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/sockets.h"
21 #include "virtio-9p.h"
22 #include "fsdev/qemu-fsdev.h"
26 #include "migration/blocker.h"
27 #include "sysemu/qtest.h"
31 static int open_fd_rc;
45 static ssize_t pdu_marshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
51 ret = pdu->s->transport->pdu_vmarshal(pdu, offset, fmt, ap);
57 static ssize_t pdu_unmarshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
63 ret = pdu->s->transport->pdu_vunmarshal(pdu, offset, fmt, ap);
69 static int omode_to_uflags(int8_t mode)
103 typedef struct DotlOpenflagMap {
108 static int dotl_to_open_flags(int flags)
112 * We have same bits for P9_DOTL_READONLY, P9_DOTL_WRONLY
113 * and P9_DOTL_NOACCESS
115 int oflags = flags & O_ACCMODE;
117 DotlOpenflagMap dotl_oflag_map[] = {
118 { P9_DOTL_CREATE, O_CREAT },
119 { P9_DOTL_EXCL, O_EXCL },
120 { P9_DOTL_NOCTTY , O_NOCTTY },
121 { P9_DOTL_TRUNC, O_TRUNC },
122 { P9_DOTL_APPEND, O_APPEND },
123 { P9_DOTL_NONBLOCK, O_NONBLOCK } ,
124 { P9_DOTL_DSYNC, O_DSYNC },
125 { P9_DOTL_FASYNC, FASYNC },
126 { P9_DOTL_DIRECT, O_DIRECT },
127 { P9_DOTL_LARGEFILE, O_LARGEFILE },
128 { P9_DOTL_DIRECTORY, O_DIRECTORY },
129 { P9_DOTL_NOFOLLOW, O_NOFOLLOW },
130 { P9_DOTL_NOATIME, O_NOATIME },
131 { P9_DOTL_SYNC, O_SYNC },
134 for (i = 0; i < ARRAY_SIZE(dotl_oflag_map); i++) {
135 if (flags & dotl_oflag_map[i].dotl_flag) {
136 oflags |= dotl_oflag_map[i].open_flag;
143 void cred_init(FsCred *credp)
151 static int get_dotl_openflags(V9fsState *s, int oflags)
155 * Filter the client open flags
157 flags = dotl_to_open_flags(oflags);
158 flags &= ~(O_NOCTTY | O_ASYNC | O_CREAT);
160 * Ignore direct disk access hint until the server supports it.
166 void v9fs_path_init(V9fsPath *path)
172 void v9fs_path_free(V9fsPath *path)
180 void GCC_FMT_ATTR(2, 3)
181 v9fs_path_sprintf(V9fsPath *path, const char *fmt, ...)
185 v9fs_path_free(path);
188 /* Bump the size for including terminating NULL */
189 path->size = g_vasprintf(&path->data, fmt, ap) + 1;
193 void v9fs_path_copy(V9fsPath *dst, const V9fsPath *src)
196 dst->size = src->size;
197 dst->data = g_memdup(src->data, src->size);
200 int v9fs_name_to_path(V9fsState *s, V9fsPath *dirpath,
201 const char *name, V9fsPath *path)
204 err = s->ops->name_to_path(&s->ctx, dirpath, name, path);
212 * Return TRUE if s1 is an ancestor of s2.
214 * E.g. "a/b" is an ancestor of "a/b/c" but not of "a/bc/d".
215 * As a special case, We treat s1 as ancestor of s2 if they are same!
217 static int v9fs_path_is_ancestor(V9fsPath *s1, V9fsPath *s2)
219 if (!strncmp(s1->data, s2->data, s1->size - 1)) {
220 if (s2->data[s1->size - 1] == '\0' || s2->data[s1->size - 1] == '/') {
227 static size_t v9fs_string_size(V9fsString *str)
233 * returns 0 if fid got re-opened, 1 if not, < 0 on error */
234 static int coroutine_fn v9fs_reopen_fid(V9fsPDU *pdu, V9fsFidState *f)
237 if (f->fid_type == P9_FID_FILE) {
238 if (f->fs.fd == -1) {
240 err = v9fs_co_open(pdu, f, f->open_flags);
241 } while (err == -EINTR && !pdu->cancelled);
243 } else if (f->fid_type == P9_FID_DIR) {
244 if (f->fs.dir.stream == NULL) {
246 err = v9fs_co_opendir(pdu, f);
247 } while (err == -EINTR && !pdu->cancelled);
253 static V9fsFidState *coroutine_fn get_fid(V9fsPDU *pdu, int32_t fid)
257 V9fsState *s = pdu->s;
259 for (f = s->fid_list; f; f = f->next) {
263 * Update the fid ref upfront so that
264 * we don't get reclaimed when we yield
269 * check whether we need to reopen the
270 * file. We might have closed the fd
271 * while trying to free up some file
274 err = v9fs_reopen_fid(pdu, f);
280 * Mark the fid as referenced so that the LRU
281 * reclaim won't close the file descriptor
283 f->flags |= FID_REFERENCED;
290 static V9fsFidState *alloc_fid(V9fsState *s, int32_t fid)
294 for (f = s->fid_list; f; f = f->next) {
295 /* If fid is already there return NULL */
301 f = g_malloc0(sizeof(V9fsFidState));
303 f->fid_type = P9_FID_NONE;
306 * Mark the fid as referenced so that the LRU
307 * reclaim won't close the file descriptor
309 f->flags |= FID_REFERENCED;
310 f->next = s->fid_list;
313 v9fs_readdir_init(&f->fs.dir);
314 v9fs_readdir_init(&f->fs_reclaim.dir);
319 static int coroutine_fn v9fs_xattr_fid_clunk(V9fsPDU *pdu, V9fsFidState *fidp)
323 if (fidp->fs.xattr.xattrwalk_fid) {
324 /* getxattr/listxattr fid */
328 * if this is fid for setxattr. clunk should
329 * result in setxattr localcall
331 if (fidp->fs.xattr.len != fidp->fs.xattr.copied_len) {
332 /* clunk after partial write */
336 if (fidp->fs.xattr.len) {
337 retval = v9fs_co_lsetxattr(pdu, &fidp->path, &fidp->fs.xattr.name,
338 fidp->fs.xattr.value,
340 fidp->fs.xattr.flags);
342 retval = v9fs_co_lremovexattr(pdu, &fidp->path, &fidp->fs.xattr.name);
345 v9fs_string_free(&fidp->fs.xattr.name);
347 g_free(fidp->fs.xattr.value);
351 static int coroutine_fn free_fid(V9fsPDU *pdu, V9fsFidState *fidp)
355 if (fidp->fid_type == P9_FID_FILE) {
356 /* If we reclaimed the fd no need to close */
357 if (fidp->fs.fd != -1) {
358 retval = v9fs_co_close(pdu, &fidp->fs);
360 } else if (fidp->fid_type == P9_FID_DIR) {
361 if (fidp->fs.dir.stream != NULL) {
362 retval = v9fs_co_closedir(pdu, &fidp->fs);
364 } else if (fidp->fid_type == P9_FID_XATTR) {
365 retval = v9fs_xattr_fid_clunk(pdu, fidp);
367 v9fs_path_free(&fidp->path);
372 static int coroutine_fn put_fid(V9fsPDU *pdu, V9fsFidState *fidp)
377 * Don't free the fid if it is in reclaim list
379 if (!fidp->ref && fidp->clunked) {
380 if (fidp->fid == pdu->s->root_fid) {
382 * if the clunked fid is root fid then we
383 * have unmounted the fs on the client side.
384 * delete the migration blocker. Ideally, this
385 * should be hooked to transport close notification
387 if (pdu->s->migration_blocker) {
388 migrate_del_blocker(pdu->s->migration_blocker);
389 error_free(pdu->s->migration_blocker);
390 pdu->s->migration_blocker = NULL;
393 return free_fid(pdu, fidp);
398 static V9fsFidState *clunk_fid(V9fsState *s, int32_t fid)
400 V9fsFidState **fidpp, *fidp;
402 for (fidpp = &s->fid_list; *fidpp; fidpp = &(*fidpp)->next) {
403 if ((*fidpp)->fid == fid) {
407 if (*fidpp == NULL) {
416 void coroutine_fn v9fs_reclaim_fd(V9fsPDU *pdu)
418 int reclaim_count = 0;
419 V9fsState *s = pdu->s;
420 V9fsFidState *f, *reclaim_list = NULL;
422 for (f = s->fid_list; f; f = f->next) {
424 * Unlink fids cannot be reclaimed. Check
425 * for them and skip them. Also skip fids
426 * currently being operated on.
428 if (f->ref || f->flags & FID_NON_RECLAIMABLE) {
432 * if it is a recently referenced fid
433 * we leave the fid untouched and clear the
434 * reference bit. We come back to it later
435 * in the next iteration. (a simple LRU without
436 * moving list elements around)
438 if (f->flags & FID_REFERENCED) {
439 f->flags &= ~FID_REFERENCED;
443 * Add fids to reclaim list.
445 if (f->fid_type == P9_FID_FILE) {
446 if (f->fs.fd != -1) {
448 * Up the reference count so that
449 * a clunk request won't free this fid
452 f->rclm_lst = reclaim_list;
454 f->fs_reclaim.fd = f->fs.fd;
458 } else if (f->fid_type == P9_FID_DIR) {
459 if (f->fs.dir.stream != NULL) {
461 * Up the reference count so that
462 * a clunk request won't free this fid
465 f->rclm_lst = reclaim_list;
467 f->fs_reclaim.dir.stream = f->fs.dir.stream;
468 f->fs.dir.stream = NULL;
472 if (reclaim_count >= open_fd_rc) {
477 * Now close the fid in reclaim list. Free them if they
478 * are already clunked.
480 while (reclaim_list) {
482 reclaim_list = f->rclm_lst;
483 if (f->fid_type == P9_FID_FILE) {
484 v9fs_co_close(pdu, &f->fs_reclaim);
485 } else if (f->fid_type == P9_FID_DIR) {
486 v9fs_co_closedir(pdu, &f->fs_reclaim);
490 * Now drop the fid reference, free it
497 static int coroutine_fn v9fs_mark_fids_unreclaim(V9fsPDU *pdu, V9fsPath *path)
500 V9fsState *s = pdu->s;
501 V9fsFidState *fidp, head_fid;
503 head_fid.next = s->fid_list;
504 for (fidp = s->fid_list; fidp; fidp = fidp->next) {
505 if (fidp->path.size != path->size) {
508 if (!memcmp(fidp->path.data, path->data, path->size)) {
509 /* Mark the fid non reclaimable. */
510 fidp->flags |= FID_NON_RECLAIMABLE;
512 /* reopen the file/dir if already closed */
513 err = v9fs_reopen_fid(pdu, fidp);
518 * Go back to head of fid list because
519 * the list could have got updated when
520 * switched to the worker thread
530 static void coroutine_fn virtfs_reset(V9fsPDU *pdu)
532 V9fsState *s = pdu->s;
536 while (s->fid_list) {
542 s->fid_list = fidp->next;
549 #define P9_QID_TYPE_DIR 0x80
550 #define P9_QID_TYPE_SYMLINK 0x02
552 #define P9_STAT_MODE_DIR 0x80000000
553 #define P9_STAT_MODE_APPEND 0x40000000
554 #define P9_STAT_MODE_EXCL 0x20000000
555 #define P9_STAT_MODE_MOUNT 0x10000000
556 #define P9_STAT_MODE_AUTH 0x08000000
557 #define P9_STAT_MODE_TMP 0x04000000
558 #define P9_STAT_MODE_SYMLINK 0x02000000
559 #define P9_STAT_MODE_LINK 0x01000000
560 #define P9_STAT_MODE_DEVICE 0x00800000
561 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
562 #define P9_STAT_MODE_SOCKET 0x00100000
563 #define P9_STAT_MODE_SETUID 0x00080000
564 #define P9_STAT_MODE_SETGID 0x00040000
565 #define P9_STAT_MODE_SETVTX 0x00010000
567 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR | \
568 P9_STAT_MODE_SYMLINK | \
569 P9_STAT_MODE_LINK | \
570 P9_STAT_MODE_DEVICE | \
571 P9_STAT_MODE_NAMED_PIPE | \
574 /* This is the algorithm from ufs in spfs */
575 static void stat_to_qid(const struct stat *stbuf, V9fsQID *qidp)
579 memset(&qidp->path, 0, sizeof(qidp->path));
580 size = MIN(sizeof(stbuf->st_ino), sizeof(qidp->path));
581 memcpy(&qidp->path, &stbuf->st_ino, size);
582 qidp->version = stbuf->st_mtime ^ (stbuf->st_size << 8);
584 if (S_ISDIR(stbuf->st_mode)) {
585 qidp->type |= P9_QID_TYPE_DIR;
587 if (S_ISLNK(stbuf->st_mode)) {
588 qidp->type |= P9_QID_TYPE_SYMLINK;
592 static int coroutine_fn fid_to_qid(V9fsPDU *pdu, V9fsFidState *fidp,
598 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
602 stat_to_qid(&stbuf, qidp);
606 V9fsPDU *pdu_alloc(V9fsState *s)
610 if (!QLIST_EMPTY(&s->free_list)) {
611 pdu = QLIST_FIRST(&s->free_list);
612 QLIST_REMOVE(pdu, next);
613 QLIST_INSERT_HEAD(&s->active_list, pdu, next);
618 void pdu_free(V9fsPDU *pdu)
620 V9fsState *s = pdu->s;
622 g_assert(!pdu->cancelled);
623 QLIST_REMOVE(pdu, next);
624 QLIST_INSERT_HEAD(&s->free_list, pdu, next);
627 static void coroutine_fn pdu_complete(V9fsPDU *pdu, ssize_t len)
629 int8_t id = pdu->id + 1; /* Response */
630 V9fsState *s = pdu->s;
634 * The 9p spec requires that successfully cancelled pdus receive no reply.
635 * Sending a reply would confuse clients because they would
636 * assume that any EINTR is the actual result of the operation,
637 * rather than a consequence of the cancellation. However, if
638 * the operation completed (succesfully or with an error other
639 * than caused be cancellation), we do send out that reply, both
640 * for efficiency and to avoid confusing the rest of the state machine
641 * that assumes passing a non-error here will mean a successful
642 * transmission of the reply.
644 bool discard = pdu->cancelled && len == -EINTR;
646 trace_v9fs_rcancel(pdu->tag, pdu->id);
655 if (s->proto_version != V9FS_PROTO_2000L) {
658 str.data = strerror(err);
659 str.size = strlen(str.data);
661 ret = pdu_marshal(pdu, len, "s", &str);
669 ret = pdu_marshal(pdu, len, "d", err);
675 if (s->proto_version == V9FS_PROTO_2000L) {
678 trace_v9fs_rerror(pdu->tag, pdu->id, err); /* Trace ERROR */
681 /* fill out the header */
682 if (pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag) < 0) {
686 /* keep these in sync */
691 pdu->s->transport->push_and_notify(pdu);
693 /* Now wakeup anybody waiting in flush for this request */
694 if (!qemu_co_queue_next(&pdu->complete)) {
699 static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
704 if (mode & P9_STAT_MODE_DIR) {
708 if (mode & P9_STAT_MODE_SYMLINK) {
711 if (mode & P9_STAT_MODE_SOCKET) {
714 if (mode & P9_STAT_MODE_NAMED_PIPE) {
717 if (mode & P9_STAT_MODE_DEVICE) {
718 if (extension->size && extension->data[0] == 'c') {
729 if (mode & P9_STAT_MODE_SETUID) {
732 if (mode & P9_STAT_MODE_SETGID) {
735 if (mode & P9_STAT_MODE_SETVTX) {
742 static int donttouch_stat(V9fsStat *stat)
744 if (stat->type == -1 &&
746 stat->qid.type == -1 &&
747 stat->qid.version == -1 &&
748 stat->qid.path == -1 &&
752 stat->length == -1 &&
759 stat->n_muid == -1) {
766 static void v9fs_stat_init(V9fsStat *stat)
768 v9fs_string_init(&stat->name);
769 v9fs_string_init(&stat->uid);
770 v9fs_string_init(&stat->gid);
771 v9fs_string_init(&stat->muid);
772 v9fs_string_init(&stat->extension);
775 static void v9fs_stat_free(V9fsStat *stat)
777 v9fs_string_free(&stat->name);
778 v9fs_string_free(&stat->uid);
779 v9fs_string_free(&stat->gid);
780 v9fs_string_free(&stat->muid);
781 v9fs_string_free(&stat->extension);
784 static uint32_t stat_to_v9mode(const struct stat *stbuf)
788 mode = stbuf->st_mode & 0777;
789 if (S_ISDIR(stbuf->st_mode)) {
790 mode |= P9_STAT_MODE_DIR;
793 if (S_ISLNK(stbuf->st_mode)) {
794 mode |= P9_STAT_MODE_SYMLINK;
797 if (S_ISSOCK(stbuf->st_mode)) {
798 mode |= P9_STAT_MODE_SOCKET;
801 if (S_ISFIFO(stbuf->st_mode)) {
802 mode |= P9_STAT_MODE_NAMED_PIPE;
805 if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
806 mode |= P9_STAT_MODE_DEVICE;
809 if (stbuf->st_mode & S_ISUID) {
810 mode |= P9_STAT_MODE_SETUID;
813 if (stbuf->st_mode & S_ISGID) {
814 mode |= P9_STAT_MODE_SETGID;
817 if (stbuf->st_mode & S_ISVTX) {
818 mode |= P9_STAT_MODE_SETVTX;
824 static int coroutine_fn stat_to_v9stat(V9fsPDU *pdu, V9fsPath *path,
825 const char *basename,
826 const struct stat *stbuf,
831 memset(v9stat, 0, sizeof(*v9stat));
833 stat_to_qid(stbuf, &v9stat->qid);
834 v9stat->mode = stat_to_v9mode(stbuf);
835 v9stat->atime = stbuf->st_atime;
836 v9stat->mtime = stbuf->st_mtime;
837 v9stat->length = stbuf->st_size;
839 v9fs_string_free(&v9stat->uid);
840 v9fs_string_free(&v9stat->gid);
841 v9fs_string_free(&v9stat->muid);
843 v9stat->n_uid = stbuf->st_uid;
844 v9stat->n_gid = stbuf->st_gid;
847 v9fs_string_free(&v9stat->extension);
849 if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
850 err = v9fs_co_readlink(pdu, path, &v9stat->extension);
854 } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
855 v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
856 S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
857 major(stbuf->st_rdev), minor(stbuf->st_rdev));
858 } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
859 v9fs_string_sprintf(&v9stat->extension, "%s %lu",
860 "HARDLINKCOUNT", (unsigned long)stbuf->st_nlink);
863 v9fs_string_sprintf(&v9stat->name, "%s", basename);
866 v9fs_string_size(&v9stat->name) +
867 v9fs_string_size(&v9stat->uid) +
868 v9fs_string_size(&v9stat->gid) +
869 v9fs_string_size(&v9stat->muid) +
870 v9fs_string_size(&v9stat->extension);
874 #define P9_STATS_MODE 0x00000001ULL
875 #define P9_STATS_NLINK 0x00000002ULL
876 #define P9_STATS_UID 0x00000004ULL
877 #define P9_STATS_GID 0x00000008ULL
878 #define P9_STATS_RDEV 0x00000010ULL
879 #define P9_STATS_ATIME 0x00000020ULL
880 #define P9_STATS_MTIME 0x00000040ULL
881 #define P9_STATS_CTIME 0x00000080ULL
882 #define P9_STATS_INO 0x00000100ULL
883 #define P9_STATS_SIZE 0x00000200ULL
884 #define P9_STATS_BLOCKS 0x00000400ULL
886 #define P9_STATS_BTIME 0x00000800ULL
887 #define P9_STATS_GEN 0x00001000ULL
888 #define P9_STATS_DATA_VERSION 0x00002000ULL
890 #define P9_STATS_BASIC 0x000007ffULL /* Mask for fields up to BLOCKS */
891 #define P9_STATS_ALL 0x00003fffULL /* Mask for All fields above */
894 static void stat_to_v9stat_dotl(V9fsState *s, const struct stat *stbuf,
895 V9fsStatDotl *v9lstat)
897 memset(v9lstat, 0, sizeof(*v9lstat));
899 v9lstat->st_mode = stbuf->st_mode;
900 v9lstat->st_nlink = stbuf->st_nlink;
901 v9lstat->st_uid = stbuf->st_uid;
902 v9lstat->st_gid = stbuf->st_gid;
903 v9lstat->st_rdev = stbuf->st_rdev;
904 v9lstat->st_size = stbuf->st_size;
905 v9lstat->st_blksize = stbuf->st_blksize;
906 v9lstat->st_blocks = stbuf->st_blocks;
907 v9lstat->st_atime_sec = stbuf->st_atime;
908 v9lstat->st_atime_nsec = stbuf->st_atim.tv_nsec;
909 v9lstat->st_mtime_sec = stbuf->st_mtime;
910 v9lstat->st_mtime_nsec = stbuf->st_mtim.tv_nsec;
911 v9lstat->st_ctime_sec = stbuf->st_ctime;
912 v9lstat->st_ctime_nsec = stbuf->st_ctim.tv_nsec;
913 /* Currently we only support BASIC fields in stat */
914 v9lstat->st_result_mask = P9_STATS_BASIC;
916 stat_to_qid(stbuf, &v9lstat->qid);
919 static void print_sg(struct iovec *sg, int cnt)
923 printf("sg[%d]: {", cnt);
924 for (i = 0; i < cnt; i++) {
928 printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
933 /* Will call this only for path name based fid */
934 static void v9fs_fix_path(V9fsPath *dst, V9fsPath *src, int len)
937 v9fs_path_init(&str);
938 v9fs_path_copy(&str, dst);
939 v9fs_path_sprintf(dst, "%s%s", src->data, str.data + len);
940 v9fs_path_free(&str);
943 static inline bool is_ro_export(FsContext *ctx)
945 return ctx->export_flags & V9FS_RDONLY;
948 static void coroutine_fn v9fs_version(void *opaque)
951 V9fsPDU *pdu = opaque;
952 V9fsState *s = pdu->s;
956 v9fs_string_init(&version);
957 err = pdu_unmarshal(pdu, offset, "ds", &s->msize, &version);
961 trace_v9fs_version(pdu->tag, pdu->id, s->msize, version.data);
965 if (!strcmp(version.data, "9P2000.u")) {
966 s->proto_version = V9FS_PROTO_2000U;
967 } else if (!strcmp(version.data, "9P2000.L")) {
968 s->proto_version = V9FS_PROTO_2000L;
970 v9fs_string_sprintf(&version, "unknown");
973 err = pdu_marshal(pdu, offset, "ds", s->msize, &version);
978 trace_v9fs_version_return(pdu->tag, pdu->id, s->msize, version.data);
980 pdu_complete(pdu, err);
981 v9fs_string_free(&version);
984 static void coroutine_fn v9fs_attach(void *opaque)
986 V9fsPDU *pdu = opaque;
987 V9fsState *s = pdu->s;
988 int32_t fid, afid, n_uname;
989 V9fsString uname, aname;
994 Error *local_err = NULL;
996 v9fs_string_init(&uname);
997 v9fs_string_init(&aname);
998 err = pdu_unmarshal(pdu, offset, "ddssd", &fid,
999 &afid, &uname, &aname, &n_uname);
1003 trace_v9fs_attach(pdu->tag, pdu->id, fid, afid, uname.data, aname.data);
1005 fidp = alloc_fid(s, fid);
1010 fidp->uid = n_uname;
1011 err = v9fs_co_name_to_path(pdu, NULL, "/", &fidp->path);
1017 err = fid_to_qid(pdu, fidp, &qid);
1025 * disable migration if we haven't done already.
1026 * attach could get called multiple times for the same export.
1028 if (!s->migration_blocker) {
1029 error_setg(&s->migration_blocker,
1030 "Migration is disabled when VirtFS export path '%s' is mounted in the guest using mount_tag '%s'",
1031 s->ctx.fs_root ? s->ctx.fs_root : "NULL", s->tag);
1032 err = migrate_add_blocker(s->migration_blocker, &local_err);
1034 error_free(local_err);
1035 error_free(s->migration_blocker);
1036 s->migration_blocker = NULL;
1043 err = pdu_marshal(pdu, offset, "Q", &qid);
1050 memcpy(&s->root_qid, &qid, sizeof(qid));
1051 trace_v9fs_attach_return(pdu->tag, pdu->id,
1052 qid.type, qid.version, qid.path);
1056 pdu_complete(pdu, err);
1057 v9fs_string_free(&uname);
1058 v9fs_string_free(&aname);
1061 static void coroutine_fn v9fs_stat(void *opaque)
1069 V9fsPDU *pdu = opaque;
1072 err = pdu_unmarshal(pdu, offset, "d", &fid);
1076 trace_v9fs_stat(pdu->tag, pdu->id, fid);
1078 fidp = get_fid(pdu, fid);
1083 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1087 basename = g_path_get_basename(fidp->path.data);
1088 err = stat_to_v9stat(pdu, &fidp->path, basename, &stbuf, &v9stat);
1093 err = pdu_marshal(pdu, offset, "wS", 0, &v9stat);
1095 v9fs_stat_free(&v9stat);
1098 trace_v9fs_stat_return(pdu->tag, pdu->id, v9stat.mode,
1099 v9stat.atime, v9stat.mtime, v9stat.length);
1101 v9fs_stat_free(&v9stat);
1105 pdu_complete(pdu, err);
1108 static void coroutine_fn v9fs_getattr(void *opaque)
1115 uint64_t request_mask;
1116 V9fsStatDotl v9stat_dotl;
1117 V9fsPDU *pdu = opaque;
1118 V9fsState *s = pdu->s;
1120 retval = pdu_unmarshal(pdu, offset, "dq", &fid, &request_mask);
1124 trace_v9fs_getattr(pdu->tag, pdu->id, fid, request_mask);
1126 fidp = get_fid(pdu, fid);
1132 * Currently we only support BASIC fields in stat, so there is no
1133 * need to look at request_mask.
1135 retval = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1139 stat_to_v9stat_dotl(s, &stbuf, &v9stat_dotl);
1141 /* fill st_gen if requested and supported by underlying fs */
1142 if (request_mask & P9_STATS_GEN) {
1143 retval = v9fs_co_st_gen(pdu, &fidp->path, stbuf.st_mode, &v9stat_dotl);
1146 /* we have valid st_gen: update result mask */
1147 v9stat_dotl.st_result_mask |= P9_STATS_GEN;
1150 /* request cancelled, e.g. by Tflush */
1153 /* failed to get st_gen: not fatal, ignore */
1157 retval = pdu_marshal(pdu, offset, "A", &v9stat_dotl);
1162 trace_v9fs_getattr_return(pdu->tag, pdu->id, v9stat_dotl.st_result_mask,
1163 v9stat_dotl.st_mode, v9stat_dotl.st_uid,
1164 v9stat_dotl.st_gid);
1168 pdu_complete(pdu, retval);
1171 /* Attribute flags */
1172 #define P9_ATTR_MODE (1 << 0)
1173 #define P9_ATTR_UID (1 << 1)
1174 #define P9_ATTR_GID (1 << 2)
1175 #define P9_ATTR_SIZE (1 << 3)
1176 #define P9_ATTR_ATIME (1 << 4)
1177 #define P9_ATTR_MTIME (1 << 5)
1178 #define P9_ATTR_CTIME (1 << 6)
1179 #define P9_ATTR_ATIME_SET (1 << 7)
1180 #define P9_ATTR_MTIME_SET (1 << 8)
1182 #define P9_ATTR_MASK 127
1184 static void coroutine_fn v9fs_setattr(void *opaque)
1191 V9fsPDU *pdu = opaque;
1193 err = pdu_unmarshal(pdu, offset, "dI", &fid, &v9iattr);
1198 trace_v9fs_setattr(pdu->tag, pdu->id, fid,
1199 v9iattr.valid, v9iattr.mode, v9iattr.uid, v9iattr.gid,
1200 v9iattr.size, v9iattr.atime_sec, v9iattr.mtime_sec);
1202 fidp = get_fid(pdu, fid);
1207 if (v9iattr.valid & P9_ATTR_MODE) {
1208 err = v9fs_co_chmod(pdu, &fidp->path, v9iattr.mode);
1213 if (v9iattr.valid & (P9_ATTR_ATIME | P9_ATTR_MTIME)) {
1214 struct timespec times[2];
1215 if (v9iattr.valid & P9_ATTR_ATIME) {
1216 if (v9iattr.valid & P9_ATTR_ATIME_SET) {
1217 times[0].tv_sec = v9iattr.atime_sec;
1218 times[0].tv_nsec = v9iattr.atime_nsec;
1220 times[0].tv_nsec = UTIME_NOW;
1223 times[0].tv_nsec = UTIME_OMIT;
1225 if (v9iattr.valid & P9_ATTR_MTIME) {
1226 if (v9iattr.valid & P9_ATTR_MTIME_SET) {
1227 times[1].tv_sec = v9iattr.mtime_sec;
1228 times[1].tv_nsec = v9iattr.mtime_nsec;
1230 times[1].tv_nsec = UTIME_NOW;
1233 times[1].tv_nsec = UTIME_OMIT;
1235 err = v9fs_co_utimensat(pdu, &fidp->path, times);
1241 * If the only valid entry in iattr is ctime we can call
1242 * chown(-1,-1) to update the ctime of the file
1244 if ((v9iattr.valid & (P9_ATTR_UID | P9_ATTR_GID)) ||
1245 ((v9iattr.valid & P9_ATTR_CTIME)
1246 && !((v9iattr.valid & P9_ATTR_MASK) & ~P9_ATTR_CTIME))) {
1247 if (!(v9iattr.valid & P9_ATTR_UID)) {
1250 if (!(v9iattr.valid & P9_ATTR_GID)) {
1253 err = v9fs_co_chown(pdu, &fidp->path, v9iattr.uid,
1259 if (v9iattr.valid & (P9_ATTR_SIZE)) {
1260 err = v9fs_co_truncate(pdu, &fidp->path, v9iattr.size);
1266 trace_v9fs_setattr_return(pdu->tag, pdu->id);
1270 pdu_complete(pdu, err);
1273 static int v9fs_walk_marshal(V9fsPDU *pdu, uint16_t nwnames, V9fsQID *qids)
1279 err = pdu_marshal(pdu, offset, "w", nwnames);
1284 for (i = 0; i < nwnames; i++) {
1285 err = pdu_marshal(pdu, offset, "Q", &qids[i]);
1294 static bool name_is_illegal(const char *name)
1296 return !*name || strchr(name, '/') != NULL;
1299 static bool not_same_qid(const V9fsQID *qid1, const V9fsQID *qid2)
1302 qid1->type != qid2->type ||
1303 qid1->version != qid2->version ||
1304 qid1->path != qid2->path;
1307 static void coroutine_fn v9fs_walk(void *opaque)
1310 V9fsQID *qids = NULL;
1312 V9fsPath dpath, path;
1316 int32_t fid, newfid;
1317 V9fsString *wnames = NULL;
1319 V9fsFidState *newfidp = NULL;
1320 V9fsPDU *pdu = opaque;
1321 V9fsState *s = pdu->s;
1324 err = pdu_unmarshal(pdu, offset, "ddw", &fid, &newfid, &nwnames);
1326 pdu_complete(pdu, err);
1331 trace_v9fs_walk(pdu->tag, pdu->id, fid, newfid, nwnames);
1333 if (nwnames && nwnames <= P9_MAXWELEM) {
1334 wnames = g_malloc0(sizeof(wnames[0]) * nwnames);
1335 qids = g_malloc0(sizeof(qids[0]) * nwnames);
1336 for (i = 0; i < nwnames; i++) {
1337 err = pdu_unmarshal(pdu, offset, "s", &wnames[i]);
1341 if (name_is_illegal(wnames[i].data)) {
1347 } else if (nwnames > P9_MAXWELEM) {
1351 fidp = get_fid(pdu, fid);
1357 v9fs_path_init(&dpath);
1358 v9fs_path_init(&path);
1360 err = fid_to_qid(pdu, fidp, &qid);
1366 * Both dpath and path initially poin to fidp.
1367 * Needed to handle request with nwnames == 0
1369 v9fs_path_copy(&dpath, &fidp->path);
1370 v9fs_path_copy(&path, &fidp->path);
1371 for (name_idx = 0; name_idx < nwnames; name_idx++) {
1372 if (not_same_qid(&pdu->s->root_qid, &qid) ||
1373 strcmp("..", wnames[name_idx].data)) {
1374 err = v9fs_co_name_to_path(pdu, &dpath, wnames[name_idx].data,
1380 err = v9fs_co_lstat(pdu, &path, &stbuf);
1384 stat_to_qid(&stbuf, &qid);
1385 v9fs_path_copy(&dpath, &path);
1387 memcpy(&qids[name_idx], &qid, sizeof(qid));
1389 if (fid == newfid) {
1390 if (fidp->fid_type != P9_FID_NONE) {
1394 v9fs_path_write_lock(s);
1395 v9fs_path_copy(&fidp->path, &path);
1396 v9fs_path_unlock(s);
1398 newfidp = alloc_fid(s, newfid);
1399 if (newfidp == NULL) {
1403 newfidp->uid = fidp->uid;
1404 v9fs_path_copy(&newfidp->path, &path);
1406 err = v9fs_walk_marshal(pdu, nwnames, qids);
1407 trace_v9fs_walk_return(pdu->tag, pdu->id, nwnames, qids);
1411 put_fid(pdu, newfidp);
1413 v9fs_path_free(&dpath);
1414 v9fs_path_free(&path);
1416 pdu_complete(pdu, err);
1417 if (nwnames && nwnames <= P9_MAXWELEM) {
1418 for (name_idx = 0; name_idx < nwnames; name_idx++) {
1419 v9fs_string_free(&wnames[name_idx]);
1426 static int32_t coroutine_fn get_iounit(V9fsPDU *pdu, V9fsPath *path)
1428 struct statfs stbuf;
1430 V9fsState *s = pdu->s;
1433 * iounit should be multiples of f_bsize (host filesystem block size
1434 * and as well as less than (client msize - P9_IOHDRSZ))
1436 if (!v9fs_co_statfs(pdu, path, &stbuf)) {
1437 iounit = stbuf.f_bsize;
1438 iounit *= (s->msize - P9_IOHDRSZ)/stbuf.f_bsize;
1441 iounit = s->msize - P9_IOHDRSZ;
1446 static void coroutine_fn v9fs_open(void *opaque)
1457 V9fsPDU *pdu = opaque;
1458 V9fsState *s = pdu->s;
1460 if (s->proto_version == V9FS_PROTO_2000L) {
1461 err = pdu_unmarshal(pdu, offset, "dd", &fid, &mode);
1464 err = pdu_unmarshal(pdu, offset, "db", &fid, &modebyte);
1470 trace_v9fs_open(pdu->tag, pdu->id, fid, mode);
1472 fidp = get_fid(pdu, fid);
1477 if (fidp->fid_type != P9_FID_NONE) {
1482 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1486 stat_to_qid(&stbuf, &qid);
1487 if (S_ISDIR(stbuf.st_mode)) {
1488 err = v9fs_co_opendir(pdu, fidp);
1492 fidp->fid_type = P9_FID_DIR;
1493 err = pdu_marshal(pdu, offset, "Qd", &qid, 0);
1499 if (s->proto_version == V9FS_PROTO_2000L) {
1500 flags = get_dotl_openflags(s, mode);
1502 flags = omode_to_uflags(mode);
1504 if (is_ro_export(&s->ctx)) {
1505 if (mode & O_WRONLY || mode & O_RDWR ||
1506 mode & O_APPEND || mode & O_TRUNC) {
1511 err = v9fs_co_open(pdu, fidp, flags);
1515 fidp->fid_type = P9_FID_FILE;
1516 fidp->open_flags = flags;
1517 if (flags & O_EXCL) {
1519 * We let the host file system do O_EXCL check
1520 * We should not reclaim such fd
1522 fidp->flags |= FID_NON_RECLAIMABLE;
1524 iounit = get_iounit(pdu, &fidp->path);
1525 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1531 trace_v9fs_open_return(pdu->tag, pdu->id,
1532 qid.type, qid.version, qid.path, iounit);
1536 pdu_complete(pdu, err);
1539 static void coroutine_fn v9fs_lcreate(void *opaque)
1541 int32_t dfid, flags, mode;
1550 V9fsPDU *pdu = opaque;
1552 v9fs_string_init(&name);
1553 err = pdu_unmarshal(pdu, offset, "dsddd", &dfid,
1554 &name, &flags, &mode, &gid);
1558 trace_v9fs_lcreate(pdu->tag, pdu->id, dfid, flags, mode, gid);
1560 if (name_is_illegal(name.data)) {
1565 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
1570 fidp = get_fid(pdu, dfid);
1575 if (fidp->fid_type != P9_FID_NONE) {
1580 flags = get_dotl_openflags(pdu->s, flags);
1581 err = v9fs_co_open2(pdu, fidp, &name, gid,
1582 flags | O_CREAT, mode, &stbuf);
1586 fidp->fid_type = P9_FID_FILE;
1587 fidp->open_flags = flags;
1588 if (flags & O_EXCL) {
1590 * We let the host file system do O_EXCL check
1591 * We should not reclaim such fd
1593 fidp->flags |= FID_NON_RECLAIMABLE;
1595 iounit = get_iounit(pdu, &fidp->path);
1596 stat_to_qid(&stbuf, &qid);
1597 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1602 trace_v9fs_lcreate_return(pdu->tag, pdu->id,
1603 qid.type, qid.version, qid.path, iounit);
1607 pdu_complete(pdu, err);
1608 v9fs_string_free(&name);
1611 static void coroutine_fn v9fs_fsync(void *opaque)
1618 V9fsPDU *pdu = opaque;
1620 err = pdu_unmarshal(pdu, offset, "dd", &fid, &datasync);
1624 trace_v9fs_fsync(pdu->tag, pdu->id, fid, datasync);
1626 fidp = get_fid(pdu, fid);
1631 err = v9fs_co_fsync(pdu, fidp, datasync);
1637 pdu_complete(pdu, err);
1640 static void coroutine_fn v9fs_clunk(void *opaque)
1646 V9fsPDU *pdu = opaque;
1647 V9fsState *s = pdu->s;
1649 err = pdu_unmarshal(pdu, offset, "d", &fid);
1653 trace_v9fs_clunk(pdu->tag, pdu->id, fid);
1655 fidp = clunk_fid(s, fid);
1661 * Bump the ref so that put_fid will
1665 err = put_fid(pdu, fidp);
1670 pdu_complete(pdu, err);
1674 * Create a QEMUIOVector for a sub-region of PDU iovecs
1676 * @qiov: uninitialized QEMUIOVector
1677 * @skip: number of bytes to skip from beginning of PDU
1678 * @size: number of bytes to include
1679 * @is_write: true - write, false - read
1681 * The resulting QEMUIOVector has heap-allocated iovecs and must be cleaned up
1682 * with qemu_iovec_destroy().
1684 static void v9fs_init_qiov_from_pdu(QEMUIOVector *qiov, V9fsPDU *pdu,
1685 size_t skip, size_t size,
1693 pdu->s->transport->init_out_iov_from_pdu(pdu, &iov, &niov, size + skip);
1695 pdu->s->transport->init_in_iov_from_pdu(pdu, &iov, &niov, size + skip);
1698 qemu_iovec_init_external(&elem, iov, niov);
1699 qemu_iovec_init(qiov, niov);
1700 qemu_iovec_concat(qiov, &elem, skip, size);
1703 static int v9fs_xattr_read(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
1704 uint64_t off, uint32_t max_count)
1708 uint64_t read_count;
1709 QEMUIOVector qiov_full;
1711 if (fidp->fs.xattr.len < off) {
1714 read_count = fidp->fs.xattr.len - off;
1716 if (read_count > max_count) {
1717 read_count = max_count;
1719 err = pdu_marshal(pdu, offset, "d", read_count);
1725 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, read_count, false);
1726 err = v9fs_pack(qiov_full.iov, qiov_full.niov, 0,
1727 ((char *)fidp->fs.xattr.value) + off,
1729 qemu_iovec_destroy(&qiov_full);
1737 static int coroutine_fn v9fs_do_readdir_with_stat(V9fsPDU *pdu,
1746 off_t saved_dir_pos;
1747 struct dirent *dent;
1749 /* save the directory position */
1750 saved_dir_pos = v9fs_co_telldir(pdu, fidp);
1751 if (saved_dir_pos < 0) {
1752 return saved_dir_pos;
1756 v9fs_path_init(&path);
1758 v9fs_readdir_lock(&fidp->fs.dir);
1760 err = v9fs_co_readdir(pdu, fidp, &dent);
1764 err = v9fs_co_name_to_path(pdu, &fidp->path, dent->d_name, &path);
1768 err = v9fs_co_lstat(pdu, &path, &stbuf);
1772 err = stat_to_v9stat(pdu, &path, dent->d_name, &stbuf, &v9stat);
1776 if ((count + v9stat.size + 2) > max_count) {
1777 v9fs_readdir_unlock(&fidp->fs.dir);
1779 /* Ran out of buffer. Set dir back to old position and return */
1780 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
1781 v9fs_stat_free(&v9stat);
1782 v9fs_path_free(&path);
1786 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
1787 len = pdu_marshal(pdu, 11 + count, "S", &v9stat);
1789 v9fs_readdir_unlock(&fidp->fs.dir);
1792 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
1793 v9fs_stat_free(&v9stat);
1794 v9fs_path_free(&path);
1798 v9fs_stat_free(&v9stat);
1799 v9fs_path_free(&path);
1800 saved_dir_pos = dent->d_off;
1803 v9fs_readdir_unlock(&fidp->fs.dir);
1805 v9fs_path_free(&path);
1812 static void coroutine_fn v9fs_read(void *opaque)
1821 V9fsPDU *pdu = opaque;
1822 V9fsState *s = pdu->s;
1824 err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &max_count);
1828 trace_v9fs_read(pdu->tag, pdu->id, fid, off, max_count);
1830 fidp = get_fid(pdu, fid);
1835 if (fidp->fid_type == P9_FID_DIR) {
1838 v9fs_co_rewinddir(pdu, fidp);
1840 count = v9fs_do_readdir_with_stat(pdu, fidp, max_count);
1845 err = pdu_marshal(pdu, offset, "d", count);
1849 err += offset + count;
1850 } else if (fidp->fid_type == P9_FID_FILE) {
1851 QEMUIOVector qiov_full;
1855 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset + 4, max_count, false);
1856 qemu_iovec_init(&qiov, qiov_full.niov);
1858 qemu_iovec_reset(&qiov);
1859 qemu_iovec_concat(&qiov, &qiov_full, count, qiov_full.size - count);
1861 print_sg(qiov.iov, qiov.niov);
1863 /* Loop in case of EINTR */
1865 len = v9fs_co_preadv(pdu, fidp, qiov.iov, qiov.niov, off);
1870 } while (len == -EINTR && !pdu->cancelled);
1872 /* IO error return the error */
1874 goto out_free_iovec;
1876 } while (count < max_count && len > 0);
1877 err = pdu_marshal(pdu, offset, "d", count);
1879 goto out_free_iovec;
1881 err += offset + count;
1883 qemu_iovec_destroy(&qiov);
1884 qemu_iovec_destroy(&qiov_full);
1885 } else if (fidp->fid_type == P9_FID_XATTR) {
1886 err = v9fs_xattr_read(s, pdu, fidp, off, max_count);
1890 trace_v9fs_read_return(pdu->tag, pdu->id, count, err);
1894 pdu_complete(pdu, err);
1897 static size_t v9fs_readdir_data_size(V9fsString *name)
1900 * Size of each dirent on the wire: size of qid (13) + size of offset (8)
1901 * size of type (1) + size of name.size (2) + strlen(name.data)
1903 return 24 + v9fs_string_size(name);
1906 static int coroutine_fn v9fs_do_readdir(V9fsPDU *pdu, V9fsFidState *fidp,
1914 off_t saved_dir_pos;
1915 struct dirent *dent;
1917 /* save the directory position */
1918 saved_dir_pos = v9fs_co_telldir(pdu, fidp);
1919 if (saved_dir_pos < 0) {
1920 return saved_dir_pos;
1924 v9fs_readdir_lock(&fidp->fs.dir);
1926 err = v9fs_co_readdir(pdu, fidp, &dent);
1930 v9fs_string_init(&name);
1931 v9fs_string_sprintf(&name, "%s", dent->d_name);
1932 if ((count + v9fs_readdir_data_size(&name)) > max_count) {
1933 v9fs_readdir_unlock(&fidp->fs.dir);
1935 /* Ran out of buffer. Set dir back to old position and return */
1936 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
1937 v9fs_string_free(&name);
1941 * Fill up just the path field of qid because the client uses
1942 * only that. To fill the entire qid structure we will have
1943 * to stat each dirent found, which is expensive
1945 size = MIN(sizeof(dent->d_ino), sizeof(qid.path));
1946 memcpy(&qid.path, &dent->d_ino, size);
1947 /* Fill the other fields with dummy values */
1951 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
1952 len = pdu_marshal(pdu, 11 + count, "Qqbs",
1954 dent->d_type, &name);
1956 v9fs_readdir_unlock(&fidp->fs.dir);
1959 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
1960 v9fs_string_free(&name);
1964 v9fs_string_free(&name);
1965 saved_dir_pos = dent->d_off;
1968 v9fs_readdir_unlock(&fidp->fs.dir);
1976 static void coroutine_fn v9fs_readdir(void *opaque)
1982 uint64_t initial_offset;
1985 V9fsPDU *pdu = opaque;
1987 retval = pdu_unmarshal(pdu, offset, "dqd", &fid,
1988 &initial_offset, &max_count);
1992 trace_v9fs_readdir(pdu->tag, pdu->id, fid, initial_offset, max_count);
1994 fidp = get_fid(pdu, fid);
1999 if (!fidp->fs.dir.stream) {
2003 if (initial_offset == 0) {
2004 v9fs_co_rewinddir(pdu, fidp);
2006 v9fs_co_seekdir(pdu, fidp, initial_offset);
2008 count = v9fs_do_readdir(pdu, fidp, max_count);
2013 retval = pdu_marshal(pdu, offset, "d", count);
2017 retval += count + offset;
2018 trace_v9fs_readdir_return(pdu->tag, pdu->id, count, retval);
2022 pdu_complete(pdu, retval);
2025 static int v9fs_xattr_write(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
2026 uint64_t off, uint32_t count,
2027 struct iovec *sg, int cnt)
2031 uint64_t write_count;
2035 if (fidp->fs.xattr.len < off) {
2039 write_count = fidp->fs.xattr.len - off;
2040 if (write_count > count) {
2041 write_count = count;
2043 err = pdu_marshal(pdu, offset, "d", write_count);
2048 fidp->fs.xattr.copied_len += write_count;
2050 * Now copy the content from sg list
2052 for (i = 0; i < cnt; i++) {
2053 if (write_count > sg[i].iov_len) {
2054 to_copy = sg[i].iov_len;
2056 to_copy = write_count;
2058 memcpy((char *)fidp->fs.xattr.value + off, sg[i].iov_base, to_copy);
2059 /* updating vs->off since we are not using below */
2061 write_count -= to_copy;
2067 static void coroutine_fn v9fs_write(void *opaque)
2077 V9fsPDU *pdu = opaque;
2078 V9fsState *s = pdu->s;
2079 QEMUIOVector qiov_full;
2082 err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &count);
2084 pdu_complete(pdu, err);
2088 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, count, true);
2089 trace_v9fs_write(pdu->tag, pdu->id, fid, off, count, qiov_full.niov);
2091 fidp = get_fid(pdu, fid);
2096 if (fidp->fid_type == P9_FID_FILE) {
2097 if (fidp->fs.fd == -1) {
2101 } else if (fidp->fid_type == P9_FID_XATTR) {
2103 * setxattr operation
2105 err = v9fs_xattr_write(s, pdu, fidp, off, count,
2106 qiov_full.iov, qiov_full.niov);
2112 qemu_iovec_init(&qiov, qiov_full.niov);
2114 qemu_iovec_reset(&qiov);
2115 qemu_iovec_concat(&qiov, &qiov_full, total, qiov_full.size - total);
2117 print_sg(qiov.iov, qiov.niov);
2119 /* Loop in case of EINTR */
2121 len = v9fs_co_pwritev(pdu, fidp, qiov.iov, qiov.niov, off);
2126 } while (len == -EINTR && !pdu->cancelled);
2128 /* IO error return the error */
2132 } while (total < count && len > 0);
2135 err = pdu_marshal(pdu, offset, "d", total);
2140 trace_v9fs_write_return(pdu->tag, pdu->id, total, err);
2142 qemu_iovec_destroy(&qiov);
2146 qemu_iovec_destroy(&qiov_full);
2147 pdu_complete(pdu, err);
2150 static void coroutine_fn v9fs_create(void *opaque)
2162 V9fsString extension;
2164 V9fsPDU *pdu = opaque;
2165 V9fsState *s = pdu->s;
2167 v9fs_path_init(&path);
2168 v9fs_string_init(&name);
2169 v9fs_string_init(&extension);
2170 err = pdu_unmarshal(pdu, offset, "dsdbs", &fid, &name,
2171 &perm, &mode, &extension);
2175 trace_v9fs_create(pdu->tag, pdu->id, fid, name.data, perm, mode);
2177 if (name_is_illegal(name.data)) {
2182 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2187 fidp = get_fid(pdu, fid);
2192 if (fidp->fid_type != P9_FID_NONE) {
2196 if (perm & P9_STAT_MODE_DIR) {
2197 err = v9fs_co_mkdir(pdu, fidp, &name, perm & 0777,
2198 fidp->uid, -1, &stbuf);
2202 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2206 v9fs_path_write_lock(s);
2207 v9fs_path_copy(&fidp->path, &path);
2208 v9fs_path_unlock(s);
2209 err = v9fs_co_opendir(pdu, fidp);
2213 fidp->fid_type = P9_FID_DIR;
2214 } else if (perm & P9_STAT_MODE_SYMLINK) {
2215 err = v9fs_co_symlink(pdu, fidp, &name,
2216 extension.data, -1 , &stbuf);
2220 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2224 v9fs_path_write_lock(s);
2225 v9fs_path_copy(&fidp->path, &path);
2226 v9fs_path_unlock(s);
2227 } else if (perm & P9_STAT_MODE_LINK) {
2228 int32_t ofid = atoi(extension.data);
2229 V9fsFidState *ofidp = get_fid(pdu, ofid);
2230 if (ofidp == NULL) {
2234 err = v9fs_co_link(pdu, ofidp, fidp, &name);
2235 put_fid(pdu, ofidp);
2239 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2241 fidp->fid_type = P9_FID_NONE;
2244 v9fs_path_write_lock(s);
2245 v9fs_path_copy(&fidp->path, &path);
2246 v9fs_path_unlock(s);
2247 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
2249 fidp->fid_type = P9_FID_NONE;
2252 } else if (perm & P9_STAT_MODE_DEVICE) {
2254 uint32_t major, minor;
2257 if (sscanf(extension.data, "%c %u %u", &ctype, &major, &minor) != 3) {
2274 nmode |= perm & 0777;
2275 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2276 makedev(major, minor), nmode, &stbuf);
2280 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2284 v9fs_path_write_lock(s);
2285 v9fs_path_copy(&fidp->path, &path);
2286 v9fs_path_unlock(s);
2287 } else if (perm & P9_STAT_MODE_NAMED_PIPE) {
2288 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2289 0, S_IFIFO | (perm & 0777), &stbuf);
2293 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2297 v9fs_path_write_lock(s);
2298 v9fs_path_copy(&fidp->path, &path);
2299 v9fs_path_unlock(s);
2300 } else if (perm & P9_STAT_MODE_SOCKET) {
2301 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2302 0, S_IFSOCK | (perm & 0777), &stbuf);
2306 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2310 v9fs_path_write_lock(s);
2311 v9fs_path_copy(&fidp->path, &path);
2312 v9fs_path_unlock(s);
2314 err = v9fs_co_open2(pdu, fidp, &name, -1,
2315 omode_to_uflags(mode)|O_CREAT, perm, &stbuf);
2319 fidp->fid_type = P9_FID_FILE;
2320 fidp->open_flags = omode_to_uflags(mode);
2321 if (fidp->open_flags & O_EXCL) {
2323 * We let the host file system do O_EXCL check
2324 * We should not reclaim such fd
2326 fidp->flags |= FID_NON_RECLAIMABLE;
2329 iounit = get_iounit(pdu, &fidp->path);
2330 stat_to_qid(&stbuf, &qid);
2331 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
2336 trace_v9fs_create_return(pdu->tag, pdu->id,
2337 qid.type, qid.version, qid.path, iounit);
2341 pdu_complete(pdu, err);
2342 v9fs_string_free(&name);
2343 v9fs_string_free(&extension);
2344 v9fs_path_free(&path);
2347 static void coroutine_fn v9fs_symlink(void *opaque)
2349 V9fsPDU *pdu = opaque;
2352 V9fsFidState *dfidp;
2360 v9fs_string_init(&name);
2361 v9fs_string_init(&symname);
2362 err = pdu_unmarshal(pdu, offset, "dssd", &dfid, &name, &symname, &gid);
2366 trace_v9fs_symlink(pdu->tag, pdu->id, dfid, name.data, symname.data, gid);
2368 if (name_is_illegal(name.data)) {
2373 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2378 dfidp = get_fid(pdu, dfid);
2379 if (dfidp == NULL) {
2383 err = v9fs_co_symlink(pdu, dfidp, &name, symname.data, gid, &stbuf);
2387 stat_to_qid(&stbuf, &qid);
2388 err = pdu_marshal(pdu, offset, "Q", &qid);
2393 trace_v9fs_symlink_return(pdu->tag, pdu->id,
2394 qid.type, qid.version, qid.path);
2396 put_fid(pdu, dfidp);
2398 pdu_complete(pdu, err);
2399 v9fs_string_free(&name);
2400 v9fs_string_free(&symname);
2403 static void coroutine_fn v9fs_flush(void *opaque)
2408 V9fsPDU *cancel_pdu = NULL;
2409 V9fsPDU *pdu = opaque;
2410 V9fsState *s = pdu->s;
2412 err = pdu_unmarshal(pdu, offset, "w", &tag);
2414 pdu_complete(pdu, err);
2417 trace_v9fs_flush(pdu->tag, pdu->id, tag);
2419 if (pdu->tag == tag) {
2420 warn_report("the guest sent a self-referencing 9P flush request");
2422 QLIST_FOREACH(cancel_pdu, &s->active_list, next) {
2423 if (cancel_pdu->tag == tag) {
2429 cancel_pdu->cancelled = 1;
2431 * Wait for pdu to complete.
2433 qemu_co_queue_wait(&cancel_pdu->complete, NULL);
2434 if (!qemu_co_queue_next(&cancel_pdu->complete)) {
2435 cancel_pdu->cancelled = 0;
2436 pdu_free(cancel_pdu);
2439 pdu_complete(pdu, 7);
2442 static void coroutine_fn v9fs_link(void *opaque)
2444 V9fsPDU *pdu = opaque;
2445 int32_t dfid, oldfid;
2446 V9fsFidState *dfidp, *oldfidp;
2451 v9fs_string_init(&name);
2452 err = pdu_unmarshal(pdu, offset, "dds", &dfid, &oldfid, &name);
2456 trace_v9fs_link(pdu->tag, pdu->id, dfid, oldfid, name.data);
2458 if (name_is_illegal(name.data)) {
2463 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2468 dfidp = get_fid(pdu, dfid);
2469 if (dfidp == NULL) {
2474 oldfidp = get_fid(pdu, oldfid);
2475 if (oldfidp == NULL) {
2479 err = v9fs_co_link(pdu, oldfidp, dfidp, &name);
2483 put_fid(pdu, oldfidp);
2485 put_fid(pdu, dfidp);
2487 v9fs_string_free(&name);
2488 pdu_complete(pdu, err);
2491 /* Only works with path name based fid */
2492 static void coroutine_fn v9fs_remove(void *opaque)
2498 V9fsPDU *pdu = opaque;
2500 err = pdu_unmarshal(pdu, offset, "d", &fid);
2504 trace_v9fs_remove(pdu->tag, pdu->id, fid);
2506 fidp = get_fid(pdu, fid);
2511 /* if fs driver is not path based, return EOPNOTSUPP */
2512 if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
2517 * IF the file is unlinked, we cannot reopen
2518 * the file later. So don't reclaim fd
2520 err = v9fs_mark_fids_unreclaim(pdu, &fidp->path);
2524 err = v9fs_co_remove(pdu, &fidp->path);
2529 /* For TREMOVE we need to clunk the fid even on failed remove */
2530 clunk_fid(pdu->s, fidp->fid);
2533 pdu_complete(pdu, err);
2536 static void coroutine_fn v9fs_unlinkat(void *opaque)
2540 int32_t dfid, flags, rflags = 0;
2543 V9fsFidState *dfidp;
2544 V9fsPDU *pdu = opaque;
2546 v9fs_string_init(&name);
2547 err = pdu_unmarshal(pdu, offset, "dsd", &dfid, &name, &flags);
2552 if (name_is_illegal(name.data)) {
2557 if (!strcmp(".", name.data)) {
2562 if (!strcmp("..", name.data)) {
2567 if (flags & ~P9_DOTL_AT_REMOVEDIR) {
2572 if (flags & P9_DOTL_AT_REMOVEDIR) {
2573 rflags |= AT_REMOVEDIR;
2576 dfidp = get_fid(pdu, dfid);
2577 if (dfidp == NULL) {
2582 * IF the file is unlinked, we cannot reopen
2583 * the file later. So don't reclaim fd
2585 v9fs_path_init(&path);
2586 err = v9fs_co_name_to_path(pdu, &dfidp->path, name.data, &path);
2590 err = v9fs_mark_fids_unreclaim(pdu, &path);
2594 err = v9fs_co_unlinkat(pdu, &dfidp->path, &name, rflags);
2599 put_fid(pdu, dfidp);
2600 v9fs_path_free(&path);
2602 pdu_complete(pdu, err);
2603 v9fs_string_free(&name);
2607 /* Only works with path name based fid */
2608 static int coroutine_fn v9fs_complete_rename(V9fsPDU *pdu, V9fsFidState *fidp,
2614 V9fsFidState *tfidp;
2615 V9fsState *s = pdu->s;
2616 V9fsFidState *dirfidp = NULL;
2618 v9fs_path_init(&new_path);
2619 if (newdirfid != -1) {
2620 dirfidp = get_fid(pdu, newdirfid);
2621 if (dirfidp == NULL) {
2625 if (fidp->fid_type != P9_FID_NONE) {
2629 err = v9fs_co_name_to_path(pdu, &dirfidp->path, name->data, &new_path);
2634 char *dir_name = g_path_get_dirname(fidp->path.data);
2637 v9fs_path_init(&dir_path);
2638 v9fs_path_sprintf(&dir_path, "%s", dir_name);
2641 err = v9fs_co_name_to_path(pdu, &dir_path, name->data, &new_path);
2642 v9fs_path_free(&dir_path);
2647 err = v9fs_co_rename(pdu, &fidp->path, &new_path);
2652 * Fixup fid's pointing to the old name to
2653 * start pointing to the new name
2655 for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
2656 if (v9fs_path_is_ancestor(&fidp->path, &tfidp->path)) {
2657 /* replace the name */
2658 v9fs_fix_path(&tfidp->path, &new_path, strlen(fidp->path.data));
2663 put_fid(pdu, dirfidp);
2665 v9fs_path_free(&new_path);
2670 /* Only works with path name based fid */
2671 static void coroutine_fn v9fs_rename(void *opaque)
2679 V9fsPDU *pdu = opaque;
2680 V9fsState *s = pdu->s;
2682 v9fs_string_init(&name);
2683 err = pdu_unmarshal(pdu, offset, "dds", &fid, &newdirfid, &name);
2688 if (name_is_illegal(name.data)) {
2693 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2698 fidp = get_fid(pdu, fid);
2703 if (fidp->fid_type != P9_FID_NONE) {
2707 /* if fs driver is not path based, return EOPNOTSUPP */
2708 if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
2712 v9fs_path_write_lock(s);
2713 err = v9fs_complete_rename(pdu, fidp, newdirfid, &name);
2714 v9fs_path_unlock(s);
2721 pdu_complete(pdu, err);
2722 v9fs_string_free(&name);
2725 static int coroutine_fn v9fs_fix_fid_paths(V9fsPDU *pdu, V9fsPath *olddir,
2726 V9fsString *old_name,
2728 V9fsString *new_name)
2730 V9fsFidState *tfidp;
2731 V9fsPath oldpath, newpath;
2732 V9fsState *s = pdu->s;
2735 v9fs_path_init(&oldpath);
2736 v9fs_path_init(&newpath);
2737 err = v9fs_co_name_to_path(pdu, olddir, old_name->data, &oldpath);
2741 err = v9fs_co_name_to_path(pdu, newdir, new_name->data, &newpath);
2747 * Fixup fid's pointing to the old name to
2748 * start pointing to the new name
2750 for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
2751 if (v9fs_path_is_ancestor(&oldpath, &tfidp->path)) {
2752 /* replace the name */
2753 v9fs_fix_path(&tfidp->path, &newpath, strlen(oldpath.data));
2757 v9fs_path_free(&oldpath);
2758 v9fs_path_free(&newpath);
2762 static int coroutine_fn v9fs_complete_renameat(V9fsPDU *pdu, int32_t olddirfid,
2763 V9fsString *old_name,
2765 V9fsString *new_name)
2768 V9fsState *s = pdu->s;
2769 V9fsFidState *newdirfidp = NULL, *olddirfidp = NULL;
2771 olddirfidp = get_fid(pdu, olddirfid);
2772 if (olddirfidp == NULL) {
2776 if (newdirfid != -1) {
2777 newdirfidp = get_fid(pdu, newdirfid);
2778 if (newdirfidp == NULL) {
2783 newdirfidp = get_fid(pdu, olddirfid);
2786 err = v9fs_co_renameat(pdu, &olddirfidp->path, old_name,
2787 &newdirfidp->path, new_name);
2791 if (s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT) {
2792 /* Only for path based fid we need to do the below fixup */
2793 err = v9fs_fix_fid_paths(pdu, &olddirfidp->path, old_name,
2794 &newdirfidp->path, new_name);
2798 put_fid(pdu, olddirfidp);
2801 put_fid(pdu, newdirfidp);
2806 static void coroutine_fn v9fs_renameat(void *opaque)
2810 V9fsPDU *pdu = opaque;
2811 V9fsState *s = pdu->s;
2812 int32_t olddirfid, newdirfid;
2813 V9fsString old_name, new_name;
2815 v9fs_string_init(&old_name);
2816 v9fs_string_init(&new_name);
2817 err = pdu_unmarshal(pdu, offset, "dsds", &olddirfid,
2818 &old_name, &newdirfid, &new_name);
2823 if (name_is_illegal(old_name.data) || name_is_illegal(new_name.data)) {
2828 if (!strcmp(".", old_name.data) || !strcmp("..", old_name.data) ||
2829 !strcmp(".", new_name.data) || !strcmp("..", new_name.data)) {
2834 v9fs_path_write_lock(s);
2835 err = v9fs_complete_renameat(pdu, olddirfid,
2836 &old_name, newdirfid, &new_name);
2837 v9fs_path_unlock(s);
2843 pdu_complete(pdu, err);
2844 v9fs_string_free(&old_name);
2845 v9fs_string_free(&new_name);
2848 static void coroutine_fn v9fs_wstat(void *opaque)
2857 V9fsPDU *pdu = opaque;
2858 V9fsState *s = pdu->s;
2860 v9fs_stat_init(&v9stat);
2861 err = pdu_unmarshal(pdu, offset, "dwS", &fid, &unused, &v9stat);
2865 trace_v9fs_wstat(pdu->tag, pdu->id, fid,
2866 v9stat.mode, v9stat.atime, v9stat.mtime);
2868 fidp = get_fid(pdu, fid);
2873 /* do we need to sync the file? */
2874 if (donttouch_stat(&v9stat)) {
2875 err = v9fs_co_fsync(pdu, fidp, 0);
2878 if (v9stat.mode != -1) {
2880 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
2884 v9_mode = stat_to_v9mode(&stbuf);
2885 if ((v9stat.mode & P9_STAT_MODE_TYPE_BITS) !=
2886 (v9_mode & P9_STAT_MODE_TYPE_BITS)) {
2887 /* Attempting to change the type */
2891 err = v9fs_co_chmod(pdu, &fidp->path,
2892 v9mode_to_mode(v9stat.mode,
2893 &v9stat.extension));
2898 if (v9stat.mtime != -1 || v9stat.atime != -1) {
2899 struct timespec times[2];
2900 if (v9stat.atime != -1) {
2901 times[0].tv_sec = v9stat.atime;
2902 times[0].tv_nsec = 0;
2904 times[0].tv_nsec = UTIME_OMIT;
2906 if (v9stat.mtime != -1) {
2907 times[1].tv_sec = v9stat.mtime;
2908 times[1].tv_nsec = 0;
2910 times[1].tv_nsec = UTIME_OMIT;
2912 err = v9fs_co_utimensat(pdu, &fidp->path, times);
2917 if (v9stat.n_gid != -1 || v9stat.n_uid != -1) {
2918 err = v9fs_co_chown(pdu, &fidp->path, v9stat.n_uid, v9stat.n_gid);
2923 if (v9stat.name.size != 0) {
2924 v9fs_path_write_lock(s);
2925 err = v9fs_complete_rename(pdu, fidp, -1, &v9stat.name);
2926 v9fs_path_unlock(s);
2931 if (v9stat.length != -1) {
2932 err = v9fs_co_truncate(pdu, &fidp->path, v9stat.length);
2941 v9fs_stat_free(&v9stat);
2942 pdu_complete(pdu, err);
2945 static int v9fs_fill_statfs(V9fsState *s, V9fsPDU *pdu, struct statfs *stbuf)
2957 int32_t bsize_factor;
2960 * compute bsize factor based on host file system block size
2963 bsize_factor = (s->msize - P9_IOHDRSZ)/stbuf->f_bsize;
2964 if (!bsize_factor) {
2967 f_type = stbuf->f_type;
2968 f_bsize = stbuf->f_bsize;
2969 f_bsize *= bsize_factor;
2971 * f_bsize is adjusted(multiplied) by bsize factor, so we need to
2972 * adjust(divide) the number of blocks, free blocks and available
2973 * blocks by bsize factor
2975 f_blocks = stbuf->f_blocks/bsize_factor;
2976 f_bfree = stbuf->f_bfree/bsize_factor;
2977 f_bavail = stbuf->f_bavail/bsize_factor;
2978 f_files = stbuf->f_files;
2979 f_ffree = stbuf->f_ffree;
2980 fsid_val = (unsigned int) stbuf->f_fsid.__val[0] |
2981 (unsigned long long)stbuf->f_fsid.__val[1] << 32;
2982 f_namelen = stbuf->f_namelen;
2984 return pdu_marshal(pdu, offset, "ddqqqqqqd",
2985 f_type, f_bsize, f_blocks, f_bfree,
2986 f_bavail, f_files, f_ffree,
2987 fsid_val, f_namelen);
2990 static void coroutine_fn v9fs_statfs(void *opaque)
2996 struct statfs stbuf;
2997 V9fsPDU *pdu = opaque;
2998 V9fsState *s = pdu->s;
3000 retval = pdu_unmarshal(pdu, offset, "d", &fid);
3004 fidp = get_fid(pdu, fid);
3009 retval = v9fs_co_statfs(pdu, &fidp->path, &stbuf);
3013 retval = v9fs_fill_statfs(s, pdu, &stbuf);
3021 pdu_complete(pdu, retval);
3024 static void coroutine_fn v9fs_mknod(void *opaque)
3037 V9fsPDU *pdu = opaque;
3039 v9fs_string_init(&name);
3040 err = pdu_unmarshal(pdu, offset, "dsdddd", &fid, &name, &mode,
3041 &major, &minor, &gid);
3045 trace_v9fs_mknod(pdu->tag, pdu->id, fid, mode, major, minor);
3047 if (name_is_illegal(name.data)) {
3052 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3057 fidp = get_fid(pdu, fid);
3062 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, gid,
3063 makedev(major, minor), mode, &stbuf);
3067 stat_to_qid(&stbuf, &qid);
3068 err = pdu_marshal(pdu, offset, "Q", &qid);
3073 trace_v9fs_mknod_return(pdu->tag, pdu->id,
3074 qid.type, qid.version, qid.path);
3078 pdu_complete(pdu, err);
3079 v9fs_string_free(&name);
3083 * Implement posix byte range locking code
3084 * Server side handling of locking code is very simple, because 9p server in
3085 * QEMU can handle only one client. And most of the lock handling
3086 * (like conflict, merging) etc is done by the VFS layer itself, so no need to
3087 * do any thing in * qemu 9p server side lock code path.
3088 * So when a TLOCK request comes, always return success
3090 static void coroutine_fn v9fs_lock(void *opaque)
3096 int32_t fid, err = 0;
3097 V9fsPDU *pdu = opaque;
3099 v9fs_string_init(&flock.client_id);
3100 err = pdu_unmarshal(pdu, offset, "dbdqqds", &fid, &flock.type,
3101 &flock.flags, &flock.start, &flock.length,
3102 &flock.proc_id, &flock.client_id);
3106 trace_v9fs_lock(pdu->tag, pdu->id, fid,
3107 flock.type, flock.start, flock.length);
3110 /* We support only block flag now (that too ignored currently) */
3111 if (flock.flags & ~P9_LOCK_FLAGS_BLOCK) {
3115 fidp = get_fid(pdu, fid);
3120 err = v9fs_co_fstat(pdu, fidp, &stbuf);
3124 err = pdu_marshal(pdu, offset, "b", P9_LOCK_SUCCESS);
3129 trace_v9fs_lock_return(pdu->tag, pdu->id, P9_LOCK_SUCCESS);
3133 pdu_complete(pdu, err);
3134 v9fs_string_free(&flock.client_id);
3138 * When a TGETLOCK request comes, always return success because all lock
3139 * handling is done by client's VFS layer.
3141 static void coroutine_fn v9fs_getlock(void *opaque)
3147 int32_t fid, err = 0;
3148 V9fsPDU *pdu = opaque;
3150 v9fs_string_init(&glock.client_id);
3151 err = pdu_unmarshal(pdu, offset, "dbqqds", &fid, &glock.type,
3152 &glock.start, &glock.length, &glock.proc_id,
3157 trace_v9fs_getlock(pdu->tag, pdu->id, fid,
3158 glock.type, glock.start, glock.length);
3160 fidp = get_fid(pdu, fid);
3165 err = v9fs_co_fstat(pdu, fidp, &stbuf);
3169 glock.type = P9_LOCK_TYPE_UNLCK;
3170 err = pdu_marshal(pdu, offset, "bqqds", glock.type,
3171 glock.start, glock.length, glock.proc_id,
3177 trace_v9fs_getlock_return(pdu->tag, pdu->id, glock.type, glock.start,
3178 glock.length, glock.proc_id);
3182 pdu_complete(pdu, err);
3183 v9fs_string_free(&glock.client_id);
3186 static void coroutine_fn v9fs_mkdir(void *opaque)
3188 V9fsPDU *pdu = opaque;
3199 v9fs_string_init(&name);
3200 err = pdu_unmarshal(pdu, offset, "dsdd", &fid, &name, &mode, &gid);
3204 trace_v9fs_mkdir(pdu->tag, pdu->id, fid, name.data, mode, gid);
3206 if (name_is_illegal(name.data)) {
3211 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3216 fidp = get_fid(pdu, fid);
3221 err = v9fs_co_mkdir(pdu, fidp, &name, mode, fidp->uid, gid, &stbuf);
3225 stat_to_qid(&stbuf, &qid);
3226 err = pdu_marshal(pdu, offset, "Q", &qid);
3231 trace_v9fs_mkdir_return(pdu->tag, pdu->id,
3232 qid.type, qid.version, qid.path, err);
3236 pdu_complete(pdu, err);
3237 v9fs_string_free(&name);
3240 static void coroutine_fn v9fs_xattrwalk(void *opaque)
3246 int32_t fid, newfid;
3247 V9fsFidState *file_fidp;
3248 V9fsFidState *xattr_fidp = NULL;
3249 V9fsPDU *pdu = opaque;
3250 V9fsState *s = pdu->s;
3252 v9fs_string_init(&name);
3253 err = pdu_unmarshal(pdu, offset, "dds", &fid, &newfid, &name);
3257 trace_v9fs_xattrwalk(pdu->tag, pdu->id, fid, newfid, name.data);
3259 file_fidp = get_fid(pdu, fid);
3260 if (file_fidp == NULL) {
3264 xattr_fidp = alloc_fid(s, newfid);
3265 if (xattr_fidp == NULL) {
3269 v9fs_path_copy(&xattr_fidp->path, &file_fidp->path);
3270 if (!v9fs_string_size(&name)) {
3272 * listxattr request. Get the size first
3274 size = v9fs_co_llistxattr(pdu, &xattr_fidp->path, NULL, 0);
3277 clunk_fid(s, xattr_fidp->fid);
3281 * Read the xattr value
3283 xattr_fidp->fs.xattr.len = size;
3284 xattr_fidp->fid_type = P9_FID_XATTR;
3285 xattr_fidp->fs.xattr.xattrwalk_fid = true;
3286 xattr_fidp->fs.xattr.value = g_malloc0(size);
3288 err = v9fs_co_llistxattr(pdu, &xattr_fidp->path,
3289 xattr_fidp->fs.xattr.value,
3290 xattr_fidp->fs.xattr.len);
3292 clunk_fid(s, xattr_fidp->fid);
3296 err = pdu_marshal(pdu, offset, "q", size);
3303 * specific xattr fid. We check for xattr
3304 * presence also collect the xattr size
3306 size = v9fs_co_lgetxattr(pdu, &xattr_fidp->path,
3310 clunk_fid(s, xattr_fidp->fid);
3314 * Read the xattr value
3316 xattr_fidp->fs.xattr.len = size;
3317 xattr_fidp->fid_type = P9_FID_XATTR;
3318 xattr_fidp->fs.xattr.xattrwalk_fid = true;
3319 xattr_fidp->fs.xattr.value = g_malloc0(size);
3321 err = v9fs_co_lgetxattr(pdu, &xattr_fidp->path,
3322 &name, xattr_fidp->fs.xattr.value,
3323 xattr_fidp->fs.xattr.len);
3325 clunk_fid(s, xattr_fidp->fid);
3329 err = pdu_marshal(pdu, offset, "q", size);
3335 trace_v9fs_xattrwalk_return(pdu->tag, pdu->id, size);
3337 put_fid(pdu, file_fidp);
3339 put_fid(pdu, xattr_fidp);
3342 pdu_complete(pdu, err);
3343 v9fs_string_free(&name);
3346 static void coroutine_fn v9fs_xattrcreate(void *opaque)
3348 int flags, rflags = 0;
3354 V9fsFidState *file_fidp;
3355 V9fsFidState *xattr_fidp;
3356 V9fsPDU *pdu = opaque;
3358 v9fs_string_init(&name);
3359 err = pdu_unmarshal(pdu, offset, "dsqd", &fid, &name, &size, &flags);
3363 trace_v9fs_xattrcreate(pdu->tag, pdu->id, fid, name.data, size, flags);
3365 if (flags & ~(P9_XATTR_CREATE | P9_XATTR_REPLACE)) {
3370 if (flags & P9_XATTR_CREATE) {
3371 rflags |= XATTR_CREATE;
3374 if (flags & P9_XATTR_REPLACE) {
3375 rflags |= XATTR_REPLACE;
3378 if (size > XATTR_SIZE_MAX) {
3383 file_fidp = get_fid(pdu, fid);
3384 if (file_fidp == NULL) {
3388 if (file_fidp->fid_type != P9_FID_NONE) {
3393 /* Make the file fid point to xattr */
3394 xattr_fidp = file_fidp;
3395 xattr_fidp->fid_type = P9_FID_XATTR;
3396 xattr_fidp->fs.xattr.copied_len = 0;
3397 xattr_fidp->fs.xattr.xattrwalk_fid = false;
3398 xattr_fidp->fs.xattr.len = size;
3399 xattr_fidp->fs.xattr.flags = rflags;
3400 v9fs_string_init(&xattr_fidp->fs.xattr.name);
3401 v9fs_string_copy(&xattr_fidp->fs.xattr.name, &name);
3402 xattr_fidp->fs.xattr.value = g_malloc0(size);
3405 put_fid(pdu, file_fidp);
3407 pdu_complete(pdu, err);
3408 v9fs_string_free(&name);
3411 static void coroutine_fn v9fs_readlink(void *opaque)
3413 V9fsPDU *pdu = opaque;
3420 err = pdu_unmarshal(pdu, offset, "d", &fid);
3424 trace_v9fs_readlink(pdu->tag, pdu->id, fid);
3425 fidp = get_fid(pdu, fid);
3431 v9fs_string_init(&target);
3432 err = v9fs_co_readlink(pdu, &fidp->path, &target);
3436 err = pdu_marshal(pdu, offset, "s", &target);
3438 v9fs_string_free(&target);
3442 trace_v9fs_readlink_return(pdu->tag, pdu->id, target.data);
3443 v9fs_string_free(&target);
3447 pdu_complete(pdu, err);
3450 static CoroutineEntry *pdu_co_handlers[] = {
3451 [P9_TREADDIR] = v9fs_readdir,
3452 [P9_TSTATFS] = v9fs_statfs,
3453 [P9_TGETATTR] = v9fs_getattr,
3454 [P9_TSETATTR] = v9fs_setattr,
3455 [P9_TXATTRWALK] = v9fs_xattrwalk,
3456 [P9_TXATTRCREATE] = v9fs_xattrcreate,
3457 [P9_TMKNOD] = v9fs_mknod,
3458 [P9_TRENAME] = v9fs_rename,
3459 [P9_TLOCK] = v9fs_lock,
3460 [P9_TGETLOCK] = v9fs_getlock,
3461 [P9_TRENAMEAT] = v9fs_renameat,
3462 [P9_TREADLINK] = v9fs_readlink,
3463 [P9_TUNLINKAT] = v9fs_unlinkat,
3464 [P9_TMKDIR] = v9fs_mkdir,
3465 [P9_TVERSION] = v9fs_version,
3466 [P9_TLOPEN] = v9fs_open,
3467 [P9_TATTACH] = v9fs_attach,
3468 [P9_TSTAT] = v9fs_stat,
3469 [P9_TWALK] = v9fs_walk,
3470 [P9_TCLUNK] = v9fs_clunk,
3471 [P9_TFSYNC] = v9fs_fsync,
3472 [P9_TOPEN] = v9fs_open,
3473 [P9_TREAD] = v9fs_read,
3475 [P9_TAUTH] = v9fs_auth,
3477 [P9_TFLUSH] = v9fs_flush,
3478 [P9_TLINK] = v9fs_link,
3479 [P9_TSYMLINK] = v9fs_symlink,
3480 [P9_TCREATE] = v9fs_create,
3481 [P9_TLCREATE] = v9fs_lcreate,
3482 [P9_TWRITE] = v9fs_write,
3483 [P9_TWSTAT] = v9fs_wstat,
3484 [P9_TREMOVE] = v9fs_remove,
3487 static void coroutine_fn v9fs_op_not_supp(void *opaque)
3489 V9fsPDU *pdu = opaque;
3490 pdu_complete(pdu, -EOPNOTSUPP);
3493 static void coroutine_fn v9fs_fs_ro(void *opaque)
3495 V9fsPDU *pdu = opaque;
3496 pdu_complete(pdu, -EROFS);
3499 static inline bool is_read_only_op(V9fsPDU *pdu)
3526 void pdu_submit(V9fsPDU *pdu, P9MsgHeader *hdr)
3529 CoroutineEntry *handler;
3530 V9fsState *s = pdu->s;
3532 pdu->size = le32_to_cpu(hdr->size_le);
3534 pdu->tag = le16_to_cpu(hdr->tag_le);
3536 if (pdu->id >= ARRAY_SIZE(pdu_co_handlers) ||
3537 (pdu_co_handlers[pdu->id] == NULL)) {
3538 handler = v9fs_op_not_supp;
3539 } else if (is_ro_export(&s->ctx) && !is_read_only_op(pdu)) {
3540 handler = v9fs_fs_ro;
3542 handler = pdu_co_handlers[pdu->id];
3545 qemu_co_queue_init(&pdu->complete);
3546 co = qemu_coroutine_create(handler, pdu);
3547 qemu_coroutine_enter(co);
3550 /* Returns 0 on success, 1 on failure. */
3551 int v9fs_device_realize_common(V9fsState *s, const V9fsTransport *t,
3560 assert(!s->transport);
3563 /* initialize pdu allocator */
3564 QLIST_INIT(&s->free_list);
3565 QLIST_INIT(&s->active_list);
3566 for (i = 0; i < MAX_REQ; i++) {
3567 QLIST_INSERT_HEAD(&s->free_list, &s->pdus[i], next);
3572 v9fs_path_init(&path);
3574 fse = get_fsdev_fsentry(s->fsconf.fsdev_id);
3577 /* We don't have a fsdev identified by fsdev_id */
3578 error_setg(errp, "9pfs device couldn't find fsdev with the "
3580 s->fsconf.fsdev_id ? s->fsconf.fsdev_id : "NULL");
3584 if (!s->fsconf.tag) {
3585 /* we haven't specified a mount_tag */
3586 error_setg(errp, "fsdev with id %s needs mount_tag arguments",
3587 s->fsconf.fsdev_id);
3591 s->ctx.export_flags = fse->export_flags;
3592 s->ctx.fs_root = g_strdup(fse->path);
3593 s->ctx.exops.get_st_gen = NULL;
3594 len = strlen(s->fsconf.tag);
3595 if (len > MAX_TAG_LEN - 1) {
3596 error_setg(errp, "mount tag '%s' (%d bytes) is longer than "
3597 "maximum (%d bytes)", s->fsconf.tag, len, MAX_TAG_LEN - 1);
3601 s->tag = g_strdup(s->fsconf.tag);
3606 s->ctx.fmode = fse->fmode;
3607 s->ctx.dmode = fse->dmode;
3610 qemu_co_rwlock_init(&s->rename_lock);
3612 if (s->ops->init(&s->ctx, errp) < 0) {
3613 error_prepend(errp, "cannot initialize fsdev '%s': ",
3614 s->fsconf.fsdev_id);
3619 * Check details of export path, We need to use fs driver
3620 * call back to do that. Since we are in the init path, we don't
3621 * use co-routines here.
3623 if (s->ops->name_to_path(&s->ctx, NULL, "/", &path) < 0) {
3625 "error in converting name to path %s", strerror(errno));
3628 if (s->ops->lstat(&s->ctx, &path, &stat)) {
3629 error_setg(errp, "share path %s does not exist", fse->path);
3631 } else if (!S_ISDIR(stat.st_mode)) {
3632 error_setg(errp, "share path %s is not a directory", fse->path);
3636 s->ctx.fst = &fse->fst;
3637 fsdev_throttle_init(s->ctx.fst);
3639 v9fs_path_free(&path);
3644 if (s->ops && s->ops->cleanup && s->ctx.private) {
3645 s->ops->cleanup(&s->ctx);
3648 g_free(s->ctx.fs_root);
3649 v9fs_path_free(&path);
3654 void v9fs_device_unrealize_common(V9fsState *s, Error **errp)
3656 if (s->ops->cleanup) {
3657 s->ops->cleanup(&s->ctx);
3659 fsdev_throttle_cleanup(s->ctx.fst);
3661 g_free(s->ctx.fs_root);
3664 typedef struct VirtfsCoResetData {
3667 } VirtfsCoResetData;
3669 static void coroutine_fn virtfs_co_reset(void *opaque)
3671 VirtfsCoResetData *data = opaque;
3673 virtfs_reset(&data->pdu);
3677 void v9fs_reset(V9fsState *s)
3679 VirtfsCoResetData data = { .pdu = { .s = s }, .done = false };
3682 while (!QLIST_EMPTY(&s->active_list)) {
3683 aio_poll(qemu_get_aio_context(), true);
3686 co = qemu_coroutine_create(virtfs_co_reset, &data);
3687 qemu_coroutine_enter(co);
3689 while (!data.done) {
3690 aio_poll(qemu_get_aio_context(), true);
3694 static void __attribute__((__constructor__)) v9fs_set_fd_limit(void)
3697 if (getrlimit(RLIMIT_NOFILE, &rlim) < 0) {
3698 error_report("Failed to get the resource limit");
3701 open_fd_hw = rlim.rlim_cur - MIN(400, rlim.rlim_cur/3);
3702 open_fd_rc = rlim.rlim_cur/2;