xref: /qemu/hw/9pfs/9p-local.c (revision a6c76285)
1 /*
2  * 9p Posix callback
3  *
4  * Copyright IBM, Corp. 2010
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "9p.h"
16 #include "9p-local.h"
17 #include "9p-xattr.h"
18 #include "9p-util.h"
19 #include "fsdev/qemu-fsdev.h"   /* local_ops */
20 #include <arpa/inet.h>
21 #include <pwd.h>
22 #include <grp.h>
23 #include <sys/socket.h>
24 #include <sys/un.h>
25 #include "qemu/xattr.h"
26 #include "qemu/cutils.h"
27 #include "qemu/error-report.h"
28 #include <libgen.h>
29 #include <linux/fs.h>
30 #ifdef CONFIG_LINUX_MAGIC_H
31 #include <linux/magic.h>
32 #endif
33 #include <sys/ioctl.h>
34 
35 #ifndef XFS_SUPER_MAGIC
36 #define XFS_SUPER_MAGIC  0x58465342
37 #endif
38 #ifndef EXT2_SUPER_MAGIC
39 #define EXT2_SUPER_MAGIC 0xEF53
40 #endif
41 #ifndef REISERFS_SUPER_MAGIC
42 #define REISERFS_SUPER_MAGIC 0x52654973
43 #endif
44 #ifndef BTRFS_SUPER_MAGIC
45 #define BTRFS_SUPER_MAGIC 0x9123683E
46 #endif
47 
48 typedef struct {
49     int mountfd;
50 } LocalData;
51 
52 int local_open_nofollow(FsContext *fs_ctx, const char *path, int flags,
53                         mode_t mode)
54 {
55     LocalData *data = fs_ctx->private;
56 
57     /* All paths are relative to the path data->mountfd points to */
58     while (*path == '/') {
59         path++;
60     }
61 
62     return relative_openat_nofollow(data->mountfd, path, flags, mode);
63 }
64 
65 int local_opendir_nofollow(FsContext *fs_ctx, const char *path)
66 {
67     return local_open_nofollow(fs_ctx, path, O_DIRECTORY | O_RDONLY, 0);
68 }
69 
70 static void renameat_preserve_errno(int odirfd, const char *opath, int ndirfd,
71                                     const char *npath)
72 {
73     int serrno = errno;
74     renameat(odirfd, opath, ndirfd, npath);
75     errno = serrno;
76 }
77 
78 static void unlinkat_preserve_errno(int dirfd, const char *path, int flags)
79 {
80     int serrno = errno;
81     unlinkat(dirfd, path, flags);
82     errno = serrno;
83 }
84 
85 #define VIRTFS_META_DIR ".virtfs_metadata"
86 
87 static FILE *local_fopenat(int dirfd, const char *name, const char *mode)
88 {
89     int fd, o_mode = 0;
90     FILE *fp;
91     int flags;
92     /*
93      * only supports two modes
94      */
95     if (mode[0] == 'r') {
96         flags = O_RDONLY;
97     } else if (mode[0] == 'w') {
98         flags = O_WRONLY | O_TRUNC | O_CREAT;
99         o_mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH;
100     } else {
101         return NULL;
102     }
103     fd = openat_file(dirfd, name, flags, o_mode);
104     if (fd == -1) {
105         return NULL;
106     }
107     fp = fdopen(fd, mode);
108     if (!fp) {
109         close(fd);
110     }
111     return fp;
112 }
113 
114 #define ATTR_MAX 100
115 static void local_mapped_file_attr(int dirfd, const char *name,
116                                    struct stat *stbuf)
117 {
118     FILE *fp;
119     char buf[ATTR_MAX];
120     int map_dirfd;
121 
122     map_dirfd = openat_dir(dirfd, VIRTFS_META_DIR);
123     if (map_dirfd == -1) {
124         return;
125     }
126 
127     fp = local_fopenat(map_dirfd, name, "r");
128     close_preserve_errno(map_dirfd);
129     if (!fp) {
130         return;
131     }
132     memset(buf, 0, ATTR_MAX);
133     while (fgets(buf, ATTR_MAX, fp)) {
134         if (!strncmp(buf, "virtfs.uid", 10)) {
135             stbuf->st_uid = atoi(buf+11);
136         } else if (!strncmp(buf, "virtfs.gid", 10)) {
137             stbuf->st_gid = atoi(buf+11);
138         } else if (!strncmp(buf, "virtfs.mode", 11)) {
139             stbuf->st_mode = atoi(buf+12);
140         } else if (!strncmp(buf, "virtfs.rdev", 11)) {
141             stbuf->st_rdev = atoi(buf+12);
142         }
143         memset(buf, 0, ATTR_MAX);
144     }
145     fclose(fp);
146 }
147 
148 static int local_lstat(FsContext *fs_ctx, V9fsPath *fs_path, struct stat *stbuf)
149 {
150     int err = -1;
151     char *dirpath = g_path_get_dirname(fs_path->data);
152     char *name = g_path_get_basename(fs_path->data);
153     int dirfd;
154 
155     dirfd = local_opendir_nofollow(fs_ctx, dirpath);
156     if (dirfd == -1) {
157         goto out;
158     }
159 
160     err = fstatat(dirfd, name, stbuf, AT_SYMLINK_NOFOLLOW);
161     if (err) {
162         goto err_out;
163     }
164     if (fs_ctx->export_flags & V9FS_SM_MAPPED) {
165         /* Actual credentials are part of extended attrs */
166         uid_t tmp_uid;
167         gid_t tmp_gid;
168         mode_t tmp_mode;
169         dev_t tmp_dev;
170 
171         if (fgetxattrat_nofollow(dirfd, name, "user.virtfs.uid", &tmp_uid,
172                                  sizeof(uid_t)) > 0) {
173             stbuf->st_uid = le32_to_cpu(tmp_uid);
174         }
175         if (fgetxattrat_nofollow(dirfd, name, "user.virtfs.gid", &tmp_gid,
176                                  sizeof(gid_t)) > 0) {
177             stbuf->st_gid = le32_to_cpu(tmp_gid);
178         }
179         if (fgetxattrat_nofollow(dirfd, name, "user.virtfs.mode", &tmp_mode,
180                                  sizeof(mode_t)) > 0) {
181             stbuf->st_mode = le32_to_cpu(tmp_mode);
182         }
183         if (fgetxattrat_nofollow(dirfd, name, "user.virtfs.rdev", &tmp_dev,
184                                  sizeof(dev_t)) > 0) {
185             stbuf->st_rdev = le64_to_cpu(tmp_dev);
186         }
187     } else if (fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
188         local_mapped_file_attr(dirfd, name, stbuf);
189     }
190 
191 err_out:
192     close_preserve_errno(dirfd);
193 out:
194     g_free(name);
195     g_free(dirpath);
196     return err;
197 }
198 
199 static int local_set_mapped_file_attrat(int dirfd, const char *name,
200                                         FsCred *credp)
201 {
202     FILE *fp;
203     int ret;
204     char buf[ATTR_MAX];
205     int uid = -1, gid = -1, mode = -1, rdev = -1;
206     int map_dirfd;
207 
208     ret = mkdirat(dirfd, VIRTFS_META_DIR, 0700);
209     if (ret < 0 && errno != EEXIST) {
210         return -1;
211     }
212 
213     map_dirfd = openat_dir(dirfd, VIRTFS_META_DIR);
214     if (map_dirfd == -1) {
215         return -1;
216     }
217 
218     fp = local_fopenat(map_dirfd, name, "r");
219     if (!fp) {
220         if (errno == ENOENT) {
221             goto update_map_file;
222         } else {
223             close_preserve_errno(map_dirfd);
224             return -1;
225         }
226     }
227     memset(buf, 0, ATTR_MAX);
228     while (fgets(buf, ATTR_MAX, fp)) {
229         if (!strncmp(buf, "virtfs.uid", 10)) {
230             uid = atoi(buf + 11);
231         } else if (!strncmp(buf, "virtfs.gid", 10)) {
232             gid = atoi(buf + 11);
233         } else if (!strncmp(buf, "virtfs.mode", 11)) {
234             mode = atoi(buf + 12);
235         } else if (!strncmp(buf, "virtfs.rdev", 11)) {
236             rdev = atoi(buf + 12);
237         }
238         memset(buf, 0, ATTR_MAX);
239     }
240     fclose(fp);
241 
242 update_map_file:
243     fp = local_fopenat(map_dirfd, name, "w");
244     close_preserve_errno(map_dirfd);
245     if (!fp) {
246         return -1;
247     }
248 
249     if (credp->fc_uid != -1) {
250         uid = credp->fc_uid;
251     }
252     if (credp->fc_gid != -1) {
253         gid = credp->fc_gid;
254     }
255     if (credp->fc_mode != -1) {
256         mode = credp->fc_mode;
257     }
258     if (credp->fc_rdev != -1) {
259         rdev = credp->fc_rdev;
260     }
261 
262     if (uid != -1) {
263         fprintf(fp, "virtfs.uid=%d\n", uid);
264     }
265     if (gid != -1) {
266         fprintf(fp, "virtfs.gid=%d\n", gid);
267     }
268     if (mode != -1) {
269         fprintf(fp, "virtfs.mode=%d\n", mode);
270     }
271     if (rdev != -1) {
272         fprintf(fp, "virtfs.rdev=%d\n", rdev);
273     }
274     fclose(fp);
275 
276     return 0;
277 }
278 
279 static int fchmodat_nofollow(int dirfd, const char *name, mode_t mode)
280 {
281     int fd, ret;
282 
283     /* FIXME: this should be handled with fchmodat(AT_SYMLINK_NOFOLLOW).
284      * Unfortunately, the linux kernel doesn't implement it yet. As an
285      * alternative, let's open the file and use fchmod() instead. This
286      * may fail depending on the permissions of the file, but it is the
287      * best we can do to avoid TOCTTOU. We first try to open read-only
288      * in case name points to a directory. If that fails, we try write-only
289      * in case name doesn't point to a directory.
290      */
291     fd = openat_file(dirfd, name, O_RDONLY, 0);
292     if (fd == -1) {
293         /* In case the file is writable-only and isn't a directory. */
294         if (errno == EACCES) {
295             fd = openat_file(dirfd, name, O_WRONLY, 0);
296         }
297         if (fd == -1 && errno == EISDIR) {
298             errno = EACCES;
299         }
300     }
301     if (fd == -1) {
302         return -1;
303     }
304     ret = fchmod(fd, mode);
305     close_preserve_errno(fd);
306     return ret;
307 }
308 
309 static int local_set_xattrat(int dirfd, const char *path, FsCred *credp)
310 {
311     int err;
312 
313     if (credp->fc_uid != -1) {
314         uint32_t tmp_uid = cpu_to_le32(credp->fc_uid);
315         err = fsetxattrat_nofollow(dirfd, path, "user.virtfs.uid", &tmp_uid,
316                                    sizeof(uid_t), 0);
317         if (err) {
318             return err;
319         }
320     }
321     if (credp->fc_gid != -1) {
322         uint32_t tmp_gid = cpu_to_le32(credp->fc_gid);
323         err = fsetxattrat_nofollow(dirfd, path, "user.virtfs.gid", &tmp_gid,
324                                    sizeof(gid_t), 0);
325         if (err) {
326             return err;
327         }
328     }
329     if (credp->fc_mode != -1) {
330         uint32_t tmp_mode = cpu_to_le32(credp->fc_mode);
331         err = fsetxattrat_nofollow(dirfd, path, "user.virtfs.mode", &tmp_mode,
332                                    sizeof(mode_t), 0);
333         if (err) {
334             return err;
335         }
336     }
337     if (credp->fc_rdev != -1) {
338         uint64_t tmp_rdev = cpu_to_le64(credp->fc_rdev);
339         err = fsetxattrat_nofollow(dirfd, path, "user.virtfs.rdev", &tmp_rdev,
340                                    sizeof(dev_t), 0);
341         if (err) {
342             return err;
343         }
344     }
345     return 0;
346 }
347 
348 static int local_set_cred_passthrough(FsContext *fs_ctx, int dirfd,
349                                       const char *name, FsCred *credp)
350 {
351     if (fchownat(dirfd, name, credp->fc_uid, credp->fc_gid,
352                  AT_SYMLINK_NOFOLLOW) < 0) {
353         /*
354          * If we fail to change ownership and if we are
355          * using security model none. Ignore the error
356          */
357         if ((fs_ctx->export_flags & V9FS_SEC_MASK) != V9FS_SM_NONE) {
358             return -1;
359         }
360     }
361 
362     return fchmodat_nofollow(dirfd, name, credp->fc_mode & 07777);
363 }
364 
365 static ssize_t local_readlink(FsContext *fs_ctx, V9fsPath *fs_path,
366                               char *buf, size_t bufsz)
367 {
368     ssize_t tsize = -1;
369 
370     if ((fs_ctx->export_flags & V9FS_SM_MAPPED) ||
371         (fs_ctx->export_flags & V9FS_SM_MAPPED_FILE)) {
372         int fd;
373 
374         fd = local_open_nofollow(fs_ctx, fs_path->data, O_RDONLY, 0);
375         if (fd == -1) {
376             return -1;
377         }
378         do {
379             tsize = read(fd, (void *)buf, bufsz);
380         } while (tsize == -1 && errno == EINTR);
381         close_preserve_errno(fd);
382     } else if ((fs_ctx->export_flags & V9FS_SM_PASSTHROUGH) ||
383                (fs_ctx->export_flags & V9FS_SM_NONE)) {
384         char *dirpath = g_path_get_dirname(fs_path->data);
385         char *name = g_path_get_basename(fs_path->data);
386         int dirfd;
387 
388         dirfd = local_opendir_nofollow(fs_ctx, dirpath);
389         if (dirfd == -1) {
390             goto out;
391         }
392 
393         tsize = readlinkat(dirfd, name, buf, bufsz);
394         close_preserve_errno(dirfd);
395     out:
396         g_free(name);
397         g_free(dirpath);
398     }
399     return tsize;
400 }
401 
402 static int local_close(FsContext *ctx, V9fsFidOpenState *fs)
403 {
404     return close(fs->fd);
405 }
406 
407 static int local_closedir(FsContext *ctx, V9fsFidOpenState *fs)
408 {
409     return closedir(fs->dir.stream);
410 }
411 
412 static int local_open(FsContext *ctx, V9fsPath *fs_path,
413                       int flags, V9fsFidOpenState *fs)
414 {
415     int fd;
416 
417     fd = local_open_nofollow(ctx, fs_path->data, flags, 0);
418     if (fd == -1) {
419         return -1;
420     }
421     fs->fd = fd;
422     return fs->fd;
423 }
424 
425 static int local_opendir(FsContext *ctx,
426                          V9fsPath *fs_path, V9fsFidOpenState *fs)
427 {
428     int dirfd;
429     DIR *stream;
430 
431     dirfd = local_opendir_nofollow(ctx, fs_path->data);
432     if (dirfd == -1) {
433         return -1;
434     }
435 
436     stream = fdopendir(dirfd);
437     if (!stream) {
438         close(dirfd);
439         return -1;
440     }
441     fs->dir.stream = stream;
442     return 0;
443 }
444 
445 static void local_rewinddir(FsContext *ctx, V9fsFidOpenState *fs)
446 {
447     rewinddir(fs->dir.stream);
448 }
449 
450 static off_t local_telldir(FsContext *ctx, V9fsFidOpenState *fs)
451 {
452     return telldir(fs->dir.stream);
453 }
454 
455 static struct dirent *local_readdir(FsContext *ctx, V9fsFidOpenState *fs)
456 {
457     struct dirent *entry;
458 
459 again:
460     entry = readdir(fs->dir.stream);
461     if (!entry) {
462         return NULL;
463     }
464 
465     if (ctx->export_flags & V9FS_SM_MAPPED) {
466         entry->d_type = DT_UNKNOWN;
467     } else if (ctx->export_flags & V9FS_SM_MAPPED_FILE) {
468         if (!strcmp(entry->d_name, VIRTFS_META_DIR)) {
469             /* skp the meta data directory */
470             goto again;
471         }
472         entry->d_type = DT_UNKNOWN;
473     }
474 
475     return entry;
476 }
477 
478 static void local_seekdir(FsContext *ctx, V9fsFidOpenState *fs, off_t off)
479 {
480     seekdir(fs->dir.stream, off);
481 }
482 
483 static ssize_t local_preadv(FsContext *ctx, V9fsFidOpenState *fs,
484                             const struct iovec *iov,
485                             int iovcnt, off_t offset)
486 {
487 #ifdef CONFIG_PREADV
488     return preadv(fs->fd, iov, iovcnt, offset);
489 #else
490     int err = lseek(fs->fd, offset, SEEK_SET);
491     if (err == -1) {
492         return err;
493     } else {
494         return readv(fs->fd, iov, iovcnt);
495     }
496 #endif
497 }
498 
499 static ssize_t local_pwritev(FsContext *ctx, V9fsFidOpenState *fs,
500                              const struct iovec *iov,
501                              int iovcnt, off_t offset)
502 {
503     ssize_t ret;
504 #ifdef CONFIG_PREADV
505     ret = pwritev(fs->fd, iov, iovcnt, offset);
506 #else
507     int err = lseek(fs->fd, offset, SEEK_SET);
508     if (err == -1) {
509         return err;
510     } else {
511         ret = writev(fs->fd, iov, iovcnt);
512     }
513 #endif
514 #ifdef CONFIG_SYNC_FILE_RANGE
515     if (ret > 0 && ctx->export_flags & V9FS_IMMEDIATE_WRITEOUT) {
516         /*
517          * Initiate a writeback. This is not a data integrity sync.
518          * We want to ensure that we don't leave dirty pages in the cache
519          * after write when writeout=immediate is sepcified.
520          */
521         sync_file_range(fs->fd, offset, ret,
522                         SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE);
523     }
524 #endif
525     return ret;
526 }
527 
528 static int local_chmod(FsContext *fs_ctx, V9fsPath *fs_path, FsCred *credp)
529 {
530     char *dirpath = g_path_get_dirname(fs_path->data);
531     char *name = g_path_get_basename(fs_path->data);
532     int ret = -1;
533     int dirfd;
534 
535     dirfd = local_opendir_nofollow(fs_ctx, dirpath);
536     if (dirfd == -1) {
537         goto out;
538     }
539 
540     if (fs_ctx->export_flags & V9FS_SM_MAPPED) {
541         ret = local_set_xattrat(dirfd, name, credp);
542     } else if (fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
543         ret = local_set_mapped_file_attrat(dirfd, name, credp);
544     } else if (fs_ctx->export_flags & V9FS_SM_PASSTHROUGH ||
545                fs_ctx->export_flags & V9FS_SM_NONE) {
546         ret = fchmodat_nofollow(dirfd, name, credp->fc_mode);
547     }
548     close_preserve_errno(dirfd);
549 
550 out:
551     g_free(dirpath);
552     g_free(name);
553     return ret;
554 }
555 
556 static int local_mknod(FsContext *fs_ctx, V9fsPath *dir_path,
557                        const char *name, FsCred *credp)
558 {
559     int err = -1;
560     int dirfd;
561 
562     dirfd = local_opendir_nofollow(fs_ctx, dir_path->data);
563     if (dirfd == -1) {
564         return -1;
565     }
566 
567     if (fs_ctx->export_flags & V9FS_SM_MAPPED ||
568         fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
569         err = mknodat(dirfd, name, SM_LOCAL_MODE_BITS | S_IFREG, 0);
570         if (err == -1) {
571             goto out;
572         }
573 
574         if (fs_ctx->export_flags & V9FS_SM_MAPPED) {
575             err = local_set_xattrat(dirfd, name, credp);
576         } else {
577             err = local_set_mapped_file_attrat(dirfd, name, credp);
578         }
579         if (err == -1) {
580             goto err_end;
581         }
582     } else if (fs_ctx->export_flags & V9FS_SM_PASSTHROUGH ||
583                fs_ctx->export_flags & V9FS_SM_NONE) {
584         err = mknodat(dirfd, name, credp->fc_mode, credp->fc_rdev);
585         if (err == -1) {
586             goto out;
587         }
588         err = local_set_cred_passthrough(fs_ctx, dirfd, name, credp);
589         if (err == -1) {
590             goto err_end;
591         }
592     }
593     goto out;
594 
595 err_end:
596     unlinkat_preserve_errno(dirfd, name, 0);
597 out:
598     close_preserve_errno(dirfd);
599     return err;
600 }
601 
602 static int local_mkdir(FsContext *fs_ctx, V9fsPath *dir_path,
603                        const char *name, FsCred *credp)
604 {
605     int err = -1;
606     int dirfd;
607 
608     dirfd = local_opendir_nofollow(fs_ctx, dir_path->data);
609     if (dirfd == -1) {
610         return -1;
611     }
612 
613     if (fs_ctx->export_flags & V9FS_SM_MAPPED ||
614         fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
615         err = mkdirat(dirfd, name, SM_LOCAL_DIR_MODE_BITS);
616         if (err == -1) {
617             goto out;
618         }
619         credp->fc_mode = credp->fc_mode | S_IFDIR;
620 
621         if (fs_ctx->export_flags & V9FS_SM_MAPPED) {
622             err = local_set_xattrat(dirfd, name, credp);
623         } else {
624             err = local_set_mapped_file_attrat(dirfd, name, credp);
625         }
626         if (err == -1) {
627             goto err_end;
628         }
629     } else if (fs_ctx->export_flags & V9FS_SM_PASSTHROUGH ||
630                fs_ctx->export_flags & V9FS_SM_NONE) {
631         err = mkdirat(dirfd, name, credp->fc_mode);
632         if (err == -1) {
633             goto out;
634         }
635         err = local_set_cred_passthrough(fs_ctx, dirfd, name, credp);
636         if (err == -1) {
637             goto err_end;
638         }
639     }
640     goto out;
641 
642 err_end:
643     unlinkat_preserve_errno(dirfd, name, AT_REMOVEDIR);
644 out:
645     close_preserve_errno(dirfd);
646     return err;
647 }
648 
649 static int local_fstat(FsContext *fs_ctx, int fid_type,
650                        V9fsFidOpenState *fs, struct stat *stbuf)
651 {
652     int err, fd;
653 
654     if (fid_type == P9_FID_DIR) {
655         fd = dirfd(fs->dir.stream);
656     } else {
657         fd = fs->fd;
658     }
659 
660     err = fstat(fd, stbuf);
661     if (err) {
662         return err;
663     }
664     if (fs_ctx->export_flags & V9FS_SM_MAPPED) {
665         /* Actual credentials are part of extended attrs */
666         uid_t tmp_uid;
667         gid_t tmp_gid;
668         mode_t tmp_mode;
669         dev_t tmp_dev;
670 
671         if (fgetxattr(fd, "user.virtfs.uid", &tmp_uid, sizeof(uid_t)) > 0) {
672             stbuf->st_uid = le32_to_cpu(tmp_uid);
673         }
674         if (fgetxattr(fd, "user.virtfs.gid", &tmp_gid, sizeof(gid_t)) > 0) {
675             stbuf->st_gid = le32_to_cpu(tmp_gid);
676         }
677         if (fgetxattr(fd, "user.virtfs.mode", &tmp_mode, sizeof(mode_t)) > 0) {
678             stbuf->st_mode = le32_to_cpu(tmp_mode);
679         }
680         if (fgetxattr(fd, "user.virtfs.rdev", &tmp_dev, sizeof(dev_t)) > 0) {
681             stbuf->st_rdev = le64_to_cpu(tmp_dev);
682         }
683     } else if (fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
684         errno = EOPNOTSUPP;
685         return -1;
686     }
687     return err;
688 }
689 
690 static int local_open2(FsContext *fs_ctx, V9fsPath *dir_path, const char *name,
691                        int flags, FsCred *credp, V9fsFidOpenState *fs)
692 {
693     int fd = -1;
694     int err = -1;
695     int dirfd;
696 
697     /*
698      * Mark all the open to not follow symlinks
699      */
700     flags |= O_NOFOLLOW;
701 
702     dirfd = local_opendir_nofollow(fs_ctx, dir_path->data);
703     if (dirfd == -1) {
704         return -1;
705     }
706 
707     /* Determine the security model */
708     if (fs_ctx->export_flags & V9FS_SM_MAPPED ||
709         fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
710         fd = openat_file(dirfd, name, flags, SM_LOCAL_MODE_BITS);
711         if (fd == -1) {
712             goto out;
713         }
714         credp->fc_mode = credp->fc_mode|S_IFREG;
715         if (fs_ctx->export_flags & V9FS_SM_MAPPED) {
716             /* Set cleint credentials in xattr */
717             err = local_set_xattrat(dirfd, name, credp);
718         } else {
719             err = local_set_mapped_file_attrat(dirfd, name, credp);
720         }
721         if (err == -1) {
722             goto err_end;
723         }
724     } else if ((fs_ctx->export_flags & V9FS_SM_PASSTHROUGH) ||
725                (fs_ctx->export_flags & V9FS_SM_NONE)) {
726         fd = openat_file(dirfd, name, flags, credp->fc_mode);
727         if (fd == -1) {
728             goto out;
729         }
730         err = local_set_cred_passthrough(fs_ctx, dirfd, name, credp);
731         if (err == -1) {
732             goto err_end;
733         }
734     }
735     err = fd;
736     fs->fd = fd;
737     goto out;
738 
739 err_end:
740     unlinkat_preserve_errno(dirfd, name,
741                             flags & O_DIRECTORY ? AT_REMOVEDIR : 0);
742     close_preserve_errno(fd);
743 out:
744     close_preserve_errno(dirfd);
745     return err;
746 }
747 
748 
749 static int local_symlink(FsContext *fs_ctx, const char *oldpath,
750                          V9fsPath *dir_path, const char *name, FsCred *credp)
751 {
752     int err = -1;
753     int dirfd;
754 
755     dirfd = local_opendir_nofollow(fs_ctx, dir_path->data);
756     if (dirfd == -1) {
757         return -1;
758     }
759 
760     /* Determine the security model */
761     if (fs_ctx->export_flags & V9FS_SM_MAPPED ||
762         fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
763         int fd;
764         ssize_t oldpath_size, write_size;
765 
766         fd = openat_file(dirfd, name, O_CREAT | O_EXCL | O_RDWR,
767                          SM_LOCAL_MODE_BITS);
768         if (fd == -1) {
769             goto out;
770         }
771         /* Write the oldpath (target) to the file. */
772         oldpath_size = strlen(oldpath);
773         do {
774             write_size = write(fd, (void *)oldpath, oldpath_size);
775         } while (write_size == -1 && errno == EINTR);
776         close_preserve_errno(fd);
777 
778         if (write_size != oldpath_size) {
779             goto err_end;
780         }
781         /* Set cleint credentials in symlink's xattr */
782         credp->fc_mode = credp->fc_mode | S_IFLNK;
783 
784         if (fs_ctx->export_flags & V9FS_SM_MAPPED) {
785             err = local_set_xattrat(dirfd, name, credp);
786         } else {
787             err = local_set_mapped_file_attrat(dirfd, name, credp);
788         }
789         if (err == -1) {
790             goto err_end;
791         }
792     } else if (fs_ctx->export_flags & V9FS_SM_PASSTHROUGH ||
793                fs_ctx->export_flags & V9FS_SM_NONE) {
794         err = symlinkat(oldpath, dirfd, name);
795         if (err) {
796             goto out;
797         }
798         err = fchownat(dirfd, name, credp->fc_uid, credp->fc_gid,
799                        AT_SYMLINK_NOFOLLOW);
800         if (err == -1) {
801             /*
802              * If we fail to change ownership and if we are
803              * using security model none. Ignore the error
804              */
805             if ((fs_ctx->export_flags & V9FS_SEC_MASK) != V9FS_SM_NONE) {
806                 goto err_end;
807             } else {
808                 err = 0;
809             }
810         }
811     }
812     goto out;
813 
814 err_end:
815     unlinkat_preserve_errno(dirfd, name, 0);
816 out:
817     close_preserve_errno(dirfd);
818     return err;
819 }
820 
821 static int local_link(FsContext *ctx, V9fsPath *oldpath,
822                       V9fsPath *dirpath, const char *name)
823 {
824     char *odirpath = g_path_get_dirname(oldpath->data);
825     char *oname = g_path_get_basename(oldpath->data);
826     int ret = -1;
827     int odirfd, ndirfd;
828 
829     odirfd = local_opendir_nofollow(ctx, odirpath);
830     if (odirfd == -1) {
831         goto out;
832     }
833 
834     ndirfd = local_opendir_nofollow(ctx, dirpath->data);
835     if (ndirfd == -1) {
836         close_preserve_errno(odirfd);
837         goto out;
838     }
839 
840     ret = linkat(odirfd, oname, ndirfd, name, 0);
841     if (ret < 0) {
842         goto out_close;
843     }
844 
845     /* now link the virtfs_metadata files */
846     if (ctx->export_flags & V9FS_SM_MAPPED_FILE) {
847         int omap_dirfd, nmap_dirfd;
848 
849         ret = mkdirat(ndirfd, VIRTFS_META_DIR, 0700);
850         if (ret < 0 && errno != EEXIST) {
851             goto err_undo_link;
852         }
853 
854         omap_dirfd = openat_dir(odirfd, VIRTFS_META_DIR);
855         if (omap_dirfd == -1) {
856             goto err;
857         }
858 
859         nmap_dirfd = openat_dir(ndirfd, VIRTFS_META_DIR);
860         if (nmap_dirfd == -1) {
861             close_preserve_errno(omap_dirfd);
862             goto err;
863         }
864 
865         ret = linkat(omap_dirfd, oname, nmap_dirfd, name, 0);
866         close_preserve_errno(nmap_dirfd);
867         close_preserve_errno(omap_dirfd);
868         if (ret < 0 && errno != ENOENT) {
869             goto err_undo_link;
870         }
871 
872         ret = 0;
873     }
874     goto out_close;
875 
876 err:
877     ret = -1;
878 err_undo_link:
879     unlinkat_preserve_errno(ndirfd, name, 0);
880 out_close:
881     close_preserve_errno(ndirfd);
882     close_preserve_errno(odirfd);
883 out:
884     g_free(oname);
885     g_free(odirpath);
886     return ret;
887 }
888 
889 static int local_truncate(FsContext *ctx, V9fsPath *fs_path, off_t size)
890 {
891     int fd, ret;
892 
893     fd = local_open_nofollow(ctx, fs_path->data, O_WRONLY, 0);
894     if (fd == -1) {
895         return -1;
896     }
897     ret = ftruncate(fd, size);
898     close_preserve_errno(fd);
899     return ret;
900 }
901 
902 static int local_chown(FsContext *fs_ctx, V9fsPath *fs_path, FsCred *credp)
903 {
904     char *dirpath = g_path_get_dirname(fs_path->data);
905     char *name = g_path_get_basename(fs_path->data);
906     int ret = -1;
907     int dirfd;
908 
909     dirfd = local_opendir_nofollow(fs_ctx, dirpath);
910     if (dirfd == -1) {
911         goto out;
912     }
913 
914     if ((credp->fc_uid == -1 && credp->fc_gid == -1) ||
915         (fs_ctx->export_flags & V9FS_SM_PASSTHROUGH) ||
916         (fs_ctx->export_flags & V9FS_SM_NONE)) {
917         ret = fchownat(dirfd, name, credp->fc_uid, credp->fc_gid,
918                        AT_SYMLINK_NOFOLLOW);
919     } else if (fs_ctx->export_flags & V9FS_SM_MAPPED) {
920         ret = local_set_xattrat(dirfd, name, credp);
921     } else if (fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
922         ret = local_set_mapped_file_attrat(dirfd, name, credp);
923     }
924 
925     close_preserve_errno(dirfd);
926 out:
927     g_free(name);
928     g_free(dirpath);
929     return ret;
930 }
931 
932 static int local_utimensat(FsContext *s, V9fsPath *fs_path,
933                            const struct timespec *buf)
934 {
935     char *dirpath = g_path_get_dirname(fs_path->data);
936     char *name = g_path_get_basename(fs_path->data);
937     int dirfd, ret = -1;
938 
939     dirfd = local_opendir_nofollow(s, dirpath);
940     if (dirfd == -1) {
941         goto out;
942     }
943 
944     ret = utimensat(dirfd, name, buf, AT_SYMLINK_NOFOLLOW);
945     close_preserve_errno(dirfd);
946 out:
947     g_free(dirpath);
948     g_free(name);
949     return ret;
950 }
951 
952 static int local_unlinkat_common(FsContext *ctx, int dirfd, const char *name,
953                                  int flags)
954 {
955     int ret = -1;
956 
957     if (ctx->export_flags & V9FS_SM_MAPPED_FILE) {
958         int map_dirfd;
959 
960         if (flags == AT_REMOVEDIR) {
961             int fd;
962 
963             fd = openat_dir(dirfd, name);
964             if (fd == -1) {
965                 goto err_out;
966             }
967             /*
968              * If directory remove .virtfs_metadata contained in the
969              * directory
970              */
971             ret = unlinkat(fd, VIRTFS_META_DIR, AT_REMOVEDIR);
972             close_preserve_errno(fd);
973             if (ret < 0 && errno != ENOENT) {
974                 /*
975                  * We didn't had the .virtfs_metadata file. May be file created
976                  * in non-mapped mode ?. Ignore ENOENT.
977                  */
978                 goto err_out;
979             }
980         }
981         /*
982          * Now remove the name from parent directory
983          * .virtfs_metadata directory.
984          */
985         map_dirfd = openat_dir(dirfd, VIRTFS_META_DIR);
986         ret = unlinkat(map_dirfd, name, 0);
987         close_preserve_errno(map_dirfd);
988         if (ret < 0 && errno != ENOENT) {
989             /*
990              * We didn't had the .virtfs_metadata file. May be file created
991              * in non-mapped mode ?. Ignore ENOENT.
992              */
993             goto err_out;
994         }
995     }
996 
997     ret = unlinkat(dirfd, name, flags);
998 err_out:
999     return ret;
1000 }
1001 
1002 static int local_remove(FsContext *ctx, const char *path)
1003 {
1004     struct stat stbuf;
1005     char *dirpath = g_path_get_dirname(path);
1006     char *name = g_path_get_basename(path);
1007     int flags = 0;
1008     int dirfd;
1009     int err = -1;
1010 
1011     dirfd = local_opendir_nofollow(ctx, dirpath);
1012     if (dirfd == -1) {
1013         goto out;
1014     }
1015 
1016     if (fstatat(dirfd, path, &stbuf, AT_SYMLINK_NOFOLLOW) < 0) {
1017         goto err_out;
1018     }
1019 
1020     if (S_ISDIR(stbuf.st_mode)) {
1021         flags |= AT_REMOVEDIR;
1022     }
1023 
1024     err = local_unlinkat_common(ctx, dirfd, name, flags);
1025 err_out:
1026     close_preserve_errno(dirfd);
1027 out:
1028     g_free(name);
1029     g_free(dirpath);
1030     return err;
1031 }
1032 
1033 static int local_fsync(FsContext *ctx, int fid_type,
1034                        V9fsFidOpenState *fs, int datasync)
1035 {
1036     int fd;
1037 
1038     if (fid_type == P9_FID_DIR) {
1039         fd = dirfd(fs->dir.stream);
1040     } else {
1041         fd = fs->fd;
1042     }
1043 
1044     if (datasync) {
1045         return qemu_fdatasync(fd);
1046     } else {
1047         return fsync(fd);
1048     }
1049 }
1050 
1051 static int local_statfs(FsContext *s, V9fsPath *fs_path, struct statfs *stbuf)
1052 {
1053     int fd, ret;
1054 
1055     fd = local_open_nofollow(s, fs_path->data, O_RDONLY, 0);
1056     if (fd == -1) {
1057         return -1;
1058     }
1059     ret = fstatfs(fd, stbuf);
1060     close_preserve_errno(fd);
1061     return ret;
1062 }
1063 
1064 static ssize_t local_lgetxattr(FsContext *ctx, V9fsPath *fs_path,
1065                                const char *name, void *value, size_t size)
1066 {
1067     char *path = fs_path->data;
1068 
1069     return v9fs_get_xattr(ctx, path, name, value, size);
1070 }
1071 
1072 static ssize_t local_llistxattr(FsContext *ctx, V9fsPath *fs_path,
1073                                 void *value, size_t size)
1074 {
1075     char *path = fs_path->data;
1076 
1077     return v9fs_list_xattr(ctx, path, value, size);
1078 }
1079 
1080 static int local_lsetxattr(FsContext *ctx, V9fsPath *fs_path, const char *name,
1081                            void *value, size_t size, int flags)
1082 {
1083     char *path = fs_path->data;
1084 
1085     return v9fs_set_xattr(ctx, path, name, value, size, flags);
1086 }
1087 
1088 static int local_lremovexattr(FsContext *ctx, V9fsPath *fs_path,
1089                               const char *name)
1090 {
1091     char *path = fs_path->data;
1092 
1093     return v9fs_remove_xattr(ctx, path, name);
1094 }
1095 
1096 static int local_name_to_path(FsContext *ctx, V9fsPath *dir_path,
1097                               const char *name, V9fsPath *target)
1098 {
1099     if (dir_path) {
1100         v9fs_path_sprintf(target, "%s/%s", dir_path->data, name);
1101     } else {
1102         v9fs_path_sprintf(target, "%s", name);
1103     }
1104     return 0;
1105 }
1106 
1107 static int local_renameat(FsContext *ctx, V9fsPath *olddir,
1108                           const char *old_name, V9fsPath *newdir,
1109                           const char *new_name)
1110 {
1111     int ret;
1112     int odirfd, ndirfd;
1113 
1114     odirfd = local_opendir_nofollow(ctx, olddir->data);
1115     if (odirfd == -1) {
1116         return -1;
1117     }
1118 
1119     ndirfd = local_opendir_nofollow(ctx, newdir->data);
1120     if (ndirfd == -1) {
1121         close_preserve_errno(odirfd);
1122         return -1;
1123     }
1124 
1125     ret = renameat(odirfd, old_name, ndirfd, new_name);
1126     if (ret < 0) {
1127         goto out;
1128     }
1129 
1130     if (ctx->export_flags & V9FS_SM_MAPPED_FILE) {
1131         int omap_dirfd, nmap_dirfd;
1132 
1133         ret = mkdirat(ndirfd, VIRTFS_META_DIR, 0700);
1134         if (ret < 0 && errno != EEXIST) {
1135             goto err_undo_rename;
1136         }
1137 
1138         omap_dirfd = openat_dir(odirfd, VIRTFS_META_DIR);
1139         if (omap_dirfd == -1) {
1140             goto err;
1141         }
1142 
1143         nmap_dirfd = openat_dir(ndirfd, VIRTFS_META_DIR);
1144         if (nmap_dirfd == -1) {
1145             close_preserve_errno(omap_dirfd);
1146             goto err;
1147         }
1148 
1149         /* rename the .virtfs_metadata files */
1150         ret = renameat(omap_dirfd, old_name, nmap_dirfd, new_name);
1151         close_preserve_errno(nmap_dirfd);
1152         close_preserve_errno(omap_dirfd);
1153         if (ret < 0 && errno != ENOENT) {
1154             goto err_undo_rename;
1155         }
1156 
1157         ret = 0;
1158     }
1159     goto out;
1160 
1161 err:
1162     ret = -1;
1163 err_undo_rename:
1164     renameat_preserve_errno(ndirfd, new_name, odirfd, old_name);
1165 out:
1166     close_preserve_errno(ndirfd);
1167     close_preserve_errno(odirfd);
1168     return ret;
1169 }
1170 
1171 static void v9fs_path_init_dirname(V9fsPath *path, const char *str)
1172 {
1173     path->data = g_path_get_dirname(str);
1174     path->size = strlen(path->data) + 1;
1175 }
1176 
1177 static int local_rename(FsContext *ctx, const char *oldpath,
1178                         const char *newpath)
1179 {
1180     int err;
1181     char *oname = g_path_get_basename(oldpath);
1182     char *nname = g_path_get_basename(newpath);
1183     V9fsPath olddir, newdir;
1184 
1185     v9fs_path_init_dirname(&olddir, oldpath);
1186     v9fs_path_init_dirname(&newdir, newpath);
1187 
1188     err = local_renameat(ctx, &olddir, oname, &newdir, nname);
1189 
1190     v9fs_path_free(&newdir);
1191     v9fs_path_free(&olddir);
1192     g_free(nname);
1193     g_free(oname);
1194 
1195     return err;
1196 }
1197 
1198 static int local_unlinkat(FsContext *ctx, V9fsPath *dir,
1199                           const char *name, int flags)
1200 {
1201     int ret;
1202     int dirfd;
1203 
1204     dirfd = local_opendir_nofollow(ctx, dir->data);
1205     if (dirfd == -1) {
1206         return -1;
1207     }
1208 
1209     ret = local_unlinkat_common(ctx, dirfd, name, flags);
1210     close_preserve_errno(dirfd);
1211     return ret;
1212 }
1213 
1214 static int local_ioc_getversion(FsContext *ctx, V9fsPath *path,
1215                                 mode_t st_mode, uint64_t *st_gen)
1216 {
1217 #ifdef FS_IOC_GETVERSION
1218     int err;
1219     V9fsFidOpenState fid_open;
1220 
1221     /*
1222      * Do not try to open special files like device nodes, fifos etc
1223      * We can get fd for regular files and directories only
1224      */
1225     if (!S_ISREG(st_mode) && !S_ISDIR(st_mode)) {
1226         errno = ENOTTY;
1227         return -1;
1228     }
1229     err = local_open(ctx, path, O_RDONLY, &fid_open);
1230     if (err < 0) {
1231         return err;
1232     }
1233     err = ioctl(fid_open.fd, FS_IOC_GETVERSION, st_gen);
1234     local_close(ctx, &fid_open);
1235     return err;
1236 #else
1237     errno = ENOTTY;
1238     return -1;
1239 #endif
1240 }
1241 
1242 static int local_init(FsContext *ctx)
1243 {
1244     struct statfs stbuf;
1245     LocalData *data = g_malloc(sizeof(*data));
1246 
1247     data->mountfd = open(ctx->fs_root, O_DIRECTORY | O_RDONLY);
1248     if (data->mountfd == -1) {
1249         goto err;
1250     }
1251 
1252 #ifdef FS_IOC_GETVERSION
1253     /*
1254      * use ioc_getversion only if the ioctl is definied
1255      */
1256     if (fstatfs(data->mountfd, &stbuf) < 0) {
1257         close_preserve_errno(data->mountfd);
1258         goto err;
1259     }
1260     switch (stbuf.f_type) {
1261     case EXT2_SUPER_MAGIC:
1262     case BTRFS_SUPER_MAGIC:
1263     case REISERFS_SUPER_MAGIC:
1264     case XFS_SUPER_MAGIC:
1265         ctx->exops.get_st_gen = local_ioc_getversion;
1266         break;
1267     }
1268 #endif
1269 
1270     if (ctx->export_flags & V9FS_SM_PASSTHROUGH) {
1271         ctx->xops = passthrough_xattr_ops;
1272     } else if (ctx->export_flags & V9FS_SM_MAPPED) {
1273         ctx->xops = mapped_xattr_ops;
1274     } else if (ctx->export_flags & V9FS_SM_NONE) {
1275         ctx->xops = none_xattr_ops;
1276     } else if (ctx->export_flags & V9FS_SM_MAPPED_FILE) {
1277         /*
1278          * xattr operation for mapped-file and passthrough
1279          * remain same.
1280          */
1281         ctx->xops = passthrough_xattr_ops;
1282     }
1283     ctx->export_flags |= V9FS_PATHNAME_FSCONTEXT;
1284 
1285     ctx->private = data;
1286     return 0;
1287 
1288 err:
1289     g_free(data);
1290     return -1;
1291 }
1292 
1293 static void local_cleanup(FsContext *ctx)
1294 {
1295     LocalData *data = ctx->private;
1296 
1297     close(data->mountfd);
1298     g_free(data);
1299 }
1300 
1301 static int local_parse_opts(QemuOpts *opts, struct FsDriverEntry *fse)
1302 {
1303     const char *sec_model = qemu_opt_get(opts, "security_model");
1304     const char *path = qemu_opt_get(opts, "path");
1305     Error *err = NULL;
1306 
1307     if (!sec_model) {
1308         error_report("Security model not specified, local fs needs security model");
1309         error_printf("valid options are:"
1310                      "\tsecurity_model=[passthrough|mapped-xattr|mapped-file|none]\n");
1311         return -1;
1312     }
1313 
1314     if (!strcmp(sec_model, "passthrough")) {
1315         fse->export_flags |= V9FS_SM_PASSTHROUGH;
1316     } else if (!strcmp(sec_model, "mapped") ||
1317                !strcmp(sec_model, "mapped-xattr")) {
1318         fse->export_flags |= V9FS_SM_MAPPED;
1319     } else if (!strcmp(sec_model, "none")) {
1320         fse->export_flags |= V9FS_SM_NONE;
1321     } else if (!strcmp(sec_model, "mapped-file")) {
1322         fse->export_flags |= V9FS_SM_MAPPED_FILE;
1323     } else {
1324         error_report("Invalid security model %s specified", sec_model);
1325         error_printf("valid options are:"
1326                      "\t[passthrough|mapped-xattr|mapped-file|none]\n");
1327         return -1;
1328     }
1329 
1330     if (!path) {
1331         error_report("fsdev: No path specified");
1332         return -1;
1333     }
1334 
1335     fsdev_throttle_parse_opts(opts, &fse->fst, &err);
1336     if (err) {
1337         error_reportf_err(err, "Throttle configuration is not valid: ");
1338         return -1;
1339     }
1340 
1341     fse->path = g_strdup(path);
1342 
1343     return 0;
1344 }
1345 
1346 FileOperations local_ops = {
1347     .parse_opts = local_parse_opts,
1348     .init  = local_init,
1349     .cleanup = local_cleanup,
1350     .lstat = local_lstat,
1351     .readlink = local_readlink,
1352     .close = local_close,
1353     .closedir = local_closedir,
1354     .open = local_open,
1355     .opendir = local_opendir,
1356     .rewinddir = local_rewinddir,
1357     .telldir = local_telldir,
1358     .readdir = local_readdir,
1359     .seekdir = local_seekdir,
1360     .preadv = local_preadv,
1361     .pwritev = local_pwritev,
1362     .chmod = local_chmod,
1363     .mknod = local_mknod,
1364     .mkdir = local_mkdir,
1365     .fstat = local_fstat,
1366     .open2 = local_open2,
1367     .symlink = local_symlink,
1368     .link = local_link,
1369     .truncate = local_truncate,
1370     .rename = local_rename,
1371     .chown = local_chown,
1372     .utimensat = local_utimensat,
1373     .remove = local_remove,
1374     .fsync = local_fsync,
1375     .statfs = local_statfs,
1376     .lgetxattr = local_lgetxattr,
1377     .llistxattr = local_llistxattr,
1378     .lsetxattr = local_lsetxattr,
1379     .lremovexattr = local_lremovexattr,
1380     .name_to_path = local_name_to_path,
1381     .renameat  = local_renameat,
1382     .unlinkat = local_unlinkat,
1383 };
1384