9fs_co_closedir(pdu, &fidp->fs);
368 } else if (fidp->fid_type == P9_FID_XATTR) {
369 retval = v9fs_xattr_fid_clunk(pdu, fidp);
371 v9fs_path_free(&fidp->path);
376 static int coroutine_fn put_fid(V9fsPDU *pdu, V9fsFidState *fidp)
381 * Don't free the fid if it is in reclaim list
383 if (!fidp->ref && fidp->clunked) {
384 if (fidp->fid == pdu->s->root_fid) {
386 * if the clunked fid is root fid then we
387 * have unmounted the fs on the client side.
388 * delete the migration blocker. Ideally, this
389 * should be hooked to transport close notification
391 if (pdu->s->migration_blocker) {
392 migrate_del_blocker(pdu->s->migration_blocker);
393 error_free(pdu->s->migration_blocker);
394 pdu->s->migration_blocker = NULL;
397 return free_fid(pdu, fidp);
402 static V9fsFidState *clunk_fid(V9fsState *s, int32_t fid)
404 V9fsFidState **fidpp, *fidp;
406 for (fidpp = &s->fid_list; *fidpp; fidpp = &(*fidpp)->next) {
407 if ((*fidpp)->fid == fid) {
411 if (*fidpp == NULL) {
420 void coroutine_fn v9fs_reclaim_fd(V9fsPDU *pdu)
422 int reclaim_count = 0;
423 V9fsState *s = pdu->s;
424 V9fsFidState *f, *reclaim_list = NULL;
426 for (f = s->fid_list; f; f = f->next) {
428 * Unlink fids cannot be reclaimed. Check
429 * for them and skip them. Also skip fids
430 * currently being operated on.
432 if (f->ref || f->flags & FID_NON_RECLAIMABLE) {
436 * if it is a recently referenced fid
437 * we leave the fid untouched and clear the
438 * reference bit. We come back to it later
439 * in the next iteration. (a simple LRU without
440 * moving list elements around)
442 if (f->flags & FID_REFERENCED) {
443 f->flags &= ~FID_REFERENCED;
447 * Add fids to reclaim list.
449 if (f->fid_type == P9_FID_FILE) {
450 if (f->fs.fd != -1) {
452 * Up the reference count so that
453 * a clunk request won't free this fid
456 f->rclm_lst = reclaim_list;
458 f->fs_reclaim.fd = f->fs.fd;
462 } else if (f->fid_type == P9_FID_DIR) {
463 if (f->fs.dir.stream != NULL) {
465 * Up the reference count so that
466 * a clunk request won't free this fid
469 f->rclm_lst = reclaim_list;
471 f->fs_reclaim.dir.stream = f->fs.dir.stream;
472 f->fs.dir.stream = NULL;
476 if (reclaim_count >= open_fd_rc) {
481 * Now close the fid in reclaim list. Free them if they
482 * are already clunked.
484 while (reclaim_list) {
486 reclaim_list = f->rclm_lst;
487 if (f->fid_type == P9_FID_FILE) {
488 v9fs_co_close(pdu, &f->fs_reclaim);
489 } else if (f->fid_type == P9_FID_DIR) {
490 v9fs_co_closedir(pdu, &f->fs_reclaim);
494 * Now drop the fid reference, free it
501 static int coroutine_fn v9fs_mark_fids_unreclaim(V9fsPDU *pdu, V9fsPath *path)
504 V9fsState *s = pdu->s;
505 V9fsFidState *fidp, head_fid;
507 head_fid.next = s->fid_list;
508 for (fidp = s->fid_list; fidp; fidp = fidp->next) {
509 if (fidp->path.size != path->size) {
512 if (!memcmp(fidp->path.data, path->data, path->size)) {
513 /* Mark the fid non reclaimable. */
514 fidp->flags |= FID_NON_RECLAIMABLE;
516 /* reopen the file/dir if already closed */
517 err = v9fs_reopen_fid(pdu, fidp);
522 * Go back to head of fid list because
523 * the list could have got updated when
524 * switched to the worker thread
534 static void coroutine_fn virtfs_reset(V9fsPDU *pdu)
536 V9fsState *s = pdu->s;
540 while (s->fid_list) {
546 s->fid_list = fidp->next;
553 #define P9_QID_TYPE_DIR 0x80
554 #define P9_QID_TYPE_SYMLINK 0x02
556 #define P9_STAT_MODE_DIR 0x80000000
557 #define P9_STAT_MODE_APPEND 0x40000000
558 #define P9_STAT_MODE_EXCL 0x20000000
559 #define P9_STAT_MODE_MOUNT 0x10000000
560 #define P9_STAT_MODE_AUTH 0x08000000
561 #define P9_STAT_MODE_TMP 0x04000000
562 #define P9_STAT_MODE_SYMLINK 0x02000000
563 #define P9_STAT_MODE_LINK 0x01000000
564 #define P9_STAT_MODE_DEVICE 0x00800000
565 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
566 #define P9_STAT_MODE_SOCKET 0x00100000
567 #define P9_STAT_MODE_SETUID 0x00080000
568 #define P9_STAT_MODE_SETGID 0x00040000
569 #define P9_STAT_MODE_SETVTX 0x00010000
571 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR | \
572 P9_STAT_MODE_SYMLINK | \
573 P9_STAT_MODE_LINK | \
574 P9_STAT_MODE_DEVICE | \
575 P9_STAT_MODE_NAMED_PIPE | \
578 /* Mirrors all bits of a byte. So e.g. binary 10100000 would become 00000101. */
579 static inline uint8_t mirror8bit(uint8_t byte)
581 return (byte * 0x0202020202ULL & 0x010884422010ULL) % 1023;
584 /* Same as mirror8bit() just for a 64 bit data type instead for a byte. */
585 static inline uint64_t mirror64bit(uint64_t value)
587 return ((uint64_t)mirror8bit(value & 0xff) << 56) |
588 ((uint64_t)mirror8bit((value >> 8) & 0xff) << 48) |
589 ((uint64_t)mirror8bit((value >> 16) & 0xff) << 40) |
590 ((uint64_t)mirror8bit((value >> 24) & 0xff) << 32) |
591 ((uint64_t)mirror8bit((value >> 32) & 0xff) << 24) |
592 ((uint64_t)mirror8bit((value >> 40) & 0xff) << 16) |
593 ((uint64_t)mirror8bit((value >> 48) & 0xff) << 8) |
594 ((uint64_t)mirror8bit((value >> 56) & 0xff));
598 * @brief Parameter k for the Exponential Golomb algorihm to be used.
600 * The smaller this value, the smaller the minimum bit count for the Exp.
601 * Golomb generated affixes will be (at lowest index) however for the
602 * price of having higher maximum bit count of generated affixes (at highest
603 * index). Likewise increasing this parameter yields in smaller maximum bit
604 * count for the price of having higher minimum bit count.
606 * In practice that means: a good value for k depends on the expected amount
607 * of devices to be exposed by one export. For a small amount of devices k
608 * should be small, for a large amount of devices k might be increased
609 * instead. The default of k=0 should be fine for most users though.
611 * @b IMPORTANT: In case this ever becomes a runtime parameter; the value of
612 * k should not change as long as guest is still running! Because that would
613 * cause completely different inode numbers to be generated on guest.
615 #define EXP_GOLOMB_K 0
618 * @brief Exponential Golomb algorithm for arbitrary k (including k=0).
620 * The Exponential Golomb algorithm generates @b prefixes (@b not suffixes!)
621 * with growing length and with the mathematical property of being
622 * "prefix-free". The latter means the generated prefixes can be prepended
623 * in front of arbitrary numbers and the resulting concatenated numbers are
624 * guaranteed to be always unique.
626 * This is a minor adjustment to the original Exp. Golomb algorithm in the
627 * sense that lowest allowed index (@param n) starts with 1, not with zero.
629 * @param n - natural number (or index) of the prefix to be generated
631 * @param k - parameter k of Exp. Golomb algorithm to be used
632 * (see comment on EXP_GOLOMB_K macro for details about k)
634 static VariLenAffix expGolombEncode(uint64_t n, int k)
636 const uint64_t value = n + (1 << k) - 1;
637 const int bits = (int) log2(value) + 1;
638 return (VariLenAffix) {
639 .type = AffixType_Prefix,
641 .bits = bits + MAX((bits - 1 - k), 0)
646 * @brief Converts a suffix into a prefix, or a prefix into a suffix.
648 * Simply mirror all bits of the affix value, for the purpose to preserve
649 * respectively the mathematical "prefix-free" or "suffix-free" property
650 * after the conversion.
652 * If a passed prefix is suitable to create unique numbers, then the
653 * returned suffix is suitable to create unique numbers as well (and vice
656 static VariLenAffix invertAffix(const VariLenAffix *affix)
658 return (VariLenAffix) {
660 (affix->type == AffixType_Suffix) ?
661 AffixType_Prefix : AffixType_Suffix,
663 mirror64bit(affix->value) >>
664 ((sizeof(affix->value) * 8) - affix->bits),
670 * @brief Generates suffix numbers with "suffix-free" property.
672 * This is just a wrapper function on top of the Exp. Golomb algorithm.
674 * Since the Exp. Golomb algorithm generates prefixes, but we need suffixes,
675 * this function converts the Exp. Golomb prefixes into appropriate suffixes
676 * which are still suitable for generating unique numbers.
678 * @param n - natural number (or index) of the suffix to be generated
681 static VariLenAffix affixForIndex(uint64_t index)
684 prefix = expGolombEncode(index, EXP_GOLOMB_K);
685 return invertAffix(&prefix); /* convert prefix to suffix */
688 /* creative abuse of tb_hash_func7, which is based on xxhash */
689 static uint32_t qpp_hash(QppEntry e)
691 return qemu_xxhash7(e.ino_prefix, e.dev, 0, 0, 0);
694 static uint32_t qpf_hash(QpfEntry e)
696 return qemu_xxhash7(e.ino, e.dev, 0, 0, 0);
699 static bool qpd_cmp_func(const void *obj, const void *userp)
701 const QpdEntry *e1 = obj, *e2 = userp;
702 return e1->dev == e2->dev;
705 static bool qpp_cmp_func(const void *obj, const void *userp)
707 const QppEntry *e1 = obj, *e2 = userp;
708 return e1->dev == e2->dev && e1->ino_prefix == e2->ino_prefix;
711 static bool qpf_cmp_func(const void *obj, const void *userp)
713 const QpfEntry *e1 = obj, *e2 = userp;
714 return e1->dev == e2->dev && e1->ino == e2->ino;
717 static void qp_table_remove(void *p, uint32_t h, void *up)
722 static void qp_table_destroy(struct qht *ht)
724 if (!ht || !ht->map) {
727 qht_iter(ht, qp_table_remove, NULL);
731 static void qpd_table_init(struct qht *ht)
733 qht_init(ht, qpd_cmp_func, 1, QHT_MODE_AUTO_RESIZE);
736 static void qpp_table_init(struct qht *ht)
738 qht_init(ht, qpp_cmp_func, 1, QHT_MODE_AUTO_RESIZE);
741 static void qpf_table_init(struct qht *ht)
743 qht_init(ht, qpf_cmp_func, 1 << 16, QHT_MODE_AUTO_RESIZE);
747 * Returns how many (high end) bits of inode numbers of the passed fs
748 * device shall be used (in combination with the device number) to
749 * generate hash values for qpp_table entries.
751 * This function is required if variable length suffixes are used for inode
752 * number mapping on guest level. Since a device may end up having multiple
753 * entries in qpp_table, each entry most probably with a different suffix
754 * length, we thus need this function in conjunction with qpd_table to
755 * "agree" about a fix amount of bits (per device) to be always used for
756 * generating hash values for the purpose of accessing qpp_table in order
757 * get consistent behaviour when accessing qpp_table.
759 static int qid_inode_prefix_hash_bits(V9fsPDU *pdu, dev_t dev)
767 val = qht_lookup(&pdu->s->qpd_table, &lookup, hash);
769 val = g_malloc0(sizeof(QpdEntry));
771 affix = affixForIndex(pdu->s->qp_affix_next);
772 val->prefix_bits = affix.bits;
773 qht_insert(&pdu->s->qpd_table, val, hash, NULL);
774 pdu->s->qp_ndevices++;
776 return val->prefix_bits;
780 * @brief Slow / full mapping host inode nr -> guest inode nr.
782 * This function performs a slower and much more costly remapping of an
783 * original file inode number on host to an appropriate different inode
784 * number on guest. For every (dev, inode) combination on host a new
785 * sequential number is generated, cached and exposed as inode number on
788 * This is just a "last resort" fallback solution if the much faster/cheaper
789 * qid_path_suffixmap() failed. In practice this slow / full mapping is not
790 * expected ever to be used at all though.
792 * @see qid_path_suffixmap() for details
795 static int qid_path_fullmap(V9fsPDU *pdu, const struct stat *stbuf,
799 .dev = stbuf->st_dev,
802 uint32_t hash = qpf_hash(lookup);
805 val = qht_lookup(&pdu->s->qpf_table, &lookup, hash);
808 if (pdu->s->qp_fullpath_next == 0) {
809 /* no more files can be mapped :'( */
811 "9p: No more prefixes available for remapping inodes from "
817 val = g_malloc0(sizeof(QppEntry));
820 /* new unique inode and device combo */
821 affix = affixForIndex(
822 1ULL << (sizeof(pdu->s->qp_affix_next) * 8)
824 val->path = (pdu->s->qp_fullpath_next++ << affix.bits) | affix.value;
825 pdu->s->qp_fullpath_next &= ((1ULL << (64 - affix.bits)) - 1);
826 qht_insert(&pdu->s->qpf_table, val, hash, NULL);
834 * @brief Quick mapping host inode nr -> guest inode nr.
836 * This function performs quick remapping of an original file inode number
837 * on host to an appropriate different inode number on guest. This remapping
838 * of inodes is required to avoid inode nr collisions on guest which would
839 * happen if the 9p export contains more than 1 exported file system (or
840 * more than 1 file system data set), because unlike on host level where the
841 * files would have different device nrs, all files exported by 9p would
842 * share the same device nr on guest (the device nr of the virtual 9p device
845 * Inode remapping is performed by chopping off high end bits of the original
846 * inode number from host, shifting the result upwards and then assigning a
847 * generated suffix number for the low end bits, where the same suffix number
848 * will be shared by all inodes with the same device id AND the same high end
849 * bits that have been chopped off. That approach utilizes the fact that inode
850 * numbers very likely share the same high end bits (i.e. due to their common
851 * sequential generation by file systems) and hence we only have to generate
852 * and track a very limited amount of suffixes in practice due to that.
854 * We generate variable size suffixes for that purpose. The 1st generated
855 * suffix will only have 1 bit and hence we only need to chop off 1 bit from
856 * the original inode number. The subsequent suffixes being generated will
857 * grow in (bit) size subsequently, i.e. the 2nd and 3rd suffix being
858 * generated will have 3 bits and hence we have to chop off 3 bits from their
859 * original inodes, and so on. That approach of using variable length suffixes
860 * (i.e. over fixed size ones) utilizes the fact that in practice only a very
861 * limited amount of devices are shared by the same export (e.g. typically
862 * less than 2 dozen devices per 9p export), so in practice we need to chop
863 * off less bits than with fixed size prefixes and yet are flexible to add
864 * new devices at runtime below host's export directory at any time without
865 * having to reboot guest nor requiring to reconfigure guest for that. And due
866 * to the very limited amount of original high end bits that we chop off that
867 * way, the total amount of suffixes we need to generate is less than by using
868 * fixed size prefixes and hence it also improves performance of the inode
869 * remapping algorithm, and finally has the nice side effect that the inode
870 * numbers on guest will be much smaller & human friendly. ;-)
872 static int qid_path_suffixmap(V9fsPDU *pdu, const struct stat *stbuf,
875 const int ino_hash_bits = qid_inode_prefix_hash_bits(pdu, stbuf->st_dev);
877 .dev = stbuf->st_dev,
878 .ino_prefix = (uint16_t) (stbuf->st_ino >> (64 - ino_hash_bits))
880 uint32_t hash = qpp_hash(lookup);
882 val = qht_lookup(&pdu->s->qpp_table, &lookup, hash);
885 if (pdu->s->qp_affix_next == 0) {
886 /* we ran out of affixes */
888 "9p: Potential degraded performance of inode remapping"
893 val = g_malloc0(sizeof(QppEntry));
896 /* new unique inode affix and device combo */
897 val->qp_affix_index = pdu->s->qp_affix_next++;
898 val->qp_affix = affixForIndex(val->qp_affix_index);
899 qht_insert(&pdu->s->qpp_table, val, hash, NULL);
901 /* assuming generated affix to be suffix type, not prefix */
902 *path = (stbuf->st_ino << val->qp_affix.bits) | val->qp_affix.value;
906 static int stat_to_qid(V9fsPDU *pdu, const struct stat *stbuf, V9fsQID *qidp)
911 if (pdu->s->ctx.export_flags & V9FS_REMAP_INODES) {
912 /* map inode+device to qid path (fast path) */
913 err = qid_path_suffixmap(pdu, stbuf, &qidp->path);
914 if (err == -ENFILE) {
915 /* fast path didn't work, fall back to full map */
916 err = qid_path_fullmap(pdu, stbuf, &qidp->path);
922 if (pdu->s->dev_id != stbuf->st_dev) {
923 if (pdu->s->ctx.export_flags & V9FS_FORBID_MULTIDEVS) {
925 "9p: Multiple devices detected in same VirtFS export. "
926 "Access of guest to additional devices is (partly) "
927 "denied due to virtfs option 'multidevs=forbid' being "
933 "9p: Multiple devices detected in same VirtFS export, "
934 "which might lead to file ID collisions and severe "
935 "misbehaviours on guest! You should either use a "
936 "separate export for each device shared from host or "
937 "use virtfs option 'multidevs=remap'!"
941 memset(&qidp->path, 0, sizeof(qidp->path));
942 size = MIN(sizeof(stbuf->st_ino), sizeof(qidp->path));
943 memcpy(&qidp->path, &stbuf->st_ino, size);
946 qidp->version = stbuf->st_mtime ^ (stbuf->st_size << 8);
948 if (S_ISDIR(stbuf->st_mode)) {
949 qidp->type |= P9_QID_TYPE_DIR;
951 if (S_ISLNK(stbuf->st_mode)) {
952 qidp->type |= P9_QID_TYPE_SYMLINK;
958 static int coroutine_fn fid_to_qid(V9fsPDU *pdu, V9fsFidState *fidp,
964 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
968 err = stat_to_qid(pdu, &stbuf, qidp);
975 V9fsPDU *pdu_alloc(V9fsState *s)
979 if (!QLIST_EMPTY(&s->free_list)) {
980 pdu = QLIST_FIRST(&s->free_list);
981 QLIST_REMOVE(pdu, next);
982 QLIST_INSERT_HEAD(&s->active_list, pdu, next);
987 void pdu_free(V9fsPDU *pdu)
989 V9fsState *s = pdu->s;
991 g_assert(!pdu->cancelled);
992 QLIST_REMOVE(pdu, next);
993 QLIST_INSERT_HEAD(&s->free_list, pdu, next);
996 static void coroutine_fn pdu_complete(V9fsPDU *pdu, ssize_t len)
998 int8_t id = pdu->id + 1; /* Response */
999 V9fsState *s = pdu->s;
1003 * The 9p spec requires that successfully cancelled pdus receive no reply.
1004 * Sending a reply would confuse clients because they would
1005 * assume that any EINTR is the actual result of the operation,
1006 * rather than a consequence of the cancellation. However, if
1007 * the operation completed (succesfully or with an error other
1008 * than caused be cancellation), we do send out that reply, both
1009 * for efficiency and to avoid confusing the rest of the state machine
1010 * that assumes passing a non-error here will mean a successful
1011 * transmission of the reply.
1013 bool discard = pdu->cancelled && len == -EINTR;
1015 trace_v9fs_rcancel(pdu->tag, pdu->id);
1024 if (s->proto_version != V9FS_PROTO_2000L) {
1027 str.data = strerror(err);
1028 str.size = strlen(str.data);
1030 ret = pdu_marshal(pdu, len, "s", &str);
1038 ret = pdu_marshal(pdu, len, "d", err);
1044 if (s->proto_version == V9FS_PROTO_2000L) {
1047 trace_v9fs_rerror(pdu->tag, pdu->id, err); /* Trace ERROR */
1050 /* fill out the header */
1051 if (pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag) < 0) {
1055 /* keep these in sync */
1060 pdu->s->transport->push_and_notify(pdu);
1062 /* Now wakeup anybody waiting in flush for this request */
1063 if (!qemu_co_queue_next(&pdu->complete)) {
1068 static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
1073 if (mode & P9_STAT_MODE_DIR) {
1077 if (mode & P9_STAT_MODE_SYMLINK) {
1080 if (mode & P9_STAT_MODE_SOCKET) {
1083 if (mode & P9_STAT_MODE_NAMED_PIPE) {
1086 if (mode & P9_STAT_MODE_DEVICE) {
1087 if (extension->size && extension->data[0] == 'c') {
1098 if (mode & P9_STAT_MODE_SETUID) {
1101 if (mode & P9_STAT_MODE_SETGID) {
1104 if (mode & P9_STAT_MODE_SETVTX) {
1111 static int donttouch_stat(V9fsStat *stat)
1113 if (stat->type == -1 &&
1115 stat->qid.type == 0xff &&
1116 stat->qid.version == (uint32_t) -1 &&
1117 stat->qid.path == (uint64_t) -1 &&
1119 stat->atime == -1 &&
1120 stat->mtime == -1 &&
1121 stat->length == -1 &&
1126 stat->n_uid == -1 &&
1127 stat->n_gid == -1 &&
1128 stat->n_muid == -1) {
1135 static void v9fs_stat_init(V9fsStat *stat)
1137 v9fs_string_init(&stat->name);
1138 v9fs_string_init(&stat->uid);
1139 v9fs_string_init(&stat->gid);
1140 v9fs_string_init(&stat->muid);
1141 v9fs_string_init(&stat->extension);
1144 static void v9fs_stat_free(V9fsStat *stat)
1146 v9fs_string_free(&stat->name);
1147 v9fs_string_free(&stat->uid);
1148 v9fs_string_free(&stat->gid);
1149 v9fs_string_free(&stat->muid);
1150 v9fs_string_free(&stat->extension);
1153 static uint32_t stat_to_v9mode(const struct stat *stbuf)
1157 mode = stbuf->st_mode & 0777;
1158 if (S_ISDIR(stbuf->st_mode)) {
1159 mode |= P9_STAT_MODE_DIR;
1162 if (S_ISLNK(stbuf->st_mode)) {
1163 mode |= P9_STAT_MODE_SYMLINK;
1166 if (S_ISSOCK(stbuf->st_mode)) {
1167 mode |= P9_STAT_MODE_SOCKET;
1170 if (S_ISFIFO(stbuf->st_mode)) {
1171 mode |= P9_STAT_MODE_NAMED_PIPE;
1174 if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
1175 mode |= P9_STAT_MODE_DEVICE;
1178 if (stbuf->st_mode & S_ISUID) {
1179 mode |= P9_STAT_MODE_SETUID;
1182 if (stbuf->st_mode & S_ISGID) {
1183 mode |= P9_STAT_MODE_SETGID;
1186 if (stbuf->st_mode & S_ISVTX) {
1187 mode |= P9_STAT_MODE_SETVTX;
1193 static int coroutine_fn stat_to_v9stat(V9fsPDU *pdu, V9fsPath *path,
1194 const char *basename,
1195 const struct stat *stbuf,
1200 memset(v9stat, 0, sizeof(*v9stat));
1202 err = stat_to_qid(pdu, stbuf, &v9stat->qid);
1206 v9stat->mode = stat_to_v9mode(stbuf);
1207 v9stat->atime = stbuf->st_atime;
1208 v9stat->mtime = stbuf->st_mtime;
1209 v9stat->length = stbuf->st_size;
1211 v9fs_string_free(&v9stat->uid);
1212 v9fs_string_free(&v9stat->gid);
1213 v9fs_string_free(&v9stat->muid);
1215 v9stat->n_uid = stbuf->st_uid;
1216 v9stat->n_gid = stbuf->st_gid;
1219 v9fs_string_free(&v9stat->extension);
1221 if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
1222 err = v9fs_co_readlink(pdu, path, &v9stat->extension);
1226 } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
1227 v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
1228 S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
1229 major(stbuf->st_rdev), minor(stbuf->st_rdev));
1230 } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
1231 v9fs_string_sprintf(&v9stat->extension, "%s %lu",
1232 "HARDLINKCOUNT", (unsigned long)stbuf->st_nlink);
1235 v9fs_string_sprintf(&v9stat->name, "%s", basename);
1238 v9fs_string_size(&v9stat->name) +
1239 v9fs_string_size(&v9stat->uid) +
1240 v9fs_string_size(&v9stat->gid) +
1241 v9fs_string_size(&v9stat->muid) +
1242 v9fs_string_size(&v9stat->extension);
1246 #define P9_STATS_MODE 0x00000001ULL
1247 #define P9_STATS_NLINK 0x00000002ULL
1248 #define P9_STATS_UID 0x00000004ULL
1249 #define P9_STATS_GID 0x00000008ULL
1250 #define P9_STATS_RDEV 0x00000010ULL
1251 #define P9_STATS_ATIME 0x00000020ULL
1252 #define P9_STATS_MTIME 0x00000040ULL
1253 #define P9_STATS_CTIME 0x00000080ULL
1254 #define P9_STATS_INO 0x00000100ULL
1255 #define P9_STATS_SIZE 0x00000200ULL
1256 #define P9_STATS_BLOCKS 0x00000400ULL
1258 #define P9_STATS_BTIME 0x00000800ULL
1259 #define P9_STATS_GEN 0x00001000ULL
1260 #define P9_STATS_DATA_VERSION 0x00002000ULL
1262 #define P9_STATS_BASIC 0x000007ffULL /* Mask for fields up to BLOCKS */
1263 #define P9_STATS_ALL 0x00003fffULL /* Mask for All fields above */
1266 static int stat_to_v9stat_dotl(V9fsPDU *pdu, const struct stat *stbuf,
1267 V9fsStatDotl *v9lstat)
1269 memset(v9lstat, 0, sizeof(*v9lstat));
1271 v9lstat->st_mode = stbuf->st_mode;
1272 v9lstat->st_nlink = stbuf->st_nlink;
1273 v9lstat->st_uid = stbuf->st_uid;
1274 v9lstat->st_gid = stbuf->st_gid;
1275 v9lstat->st_rdev = stbuf->st_rdev;
1276 v9lstat->st_size = stbuf->st_size;
1277 v9lstat->st_blksize = stbuf->st_blksize;
1278 v9lstat->st_blocks = stbuf->st_blocks;
1279 v9lstat->st_atime_sec = stbuf->st_atime;
1280 v9lstat->st_atime_nsec = stbuf->st_atim.tv_nsec;
1281 v9lstat->st_mtime_sec = stbuf->st_mtime;
1282 v9lstat->st_mtime_nsec = stbuf->st_mtim.tv_nsec;
1283 v9lstat->st_ctime_sec = stbuf->st_ctime;
1284 v9lstat->st_ctime_nsec = stbuf->st_ctim.tv_nsec;
1285 /* Currently we only support BASIC fields in stat */
1286 v9lstat->st_result_mask = P9_STATS_BASIC;
1288 return stat_to_qid(pdu, stbuf, &v9lstat->qid);
1291 static void print_sg(struct iovec *sg, int cnt)
1295 printf("sg[%d]: {", cnt);
1296 for (i = 0; i < cnt; i++) {
1300 printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
1305 /* Will call this only for path name based fid */
1306 static void v9fs_fix_path(V9fsPath *dst, V9fsPath *src, int len)
1309 v9fs_path_init(&str);
1310 v9fs_path_copy(&str, dst);
1311 v9fs_path_sprintf(dst, "%s%s", src->data, str.data + len);
1312 v9fs_path_free(&str);
1315 static inline bool is_ro_export(FsContext *ctx)
1317 return ctx->export_flags & V9FS_RDONLY;
1320 static void coroutine_fn v9fs_version(void *opaque)
1323 V9fsPDU *pdu = opaque;
1324 V9fsState *s = pdu->s;
1328 v9fs_string_init(&version);
1329 err = pdu_unmarshal(pdu, offset, "ds", &s->msize, &version);
1333 trace_v9fs_version(pdu->tag, pdu->id, s->msize, version.data);
1337 if (!strcmp(version.data, "9P2000.u")) {
1338 s->proto_version = V9FS_PROTO_2000U;
1339 } else if (!strcmp(version.data, "9P2000.L")) {
1340 s->proto_version = V9FS_PROTO_2000L;
1342 v9fs_string_sprintf(&version, "unknown");
1343 /* skip min. msize check, reporting invalid version has priority */
1347 if (s->msize < P9_MIN_MSIZE) {
1350 "9pfs: Client requested msize < minimum msize ("
1351 stringify(P9_MIN_MSIZE) ") supported by this server."
1357 err = pdu_marshal(pdu, offset, "ds", s->msize, &version);
1362 trace_v9fs_version_return(pdu->tag, pdu->id, s->msize, version.data);
1364 pdu_complete(pdu, err);
1365 v9fs_string_free(&version);
1368 static void coroutine_fn v9fs_attach(void *opaque)
1370 V9fsPDU *pdu = opaque;
1371 V9fsState *s = pdu->s;
1372 int32_t fid, afid, n_uname;
1373 V9fsString uname, aname;
1379 v9fs_string_init(&uname);
1380 v9fs_string_init(&aname);
1381 err = pdu_unmarshal(pdu, offset, "ddssd", &fid,
1382 &afid, &uname, &aname, &n_uname);
1386 trace_v9fs_attach(pdu->tag, pdu->id, fid, afid, uname.data, aname.data);
1388 fidp = alloc_fid(s, fid);
1393 fidp->uid = n_uname;
1394 err = v9fs_co_name_to_path(pdu, NULL, "/", &fidp->path);
1400 err = fid_to_qid(pdu, fidp, &qid);
1408 * disable migration if we haven't done already.
1409 * attach could get called multiple times for the same export.
1411 if (!s->migration_blocker) {
1412 error_setg(&s->migration_blocker,
1413 "Migration is disabled when VirtFS export path '%s' is mounted in the guest using mount_tag '%s'",
1414 s->ctx.fs_root ? s->ctx.fs_root : "NULL", s->tag);
1415 err = migrate_add_blocker(s->migration_blocker, NULL);
1417 error_free(s->migration_blocker);
1418 s->migration_blocker = NULL;
1425 err = pdu_marshal(pdu, offset, "Q", &qid);
1432 memcpy(&s->root_qid, &qid, sizeof(qid));
1433 trace_v9fs_attach_return(pdu->tag, pdu->id,
1434 qid.type, qid.version, qid.path);
1438 pdu_complete(pdu, err);
1439 v9fs_string_free(&uname);
1440 v9fs_string_free(&aname);
1443 static void coroutine_fn v9fs_stat(void *opaque)
1451 V9fsPDU *pdu = opaque;
1454 err = pdu_unmarshal(pdu, offset, "d", &fid);
1458 trace_v9fs_stat(pdu->tag, pdu->id, fid);
1460 fidp = get_fid(pdu, fid);
1465 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1469 basename = g_path_get_basename(fidp->path.data);
1470 err = stat_to_v9stat(pdu, &fidp->path, basename, &stbuf, &v9stat);
1475 err = pdu_marshal(pdu, offset, "wS", 0, &v9stat);
1477 v9fs_stat_free(&v9stat);
1480 trace_v9fs_stat_return(pdu->tag, pdu->id, v9stat.mode,
1481 v9stat.atime, v9stat.mtime, v9stat.length);
1483 v9fs_stat_free(&v9stat);
1487 pdu_complete(pdu, err);
1490 static void coroutine_fn v9fs_getattr(void *opaque)
1497 uint64_t request_mask;
1498 V9fsStatDotl v9stat_dotl;
1499 V9fsPDU *pdu = opaque;
1501 retval = pdu_unmarshal(pdu, offset, "dq", &fid, &request_mask);
1505 trace_v9fs_getattr(pdu->tag, pdu->id, fid, request_mask);
1507 fidp = get_fid(pdu, fid);
1513 * Currently we only support BASIC fields in stat, so there is no
1514 * need to look at request_mask.
1516 retval = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1520 retval = stat_to_v9stat_dotl(pdu, &stbuf, &v9stat_dotl);
1525 /* fill st_gen if requested and supported by underlying fs */
1526 if (request_mask & P9_STATS_GEN) {
1527 retval = v9fs_co_st_gen(pdu, &fidp->path, stbuf.st_mode, &v9stat_dotl);
1530 /* we have valid st_gen: update result mask */
1531 v9stat_dotl.st_result_mask |= P9_STATS_GEN;
1534 /* request cancelled, e.g. by Tflush */
1537 /* failed to get st_gen: not fatal, ignore */
1541 retval = pdu_marshal(pdu, offset, "A", &v9stat_dotl);
1546 trace_v9fs_getattr_return(pdu->tag, pdu->id, v9stat_dotl.st_result_mask,
1547 v9stat_dotl.st_mode, v9stat_dotl.st_uid,
1548 v9stat_dotl.st_gid);
1552 pdu_complete(pdu, retval);
1555 /* Attribute flags */
1556 #define P9_ATTR_MODE (1 << 0)
1557 #define P9_ATTR_UID (1 << 1)
1558 #define P9_ATTR_GID (1 << 2)
1559 #define P9_ATTR_SIZE (1 << 3)
1560 #define P9_ATTR_ATIME (1 << 4)
1561 #define P9_ATTR_MTIME (1 << 5)
1562 #define P9_ATTR_CTIME (1 << 6)
1563 #define P9_ATTR_ATIME_SET (1 << 7)
1564 #define P9_ATTR_MTIME_SET (1 << 8)
1566 #define P9_ATTR_MASK 127
1568 static void coroutine_fn v9fs_setattr(void *opaque)
1575 V9fsPDU *pdu = opaque;
1577 err = pdu_unmarshal(pdu, offset, "dI", &fid, &v9iattr);
1582 trace_v9fs_setattr(pdu->tag, pdu->id, fid,
1583 v9iattr.valid, v9iattr.mode, v9iattr.uid, v9iattr.gid,
1584 v9iattr.size, v9iattr.atime_sec, v9iattr.mtime_sec);
1586 fidp = get_fid(pdu, fid);
1591 if (v9iattr.valid & P9_ATTR_MODE) {
1592 err = v9fs_co_chmod(pdu, &fidp->path, v9iattr.mode);
1597 if (v9iattr.valid & (P9_ATTR_ATIME | P9_ATTR_MTIME)) {
1598 struct timespec times[2];
1599 if (v9iattr.valid & P9_ATTR_ATIME) {
1600 if (v9iattr.valid & P9_ATTR_ATIME_SET) {
1601 times[0].tv_sec = v9iattr.atime_sec;
1602 times[0].tv_nsec = v9iattr.atime_nsec;
1604 times[0].tv_nsec = UTIME_NOW;
1607 times[0].tv_nsec = UTIME_OMIT;
1609 if (v9iattr.valid & P9_ATTR_MTIME) {
1610 if (v9iattr.valid & P9_ATTR_MTIME_SET) {
1611 times[1].tv_sec = v9iattr.mtime_sec;
1612 times[1].tv_nsec = v9iattr.mtime_nsec;
1614 times[1].tv_nsec = UTIME_NOW;
1617 times[1].tv_nsec = UTIME_OMIT;
1619 err = v9fs_co_utimensat(pdu, &fidp->path, times);
1625 * If the only valid entry in iattr is ctime we can call
1626 * chown(-1,-1) to update the ctime of the file
1628 if ((v9iattr.valid & (P9_ATTR_UID | P9_ATTR_GID)) ||
1629 ((v9iattr.valid & P9_ATTR_CTIME)
1630 && !((v9iattr.valid & P9_ATTR_MASK) & ~P9_ATTR_CTIME))) {
1631 if (!(v9iattr.valid & P9_ATTR_UID)) {
1634 if (!(v9iattr.valid & P9_ATTR_GID)) {
1637 err = v9fs_co_chown(pdu, &fidp->path, v9iattr.uid,
1643 if (v9iattr.valid & (P9_ATTR_SIZE)) {
1644 err = v9fs_co_truncate(pdu, &fidp->path, v9iattr.size);
1650 trace_v9fs_setattr_return(pdu->tag, pdu->id);
1654 pdu_complete(pdu, err);
1657 static int v9fs_walk_marshal(V9fsPDU *pdu, uint16_t nwnames, V9fsQID *qids)
1663 err = pdu_marshal(pdu, offset, "w", nwnames);
1668 for (i = 0; i < nwnames; i++) {
1669 err = pdu_marshal(pdu, offset, "Q", &qids[i]);
1678 static bool name_is_illegal(const char *name)
1680 return !*name || strchr(name, '/') != NULL;
1683 static bool not_same_qid(const V9fsQID *qid1, const V9fsQID *qid2)
1686 qid1->type != qid2->type ||
1687 qid1->version != qid2->version ||
1688 qid1->path != qid2->path;
1691 static void coroutine_fn v9fs_walk(void *opaque)
1694 V9fsQID *qids = NULL;
1696 V9fsPath dpath, path;
1700 int32_t fid, newfid;
1701 V9fsString *wnames = NULL;
1703 V9fsFidState *newfidp = NULL;
1704 V9fsPDU *pdu = opaque;
1705 V9fsState *s = pdu->s;
1708 err = pdu_unmarshal(pdu, offset, "ddw", &fid, &newfid, &nwnames);
1710 pdu_complete(pdu, err);
1715 trace_v9fs_walk(pdu->tag, pdu->id, fid, newfid, nwnames);
1717 if (nwnames && nwnames <= P9_MAXWELEM) {
1718 wnames = g_new0(V9fsString, nwnames);
1719 qids = g_new0(V9fsQID, nwnames);
1720 for (i = 0; i < nwnames; i++) {
1721 err = pdu_unmarshal(pdu, offset, "s", &wnames[i]);
1725 if (name_is_illegal(wnames[i].data)) {
1731 } else if (nwnames > P9_MAXWELEM) {
1735 fidp = get_fid(pdu, fid);
1741 v9fs_path_init(&dpath);
1742 v9fs_path_init(&path);
1744 err = fid_to_qid(pdu, fidp, &qid);
1750 * Both dpath and path initially poin to fidp.
1751 * Needed to handle request with nwnames == 0
1753 v9fs_path_copy(&dpath, &fidp->path);
1754 v9fs_path_copy(&path, &fidp->path);
1755 for (name_idx = 0; name_idx < nwnames; name_idx++) {
1756 if (not_same_qid(&pdu->s->root_qid, &qid) ||
1757 strcmp("..", wnames[name_idx].data)) {
1758 err = v9fs_co_name_to_path(pdu, &dpath, wnames[name_idx].data,
1764 err = v9fs_co_lstat(pdu, &path, &stbuf);
1768 err = stat_to_qid(pdu, &stbuf, &qid);
1772 v9fs_path_copy(&dpath, &path);
1774 memcpy(&qids[name_idx], &qid, sizeof(qid));
1776 if (fid == newfid) {
1777 if (fidp->fid_type != P9_FID_NONE) {
1781 v9fs_path_write_lock(s);
1782 v9fs_path_copy(&fidp->path, &path);
1783 v9fs_path_unlock(s);
1785 newfidp = alloc_fid(s, newfid);
1786 if (newfidp == NULL) {
1790 newfidp->uid = fidp->uid;
1791 v9fs_path_copy(&newfidp->path, &path);
1793 err = v9fs_walk_marshal(pdu, nwnames, qids);
1794 trace_v9fs_walk_return(pdu->tag, pdu->id, nwnames, qids);
1798 put_fid(pdu, newfidp);
1800 v9fs_path_free(&dpath);
1801 v9fs_path_free(&path);
1803 pdu_complete(pdu, err);
1804 if (nwnames && nwnames <= P9_MAXWELEM) {
1805 for (name_idx = 0; name_idx < nwnames; name_idx++) {
1806 v9fs_string_free(&wnames[name_idx]);
1813 static int32_t coroutine_fn get_iounit(V9fsPDU *pdu, V9fsPath *path)
1815 struct statfs stbuf;
1817 V9fsState *s = pdu->s;
1820 * iounit should be multiples of f_bsize (host filesystem block size
1821 * and as well as less than (client msize - P9_IOHDRSZ))
1823 if (!v9fs_co_statfs(pdu, path, &stbuf)) {
1824 if (stbuf.f_bsize) {
1825 iounit = stbuf.f_bsize;
1826 iounit *= (s->msize - P9_IOHDRSZ) / stbuf.f_bsize;
1830 iounit = s->msize - P9_IOHDRSZ;
1835 static void coroutine_fn v9fs_open(void *opaque)
1846 V9fsPDU *pdu = opaque;
1847 V9fsState *s = pdu->s;
1849 if (s->proto_version == V9FS_PROTO_2000L) {
1850 err = pdu_unmarshal(pdu, offset, "dd", &fid, &mode);
1853 err = pdu_unmarshal(pdu, offset, "db", &fid, &modebyte);
1859 trace_v9fs_open(pdu->tag, pdu->id, fid, mode);
1861 fidp = get_fid(pdu, fid);
1866 if (fidp->fid_type != P9_FID_NONE) {
1871 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1875 err = stat_to_qid(pdu, &stbuf, &qid);
1879 if (S_ISDIR(stbuf.st_mode)) {
1880 err = v9fs_co_opendir(pdu, fidp);
1884 fidp->fid_type = P9_FID_DIR;
1885 err = pdu_marshal(pdu, offset, "Qd", &qid, 0);
1891 if (s->proto_version == V9FS_PROTO_2000L) {
1892 flags = get_dotl_openflags(s, mode);
1894 flags = omode_to_uflags(mode);
1896 if (is_ro_export(&s->ctx)) {
1897 if (mode & O_WRONLY || mode & O_RDWR ||
1898 mode & O_APPEND || mode & O_TRUNC) {
1903 err = v9fs_co_open(pdu, fidp, flags);
1907 fidp->fid_type = P9_FID_FILE;
1908 fidp->open_flags = flags;
1909 if (flags & O_EXCL) {
1911 * We let the host file system do O_EXCL check
1912 * We should not reclaim such fd
1914 fidp->flags |= FID_NON_RECLAIMABLE;
1916 iounit = get_iounit(pdu, &fidp->path);
1917 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1923 trace_v9fs_open_return(pdu->tag, pdu->id,
1924 qid.type, qid.version, qid.path, iounit);
1928 pdu_complete(pdu, err);
1931 static void coroutine_fn v9fs_lcreate(void *opaque)
1933 int32_t dfid, flags, mode;
1942 V9fsPDU *pdu = opaque;
1944 v9fs_string_init(&name);
1945 err = pdu_unmarshal(pdu, offset, "dsddd", &dfid,
1946 &name, &flags, &mode, &gid);
1950 trace_v9fs_lcreate(pdu->tag, pdu->id, dfid, flags, mode, gid);
1952 if (name_is_illegal(name.data)) {
1957 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
1962 fidp = get_fid(pdu, dfid);
1967 if (fidp->fid_type != P9_FID_NONE) {
1972 flags = get_dotl_openflags(pdu->s, flags);
1973 err = v9fs_co_open2(pdu, fidp, &name, gid,
1974 flags | O_CREAT, mode, &stbuf);
1978 fidp->fid_type = P9_FID_FILE;
1979 fidp->open_flags = flags;
1980 if (flags & O_EXCL) {
1982 * We let the host file system do O_EXCL check
1983 * We should not reclaim such fd
1985 fidp->flags |= FID_NON_RECLAIMABLE;
1987 iounit = get_iounit(pdu, &fidp->path);
1988 err = stat_to_qid(pdu, &stbuf, &qid);
1992 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1997 trace_v9fs_lcreate_return(pdu->tag, pdu->id,
1998 qid.type, qid.version, qid.path, iounit);
2002 pdu_complete(pdu, err);
2003 v9fs_string_free(&name);
2006 static void coroutine_fn v9fs_fsync(void *opaque)
2013 V9fsPDU *pdu = opaque;
2015 err = pdu_unmarshal(pdu, offset, "dd", &fid, &datasync);
2019 trace_v9fs_fsync(pdu->tag, pdu->id, fid, datasync);
2021 fidp = get_fid(pdu, fid);
2026 err = v9fs_co_fsync(pdu, fidp, datasync);
2032 pdu_complete(pdu, err);
2035 static void coroutine_fn v9fs_clunk(void *opaque)
2041 V9fsPDU *pdu = opaque;
2042 V9fsState *s = pdu->s;
2044 err = pdu_unmarshal(pdu, offset, "d", &fid);
2048 trace_v9fs_clunk(pdu->tag, pdu->id, fid);
2050 fidp = clunk_fid(s, fid);
2056 * Bump the ref so that put_fid will
2060 err = put_fid(pdu, fidp);
2065 pdu_complete(pdu, err);
2069 * Create a QEMUIOVector for a sub-region of PDU iovecs
2071 * @qiov: uninitialized QEMUIOVector
2072 * @skip: number of bytes to skip from beginning of PDU
2073 * @size: number of bytes to include
2074 * @is_write: true - write, false - read
2076 * The resulting QEMUIOVector has heap-allocated iovecs and must be cleaned up
2077 * with qemu_iovec_destroy().
2079 static void v9fs_init_qiov_from_pdu(QEMUIOVector *qiov, V9fsPDU *pdu,
2080 size_t skip, size_t size,
2088 pdu->s->transport->init_out_iov_from_pdu(pdu, &iov, &niov, size + skip);
2090 pdu->s->transport->init_in_iov_from_pdu(pdu, &iov, &niov, size + skip);
2093 qemu_iovec_init_external(&elem, iov, niov);
2094 qemu_iovec_init(qiov, niov);
2095 qemu_iovec_concat(qiov, &elem, skip, size);
2098 static int v9fs_xattr_read(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
2099 uint64_t off, uint32_t max_count)
2103 uint64_t read_count;
2104 QEMUIOVector qiov_full;
2106 if (fidp->fs.xattr.len < off) {
2109 read_count = fidp->fs.xattr.len - off;
2111 if (read_count > max_count) {
2112 read_count = max_count;
2114 err = pdu_marshal(pdu, offset, "d", read_count);
2120 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, read_count, false);
2121 err = v9fs_pack(qiov_full.iov, qiov_full.niov, 0,
2122 ((char *)fidp->fs.xattr.value) + off,
2124 qemu_iovec_destroy(&qiov_full);
2132 static int coroutine_fn v9fs_do_readdir_with_stat(V9fsPDU *pdu,
2141 off_t saved_dir_pos;
2142 struct dirent *dent;
2144 /* save the directory position */
2145 saved_dir_pos = v9fs_co_telldir(pdu, fidp);
2146 if (saved_dir_pos < 0) {
2147 return saved_dir_pos;
2151 v9fs_path_init(&path);
2153 v9fs_readdir_lock(&fidp->fs.dir);
2155 err = v9fs_co_readdir(pdu, fidp, &dent);
2159 err = v9fs_co_name_to_path(pdu, &fidp->path, dent->d_name, &path);
2163 err = v9fs_co_lstat(pdu, &path, &stbuf);
2167 err = stat_to_v9stat(pdu, &path, dent->d_name, &stbuf, &v9stat);
2171 if ((count + v9stat.size + 2) > max_count) {
2172 v9fs_readdir_unlock(&fidp->fs.dir);
2174 /* Ran out of buffer. Set dir back to old position and return */
2175 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2176 v9fs_stat_free(&v9stat);
2177 v9fs_path_free(&path);
2181 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2182 len = pdu_marshal(pdu, 11 + count, "S", &v9stat);
2184 v9fs_readdir_unlock(&fidp->fs.dir);
2187 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2188 v9fs_stat_free(&v9stat);
2189 v9fs_path_free(&path);
2193 v9fs_stat_free(&v9stat);
2194 v9fs_path_free(&path);
2195 saved_dir_pos = dent->d_off;
2198 v9fs_readdir_unlock(&fidp->fs.dir);
2200 v9fs_path_free(&path);
2207 static void coroutine_fn v9fs_read(void *opaque)
2216 V9fsPDU *pdu = opaque;
2217 V9fsState *s = pdu->s;
2219 err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &max_count);
2223 trace_v9fs_read(pdu->tag, pdu->id, fid, off, max_count);
2225 fidp = get_fid(pdu, fid);
2230 if (fidp->fid_type == P9_FID_DIR) {
2231 if (s->proto_version != V9FS_PROTO_2000U) {
2233 "9p: bad client: T_read request on directory only expected "
2234 "with 9P2000.u protocol version"
2240 v9fs_co_rewinddir(pdu, fidp);
2242 count = v9fs_do_readdir_with_stat(pdu, fidp, max_count);
2247 err = pdu_marshal(pdu, offset, "d", count);
2251 err += offset + count;
2252 } else if (fidp->fid_type == P9_FID_FILE) {
2253 QEMUIOVector qiov_full;
2257 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset + 4, max_count, false);
2258 qemu_iovec_init(&qiov, qiov_full.niov);
2260 qemu_iovec_reset(&qiov);
2261 qemu_iovec_concat(&qiov, &qiov_full, count, qiov_full.size - count);
2263 print_sg(qiov.iov, qiov.niov);
2265 /* Loop in case of EINTR */
2267 len = v9fs_co_preadv(pdu, fidp, qiov.iov, qiov.niov, off);
2272 } while (len == -EINTR && !pdu->cancelled);
2274 /* IO error return the error */
2276 goto out_free_iovec;
2278 } while (count < max_count && len > 0);
2279 err = pdu_marshal(pdu, offset, "d", count);
2281 goto out_free_iovec;
2283 err += offset + count;
2285 qemu_iovec_destroy(&qiov);
2286 qemu_iovec_destroy(&qiov_full);
2287 } else if (fidp->fid_type == P9_FID_XATTR) {
2288 err = v9fs_xattr_read(s, pdu, fidp, off, max_count);
2292 trace_v9fs_read_return(pdu->tag, pdu->id, count, err);
2296 pdu_complete(pdu, err);
2300 * Returns size required in Rreaddir response for the passed dirent @p name.
2302 * @param name - directory entry's name (i.e. file name, directory name)
2303 * @returns required size in bytes
2305 size_t v9fs_readdir_response_size(V9fsString *name)
2308 * Size of each dirent on the wire: size of qid (13) + size of offset (8)
2309 * size of type (1) + size of name.size (2) + strlen(name.data)
2311 return 24 + v9fs_string_size(name);
2314 static void v9fs_free_dirents(struct V9fsDirEnt *e)
2316 struct V9fsDirEnt *next = NULL;
2318 for (; e; e = next) {
2326 static int coroutine_fn v9fs_do_readdir(V9fsPDU *pdu, V9fsFidState *fidp,
2327 off_t offset, int32_t max_count)
2334 struct dirent *dent;
2336 struct V9fsDirEnt *entries = NULL;
2339 * inode remapping requires the device id, which in turn might be
2340 * different for different directory entries, so if inode remapping is
2341 * enabled we have to make a full stat for each directory entry
2343 const bool dostat = pdu->s->ctx.export_flags & V9FS_REMAP_INODES;
2346 * Fetch all required directory entries altogether on a background IO
2347 * thread from fs driver. We don't want to do that for each entry
2348 * individually, because hopping between threads (this main IO thread
2349 * and background IO driver thread) would sum up to huge latencies.
2351 count = v9fs_co_readdir_many(pdu, fidp, &entries, offset, max_count,
2360 for (struct V9fsDirEnt *e = entries; e; e = e->next) {
2363 if (pdu->s->ctx.export_flags & V9FS_REMAP_INODES) {
2365 /* e->st should never be NULL, but just to be sure */
2372 err = stat_to_qid(pdu, st, &qid);
2378 * Fill up just the path field of qid because the client uses
2379 * only that. To fill the entire qid structure we will have
2380 * to stat each dirent found, which is expensive. For the
2381 * latter reason we don't call stat_to_qid() here. Only drawback
2382 * is that no multi-device export detection of stat_to_qid()
2383 * would be done and provided as error to the user here. But
2384 * user would get that error anyway when accessing those
2385 * files/dirs through other ways.
2387 size = MIN(sizeof(dent->d_ino), sizeof(qid.path));
2388 memcpy(&qid.path, &dent->d_ino, size);
2389 /* Fill the other fields with dummy values */
2394 v9fs_string_init(&name);
2395 v9fs_string_sprintf(&name, "%s", dent->d_name);
2397 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2398 len = pdu_marshal(pdu, 11 + count, "Qqbs",
2400 dent->d_type, &name);
2402 v9fs_string_free(&name);
2413 v9fs_free_dirents(entries);
2420 static void coroutine_fn v9fs_readdir(void *opaque)
2426 uint64_t initial_offset;
2429 V9fsPDU *pdu = opaque;
2430 V9fsState *s = pdu->s;
2432 retval = pdu_unmarshal(pdu, offset, "dqd", &fid,
2433 &initial_offset, &max_count);
2437 trace_v9fs_readdir(pdu->tag, pdu->id, fid, initial_offset, max_count);
2439 /* Enough space for a R_readdir header: size[4] Rreaddir tag[2] count[4] */
2440 if (max_count > s->msize - 11) {
2441 max_count = s->msize - 11;
2443 "9p: bad client: T_readdir with count > msize - 11"
2447 fidp = get_fid(pdu, fid);
2452 if (!fidp->fs.dir.stream) {
2456 if (s->proto_version != V9FS_PROTO_2000L) {
2458 "9p: bad client: T_readdir request only expected with 9P2000.L "
2461 retval = -EOPNOTSUPP;
2464 count = v9fs_do_readdir(pdu, fidp, (off_t) initial_offset, max_count);
2469 retval = pdu_marshal(pdu, offset, "d", count);
2473 retval += count + offset;
2474 trace_v9fs_readdir_return(pdu->tag, pdu->id, count, retval);
2478 pdu_complete(pdu, retval);
2481 static int v9fs_xattr_write(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
2482 uint64_t off, uint32_t count,
2483 struct iovec *sg, int cnt)
2487 uint64_t write_count;
2491 if (fidp->fs.xattr.len < off) {
2494 write_count = fidp->fs.xattr.len - off;
2495 if (write_count > count) {
2496 write_count = count;
2498 err = pdu_marshal(pdu, offset, "d", write_count);
2503 fidp->fs.xattr.copied_len += write_count;
2505 * Now copy the content from sg list
2507 for (i = 0; i < cnt; i++) {
2508 if (write_count > sg[i].iov_len) {
2509 to_copy = sg[i].iov_len;
2511 to_copy = write_count;
2513 memcpy((char *)fidp->fs.xattr.value + off, sg[i].iov_base, to_copy);
2514 /* updating vs->off since we are not using below */
2516 write_count -= to_copy;
2522 static void coroutine_fn v9fs_write(void *opaque)
2532 V9fsPDU *pdu = opaque;
2533 V9fsState *s = pdu->s;
2534 QEMUIOVector qiov_full;
2537 err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &count);
2539 pdu_complete(pdu, err);
2543 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, count, true);
2544 trace_v9fs_write(pdu->tag, pdu->id, fid, off, count, qiov_full.niov);
2546 fidp = get_fid(pdu, fid);
2551 if (fidp->fid_type == P9_FID_FILE) {
2552 if (fidp->fs.fd == -1) {
2556 } else if (fidp->fid_type == P9_FID_XATTR) {
2558 * setxattr operation
2560 err = v9fs_xattr_write(s, pdu, fidp, off, count,
2561 qiov_full.iov, qiov_full.niov);
2567 qemu_iovec_init(&qiov, qiov_full.niov);
2569 qemu_iovec_reset(&qiov);
2570 qemu_iovec_concat(&qiov, &qiov_full, total, qiov_full.size - total);
2572 print_sg(qiov.iov, qiov.niov);
2574 /* Loop in case of EINTR */
2576 len = v9fs_co_pwritev(pdu, fidp, qiov.iov, qiov.niov, off);
2581 } while (len == -EINTR && !pdu->cancelled);
2583 /* IO error return the error */
2587 } while (total < count && len > 0);
2590 err = pdu_marshal(pdu, offset, "d", total);
2595 trace_v9fs_write_return(pdu->tag, pdu->id, total, err);
2597 qemu_iovec_destroy(&qiov);
2601 qemu_iovec_destroy(&qiov_full);
2602 pdu_complete(pdu, err);
2605 static void coroutine_fn v9fs_create(void *opaque)
2617 V9fsString extension;
2619 V9fsPDU *pdu = opaque;
2620 V9fsState *s = pdu->s;
2622 v9fs_path_init(&path);
2623 v9fs_string_init(&name);
2624 v9fs_string_init(&extension);
2625 err = pdu_unmarshal(pdu, offset, "dsdbs", &fid, &name,
2626 &perm, &mode, &extension);
2630 trace_v9fs_create(pdu->tag, pdu->id, fid, name.data, perm, mode);
2632 if (name_is_illegal(name.data)) {
2637 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2642 fidp = get_fid(pdu, fid);
2647 if (fidp->fid_type != P9_FID_NONE) {
2651 if (perm & P9_STAT_MODE_DIR) {
2652 err = v9fs_co_mkdir(pdu, fidp, &name, perm & 0777,
2653 fidp->uid, -1, &stbuf);
2657 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2661 v9fs_path_write_lock(s);
2662 v9fs_path_copy(&fidp->path, &path);
2663 v9fs_path_unlock(s);
2664 err = v9fs_co_opendir(pdu, fidp);
2668 fidp->fid_type = P9_FID_DIR;
2669 } else if (perm & P9_STAT_MODE_SYMLINK) {
2670 err = v9fs_co_symlink(pdu, fidp, &name,
2671 extension.data, -1 , &stbuf);
2675 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2679 v9fs_path_write_lock(s);
2680 v9fs_path_copy(&fidp->path, &path);
2681 v9fs_path_unlock(s);
2682 } else if (perm & P9_STAT_MODE_LINK) {
2683 int32_t ofid = atoi(extension.data);
2684 V9fsFidState *ofidp = get_fid(pdu, ofid);
2685 if (ofidp == NULL) {
2689 err = v9fs_co_link(pdu, ofidp, fidp, &name);
2690 put_fid(pdu, ofidp);
2694 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2696 fidp->fid_type = P9_FID_NONE;
2699 v9fs_path_write_lock(s);
2700 v9fs_path_copy(&fidp->path, &path);
2701 v9fs_path_unlock(s);
2702 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
2704 fidp->fid_type = P9_FID_NONE;
2707 } else if (perm & P9_STAT_MODE_DEVICE) {
2709 uint32_t major, minor;
2712 if (sscanf(extension.data, "%c %u %u", &ctype, &major, &minor) != 3) {
2729 nmode |= perm & 0777;
2730 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2731 makedev(major, minor), nmode, &stbuf);
2735 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2739 v9fs_path_write_lock(s);
2740 v9fs_path_copy(&fidp->path, &path);
2741 v9fs_path_unlock(s);
2742 } else if (perm & P9_STAT_MODE_NAMED_PIPE) {
2743 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2744 0, S_IFIFO | (perm & 0777), &stbuf);
2748 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2752 v9fs_path_write_lock(s);
2753 v9fs_path_copy(&fidp->path, &path);
2754 v9fs_path_unlock(s);
2755 } else if (perm & P9_STAT_MODE_SOCKET) {
2756 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2757 0, S_IFSOCK | (perm & 0777), &stbuf);
2761 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2765 v9fs_path_write_lock(s);
2766 v9fs_path_copy(&fidp->path, &path);
2767 v9fs_path_unlock(s);
2769 err = v9fs_co_open2(pdu, fidp, &name, -1,
2770 omode_to_uflags(mode)|O_CREAT, perm, &stbuf);
2774 fidp->fid_type = P9_FID_FILE;
2775 fidp->open_flags = omode_to_uflags(mode);
2776 if (fidp->open_flags & O_EXCL) {
2778 * We let the host file system do O_EXCL check
2779 * We should not reclaim such fd
2781 fidp->flags |= FID_NON_RECLAIMABLE;
2784 iounit = get_iounit(pdu, &fidp->path);
2785 err = stat_to_qid(pdu, &stbuf, &qid);
2789 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
2794 trace_v9fs_create_return(pdu->tag, pdu->id,
2795 qid.type, qid.version, qid.path, iounit);
2799 pdu_complete(pdu, err);
2800 v9fs_string_free(&name);
2801 v9fs_string_free(&extension);
2802 v9fs_path_free(&path);
2805 static void coroutine_fn v9fs_symlink(void *opaque)
2807 V9fsPDU *pdu = opaque;
2810 V9fsFidState *dfidp;
2818 v9fs_string_init(&name);
2819 v9fs_string_init(&symname);
2820 err = pdu_unmarshal(pdu, offset, "dssd", &dfid, &name, &symname, &gid);
2824 trace_v9fs_symlink(pdu->tag, pdu->id, dfid, name.data, symname.data, gid);
2826 if (name_is_illegal(name.data)) {
2831 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2836 dfidp = get_fid(pdu, dfid);
2837 if (dfidp == NULL) {
2841 err = v9fs_co_symlink(pdu, dfidp, &name, symname.data, gid, &stbuf);
2845 err = stat_to_qid(pdu, &stbuf, &qid);
2849 err = pdu_marshal(pdu, offset, "Q", &qid);
2854 trace_v9fs_symlink_return(pdu->tag, pdu->id,
2855 qid.type, qid.version, qid.path);
2857 put_fid(pdu, dfidp);
2859 pdu_complete(pdu, err);
2860 v9fs_string_free(&name);
2861 v9fs_string_free(&symname);
2864 static void coroutine_fn v9fs_flush(void *opaque)
2869 V9fsPDU *cancel_pdu = NULL;
2870 V9fsPDU *pdu = opaque;
2871 V9fsState *s = pdu->s;
2873 err = pdu_unmarshal(pdu, offset, "w", &tag);
2875 pdu_complete(pdu, err);
2878 trace_v9fs_flush(pdu->tag, pdu->id, tag);
2880 if (pdu->tag == tag) {
2881 warn_report("the guest sent a self-referencing 9P flush request");
2883 QLIST_FOREACH(cancel_pdu, &s->active_list, next) {
2884 if (cancel_pdu->tag == tag) {
2890 cancel_pdu->cancelled = 1;
2892 * Wait for pdu to complete.
2894 qemu_co_queue_wait(&cancel_pdu->complete, NULL);
2895 if (!qemu_co_queue_next(&cancel_pdu->complete)) {
2896 cancel_pdu->cancelled = 0;
2897 pdu_free(cancel_pdu);
2900 pdu_complete(pdu, 7);
2903 static void coroutine_fn v9fs_link(void *opaque)
2905 V9fsPDU *pdu = opaque;
2906 int32_t dfid, oldfid;
2907 V9fsFidState *dfidp, *oldfidp;
2912 v9fs_string_init(&name);
2913 err = pdu_unmarshal(pdu, offset, "dds", &dfid, &oldfid, &name);
2917 trace_v9fs_link(pdu->tag, pdu->id, dfid, oldfid, name.data);
2919 if (name_is_illegal(name.data)) {
2924 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2929 dfidp = get_fid(pdu, dfid);
2930 if (dfidp == NULL) {
2935 oldfidp = get_fid(pdu, oldfid);
2936 if (oldfidp == NULL) {
2940 err = v9fs_co_link(pdu, oldfidp, dfidp, &name);
2944 put_fid(pdu, oldfidp);
2946 put_fid(pdu, dfidp);
2948 v9fs_string_free(&name);
2949 pdu_complete(pdu, err);
2952 /* Only works with path name based fid */
2953 static void coroutine_fn v9fs_remove(void *opaque)
2959 V9fsPDU *pdu = opaque;
2961 err = pdu_unmarshal(pdu, offset, "d", &fid);
2965 trace_v9fs_remove(pdu->tag, pdu->id, fid);
2967 fidp = get_fid(pdu, fid);
2972 /* if fs driver is not path based, return EOPNOTSUPP */
2973 if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
2978 * IF the file is unlinked, we cannot reopen
2979 * the file later. So don't reclaim fd
2981 err = v9fs_mark_fids_unreclaim(pdu, &fidp->path);
2985 err = v9fs_co_remove(pdu, &fidp->path);
2990 /* For TREMOVE we need to clunk the fid even on failed remove */
2991 clunk_fid(pdu->s, fidp->fid);
2994 pdu_complete(pdu, err);
2997 static void coroutine_fn v9fs_unlinkat(void *opaque)
3001 int32_t dfid, flags, rflags = 0;
3004 V9fsFidState *dfidp;
3005 V9fsPDU *pdu = opaque;
3007 v9fs_string_init(&name);
3008 err = pdu_unmarshal(pdu, offset, "dsd", &dfid, &name, &flags);
3013 if (name_is_illegal(name.data)) {
3018 if (!strcmp(".", name.data)) {
3023 if (!strcmp("..", name.data)) {
3028 if (flags & ~P9_DOTL_AT_REMOVEDIR) {
3033 if (flags & P9_DOTL_AT_REMOVEDIR) {
3034 rflags |= AT_REMOVEDIR;
3037 dfidp = get_fid(pdu, dfid);
3038 if (dfidp == NULL) {
3043 * IF the file is unlinked, we cannot reopen
3044 * the file later. So don't reclaim fd
3046 v9fs_path_init(&path);
3047 err = v9fs_co_name_to_path(pdu, &dfidp->path, name.data, &path);
3051 err = v9fs_mark_fids_unreclaim(pdu, &path);
3055 err = v9fs_co_unlinkat(pdu, &dfidp->path, &name, rflags);
3060 put_fid(pdu, dfidp);
3061 v9fs_path_free(&path);
3063 pdu_complete(pdu, err);
3064 v9fs_string_free(&name);
3068 /* Only works with path name based fid */
3069 static int coroutine_fn v9fs_complete_rename(V9fsPDU *pdu, V9fsFidState *fidp,
3075 V9fsFidState *tfidp;
3076 V9fsState *s = pdu->s;
3077 V9fsFidState *dirfidp = NULL;
3079 v9fs_path_init(&new_path);
3080 if (newdirfid != -1) {
3081 dirfidp = get_fid(pdu, newdirfid);
3082 if (dirfidp == NULL) {
3085 if (fidp->fid_type != P9_FID_NONE) {
3089 err = v9fs_co_name_to_path(pdu, &dirfidp->path, name->data, &new_path);
3094 char *dir_name = g_path_get_dirname(fidp->path.data);
3097 v9fs_path_init(&dir_path);
3098 v9fs_path_sprintf(&dir_path, "%s", dir_name);
3101 err = v9fs_co_name_to_path(pdu, &dir_path, name->data, &new_path);
3102 v9fs_path_free(&dir_path);
3107 err = v9fs_co_rename(pdu, &fidp->path, &new_path);
3112 * Fixup fid's pointing to the old name to
3113 * start pointing to the new name
3115 for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
3116 if (v9fs_path_is_ancestor(&fidp->path, &tfidp->path)) {
3117 /* replace the name */
3118 v9fs_fix_path(&tfidp->path, &new_path, strlen(fidp->path.data));
3123 put_fid(pdu, dirfidp);
3125 v9fs_path_free(&new_path);
3129 /* Only works with path name based fid */
3130 static void coroutine_fn v9fs_rename(void *opaque)
3138 V9fsPDU *pdu = opaque;
3139 V9fsState *s = pdu->s;
3141 v9fs_string_init(&name);
3142 err = pdu_unmarshal(pdu, offset, "dds", &fid, &newdirfid, &name);
3147 if (name_is_illegal(name.data)) {
3152 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3157 fidp = get_fid(pdu, fid);
3162 if (fidp->fid_type != P9_FID_NONE) {
3166 /* if fs driver is not path based, return EOPNOTSUPP */
3167 if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
3171 v9fs_path_write_lock(s);
3172 err = v9fs_complete_rename(pdu, fidp, newdirfid, &name);
3173 v9fs_path_unlock(s);
3180 pdu_complete(pdu, err);
3181 v9fs_string_free(&name);
3184 static int coroutine_fn v9fs_fix_fid_paths(V9fsPDU *pdu, V9fsPath *olddir,
3185 V9fsString *old_name,
3187 V9fsString *new_name)
3189 V9fsFidState *tfidp;
3190 V9fsPath oldpath, newpath;
3191 V9fsState *s = pdu->s;
3194 v9fs_path_init(&oldpath);
3195 v9fs_path_init(&newpath);
3196 err = v9fs_co_name_to_path(pdu, olddir, old_name->data, &oldpath);
3200 err = v9fs_co_name_to_path(pdu, newdir, new_name->data, &newpath);
3206 * Fixup fid's pointing to the old name to
3207 * start pointing to the new name
3209 for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
3210 if (v9fs_path_is_ancestor(&oldpath, &tfidp->path)) {
3211 /* replace the name */
3212 v9fs_fix_path(&tfidp->path, &newpath, strlen(oldpath.data));
3216 v9fs_path_free(&oldpath);
3217 v9fs_path_free(&newpath);
3221 static int coroutine_fn v9fs_complete_renameat(V9fsPDU *pdu, int32_t olddirfid,
3222 V9fsString *old_name,
3224 V9fsString *new_name)
3227 V9fsState *s = pdu->s;
3228 V9fsFidState *newdirfidp = NULL, *olddirfidp = NULL;
3230 olddirfidp = get_fid(pdu, olddirfid);
3231 if (olddirfidp == NULL) {
3235 if (newdirfid != -1) {
3236 newdirfidp = get_fid(pdu, newdirfid);
3237 if (newdirfidp == NULL) {
3242 newdirfidp = get_fid(pdu, olddirfid);
3245 err = v9fs_co_renameat(pdu, &olddirfidp->path, old_name,
3246 &newdirfidp->path, new_name);
3250 if (s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT) {
3251 /* Only for path based fid we need to do the below fixup */
3252 err = v9fs_fix_fid_paths(pdu, &olddirfidp->path, old_name,
3253 &newdirfidp->path, new_name);
3257 put_fid(pdu, olddirfidp);
3260 put_fid(pdu, newdirfidp);
3265 static void coroutine_fn v9fs_renameat(void *opaque)
3269 V9fsPDU *pdu = opaque;
3270 V9fsState *s = pdu->s;
3271 int32_t olddirfid, newdirfid;
3272 V9fsString old_name, new_name;
3274 v9fs_string_init(&old_name);
3275 v9fs_string_init(&new_name);
3276 err = pdu_unmarshal(pdu, offset, "dsds", &olddirfid,
3277 &old_name, &newdirfid, &new_name);
3282 if (name_is_illegal(old_name.data) || name_is_illegal(new_name.data)) {
3287 if (!strcmp(".", old_name.data) || !strcmp("..", old_name.data) ||
3288 !strcmp(".", new_name.data) || !strcmp("..", new_name.data)) {
3293 v9fs_path_write_lock(s);
3294 err = v9fs_complete_renameat(pdu, olddirfid,
3295 &old_name, newdirfid, &new_name);
3296 v9fs_path_unlock(s);
3302 pdu_complete(pdu, err);
3303 v9fs_string_free(&old_name);
3304 v9fs_string_free(&new_name);
3307 static void coroutine_fn v9fs_wstat(void *opaque)
3316 V9fsPDU *pdu = opaque;
3317 V9fsState *s = pdu->s;
3319 v9fs_stat_init(&v9stat);
3320 err = pdu_unmarshal(pdu, offset, "dwS", &fid, &unused, &v9stat);
3324 trace_v9fs_wstat(pdu->tag, pdu->id, fid,
3325 v9stat.mode, v9stat.atime, v9stat.mtime);
3327 fidp = get_fid(pdu, fid);
3332 /* do we need to sync the file? */
3333 if (donttouch_stat(&v9stat)) {
3334 err = v9fs_co_fsync(pdu, fidp, 0);
3337 if (v9stat.mode != -1) {
3339 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
3343 v9_mode = stat_to_v9mode(&stbuf);
3344 if ((v9stat.mode & P9_STAT_MODE_TYPE_BITS) !=
3345 (v9_mode & P9_STAT_MODE_TYPE_BITS)) {
3346 /* Attempting to change the type */
3350 err = v9fs_co_chmod(pdu, &fidp->path,
3351 v9mode_to_mode(v9stat.mode,
3352 &v9stat.extension));
3357 if (v9stat.mtime != -1 || v9stat.atime != -1) {
3358 struct timespec times[2];
3359 if (v9stat.atime != -1) {
3360 times[0].tv_sec = v9stat.atime;
3361 times[0].tv_nsec = 0;
3363 times[0].tv_nsec = UTIME_OMIT;
3365 if (v9stat.mtime != -1) {
3366 times[1].tv_sec = v9stat.mtime;
3367 times[1].tv_nsec = 0;
3369 times[1].tv_nsec = UTIME_OMIT;
3371 err = v9fs_co_utimensat(pdu, &fidp->path, times);
3376 if (v9stat.n_gid != -1 || v9stat.n_uid != -1) {
3377 err = v9fs_co_chown(pdu, &fidp->path, v9stat.n_uid, v9stat.n_gid);
3382 if (v9stat.name.size != 0) {
3383 v9fs_path_write_lock(s);
3384 err = v9fs_complete_rename(pdu, fidp, -1, &v9stat.name);
3385 v9fs_path_unlock(s);
3390 if (v9stat.length != -1) {
3391 err = v9fs_co_truncate(pdu, &fidp->path, v9stat.length);
3400 v9fs_stat_free(&v9stat);
3401 pdu_complete(pdu, err);
3404 static int v9fs_fill_statfs(V9fsState *s, V9fsPDU *pdu, struct statfs *stbuf)
3416 int32_t bsize_factor;
3419 * compute bsize factor based on host file system block size
3422 bsize_factor = (s->msize - P9_IOHDRSZ)/stbuf->f_bsize;
3423 if (!bsize_factor) {
3426 f_type = stbuf->f_type;
3427 f_bsize = stbuf->f_bsize;
3428 f_bsize *= bsize_factor;
3430 * f_bsize is adjusted(multiplied) by bsize factor, so we need to
3431 * adjust(divide) the number of blocks, free blocks and available
3432 * blocks by bsize factor
3434 f_blocks = stbuf->f_blocks/bsize_factor;
3435 f_bfree = stbuf->f_bfree/bsize_factor;
3436 f_bavail = stbuf->f_bavail/bsize_factor;
3437 f_files = stbuf->f_files;
3438 f_ffree = stbuf->f_ffree;
3439 fsid_val = (unsigned int) stbuf->f_fsid.__val[0] |
3440 (unsigned long long)stbuf->f_fsid.__val[1] << 32;
3441 f_namelen = stbuf->f_namelen;
3443 return pdu_marshal(pdu, offset, "ddqqqqqqd",
3444 f_type, f_bsize, f_blocks, f_bfree,
3445 f_bavail, f_files, f_ffree,
3446 fsid_val, f_namelen);
3449 static void coroutine_fn v9fs_statfs(void *opaque)
3455 struct statfs stbuf;
3456 V9fsPDU *pdu = opaque;
3457 V9fsState *s = pdu->s;
3459 retval = pdu_unmarshal(pdu, offset, "d", &fid);
3463 fidp = get_fid(pdu, fid);
3468 retval = v9fs_co_statfs(pdu, &fidp->path, &stbuf);
3472 retval = v9fs_fill_statfs(s, pdu, &stbuf);
3480 pdu_complete(pdu, retval);
3483 static void coroutine_fn v9fs_mknod(void *opaque)
3496 V9fsPDU *pdu = opaque;
3498 v9fs_string_init(&name);
3499 err = pdu_unmarshal(pdu, offset, "dsdddd", &fid, &name, &mode,
3500 &major, &minor, &gid);
3504 trace_v9fs_mknod(pdu->tag, pdu->id, fid, mode, major, minor);
3506 if (name_is_illegal(name.data)) {
3511 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3516 fidp = get_fid(pdu, fid);
3521 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, gid,
3522 makedev(major, minor), mode, &stbuf);
3526 err = stat_to_qid(pdu, &stbuf, &qid);
3530 err = pdu_marshal(pdu, offset, "Q", &qid);
3535 trace_v9fs_mknod_return(pdu->tag, pdu->id,
3536 qid.type, qid.version, qid.path);
3540 pdu_complete(pdu, err);
3541 v9fs_string_free(&name);
3545 * Implement posix byte range locking code
3546 * Server side handling of locking code is very simple, because 9p server in
3547 * QEMU can handle only one client. And most of the lock handling
3548 * (like conflict, merging) etc is done by the VFS layer itself, so no need to
3549 * do any thing in * qemu 9p server side lock code path.
3550 * So when a TLOCK request comes, always return success
3552 static void coroutine_fn v9fs_lock(void *opaque)
3558 int32_t fid, err = 0;
3559 V9fsPDU *pdu = opaque;
3561 v9fs_string_init(&flock.client_id);
3562 err = pdu_unmarshal(pdu, offset, "dbdqqds", &fid, &flock.type,
3563 &flock.flags, &flock.start, &flock.length,
3564 &flock.proc_id, &flock.client_id);
3568 trace_v9fs_lock(pdu->tag, pdu->id, fid,
3569 flock.type, flock.start, flock.length);
3572 /* We support only block flag now (that too ignored currently) */
3573 if (flock.flags & ~P9_LOCK_FLAGS_BLOCK) {
3577 fidp = get_fid(pdu, fid);
3582 err = v9fs_co_fstat(pdu, fidp, &stbuf);
3586 err = pdu_marshal(pdu, offset, "b", P9_LOCK_SUCCESS);
3591 trace_v9fs_lock_return(pdu->tag, pdu->id, P9_LOCK_SUCCESS);
3595 pdu_complete(pdu, err);
3596 v9fs_string_free(&flock.client_id);
3600 * When a TGETLOCK request comes, always return success because all lock
3601 * handling is done by client's VFS layer.
3603 static void coroutine_fn v9fs_getlock(void *opaque)
3609 int32_t fid, err = 0;
3610 V9fsPDU *pdu = opaque;
3612 v9fs_string_init(&glock.client_id);
3613 err = pdu_unmarshal(pdu, offset, "dbqqds", &fid, &glock.type,
3614 &glock.start, &glock.length, &glock.proc_id,
3619 trace_v9fs_getlock(pdu->tag, pdu->id, fid,
3620 glock.type, glock.start, glock.length);
3622 fidp = get_fid(pdu, fid);
3627 err = v9fs_co_fstat(pdu, fidp, &stbuf);
3631 glock.type = P9_LOCK_TYPE_UNLCK;
3632 err = pdu_marshal(pdu, offset, "bqqds", glock.type,
3633 glock.start, glock.length, glock.proc_id,
3639 trace_v9fs_getlock_return(pdu->tag, pdu->id, glock.type, glock.start,
3640 glock.length, glock.proc_id);
3644 pdu_complete(pdu, err);
3645 v9fs_string_free(&glock.client_id);
3648 static void coroutine_fn v9fs_mkdir(void *opaque)
3650 V9fsPDU *pdu = opaque;
3661 v9fs_string_init(&name);
3662 err = pdu_unmarshal(pdu, offset, "dsdd", &fid, &name, &mode, &gid);
3666 trace_v9fs_mkdir(pdu->tag, pdu->id, fid, name.data, mode, gid);
3668 if (name_is_illegal(name.data)) {
3673 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3678 fidp = get_fid(pdu, fid);
3683 err = v9fs_co_mkdir(pdu, fidp, &name, mode, fidp->uid, gid, &stbuf);
3687 err = stat_to_qid(pdu, &stbuf, &qid);
3691 err = pdu_marshal(pdu, offset, "Q", &qid);
3696 trace_v9fs_mkdir_return(pdu->tag, pdu->id,
3697 qid.type, qid.version, qid.path, err);
3701 pdu_complete(pdu, err);
3702 v9fs_string_free(&name);
3705 static void coroutine_fn v9fs_xattrwalk(void *opaque)
3711 int32_t fid, newfid;
3712 V9fsFidState *file_fidp;
3713 V9fsFidState *xattr_fidp = NULL;
3714 V9fsPDU *pdu = opaque;
3715 V9fsState *s = pdu->s;
3717 v9fs_string_init(&name);
3718 err = pdu_unmarshal(pdu, offset, "dds", &fid, &newfid, &name);
3722 trace_v9fs_xattrwalk(pdu->tag, pdu->id, fid, newfid, name.data);
3724 file_fidp = get_fid(pdu, fid);
3725 if (file_fidp == NULL) {
3729 xattr_fidp = alloc_fid(s, newfid);
3730 if (xattr_fidp == NULL) {
3734 v9fs_path_copy(&xattr_fidp->path, &file_fidp->path);
3735 if (!v9fs_string_size(&name)) {
3737 * listxattr request. Get the size first
3739 size = v9fs_co_llistxattr(pdu, &xattr_fidp->path, NULL, 0);
3742 clunk_fid(s, xattr_fidp->fid);
3746 * Read the xattr value
3748 xattr_fidp->fs.xattr.len = size;
3749 xattr_fidp->fid_type = P9_FID_XATTR;
3750 xattr_fidp->fs.xattr.xattrwalk_fid = true;
3751 xattr_fidp->fs.xattr.value = g_malloc0(size);
3753 err = v9fs_co_llistxattr(pdu, &xattr_fidp->path,
3754 xattr_fidp->fs.xattr.value,
3755 xattr_fidp->fs.xattr.len);
3757 clunk_fid(s, xattr_fidp->fid);
3761 err = pdu_marshal(pdu, offset, "q", size);
3768 * specific xattr fid. We check for xattr
3769 * presence also collect the xattr size
3771 size = v9fs_co_lgetxattr(pdu, &xattr_fidp->path,
3775 clunk_fid(s, xattr_fidp->fid);
3779 * Read the xattr value
3781 xattr_fidp->fs.xattr.len = size;
3782 xattr_fidp->fid_type = P9_FID_XATTR;
3783 xattr_fidp->fs.xattr.xattrwalk_fid = true;
3784 xattr_fidp->fs.xattr.value = g_malloc0(size);
3786 err = v9fs_co_lgetxattr(pdu, &xattr_fidp->path,
3787 &name, xattr_fidp->fs.xattr.value,
3788 xattr_fidp->fs.xattr.len);
3790 clunk_fid(s, xattr_fidp->fid);
3794 err = pdu_marshal(pdu, offset, "q", size);
3800 trace_v9fs_xattrwalk_return(pdu->tag, pdu->id, size);
3802 put_fid(pdu, file_fidp);
3804 put_fid(pdu, xattr_fidp);
3807 pdu_complete(pdu, err);
3808 v9fs_string_free(&name);
3811 static void coroutine_fn v9fs_xattrcreate(void *opaque)
3813 int flags, rflags = 0;
3819 V9fsFidState *file_fidp;
3820 V9fsFidState *xattr_fidp;
3821 V9fsPDU *pdu = opaque;
3823 v9fs_string_init(&name);
3824 err = pdu_unmarshal(pdu, offset, "dsqd", &fid, &name, &size, &flags);
3828 trace_v9fs_xattrcreate(pdu->tag, pdu->id, fid, name.data, size, flags);
3830 if (flags & ~(P9_XATTR_CREATE | P9_XATTR_REPLACE)) {
3835 if (flags & P9_XATTR_CREATE) {
3836 rflags |= XATTR_CREATE;
3839 if (flags & P9_XATTR_REPLACE) {
3840 rflags |= XATTR_REPLACE;
3843 if (size > XATTR_SIZE_MAX) {
3848 file_fidp = get_fid(pdu, fid);
3849 if (file_fidp == NULL) {
3853 if (file_fidp->fid_type != P9_FID_NONE) {
3858 /* Make the file fid point to xattr */
3859 xattr_fidp = file_fidp;
3860 xattr_fidp->fid_type = P9_FID_XATTR;
3861 xattr_fidp->fs.xattr.copied_len = 0;
3862 xattr_fidp->fs.xattr.xattrwalk_fid = false;
3863 xattr_fidp->fs.xattr.len = size;
3864 xattr_fidp->fs.xattr.flags = rflags;
3865 v9fs_string_init(&xattr_fidp->fs.xattr.name);
3866 v9fs_string_copy(&xattr_fidp->fs.xattr.name, &name);
3867 xattr_fidp->fs.xattr.value = g_malloc0(size);
3870 put_fid(pdu, file_fidp);
3872 pdu_complete(pdu, err);
3873 v9fs_string_free(&name);
3876 static void coroutine_fn v9fs_readlink(void *opaque)
3878 V9fsPDU *pdu = opaque;
3885 err = pdu_unmarshal(pdu, offset, "d", &fid);
3889 trace_v9fs_readlink(pdu->tag, pdu->id, fid);
3890 fidp = get_fid(pdu, fid);
3896 v9fs_string_init(&target);
3897 err = v9fs_co_readlink(pdu, &fidp->path, &target);
3901 err = pdu_marshal(pdu, offset, "s", &target);
3903 v9fs_string_free(&target);
3907 trace_v9fs_readlink_return(pdu->tag, pdu->id, target.data);
3908 v9fs_string_free(&target);
3912 pdu_complete(pdu, err);
3915 static CoroutineEntry *pdu_co_handlers[] = {
3916 [P9_TREADDIR] = v9fs_readdir,
3917 [P9_TSTATFS] = v9fs_statfs,
3918 [P9_TGETATTR] = v9fs_getattr,
3919 [P9_TSETATTR] = v9fs_setattr,
3920 [P9_TXATTRWALK] = v9fs_xattrwalk,
3921 [P9_TXATTRCREATE] = v9fs_xattrcreate,
3922 [P9_TMKNOD] = v9fs_mknod,
3923 [P9_TRENAME] = v9fs_rename,
3924 [P9_TLOCK] = v9fs_lock,
3925 [P9_TGETLOCK] = v9fs_getlock,
3926 [P9_TRENAMEAT] = v9fs_renameat,
3927 [P9_TREADLINK] = v9fs_readlink,
3928 [P9_TUNLINKAT] = v9fs_unlinkat,
3929 [P9_TMKDIR] = v9fs_mkdir,
3930 [P9_TVERSION] = v9fs_version,
3931 [P9_TLOPEN] = v9fs_open,
3932 [P9_TATTACH] = v9fs_attach,
3933 [P9_TSTAT] = v9fs_stat,
3934 [P9_TWALK] = v9fs_walk,
3935 [P9_TCLUNK] = v9fs_clunk,
3936 [P9_TFSYNC] = v9fs_fsync,
3937 [P9_TOPEN] = v9fs_open,
3938 [P9_TREAD] = v9fs_read,
3940 [P9_TAUTH] = v9fs_auth,
3942 [P9_TFLUSH] = v9fs_flush,
3943 [P9_TLINK] = v9fs_link,
3944 [P9_TSYMLINK] = v9fs_symlink,
3945 [P9_TCREATE] = v9fs_create,
3946 [P9_TLCREATE] = v9fs_lcreate,
3947 [P9_TWRITE] = v9fs_write,
3948 [P9_TWSTAT] = v9fs_wstat,
3949 [P9_TREMOVE] = v9fs_remove,
3952 static void coroutine_fn v9fs_op_not_supp(void *opaque)
3954 V9fsPDU *pdu = opaque;
3955 pdu_complete(pdu, -EOPNOTSUPP);
3958 static void coroutine_fn v9fs_fs_ro(void *opaque)
3960 V9fsPDU *pdu = opaque;
3961 pdu_complete(pdu, -EROFS);
3964 static inline bool is_read_only_op(V9fsPDU *pdu)
3991 void pdu_submit(V9fsPDU *pdu, P9MsgHeader *hdr)
3994 CoroutineEntry *handler;
3995 V9fsState *s = pdu->s;
3997 pdu->size = le32_to_cpu(hdr->size_le);
3999 pdu->tag = le16_to_cpu(hdr->tag_le);
4001 if (pdu->id >= ARRAY_SIZE(pdu_co_handlers) ||
4002 (pdu_co_handlers[pdu->id] == NULL)) {
4003 handler = v9fs_op_not_supp;
4004 } else if (is_ro_export(&s->ctx) && !is_read_only_op(pdu)) {
4005 handler = v9fs_fs_ro;
4007 handler = pdu_co_handlers[pdu->id];
4010 qemu_co_queue_init(&pdu->complete);
4011 co = qemu_coroutine_create(handler, pdu);
4012 qemu_coroutine_enter(co);
4015 /* Returns 0 on success, 1 on failure. */
4016 int v9fs_device_realize_common(V9fsState *s, const V9fsTransport *t,
4026 assert(!s->transport);
4029 /* initialize pdu allocator */
4030 QLIST_INIT(&s->free_list);
4031 QLIST_INIT(&s->active_list);
4032 for (i = 0; i < MAX_REQ; i++) {
4033 QLIST_INSERT_HEAD(&s->free_list, &s->pdus[i], next);
4038 v9fs_path_init(&path);
4040 fse = get_fsdev_fsentry(s->fsconf.fsdev_id);
4043 /* We don't have a fsdev identified by fsdev_id */
4044 error_setg(errp, "9pfs device couldn't find fsdev with the "
4046 s->fsconf.fsdev_id ? s->fsconf.fsdev_id : "NULL");
4050 if (!s->fsconf.tag) {
4051 /* we haven't specified a mount_tag */
4052 error_setg(errp, "fsdev with id %s needs mount_tag arguments",
4053 s->fsconf.fsdev_id);
4057 s->ctx.export_flags = fse->export_flags;
4058 s->ctx.fs_root = g_strdup(fse->path);
4059 s->ctx.exops.get_st_gen = NULL;
4060 len = strlen(s->fsconf.tag);
4061 if (len > MAX_TAG_LEN - 1) {
4062 error_setg(errp, "mount tag '%s' (%d bytes) is longer than "
4063 "maximum (%d bytes)", s->fsconf.tag, len, MAX_TAG_LEN - 1);
4067 s->tag = g_strdup(s->fsconf.tag);
4072 s->ctx.fmode = fse->fmode;
4073 s->ctx.dmode = fse->dmode;
4076 qemu_co_rwlock_init(&s->rename_lock);
4078 if (s->ops->init(&s->ctx, errp) < 0) {
4079 error_prepend(errp, "cannot initialize fsdev '%s': ",
4080 s->fsconf.fsdev_id);
4085 * Check details of export path, We need to use fs driver
4086 * call back to do that. Since we are in the init path, we don't
4087 * use co-routines here.
4089 if (s->ops->name_to_path(&s->ctx, NULL, "/", &path) < 0) {
4091 "error in converting name to path %s", strerror(errno));
4094 if (s->ops->lstat(&s->ctx, &path, &stat)) {
4095 error_setg(errp, "share path %s does not exist", fse->path);
4097 } else if (!S_ISDIR(stat.st_mode)) {
4098 error_setg(errp, "share path %s is not a directory", fse->path);
4102 s->dev_id = stat.st_dev;
4104 /* init inode remapping : */
4105 /* hash table for variable length inode suffixes */
4106 qpd_table_init(&s->qpd_table);
4107 /* hash table for slow/full inode remapping (most users won't need it) */
4108 qpf_table_init(&s->qpf_table);
4109 /* hash table for quick inode remapping */
4110 qpp_table_init(&s->qpp_table);
4112 s->qp_affix_next = 1; /* reserve 0 to detect overflow */
4113 s->qp_fullpath_next = 1;
4115 s->ctx.fst = &fse->fst;
4116 fsdev_throttle_init(s->ctx.fst);
4121 v9fs_device_unrealize_common(s);
4123 v9fs_path_free(&path);
4127 void v9fs_device_unrealize_common(V9fsState *s)
4129 if (s->ops && s->ops->cleanup) {
4130 s->ops->cleanup(&s->ctx);
4133 fsdev_throttle_cleanup(s->ctx.fst);
4136 qp_table_destroy(&s->qpd_table);
4137 qp_table_destroy(&s->qpp_table);
4138 qp_table_destroy(&s->qpf_table);
4139 g_free(s->ctx.fs_root);
4142 typedef struct VirtfsCoResetData {
4145 } VirtfsCoResetData;
4147 static void coroutine_fn virtfs_co_reset(void *opaque)
4149 VirtfsCoResetData *data = opaque;
4151 virtfs_reset(&data->pdu);
4155 void v9fs_reset(V9fsState *s)
4157 VirtfsCoResetData data = { .pdu = { .s = s }, .done = false };
4160 while (!QLIST_EMPTY(&s->active_list)) {
4161 aio_poll(qemu_get_aio_context(), true);
4164 co = qemu_coroutine_create(virtfs_co_reset, &data);
4165 qemu_coroutine_enter(co);
4167 while (!data.done) {
4168 aio_poll(qemu_get_aio_context(), true);
4172 static void __attribute__((__constructor__)) v9fs_set_fd_limit(void)
4175 if (getrlimit(RLIMIT_NOFILE, &rlim) < 0) {
4176 error_report("Failed to get the resource limit");
4179 open_fd_hw = rlim.rlim_cur - MIN(400, rlim.rlim_cur/3);
4180 open_fd_rc = rlim.rlim_cur/2;