Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20200803' into...
[qemu.git] / hw / 9pfs / 9p.c
1 /*
2 * Virtio 9p backend
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 <glib/gprintf.h>
16 #include "hw/virtio/virtio.h"
17 #include "qapi/error.h"
18 #include "qemu/error-report.h"
19 #include "qemu/iov.h"
20 #include "qemu/main-loop.h"
21 #include "qemu/sockets.h"
22 #include "virtio-9p.h"
23 #include "fsdev/qemu-fsdev.h"
24 #include "9p-xattr.h"
25 #include "coth.h"
26 #include "trace.h"
27 #include "migration/blocker.h"
28 #include "sysemu/qtest.h"
29 #include "qemu/xxhash.h"
30 #include <math.h>
31 #include <linux/limits.h>
32
33 int open_fd_hw;
34 int total_open_fd;
35 static int open_fd_rc;
36
37 enum {
38 Oread = 0x00,
39 Owrite = 0x01,
40 Ordwr = 0x02,
41 Oexec = 0x03,
42 Oexcl = 0x04,
43 Otrunc = 0x10,
44 Orexec = 0x20,
45 Orclose = 0x40,
46 Oappend = 0x80,
47 };
48
49 static ssize_t pdu_marshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
50 {
51 ssize_t ret;
52 va_list ap;
53
54 va_start(ap, fmt);
55 ret = pdu->s->transport->pdu_vmarshal(pdu, offset, fmt, ap);
56 va_end(ap);
57
58 return ret;
59 }
60
61 static ssize_t pdu_unmarshal(V9fsPDU *pdu, size_t offset, const char *fmt, ...)
62 {
63 ssize_t ret;
64 va_list ap;
65
66 va_start(ap, fmt);
67 ret = pdu->s->transport->pdu_vunmarshal(pdu, offset, fmt, ap);
68 va_end(ap);
69
70 return ret;
71 }
72
73 static int omode_to_uflags(int8_t mode)
74 {
75 int ret = 0;
76
77 switch (mode & 3) {
78 case Oread:
79 ret = O_RDONLY;
80 break;
81 case Ordwr:
82 ret = O_RDWR;
83 break;
84 case Owrite:
85 ret = O_WRONLY;
86 break;
87 case Oexec:
88 ret = O_RDONLY;
89 break;
90 }
91
92 if (mode & Otrunc) {
93 ret |= O_TRUNC;
94 }
95
96 if (mode & Oappend) {
97 ret |= O_APPEND;
98 }
99
100 if (mode & Oexcl) {
101 ret |= O_EXCL;
102 }
103
104 return ret;
105 }
106
107 typedef struct DotlOpenflagMap {
108 int dotl_flag;
109 int open_flag;
110 } DotlOpenflagMap;
111
112 static int dotl_to_open_flags(int flags)
113 {
114 int i;
115 /*
116 * We have same bits for P9_DOTL_READONLY, P9_DOTL_WRONLY
117 * and P9_DOTL_NOACCESS
118 */
119 int oflags = flags & O_ACCMODE;
120
121 DotlOpenflagMap dotl_oflag_map[] = {
122 { P9_DOTL_CREATE, O_CREAT },
123 { P9_DOTL_EXCL, O_EXCL },
124 { P9_DOTL_NOCTTY , O_NOCTTY },
125 { P9_DOTL_TRUNC, O_TRUNC },
126 { P9_DOTL_APPEND, O_APPEND },
127 { P9_DOTL_NONBLOCK, O_NONBLOCK } ,
128 { P9_DOTL_DSYNC, O_DSYNC },
129 { P9_DOTL_FASYNC, FASYNC },
130 { P9_DOTL_DIRECT, O_DIRECT },
131 { P9_DOTL_LARGEFILE, O_LARGEFILE },
132 { P9_DOTL_DIRECTORY, O_DIRECTORY },
133 { P9_DOTL_NOFOLLOW, O_NOFOLLOW },
134 { P9_DOTL_NOATIME, O_NOATIME },
135 { P9_DOTL_SYNC, O_SYNC },
136 };
137
138 for (i = 0; i < ARRAY_SIZE(dotl_oflag_map); i++) {
139 if (flags & dotl_oflag_map[i].dotl_flag) {
140 oflags |= dotl_oflag_map[i].open_flag;
141 }
142 }
143
144 return oflags;
145 }
146
147 void cred_init(FsCred *credp)
148 {
149 credp->fc_uid = -1;
150 credp->fc_gid = -1;
151 credp->fc_mode = -1;
152 credp->fc_rdev = -1;
153 }
154
155 static int get_dotl_openflags(V9fsState *s, int oflags)
156 {
157 int flags;
158 /*
159 * Filter the client open flags
160 */
161 flags = dotl_to_open_flags(oflags);
162 flags &= ~(O_NOCTTY | O_ASYNC | O_CREAT);
163 /*
164 * Ignore direct disk access hint until the server supports it.
165 */
166 flags &= ~O_DIRECT;
167 return flags;
168 }
169
170 void v9fs_path_init(V9fsPath *path)
171 {
172 path->data = NULL;
173 path->size = 0;
174 }
175
176 void v9fs_path_free(V9fsPath *path)
177 {
178 g_free(path->data);
179 path->data = NULL;
180 path->size = 0;
181 }
182
183
184 void GCC_FMT_ATTR(2, 3)
185 v9fs_path_sprintf(V9fsPath *path, const char *fmt, ...)
186 {
187 va_list ap;
188
189 v9fs_path_free(path);
190
191 va_start(ap, fmt);
192 /* Bump the size for including terminating NULL */
193 path->size = g_vasprintf(&path->data, fmt, ap) + 1;
194 va_end(ap);
195 }
196
197 void v9fs_path_copy(V9fsPath *dst, const V9fsPath *src)
198 {
199 v9fs_path_free(dst);
200 dst->size = src->size;
201 dst->data = g_memdup(src->data, src->size);
202 }
203
204 int v9fs_name_to_path(V9fsState *s, V9fsPath *dirpath,
205 const char *name, V9fsPath *path)
206 {
207 int err;
208 err = s->ops->name_to_path(&s->ctx, dirpath, name, path);
209 if (err < 0) {
210 err = -errno;
211 }
212 return err;
213 }
214
215 /*
216 * Return TRUE if s1 is an ancestor of s2.
217 *
218 * E.g. "a/b" is an ancestor of "a/b/c" but not of "a/bc/d".
219 * As a special case, We treat s1 as ancestor of s2 if they are same!
220 */
221 static int v9fs_path_is_ancestor(V9fsPath *s1, V9fsPath *s2)
222 {
223 if (!strncmp(s1->data, s2->data, s1->size - 1)) {
224 if (s2->data[s1->size - 1] == '\0' || s2->data[s1->size - 1] == '/') {
225 return 1;
226 }
227 }
228 return 0;
229 }
230
231 static size_t v9fs_string_size(V9fsString *str)
232 {
233 return str->size;
234 }
235
236 /*
237 * returns 0 if fid got re-opened, 1 if not, < 0 on error */
238 static int coroutine_fn v9fs_reopen_fid(V9fsPDU *pdu, V9fsFidState *f)
239 {
240 int err = 1;
241 if (f->fid_type == P9_FID_FILE) {
242 if (f->fs.fd == -1) {
243 do {
244 err = v9fs_co_open(pdu, f, f->open_flags);
245 } while (err == -EINTR && !pdu->cancelled);
246 }
247 } else if (f->fid_type == P9_FID_DIR) {
248 if (f->fs.dir.stream == NULL) {
249 do {
250 err = v9fs_co_opendir(pdu, f);
251 } while (err == -EINTR && !pdu->cancelled);
252 }
253 }
254 return err;
255 }
256
257 static V9fsFidState *coroutine_fn get_fid(V9fsPDU *pdu, int32_t fid)
258 {
259 int err;
260 V9fsFidState *f;
261 V9fsState *s = pdu->s;
262
263 for (f = s->fid_list; f; f = f->next) {
264 BUG_ON(f->clunked);
265 if (f->fid == fid) {
266 /*
267 * Update the fid ref upfront so that
268 * we don't get reclaimed when we yield
269 * in open later.
270 */
271 f->ref++;
272 /*
273 * check whether we need to reopen the
274 * file. We might have closed the fd
275 * while trying to free up some file
276 * descriptors.
277 */
278 err = v9fs_reopen_fid(pdu, f);
279 if (err < 0) {
280 f->ref--;
281 return NULL;
282 }
283 /*
284 * Mark the fid as referenced so that the LRU
285 * reclaim won't close the file descriptor
286 */
287 f->flags |= FID_REFERENCED;
288 return f;
289 }
290 }
291 return NULL;
292 }
293
294 static V9fsFidState *alloc_fid(V9fsState *s, int32_t fid)
295 {
296 V9fsFidState *f;
297
298 for (f = s->fid_list; f; f = f->next) {
299 /* If fid is already there return NULL */
300 BUG_ON(f->clunked);
301 if (f->fid == fid) {
302 return NULL;
303 }
304 }
305 f = g_malloc0(sizeof(V9fsFidState));
306 f->fid = fid;
307 f->fid_type = P9_FID_NONE;
308 f->ref = 1;
309 /*
310 * Mark the fid as referenced so that the LRU
311 * reclaim won't close the file descriptor
312 */
313 f->flags |= FID_REFERENCED;
314 f->next = s->fid_list;
315 s->fid_list = f;
316
317 v9fs_readdir_init(&f->fs.dir);
318 v9fs_readdir_init(&f->fs_reclaim.dir);
319
320 return f;
321 }
322
323 static int coroutine_fn v9fs_xattr_fid_clunk(V9fsPDU *pdu, V9fsFidState *fidp)
324 {
325 int retval = 0;
326
327 if (fidp->fs.xattr.xattrwalk_fid) {
328 /* getxattr/listxattr fid */
329 goto free_value;
330 }
331 /*
332 * if this is fid for setxattr. clunk should
333 * result in setxattr localcall
334 */
335 if (fidp->fs.xattr.len != fidp->fs.xattr.copied_len) {
336 /* clunk after partial write */
337 retval = -EINVAL;
338 goto free_out;
339 }
340 if (fidp->fs.xattr.len) {
341 retval = v9fs_co_lsetxattr(pdu, &fidp->path, &fidp->fs.xattr.name,
342 fidp->fs.xattr.value,
343 fidp->fs.xattr.len,
344 fidp->fs.xattr.flags);
345 } else {
346 retval = v9fs_co_lremovexattr(pdu, &fidp->path, &fidp->fs.xattr.name);
347 }
348 free_out:
349 v9fs_string_free(&fidp->fs.xattr.name);
350 free_value:
351 g_free(fidp->fs.xattr.value);
352 return retval;
353 }
354
355 static int coroutine_fn free_fid(V9fsPDU *pdu, V9fsFidState *fidp)
356 {
357 int retval = 0;
358
359 if (fidp->fid_type == P9_FID_FILE) {
360 /* If we reclaimed the fd no need to close */
361 if (fidp->fs.fd != -1) {
362 retval = v9fs_co_close(pdu, &fidp->fs);
363 }
364 } else if (fidp->fid_type == P9_FID_DIR) {
365 if (fidp->fs.dir.stream != NULL) {
366 retval = v9fs_co_closedir(pdu, &fidp->fs);
367 }
368 } else if (fidp->fid_type == P9_FID_XATTR) {
369 retval = v9fs_xattr_fid_clunk(pdu, fidp);
370 }
371 v9fs_path_free(&fidp->path);
372 g_free(fidp);
373 return retval;
374 }
375
376 static int coroutine_fn put_fid(V9fsPDU *pdu, V9fsFidState *fidp)
377 {
378 BUG_ON(!fidp->ref);
379 fidp->ref--;
380 /*
381 * Don't free the fid if it is in reclaim list
382 */
383 if (!fidp->ref && fidp->clunked) {
384 if (fidp->fid == pdu->s->root_fid) {
385 /*
386 * if the clunked fid is root fid then we
387 * have unmounted the fs on the client side.
388 * delete the migration blocker. Ideally, this
389 * should be hooked to transport close notification
390 */
391 if (pdu->s->migration_blocker) {
392 migrate_del_blocker(pdu->s->migration_blocker);
393 error_free(pdu->s->migration_blocker);
394 pdu->s->migration_blocker = NULL;
395 }
396 }
397 return free_fid(pdu, fidp);
398 }
399 return 0;
400 }
401
402 static V9fsFidState *clunk_fid(V9fsState *s, int32_t fid)
403 {
404 V9fsFidState **fidpp, *fidp;
405
406 for (fidpp = &s->fid_list; *fidpp; fidpp = &(*fidpp)->next) {
407 if ((*fidpp)->fid == fid) {
408 break;
409 }
410 }
411 if (*fidpp == NULL) {
412 return NULL;
413 }
414 fidp = *fidpp;
415 *fidpp = fidp->next;
416 fidp->clunked = 1;
417 return fidp;
418 }
419
420 void coroutine_fn v9fs_reclaim_fd(V9fsPDU *pdu)
421 {
422 int reclaim_count = 0;
423 V9fsState *s = pdu->s;
424 V9fsFidState *f, *reclaim_list = NULL;
425
426 for (f = s->fid_list; f; f = f->next) {
427 /*
428 * Unlink fids cannot be reclaimed. Check
429 * for them and skip them. Also skip fids
430 * currently being operated on.
431 */
432 if (f->ref || f->flags & FID_NON_RECLAIMABLE) {
433 continue;
434 }
435 /*
436 * if it is a recently referenced fid
437 * we leave the fid untouched and clear the
438 * reference bit. We come back to it later
439 * in the next iteration. (a simple LRU without
440 * moving list elements around)
441 */
442 if (f->flags & FID_REFERENCED) {
443 f->flags &= ~FID_REFERENCED;
444 continue;
445 }
446 /*
447 * Add fids to reclaim list.
448 */
449 if (f->fid_type == P9_FID_FILE) {
450 if (f->fs.fd != -1) {
451 /*
452 * Up the reference count so that
453 * a clunk request won't free this fid
454 */
455 f->ref++;
456 f->rclm_lst = reclaim_list;
457 reclaim_list = f;
458 f->fs_reclaim.fd = f->fs.fd;
459 f->fs.fd = -1;
460 reclaim_count++;
461 }
462 } else if (f->fid_type == P9_FID_DIR) {
463 if (f->fs.dir.stream != NULL) {
464 /*
465 * Up the reference count so that
466 * a clunk request won't free this fid
467 */
468 f->ref++;
469 f->rclm_lst = reclaim_list;
470 reclaim_list = f;
471 f->fs_reclaim.dir.stream = f->fs.dir.stream;
472 f->fs.dir.stream = NULL;
473 reclaim_count++;
474 }
475 }
476 if (reclaim_count >= open_fd_rc) {
477 break;
478 }
479 }
480 /*
481 * Now close the fid in reclaim list. Free them if they
482 * are already clunked.
483 */
484 while (reclaim_list) {
485 f = reclaim_list;
486 reclaim_list = f->rclm_lst;
487 if (f->fid_type == P9_FID_FILE) {
488 v9fs_co_close(pdu, &f->fs_reclaim);
489 } else if (f->fid_type == P9_FID_DIR) {
490 v9fs_co_closedir(pdu, &f->fs_reclaim);
491 }
492 f->rclm_lst = NULL;
493 /*
494 * Now drop the fid reference, free it
495 * if clunked.
496 */
497 put_fid(pdu, f);
498 }
499 }
500
501 static int coroutine_fn v9fs_mark_fids_unreclaim(V9fsPDU *pdu, V9fsPath *path)
502 {
503 int err;
504 V9fsState *s = pdu->s;
505 V9fsFidState *fidp, head_fid;
506
507 head_fid.next = s->fid_list;
508 for (fidp = s->fid_list; fidp; fidp = fidp->next) {
509 if (fidp->path.size != path->size) {
510 continue;
511 }
512 if (!memcmp(fidp->path.data, path->data, path->size)) {
513 /* Mark the fid non reclaimable. */
514 fidp->flags |= FID_NON_RECLAIMABLE;
515
516 /* reopen the file/dir if already closed */
517 err = v9fs_reopen_fid(pdu, fidp);
518 if (err < 0) {
519 return err;
520 }
521 /*
522 * Go back to head of fid list because
523 * the list could have got updated when
524 * switched to the worker thread
525 */
526 if (err == 0) {
527 fidp = &head_fid;
528 }
529 }
530 }
531 return 0;
532 }
533
534 static void coroutine_fn virtfs_reset(V9fsPDU *pdu)
535 {
536 V9fsState *s = pdu->s;
537 V9fsFidState *fidp;
538
539 /* Free all fids */
540 while (s->fid_list) {
541 /* Get fid */
542 fidp = s->fid_list;
543 fidp->ref++;
544
545 /* Clunk fid */
546 s->fid_list = fidp->next;
547 fidp->clunked = 1;
548
549 put_fid(pdu, fidp);
550 }
551 }
552
553 #define P9_QID_TYPE_DIR 0x80
554 #define P9_QID_TYPE_SYMLINK 0x02
555
556 #define P9_STAT_MODE_DIR 0x80000000
557 #define P9_STAT_MODE_APPEND 0x40000000
558 #define P9_STAT_MODE_EXCL 0x20000000
559 #define P9_STAT_MODE_MOUNT 0x10000000
560 #define P9_STAT_MODE_AUTH 0x08000000
561 #define P9_STAT_MODE_TMP 0x04000000
562 #define P9_STAT_MODE_SYMLINK 0x02000000
563 #define P9_STAT_MODE_LINK 0x01000000
564 #define P9_STAT_MODE_DEVICE 0x00800000
565 #define P9_STAT_MODE_NAMED_PIPE 0x00200000
566 #define P9_STAT_MODE_SOCKET 0x00100000
567 #define P9_STAT_MODE_SETUID 0x00080000
568 #define P9_STAT_MODE_SETGID 0x00040000
569 #define P9_STAT_MODE_SETVTX 0x00010000
570
571 #define P9_STAT_MODE_TYPE_BITS (P9_STAT_MODE_DIR | \
572 P9_STAT_MODE_SYMLINK | \
573 P9_STAT_MODE_LINK | \
574 P9_STAT_MODE_DEVICE | \
575 P9_STAT_MODE_NAMED_PIPE | \
576 P9_STAT_MODE_SOCKET)
577
578 /* Mirrors all bits of a byte. So e.g. binary 10100000 would become 00000101. */
579 static inline uint8_t mirror8bit(uint8_t byte)
580 {
581 return (byte * 0x0202020202ULL & 0x010884422010ULL) % 1023;
582 }
583
584 /* Same as mirror8bit() just for a 64 bit data type instead for a byte. */
585 static inline uint64_t mirror64bit(uint64_t value)
586 {
587 return ((uint64_t)mirror8bit(value & 0xff) << 56) |
588 ((uint64_t)mirror8bit((value >> 8) & 0xff) << 48) |
589 ((uint64_t)mirror8bit((value >> 16) & 0xff) << 40) |
590 ((uint64_t)mirror8bit((value >> 24) & 0xff) << 32) |
591 ((uint64_t)mirror8bit((value >> 32) & 0xff) << 24) |
592 ((uint64_t)mirror8bit((value >> 40) & 0xff) << 16) |
593 ((uint64_t)mirror8bit((value >> 48) & 0xff) << 8) |
594 ((uint64_t)mirror8bit((value >> 56) & 0xff));
595 }
596
597 /**
598 * @brief Parameter k for the Exponential Golomb algorihm to be used.
599 *
600 * The smaller this value, the smaller the minimum bit count for the Exp.
601 * Golomb generated affixes will be (at lowest index) however for the
602 * price of having higher maximum bit count of generated affixes (at highest
603 * index). Likewise increasing this parameter yields in smaller maximum bit
604 * count for the price of having higher minimum bit count.
605 *
606 * In practice that means: a good value for k depends on the expected amount
607 * of devices to be exposed by one export. For a small amount of devices k
608 * should be small, for a large amount of devices k might be increased
609 * instead. The default of k=0 should be fine for most users though.
610 *
611 * @b IMPORTANT: In case this ever becomes a runtime parameter; the value of
612 * k should not change as long as guest is still running! Because that would
613 * cause completely different inode numbers to be generated on guest.
614 */
615 #define EXP_GOLOMB_K 0
616
617 /**
618 * @brief Exponential Golomb algorithm for arbitrary k (including k=0).
619 *
620 * The Exponential Golomb algorithm generates @b prefixes (@b not suffixes!)
621 * with growing length and with the mathematical property of being
622 * "prefix-free". The latter means the generated prefixes can be prepended
623 * in front of arbitrary numbers and the resulting concatenated numbers are
624 * guaranteed to be always unique.
625 *
626 * This is a minor adjustment to the original Exp. Golomb algorithm in the
627 * sense that lowest allowed index (@param n) starts with 1, not with zero.
628 *
629 * @param n - natural number (or index) of the prefix to be generated
630 * (1, 2, 3, ...)
631 * @param k - parameter k of Exp. Golomb algorithm to be used
632 * (see comment on EXP_GOLOMB_K macro for details about k)
633 */
634 static VariLenAffix expGolombEncode(uint64_t n, int k)
635 {
636 const uint64_t value = n + (1 << k) - 1;
637 const int bits = (int) log2(value) + 1;
638 return (VariLenAffix) {
639 .type = AffixType_Prefix,
640 .value = value,
641 .bits = bits + MAX((bits - 1 - k), 0)
642 };
643 }
644
645 /**
646 * @brief Converts a suffix into a prefix, or a prefix into a suffix.
647 *
648 * Simply mirror all bits of the affix value, for the purpose to preserve
649 * respectively the mathematical "prefix-free" or "suffix-free" property
650 * after the conversion.
651 *
652 * If a passed prefix is suitable to create unique numbers, then the
653 * returned suffix is suitable to create unique numbers as well (and vice
654 * versa).
655 */
656 static VariLenAffix invertAffix(const VariLenAffix *affix)
657 {
658 return (VariLenAffix) {
659 .type =
660 (affix->type == AffixType_Suffix) ?
661 AffixType_Prefix : AffixType_Suffix,
662 .value =
663 mirror64bit(affix->value) >>
664 ((sizeof(affix->value) * 8) - affix->bits),
665 .bits = affix->bits
666 };
667 }
668
669 /**
670 * @brief Generates suffix numbers with "suffix-free" property.
671 *
672 * This is just a wrapper function on top of the Exp. Golomb algorithm.
673 *
674 * Since the Exp. Golomb algorithm generates prefixes, but we need suffixes,
675 * this function converts the Exp. Golomb prefixes into appropriate suffixes
676 * which are still suitable for generating unique numbers.
677 *
678 * @param n - natural number (or index) of the suffix to be generated
679 * (1, 2, 3, ...)
680 */
681 static VariLenAffix affixForIndex(uint64_t index)
682 {
683 VariLenAffix prefix;
684 prefix = expGolombEncode(index, EXP_GOLOMB_K);
685 return invertAffix(&prefix); /* convert prefix to suffix */
686 }
687
688 /* creative abuse of tb_hash_func7, which is based on xxhash */
689 static uint32_t qpp_hash(QppEntry e)
690 {
691 return qemu_xxhash7(e.ino_prefix, e.dev, 0, 0, 0);
692 }
693
694 static uint32_t qpf_hash(QpfEntry e)
695 {
696 return qemu_xxhash7(e.ino, e.dev, 0, 0, 0);
697 }
698
699 static bool qpd_cmp_func(const void *obj, const void *userp)
700 {
701 const QpdEntry *e1 = obj, *e2 = userp;
702 return e1->dev == e2->dev;
703 }
704
705 static bool qpp_cmp_func(const void *obj, const void *userp)
706 {
707 const QppEntry *e1 = obj, *e2 = userp;
708 return e1->dev == e2->dev && e1->ino_prefix == e2->ino_prefix;
709 }
710
711 static bool qpf_cmp_func(const void *obj, const void *userp)
712 {
713 const QpfEntry *e1 = obj, *e2 = userp;
714 return e1->dev == e2->dev && e1->ino == e2->ino;
715 }
716
717 static void qp_table_remove(void *p, uint32_t h, void *up)
718 {
719 g_free(p);
720 }
721
722 static void qp_table_destroy(struct qht *ht)
723 {
724 if (!ht || !ht->map) {
725 return;
726 }
727 qht_iter(ht, qp_table_remove, NULL);
728 qht_destroy(ht);
729 }
730
731 static void qpd_table_init(struct qht *ht)
732 {
733 qht_init(ht, qpd_cmp_func, 1, QHT_MODE_AUTO_RESIZE);
734 }
735
736 static void qpp_table_init(struct qht *ht)
737 {
738 qht_init(ht, qpp_cmp_func, 1, QHT_MODE_AUTO_RESIZE);
739 }
740
741 static void qpf_table_init(struct qht *ht)
742 {
743 qht_init(ht, qpf_cmp_func, 1 << 16, QHT_MODE_AUTO_RESIZE);
744 }
745
746 /*
747 * Returns how many (high end) bits of inode numbers of the passed fs
748 * device shall be used (in combination with the device number) to
749 * generate hash values for qpp_table entries.
750 *
751 * This function is required if variable length suffixes are used for inode
752 * number mapping on guest level. Since a device may end up having multiple
753 * entries in qpp_table, each entry most probably with a different suffix
754 * length, we thus need this function in conjunction with qpd_table to
755 * "agree" about a fix amount of bits (per device) to be always used for
756 * generating hash values for the purpose of accessing qpp_table in order
757 * get consistent behaviour when accessing qpp_table.
758 */
759 static int qid_inode_prefix_hash_bits(V9fsPDU *pdu, dev_t dev)
760 {
761 QpdEntry lookup = {
762 .dev = dev
763 }, *val;
764 uint32_t hash = dev;
765 VariLenAffix affix;
766
767 val = qht_lookup(&pdu->s->qpd_table, &lookup, hash);
768 if (!val) {
769 val = g_malloc0(sizeof(QpdEntry));
770 *val = lookup;
771 affix = affixForIndex(pdu->s->qp_affix_next);
772 val->prefix_bits = affix.bits;
773 qht_insert(&pdu->s->qpd_table, val, hash, NULL);
774 pdu->s->qp_ndevices++;
775 }
776 return val->prefix_bits;
777 }
778
779 /**
780 * @brief Slow / full mapping host inode nr -> guest inode nr.
781 *
782 * This function performs a slower and much more costly remapping of an
783 * original file inode number on host to an appropriate different inode
784 * number on guest. For every (dev, inode) combination on host a new
785 * sequential number is generated, cached and exposed as inode number on
786 * guest.
787 *
788 * This is just a "last resort" fallback solution if the much faster/cheaper
789 * qid_path_suffixmap() failed. In practice this slow / full mapping is not
790 * expected ever to be used at all though.
791 *
792 * @see qid_path_suffixmap() for details
793 *
794 */
795 static int qid_path_fullmap(V9fsPDU *pdu, const struct stat *stbuf,
796 uint64_t *path)
797 {
798 QpfEntry lookup = {
799 .dev = stbuf->st_dev,
800 .ino = stbuf->st_ino
801 }, *val;
802 uint32_t hash = qpf_hash(lookup);
803 VariLenAffix affix;
804
805 val = qht_lookup(&pdu->s->qpf_table, &lookup, hash);
806
807 if (!val) {
808 if (pdu->s->qp_fullpath_next == 0) {
809 /* no more files can be mapped :'( */
810 error_report_once(
811 "9p: No more prefixes available for remapping inodes from "
812 "host to guest."
813 );
814 return -ENFILE;
815 }
816
817 val = g_malloc0(sizeof(QppEntry));
818 *val = lookup;
819
820 /* new unique inode and device combo */
821 affix = affixForIndex(
822 1ULL << (sizeof(pdu->s->qp_affix_next) * 8)
823 );
824 val->path = (pdu->s->qp_fullpath_next++ << affix.bits) | affix.value;
825 pdu->s->qp_fullpath_next &= ((1ULL << (64 - affix.bits)) - 1);
826 qht_insert(&pdu->s->qpf_table, val, hash, NULL);
827 }
828
829 *path = val->path;
830 return 0;
831 }
832
833 /**
834 * @brief Quick mapping host inode nr -> guest inode nr.
835 *
836 * This function performs quick remapping of an original file inode number
837 * on host to an appropriate different inode number on guest. This remapping
838 * of inodes is required to avoid inode nr collisions on guest which would
839 * happen if the 9p export contains more than 1 exported file system (or
840 * more than 1 file system data set), because unlike on host level where the
841 * files would have different device nrs, all files exported by 9p would
842 * share the same device nr on guest (the device nr of the virtual 9p device
843 * that is).
844 *
845 * Inode remapping is performed by chopping off high end bits of the original
846 * inode number from host, shifting the result upwards and then assigning a
847 * generated suffix number for the low end bits, where the same suffix number
848 * will be shared by all inodes with the same device id AND the same high end
849 * bits that have been chopped off. That approach utilizes the fact that inode
850 * numbers very likely share the same high end bits (i.e. due to their common
851 * sequential generation by file systems) and hence we only have to generate
852 * and track a very limited amount of suffixes in practice due to that.
853 *
854 * We generate variable size suffixes for that purpose. The 1st generated
855 * suffix will only have 1 bit and hence we only need to chop off 1 bit from
856 * the original inode number. The subsequent suffixes being generated will
857 * grow in (bit) size subsequently, i.e. the 2nd and 3rd suffix being
858 * generated will have 3 bits and hence we have to chop off 3 bits from their
859 * original inodes, and so on. That approach of using variable length suffixes
860 * (i.e. over fixed size ones) utilizes the fact that in practice only a very
861 * limited amount of devices are shared by the same export (e.g. typically
862 * less than 2 dozen devices per 9p export), so in practice we need to chop
863 * off less bits than with fixed size prefixes and yet are flexible to add
864 * new devices at runtime below host's export directory at any time without
865 * having to reboot guest nor requiring to reconfigure guest for that. And due
866 * to the very limited amount of original high end bits that we chop off that
867 * way, the total amount of suffixes we need to generate is less than by using
868 * fixed size prefixes and hence it also improves performance of the inode
869 * remapping algorithm, and finally has the nice side effect that the inode
870 * numbers on guest will be much smaller & human friendly. ;-)
871 */
872 static int qid_path_suffixmap(V9fsPDU *pdu, const struct stat *stbuf,
873 uint64_t *path)
874 {
875 const int ino_hash_bits = qid_inode_prefix_hash_bits(pdu, stbuf->st_dev);
876 QppEntry lookup = {
877 .dev = stbuf->st_dev,
878 .ino_prefix = (uint16_t) (stbuf->st_ino >> (64 - ino_hash_bits))
879 }, *val;
880 uint32_t hash = qpp_hash(lookup);
881
882 val = qht_lookup(&pdu->s->qpp_table, &lookup, hash);
883
884 if (!val) {
885 if (pdu->s->qp_affix_next == 0) {
886 /* we ran out of affixes */
887 warn_report_once(
888 "9p: Potential degraded performance of inode remapping"
889 );
890 return -ENFILE;
891 }
892
893 val = g_malloc0(sizeof(QppEntry));
894 *val = lookup;
895
896 /* new unique inode affix and device combo */
897 val->qp_affix_index = pdu->s->qp_affix_next++;
898 val->qp_affix = affixForIndex(val->qp_affix_index);
899 qht_insert(&pdu->s->qpp_table, val, hash, NULL);
900 }
901 /* assuming generated affix to be suffix type, not prefix */
902 *path = (stbuf->st_ino << val->qp_affix.bits) | val->qp_affix.value;
903 return 0;
904 }
905
906 static int stat_to_qid(V9fsPDU *pdu, const struct stat *stbuf, V9fsQID *qidp)
907 {
908 int err;
909 size_t size;
910
911 if (pdu->s->ctx.export_flags & V9FS_REMAP_INODES) {
912 /* map inode+device to qid path (fast path) */
913 err = qid_path_suffixmap(pdu, stbuf, &qidp->path);
914 if (err == -ENFILE) {
915 /* fast path didn't work, fall back to full map */
916 err = qid_path_fullmap(pdu, stbuf, &qidp->path);
917 }
918 if (err) {
919 return err;
920 }
921 } else {
922 if (pdu->s->dev_id != stbuf->st_dev) {
923 if (pdu->s->ctx.export_flags & V9FS_FORBID_MULTIDEVS) {
924 error_report_once(
925 "9p: Multiple devices detected in same VirtFS export. "
926 "Access of guest to additional devices is (partly) "
927 "denied due to virtfs option 'multidevs=forbid' being "
928 "effective."
929 );
930 return -ENODEV;
931 } else {
932 warn_report_once(
933 "9p: Multiple devices detected in same VirtFS export, "
934 "which might lead to file ID collisions and severe "
935 "misbehaviours on guest! You should either use a "
936 "separate export for each device shared from host or "
937 "use virtfs option 'multidevs=remap'!"
938 );
939 }
940 }
941 memset(&qidp->path, 0, sizeof(qidp->path));
942 size = MIN(sizeof(stbuf->st_ino), sizeof(qidp->path));
943 memcpy(&qidp->path, &stbuf->st_ino, size);
944 }
945
946 qidp->version = stbuf->st_mtime ^ (stbuf->st_size << 8);
947 qidp->type = 0;
948 if (S_ISDIR(stbuf->st_mode)) {
949 qidp->type |= P9_QID_TYPE_DIR;
950 }
951 if (S_ISLNK(stbuf->st_mode)) {
952 qidp->type |= P9_QID_TYPE_SYMLINK;
953 }
954
955 return 0;
956 }
957
958 static int coroutine_fn fid_to_qid(V9fsPDU *pdu, V9fsFidState *fidp,
959 V9fsQID *qidp)
960 {
961 struct stat stbuf;
962 int err;
963
964 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
965 if (err < 0) {
966 return err;
967 }
968 err = stat_to_qid(pdu, &stbuf, qidp);
969 if (err < 0) {
970 return err;
971 }
972 return 0;
973 }
974
975 static int coroutine_fn dirent_to_qid(V9fsPDU *pdu, V9fsFidState *fidp,
976 struct dirent *dent, V9fsQID *qidp)
977 {
978 struct stat stbuf;
979 V9fsPath path;
980 int err;
981
982 v9fs_path_init(&path);
983
984 err = v9fs_co_name_to_path(pdu, &fidp->path, dent->d_name, &path);
985 if (err < 0) {
986 goto out;
987 }
988 err = v9fs_co_lstat(pdu, &path, &stbuf);
989 if (err < 0) {
990 goto out;
991 }
992 err = stat_to_qid(pdu, &stbuf, qidp);
993
994 out:
995 v9fs_path_free(&path);
996 return err;
997 }
998
999 V9fsPDU *pdu_alloc(V9fsState *s)
1000 {
1001 V9fsPDU *pdu = NULL;
1002
1003 if (!QLIST_EMPTY(&s->free_list)) {
1004 pdu = QLIST_FIRST(&s->free_list);
1005 QLIST_REMOVE(pdu, next);
1006 QLIST_INSERT_HEAD(&s->active_list, pdu, next);
1007 }
1008 return pdu;
1009 }
1010
1011 void pdu_free(V9fsPDU *pdu)
1012 {
1013 V9fsState *s = pdu->s;
1014
1015 g_assert(!pdu->cancelled);
1016 QLIST_REMOVE(pdu, next);
1017 QLIST_INSERT_HEAD(&s->free_list, pdu, next);
1018 }
1019
1020 static void coroutine_fn pdu_complete(V9fsPDU *pdu, ssize_t len)
1021 {
1022 int8_t id = pdu->id + 1; /* Response */
1023 V9fsState *s = pdu->s;
1024 int ret;
1025
1026 /*
1027 * The 9p spec requires that successfully cancelled pdus receive no reply.
1028 * Sending a reply would confuse clients because they would
1029 * assume that any EINTR is the actual result of the operation,
1030 * rather than a consequence of the cancellation. However, if
1031 * the operation completed (succesfully or with an error other
1032 * than caused be cancellation), we do send out that reply, both
1033 * for efficiency and to avoid confusing the rest of the state machine
1034 * that assumes passing a non-error here will mean a successful
1035 * transmission of the reply.
1036 */
1037 bool discard = pdu->cancelled && len == -EINTR;
1038 if (discard) {
1039 trace_v9fs_rcancel(pdu->tag, pdu->id);
1040 pdu->size = 0;
1041 goto out_notify;
1042 }
1043
1044 if (len < 0) {
1045 int err = -len;
1046 len = 7;
1047
1048 if (s->proto_version != V9FS_PROTO_2000L) {
1049 V9fsString str;
1050
1051 str.data = strerror(err);
1052 str.size = strlen(str.data);
1053
1054 ret = pdu_marshal(pdu, len, "s", &str);
1055 if (ret < 0) {
1056 goto out_notify;
1057 }
1058 len += ret;
1059 id = P9_RERROR;
1060 }
1061
1062 ret = pdu_marshal(pdu, len, "d", err);
1063 if (ret < 0) {
1064 goto out_notify;
1065 }
1066 len += ret;
1067
1068 if (s->proto_version == V9FS_PROTO_2000L) {
1069 id = P9_RLERROR;
1070 }
1071 trace_v9fs_rerror(pdu->tag, pdu->id, err); /* Trace ERROR */
1072 }
1073
1074 /* fill out the header */
1075 if (pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag) < 0) {
1076 goto out_notify;
1077 }
1078
1079 /* keep these in sync */
1080 pdu->size = len;
1081 pdu->id = id;
1082
1083 out_notify:
1084 pdu->s->transport->push_and_notify(pdu);
1085
1086 /* Now wakeup anybody waiting in flush for this request */
1087 if (!qemu_co_queue_next(&pdu->complete)) {
1088 pdu_free(pdu);
1089 }
1090 }
1091
1092 static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
1093 {
1094 mode_t ret;
1095
1096 ret = mode & 0777;
1097 if (mode & P9_STAT_MODE_DIR) {
1098 ret |= S_IFDIR;
1099 }
1100
1101 if (mode & P9_STAT_MODE_SYMLINK) {
1102 ret |= S_IFLNK;
1103 }
1104 if (mode & P9_STAT_MODE_SOCKET) {
1105 ret |= S_IFSOCK;
1106 }
1107 if (mode & P9_STAT_MODE_NAMED_PIPE) {
1108 ret |= S_IFIFO;
1109 }
1110 if (mode & P9_STAT_MODE_DEVICE) {
1111 if (extension->size && extension->data[0] == 'c') {
1112 ret |= S_IFCHR;
1113 } else {
1114 ret |= S_IFBLK;
1115 }
1116 }
1117
1118 if (!(ret&~0777)) {
1119 ret |= S_IFREG;
1120 }
1121
1122 if (mode & P9_STAT_MODE_SETUID) {
1123 ret |= S_ISUID;
1124 }
1125 if (mode & P9_STAT_MODE_SETGID) {
1126 ret |= S_ISGID;
1127 }
1128 if (mode & P9_STAT_MODE_SETVTX) {
1129 ret |= S_ISVTX;
1130 }
1131
1132 return ret;
1133 }
1134
1135 static int donttouch_stat(V9fsStat *stat)
1136 {
1137 if (stat->type == -1 &&
1138 stat->dev == -1 &&
1139 stat->qid.type == 0xff &&
1140 stat->qid.version == (uint32_t) -1 &&
1141 stat->qid.path == (uint64_t) -1 &&
1142 stat->mode == -1 &&
1143 stat->atime == -1 &&
1144 stat->mtime == -1 &&
1145 stat->length == -1 &&
1146 !stat->name.size &&
1147 !stat->uid.size &&
1148 !stat->gid.size &&
1149 !stat->muid.size &&
1150 stat->n_uid == -1 &&
1151 stat->n_gid == -1 &&
1152 stat->n_muid == -1) {
1153 return 1;
1154 }
1155
1156 return 0;
1157 }
1158
1159 static void v9fs_stat_init(V9fsStat *stat)
1160 {
1161 v9fs_string_init(&stat->name);
1162 v9fs_string_init(&stat->uid);
1163 v9fs_string_init(&stat->gid);
1164 v9fs_string_init(&stat->muid);
1165 v9fs_string_init(&stat->extension);
1166 }
1167
1168 static void v9fs_stat_free(V9fsStat *stat)
1169 {
1170 v9fs_string_free(&stat->name);
1171 v9fs_string_free(&stat->uid);
1172 v9fs_string_free(&stat->gid);
1173 v9fs_string_free(&stat->muid);
1174 v9fs_string_free(&stat->extension);
1175 }
1176
1177 static uint32_t stat_to_v9mode(const struct stat *stbuf)
1178 {
1179 uint32_t mode;
1180
1181 mode = stbuf->st_mode & 0777;
1182 if (S_ISDIR(stbuf->st_mode)) {
1183 mode |= P9_STAT_MODE_DIR;
1184 }
1185
1186 if (S_ISLNK(stbuf->st_mode)) {
1187 mode |= P9_STAT_MODE_SYMLINK;
1188 }
1189
1190 if (S_ISSOCK(stbuf->st_mode)) {
1191 mode |= P9_STAT_MODE_SOCKET;
1192 }
1193
1194 if (S_ISFIFO(stbuf->st_mode)) {
1195 mode |= P9_STAT_MODE_NAMED_PIPE;
1196 }
1197
1198 if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
1199 mode |= P9_STAT_MODE_DEVICE;
1200 }
1201
1202 if (stbuf->st_mode & S_ISUID) {
1203 mode |= P9_STAT_MODE_SETUID;
1204 }
1205
1206 if (stbuf->st_mode & S_ISGID) {
1207 mode |= P9_STAT_MODE_SETGID;
1208 }
1209
1210 if (stbuf->st_mode & S_ISVTX) {
1211 mode |= P9_STAT_MODE_SETVTX;
1212 }
1213
1214 return mode;
1215 }
1216
1217 static int coroutine_fn stat_to_v9stat(V9fsPDU *pdu, V9fsPath *path,
1218 const char *basename,
1219 const struct stat *stbuf,
1220 V9fsStat *v9stat)
1221 {
1222 int err;
1223
1224 memset(v9stat, 0, sizeof(*v9stat));
1225
1226 err = stat_to_qid(pdu, stbuf, &v9stat->qid);
1227 if (err < 0) {
1228 return err;
1229 }
1230 v9stat->mode = stat_to_v9mode(stbuf);
1231 v9stat->atime = stbuf->st_atime;
1232 v9stat->mtime = stbuf->st_mtime;
1233 v9stat->length = stbuf->st_size;
1234
1235 v9fs_string_free(&v9stat->uid);
1236 v9fs_string_free(&v9stat->gid);
1237 v9fs_string_free(&v9stat->muid);
1238
1239 v9stat->n_uid = stbuf->st_uid;
1240 v9stat->n_gid = stbuf->st_gid;
1241 v9stat->n_muid = 0;
1242
1243 v9fs_string_free(&v9stat->extension);
1244
1245 if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
1246 err = v9fs_co_readlink(pdu, path, &v9stat->extension);
1247 if (err < 0) {
1248 return err;
1249 }
1250 } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
1251 v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
1252 S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
1253 major(stbuf->st_rdev), minor(stbuf->st_rdev));
1254 } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
1255 v9fs_string_sprintf(&v9stat->extension, "%s %lu",
1256 "HARDLINKCOUNT", (unsigned long)stbuf->st_nlink);
1257 }
1258
1259 v9fs_string_sprintf(&v9stat->name, "%s", basename);
1260
1261 v9stat->size = 61 +
1262 v9fs_string_size(&v9stat->name) +
1263 v9fs_string_size(&v9stat->uid) +
1264 v9fs_string_size(&v9stat->gid) +
1265 v9fs_string_size(&v9stat->muid) +
1266 v9fs_string_size(&v9stat->extension);
1267 return 0;
1268 }
1269
1270 #define P9_STATS_MODE 0x00000001ULL
1271 #define P9_STATS_NLINK 0x00000002ULL
1272 #define P9_STATS_UID 0x00000004ULL
1273 #define P9_STATS_GID 0x00000008ULL
1274 #define P9_STATS_RDEV 0x00000010ULL
1275 #define P9_STATS_ATIME 0x00000020ULL
1276 #define P9_STATS_MTIME 0x00000040ULL
1277 #define P9_STATS_CTIME 0x00000080ULL
1278 #define P9_STATS_INO 0x00000100ULL
1279 #define P9_STATS_SIZE 0x00000200ULL
1280 #define P9_STATS_BLOCKS 0x00000400ULL
1281
1282 #define P9_STATS_BTIME 0x00000800ULL
1283 #define P9_STATS_GEN 0x00001000ULL
1284 #define P9_STATS_DATA_VERSION 0x00002000ULL
1285
1286 #define P9_STATS_BASIC 0x000007ffULL /* Mask for fields up to BLOCKS */
1287 #define P9_STATS_ALL 0x00003fffULL /* Mask for All fields above */
1288
1289
1290 static int stat_to_v9stat_dotl(V9fsPDU *pdu, const struct stat *stbuf,
1291 V9fsStatDotl *v9lstat)
1292 {
1293 memset(v9lstat, 0, sizeof(*v9lstat));
1294
1295 v9lstat->st_mode = stbuf->st_mode;
1296 v9lstat->st_nlink = stbuf->st_nlink;
1297 v9lstat->st_uid = stbuf->st_uid;
1298 v9lstat->st_gid = stbuf->st_gid;
1299 v9lstat->st_rdev = stbuf->st_rdev;
1300 v9lstat->st_size = stbuf->st_size;
1301 v9lstat->st_blksize = stbuf->st_blksize;
1302 v9lstat->st_blocks = stbuf->st_blocks;
1303 v9lstat->st_atime_sec = stbuf->st_atime;
1304 v9lstat->st_atime_nsec = stbuf->st_atim.tv_nsec;
1305 v9lstat->st_mtime_sec = stbuf->st_mtime;
1306 v9lstat->st_mtime_nsec = stbuf->st_mtim.tv_nsec;
1307 v9lstat->st_ctime_sec = stbuf->st_ctime;
1308 v9lstat->st_ctime_nsec = stbuf->st_ctim.tv_nsec;
1309 /* Currently we only support BASIC fields in stat */
1310 v9lstat->st_result_mask = P9_STATS_BASIC;
1311
1312 return stat_to_qid(pdu, stbuf, &v9lstat->qid);
1313 }
1314
1315 static void print_sg(struct iovec *sg, int cnt)
1316 {
1317 int i;
1318
1319 printf("sg[%d]: {", cnt);
1320 for (i = 0; i < cnt; i++) {
1321 if (i) {
1322 printf(", ");
1323 }
1324 printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
1325 }
1326 printf("}\n");
1327 }
1328
1329 /* Will call this only for path name based fid */
1330 static void v9fs_fix_path(V9fsPath *dst, V9fsPath *src, int len)
1331 {
1332 V9fsPath str;
1333 v9fs_path_init(&str);
1334 v9fs_path_copy(&str, dst);
1335 v9fs_path_sprintf(dst, "%s%s", src->data, str.data + len);
1336 v9fs_path_free(&str);
1337 }
1338
1339 static inline bool is_ro_export(FsContext *ctx)
1340 {
1341 return ctx->export_flags & V9FS_RDONLY;
1342 }
1343
1344 static void coroutine_fn v9fs_version(void *opaque)
1345 {
1346 ssize_t err;
1347 V9fsPDU *pdu = opaque;
1348 V9fsState *s = pdu->s;
1349 V9fsString version;
1350 size_t offset = 7;
1351
1352 v9fs_string_init(&version);
1353 err = pdu_unmarshal(pdu, offset, "ds", &s->msize, &version);
1354 if (err < 0) {
1355 goto out;
1356 }
1357 trace_v9fs_version(pdu->tag, pdu->id, s->msize, version.data);
1358
1359 virtfs_reset(pdu);
1360
1361 if (!strcmp(version.data, "9P2000.u")) {
1362 s->proto_version = V9FS_PROTO_2000U;
1363 } else if (!strcmp(version.data, "9P2000.L")) {
1364 s->proto_version = V9FS_PROTO_2000L;
1365 } else {
1366 v9fs_string_sprintf(&version, "unknown");
1367 /* skip min. msize check, reporting invalid version has priority */
1368 goto marshal;
1369 }
1370
1371 if (s->msize < P9_MIN_MSIZE) {
1372 err = -EMSGSIZE;
1373 error_report(
1374 "9pfs: Client requested msize < minimum msize ("
1375 stringify(P9_MIN_MSIZE) ") supported by this server."
1376 );
1377 goto out;
1378 }
1379
1380 marshal:
1381 err = pdu_marshal(pdu, offset, "ds", s->msize, &version);
1382 if (err < 0) {
1383 goto out;
1384 }
1385 err += offset;
1386 trace_v9fs_version_return(pdu->tag, pdu->id, s->msize, version.data);
1387 out:
1388 pdu_complete(pdu, err);
1389 v9fs_string_free(&version);
1390 }
1391
1392 static void coroutine_fn v9fs_attach(void *opaque)
1393 {
1394 V9fsPDU *pdu = opaque;
1395 V9fsState *s = pdu->s;
1396 int32_t fid, afid, n_uname;
1397 V9fsString uname, aname;
1398 V9fsFidState *fidp;
1399 size_t offset = 7;
1400 V9fsQID qid;
1401 ssize_t err;
1402
1403 v9fs_string_init(&uname);
1404 v9fs_string_init(&aname);
1405 err = pdu_unmarshal(pdu, offset, "ddssd", &fid,
1406 &afid, &uname, &aname, &n_uname);
1407 if (err < 0) {
1408 goto out_nofid;
1409 }
1410 trace_v9fs_attach(pdu->tag, pdu->id, fid, afid, uname.data, aname.data);
1411
1412 fidp = alloc_fid(s, fid);
1413 if (fidp == NULL) {
1414 err = -EINVAL;
1415 goto out_nofid;
1416 }
1417 fidp->uid = n_uname;
1418 err = v9fs_co_name_to_path(pdu, NULL, "/", &fidp->path);
1419 if (err < 0) {
1420 err = -EINVAL;
1421 clunk_fid(s, fid);
1422 goto out;
1423 }
1424 err = fid_to_qid(pdu, fidp, &qid);
1425 if (err < 0) {
1426 err = -EINVAL;
1427 clunk_fid(s, fid);
1428 goto out;
1429 }
1430
1431 /*
1432 * disable migration if we haven't done already.
1433 * attach could get called multiple times for the same export.
1434 */
1435 if (!s->migration_blocker) {
1436 error_setg(&s->migration_blocker,
1437 "Migration is disabled when VirtFS export path '%s' is mounted in the guest using mount_tag '%s'",
1438 s->ctx.fs_root ? s->ctx.fs_root : "NULL", s->tag);
1439 err = migrate_add_blocker(s->migration_blocker, NULL);
1440 if (err < 0) {
1441 error_free(s->migration_blocker);
1442 s->migration_blocker = NULL;
1443 clunk_fid(s, fid);
1444 goto out;
1445 }
1446 s->root_fid = fid;
1447 }
1448
1449 err = pdu_marshal(pdu, offset, "Q", &qid);
1450 if (err < 0) {
1451 clunk_fid(s, fid);
1452 goto out;
1453 }
1454 err += offset;
1455
1456 memcpy(&s->root_qid, &qid, sizeof(qid));
1457 trace_v9fs_attach_return(pdu->tag, pdu->id,
1458 qid.type, qid.version, qid.path);
1459 out:
1460 put_fid(pdu, fidp);
1461 out_nofid:
1462 pdu_complete(pdu, err);
1463 v9fs_string_free(&uname);
1464 v9fs_string_free(&aname);
1465 }
1466
1467 static void coroutine_fn v9fs_stat(void *opaque)
1468 {
1469 int32_t fid;
1470 V9fsStat v9stat;
1471 ssize_t err = 0;
1472 size_t offset = 7;
1473 struct stat stbuf;
1474 V9fsFidState *fidp;
1475 V9fsPDU *pdu = opaque;
1476 char *basename;
1477
1478 err = pdu_unmarshal(pdu, offset, "d", &fid);
1479 if (err < 0) {
1480 goto out_nofid;
1481 }
1482 trace_v9fs_stat(pdu->tag, pdu->id, fid);
1483
1484 fidp = get_fid(pdu, fid);
1485 if (fidp == NULL) {
1486 err = -ENOENT;
1487 goto out_nofid;
1488 }
1489 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1490 if (err < 0) {
1491 goto out;
1492 }
1493 basename = g_path_get_basename(fidp->path.data);
1494 err = stat_to_v9stat(pdu, &fidp->path, basename, &stbuf, &v9stat);
1495 g_free(basename);
1496 if (err < 0) {
1497 goto out;
1498 }
1499 err = pdu_marshal(pdu, offset, "wS", 0, &v9stat);
1500 if (err < 0) {
1501 v9fs_stat_free(&v9stat);
1502 goto out;
1503 }
1504 trace_v9fs_stat_return(pdu->tag, pdu->id, v9stat.mode,
1505 v9stat.atime, v9stat.mtime, v9stat.length);
1506 err += offset;
1507 v9fs_stat_free(&v9stat);
1508 out:
1509 put_fid(pdu, fidp);
1510 out_nofid:
1511 pdu_complete(pdu, err);
1512 }
1513
1514 static void coroutine_fn v9fs_getattr(void *opaque)
1515 {
1516 int32_t fid;
1517 size_t offset = 7;
1518 ssize_t retval = 0;
1519 struct stat stbuf;
1520 V9fsFidState *fidp;
1521 uint64_t request_mask;
1522 V9fsStatDotl v9stat_dotl;
1523 V9fsPDU *pdu = opaque;
1524
1525 retval = pdu_unmarshal(pdu, offset, "dq", &fid, &request_mask);
1526 if (retval < 0) {
1527 goto out_nofid;
1528 }
1529 trace_v9fs_getattr(pdu->tag, pdu->id, fid, request_mask);
1530
1531 fidp = get_fid(pdu, fid);
1532 if (fidp == NULL) {
1533 retval = -ENOENT;
1534 goto out_nofid;
1535 }
1536 /*
1537 * Currently we only support BASIC fields in stat, so there is no
1538 * need to look at request_mask.
1539 */
1540 retval = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1541 if (retval < 0) {
1542 goto out;
1543 }
1544 retval = stat_to_v9stat_dotl(pdu, &stbuf, &v9stat_dotl);
1545 if (retval < 0) {
1546 goto out;
1547 }
1548
1549 /* fill st_gen if requested and supported by underlying fs */
1550 if (request_mask & P9_STATS_GEN) {
1551 retval = v9fs_co_st_gen(pdu, &fidp->path, stbuf.st_mode, &v9stat_dotl);
1552 switch (retval) {
1553 case 0:
1554 /* we have valid st_gen: update result mask */
1555 v9stat_dotl.st_result_mask |= P9_STATS_GEN;
1556 break;
1557 case -EINTR:
1558 /* request cancelled, e.g. by Tflush */
1559 goto out;
1560 default:
1561 /* failed to get st_gen: not fatal, ignore */
1562 break;
1563 }
1564 }
1565 retval = pdu_marshal(pdu, offset, "A", &v9stat_dotl);
1566 if (retval < 0) {
1567 goto out;
1568 }
1569 retval += offset;
1570 trace_v9fs_getattr_return(pdu->tag, pdu->id, v9stat_dotl.st_result_mask,
1571 v9stat_dotl.st_mode, v9stat_dotl.st_uid,
1572 v9stat_dotl.st_gid);
1573 out:
1574 put_fid(pdu, fidp);
1575 out_nofid:
1576 pdu_complete(pdu, retval);
1577 }
1578
1579 /* Attribute flags */
1580 #define P9_ATTR_MODE (1 << 0)
1581 #define P9_ATTR_UID (1 << 1)
1582 #define P9_ATTR_GID (1 << 2)
1583 #define P9_ATTR_SIZE (1 << 3)
1584 #define P9_ATTR_ATIME (1 << 4)
1585 #define P9_ATTR_MTIME (1 << 5)
1586 #define P9_ATTR_CTIME (1 << 6)
1587 #define P9_ATTR_ATIME_SET (1 << 7)
1588 #define P9_ATTR_MTIME_SET (1 << 8)
1589
1590 #define P9_ATTR_MASK 127
1591
1592 static void coroutine_fn v9fs_setattr(void *opaque)
1593 {
1594 int err = 0;
1595 int32_t fid;
1596 V9fsFidState *fidp;
1597 size_t offset = 7;
1598 V9fsIattr v9iattr;
1599 V9fsPDU *pdu = opaque;
1600
1601 err = pdu_unmarshal(pdu, offset, "dI", &fid, &v9iattr);
1602 if (err < 0) {
1603 goto out_nofid;
1604 }
1605
1606 trace_v9fs_setattr(pdu->tag, pdu->id, fid,
1607 v9iattr.valid, v9iattr.mode, v9iattr.uid, v9iattr.gid,
1608 v9iattr.size, v9iattr.atime_sec, v9iattr.mtime_sec);
1609
1610 fidp = get_fid(pdu, fid);
1611 if (fidp == NULL) {
1612 err = -EINVAL;
1613 goto out_nofid;
1614 }
1615 if (v9iattr.valid & P9_ATTR_MODE) {
1616 err = v9fs_co_chmod(pdu, &fidp->path, v9iattr.mode);
1617 if (err < 0) {
1618 goto out;
1619 }
1620 }
1621 if (v9iattr.valid & (P9_ATTR_ATIME | P9_ATTR_MTIME)) {
1622 struct timespec times[2];
1623 if (v9iattr.valid & P9_ATTR_ATIME) {
1624 if (v9iattr.valid & P9_ATTR_ATIME_SET) {
1625 times[0].tv_sec = v9iattr.atime_sec;
1626 times[0].tv_nsec = v9iattr.atime_nsec;
1627 } else {
1628 times[0].tv_nsec = UTIME_NOW;
1629 }
1630 } else {
1631 times[0].tv_nsec = UTIME_OMIT;
1632 }
1633 if (v9iattr.valid & P9_ATTR_MTIME) {
1634 if (v9iattr.valid & P9_ATTR_MTIME_SET) {
1635 times[1].tv_sec = v9iattr.mtime_sec;
1636 times[1].tv_nsec = v9iattr.mtime_nsec;
1637 } else {
1638 times[1].tv_nsec = UTIME_NOW;
1639 }
1640 } else {
1641 times[1].tv_nsec = UTIME_OMIT;
1642 }
1643 err = v9fs_co_utimensat(pdu, &fidp->path, times);
1644 if (err < 0) {
1645 goto out;
1646 }
1647 }
1648 /*
1649 * If the only valid entry in iattr is ctime we can call
1650 * chown(-1,-1) to update the ctime of the file
1651 */
1652 if ((v9iattr.valid & (P9_ATTR_UID | P9_ATTR_GID)) ||
1653 ((v9iattr.valid & P9_ATTR_CTIME)
1654 && !((v9iattr.valid & P9_ATTR_MASK) & ~P9_ATTR_CTIME))) {
1655 if (!(v9iattr.valid & P9_ATTR_UID)) {
1656 v9iattr.uid = -1;
1657 }
1658 if (!(v9iattr.valid & P9_ATTR_GID)) {
1659 v9iattr.gid = -1;
1660 }
1661 err = v9fs_co_chown(pdu, &fidp->path, v9iattr.uid,
1662 v9iattr.gid);
1663 if (err < 0) {
1664 goto out;
1665 }
1666 }
1667 if (v9iattr.valid & (P9_ATTR_SIZE)) {
1668 err = v9fs_co_truncate(pdu, &fidp->path, v9iattr.size);
1669 if (err < 0) {
1670 goto out;
1671 }
1672 }
1673 err = offset;
1674 trace_v9fs_setattr_return(pdu->tag, pdu->id);
1675 out:
1676 put_fid(pdu, fidp);
1677 out_nofid:
1678 pdu_complete(pdu, err);
1679 }
1680
1681 static int v9fs_walk_marshal(V9fsPDU *pdu, uint16_t nwnames, V9fsQID *qids)
1682 {
1683 int i;
1684 ssize_t err;
1685 size_t offset = 7;
1686
1687 err = pdu_marshal(pdu, offset, "w", nwnames);
1688 if (err < 0) {
1689 return err;
1690 }
1691 offset += err;
1692 for (i = 0; i < nwnames; i++) {
1693 err = pdu_marshal(pdu, offset, "Q", &qids[i]);
1694 if (err < 0) {
1695 return err;
1696 }
1697 offset += err;
1698 }
1699 return offset;
1700 }
1701
1702 static bool name_is_illegal(const char *name)
1703 {
1704 return !*name || strchr(name, '/') != NULL;
1705 }
1706
1707 static bool not_same_qid(const V9fsQID *qid1, const V9fsQID *qid2)
1708 {
1709 return
1710 qid1->type != qid2->type ||
1711 qid1->version != qid2->version ||
1712 qid1->path != qid2->path;
1713 }
1714
1715 static void coroutine_fn v9fs_walk(void *opaque)
1716 {
1717 int name_idx;
1718 V9fsQID *qids = NULL;
1719 int i, err = 0;
1720 V9fsPath dpath, path;
1721 uint16_t nwnames;
1722 struct stat stbuf;
1723 size_t offset = 7;
1724 int32_t fid, newfid;
1725 V9fsString *wnames = NULL;
1726 V9fsFidState *fidp;
1727 V9fsFidState *newfidp = NULL;
1728 V9fsPDU *pdu = opaque;
1729 V9fsState *s = pdu->s;
1730 V9fsQID qid;
1731
1732 err = pdu_unmarshal(pdu, offset, "ddw", &fid, &newfid, &nwnames);
1733 if (err < 0) {
1734 pdu_complete(pdu, err);
1735 return ;
1736 }
1737 offset += err;
1738
1739 trace_v9fs_walk(pdu->tag, pdu->id, fid, newfid, nwnames);
1740
1741 if (nwnames && nwnames <= P9_MAXWELEM) {
1742 wnames = g_new0(V9fsString, nwnames);
1743 qids = g_new0(V9fsQID, nwnames);
1744 for (i = 0; i < nwnames; i++) {
1745 err = pdu_unmarshal(pdu, offset, "s", &wnames[i]);
1746 if (err < 0) {
1747 goto out_nofid;
1748 }
1749 if (name_is_illegal(wnames[i].data)) {
1750 err = -ENOENT;
1751 goto out_nofid;
1752 }
1753 offset += err;
1754 }
1755 } else if (nwnames > P9_MAXWELEM) {
1756 err = -EINVAL;
1757 goto out_nofid;
1758 }
1759 fidp = get_fid(pdu, fid);
1760 if (fidp == NULL) {
1761 err = -ENOENT;
1762 goto out_nofid;
1763 }
1764
1765 v9fs_path_init(&dpath);
1766 v9fs_path_init(&path);
1767
1768 err = fid_to_qid(pdu, fidp, &qid);
1769 if (err < 0) {
1770 goto out;
1771 }
1772
1773 /*
1774 * Both dpath and path initially poin to fidp.
1775 * Needed to handle request with nwnames == 0
1776 */
1777 v9fs_path_copy(&dpath, &fidp->path);
1778 v9fs_path_copy(&path, &fidp->path);
1779 for (name_idx = 0; name_idx < nwnames; name_idx++) {
1780 if (not_same_qid(&pdu->s->root_qid, &qid) ||
1781 strcmp("..", wnames[name_idx].data)) {
1782 err = v9fs_co_name_to_path(pdu, &dpath, wnames[name_idx].data,
1783 &path);
1784 if (err < 0) {
1785 goto out;
1786 }
1787
1788 err = v9fs_co_lstat(pdu, &path, &stbuf);
1789 if (err < 0) {
1790 goto out;
1791 }
1792 err = stat_to_qid(pdu, &stbuf, &qid);
1793 if (err < 0) {
1794 goto out;
1795 }
1796 v9fs_path_copy(&dpath, &path);
1797 }
1798 memcpy(&qids[name_idx], &qid, sizeof(qid));
1799 }
1800 if (fid == newfid) {
1801 if (fidp->fid_type != P9_FID_NONE) {
1802 err = -EINVAL;
1803 goto out;
1804 }
1805 v9fs_path_write_lock(s);
1806 v9fs_path_copy(&fidp->path, &path);
1807 v9fs_path_unlock(s);
1808 } else {
1809 newfidp = alloc_fid(s, newfid);
1810 if (newfidp == NULL) {
1811 err = -EINVAL;
1812 goto out;
1813 }
1814 newfidp->uid = fidp->uid;
1815 v9fs_path_copy(&newfidp->path, &path);
1816 }
1817 err = v9fs_walk_marshal(pdu, nwnames, qids);
1818 trace_v9fs_walk_return(pdu->tag, pdu->id, nwnames, qids);
1819 out:
1820 put_fid(pdu, fidp);
1821 if (newfidp) {
1822 put_fid(pdu, newfidp);
1823 }
1824 v9fs_path_free(&dpath);
1825 v9fs_path_free(&path);
1826 out_nofid:
1827 pdu_complete(pdu, err);
1828 if (nwnames && nwnames <= P9_MAXWELEM) {
1829 for (name_idx = 0; name_idx < nwnames; name_idx++) {
1830 v9fs_string_free(&wnames[name_idx]);
1831 }
1832 g_free(wnames);
1833 g_free(qids);
1834 }
1835 }
1836
1837 static int32_t coroutine_fn get_iounit(V9fsPDU *pdu, V9fsPath *path)
1838 {
1839 struct statfs stbuf;
1840 int32_t iounit = 0;
1841 V9fsState *s = pdu->s;
1842
1843 /*
1844 * iounit should be multiples of f_bsize (host filesystem block size
1845 * and as well as less than (client msize - P9_IOHDRSZ))
1846 */
1847 if (!v9fs_co_statfs(pdu, path, &stbuf)) {
1848 if (stbuf.f_bsize) {
1849 iounit = stbuf.f_bsize;
1850 iounit *= (s->msize - P9_IOHDRSZ) / stbuf.f_bsize;
1851 }
1852 }
1853 if (!iounit) {
1854 iounit = s->msize - P9_IOHDRSZ;
1855 }
1856 return iounit;
1857 }
1858
1859 static void coroutine_fn v9fs_open(void *opaque)
1860 {
1861 int flags;
1862 int32_t fid;
1863 int32_t mode;
1864 V9fsQID qid;
1865 int iounit = 0;
1866 ssize_t err = 0;
1867 size_t offset = 7;
1868 struct stat stbuf;
1869 V9fsFidState *fidp;
1870 V9fsPDU *pdu = opaque;
1871 V9fsState *s = pdu->s;
1872
1873 if (s->proto_version == V9FS_PROTO_2000L) {
1874 err = pdu_unmarshal(pdu, offset, "dd", &fid, &mode);
1875 } else {
1876 uint8_t modebyte;
1877 err = pdu_unmarshal(pdu, offset, "db", &fid, &modebyte);
1878 mode = modebyte;
1879 }
1880 if (err < 0) {
1881 goto out_nofid;
1882 }
1883 trace_v9fs_open(pdu->tag, pdu->id, fid, mode);
1884
1885 fidp = get_fid(pdu, fid);
1886 if (fidp == NULL) {
1887 err = -ENOENT;
1888 goto out_nofid;
1889 }
1890 if (fidp->fid_type != P9_FID_NONE) {
1891 err = -EINVAL;
1892 goto out;
1893 }
1894
1895 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1896 if (err < 0) {
1897 goto out;
1898 }
1899 err = stat_to_qid(pdu, &stbuf, &qid);
1900 if (err < 0) {
1901 goto out;
1902 }
1903 if (S_ISDIR(stbuf.st_mode)) {
1904 err = v9fs_co_opendir(pdu, fidp);
1905 if (err < 0) {
1906 goto out;
1907 }
1908 fidp->fid_type = P9_FID_DIR;
1909 err = pdu_marshal(pdu, offset, "Qd", &qid, 0);
1910 if (err < 0) {
1911 goto out;
1912 }
1913 err += offset;
1914 } else {
1915 if (s->proto_version == V9FS_PROTO_2000L) {
1916 flags = get_dotl_openflags(s, mode);
1917 } else {
1918 flags = omode_to_uflags(mode);
1919 }
1920 if (is_ro_export(&s->ctx)) {
1921 if (mode & O_WRONLY || mode & O_RDWR ||
1922 mode & O_APPEND || mode & O_TRUNC) {
1923 err = -EROFS;
1924 goto out;
1925 }
1926 }
1927 err = v9fs_co_open(pdu, fidp, flags);
1928 if (err < 0) {
1929 goto out;
1930 }
1931 fidp->fid_type = P9_FID_FILE;
1932 fidp->open_flags = flags;
1933 if (flags & O_EXCL) {
1934 /*
1935 * We let the host file system do O_EXCL check
1936 * We should not reclaim such fd
1937 */
1938 fidp->flags |= FID_NON_RECLAIMABLE;
1939 }
1940 iounit = get_iounit(pdu, &fidp->path);
1941 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1942 if (err < 0) {
1943 goto out;
1944 }
1945 err += offset;
1946 }
1947 trace_v9fs_open_return(pdu->tag, pdu->id,
1948 qid.type, qid.version, qid.path, iounit);
1949 out:
1950 put_fid(pdu, fidp);
1951 out_nofid:
1952 pdu_complete(pdu, err);
1953 }
1954
1955 static void coroutine_fn v9fs_lcreate(void *opaque)
1956 {
1957 int32_t dfid, flags, mode;
1958 gid_t gid;
1959 ssize_t err = 0;
1960 ssize_t offset = 7;
1961 V9fsString name;
1962 V9fsFidState *fidp;
1963 struct stat stbuf;
1964 V9fsQID qid;
1965 int32_t iounit;
1966 V9fsPDU *pdu = opaque;
1967
1968 v9fs_string_init(&name);
1969 err = pdu_unmarshal(pdu, offset, "dsddd", &dfid,
1970 &name, &flags, &mode, &gid);
1971 if (err < 0) {
1972 goto out_nofid;
1973 }
1974 trace_v9fs_lcreate(pdu->tag, pdu->id, dfid, flags, mode, gid);
1975
1976 if (name_is_illegal(name.data)) {
1977 err = -ENOENT;
1978 goto out_nofid;
1979 }
1980
1981 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
1982 err = -EEXIST;
1983 goto out_nofid;
1984 }
1985
1986 fidp = get_fid(pdu, dfid);
1987 if (fidp == NULL) {
1988 err = -ENOENT;
1989 goto out_nofid;
1990 }
1991 if (fidp->fid_type != P9_FID_NONE) {
1992 err = -EINVAL;
1993 goto out;
1994 }
1995
1996 flags = get_dotl_openflags(pdu->s, flags);
1997 err = v9fs_co_open2(pdu, fidp, &name, gid,
1998 flags | O_CREAT, mode, &stbuf);
1999 if (err < 0) {
2000 goto out;
2001 }
2002 fidp->fid_type = P9_FID_FILE;
2003 fidp->open_flags = flags;
2004 if (flags & O_EXCL) {
2005 /*
2006 * We let the host file system do O_EXCL check
2007 * We should not reclaim such fd
2008 */
2009 fidp->flags |= FID_NON_RECLAIMABLE;
2010 }
2011 iounit = get_iounit(pdu, &fidp->path);
2012 err = stat_to_qid(pdu, &stbuf, &qid);
2013 if (err < 0) {
2014 goto out;
2015 }
2016 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
2017 if (err < 0) {
2018 goto out;
2019 }
2020 err += offset;
2021 trace_v9fs_lcreate_return(pdu->tag, pdu->id,
2022 qid.type, qid.version, qid.path, iounit);
2023 out:
2024 put_fid(pdu, fidp);
2025 out_nofid:
2026 pdu_complete(pdu, err);
2027 v9fs_string_free(&name);
2028 }
2029
2030 static void coroutine_fn v9fs_fsync(void *opaque)
2031 {
2032 int err;
2033 int32_t fid;
2034 int datasync;
2035 size_t offset = 7;
2036 V9fsFidState *fidp;
2037 V9fsPDU *pdu = opaque;
2038
2039 err = pdu_unmarshal(pdu, offset, "dd", &fid, &datasync);
2040 if (err < 0) {
2041 goto out_nofid;
2042 }
2043 trace_v9fs_fsync(pdu->tag, pdu->id, fid, datasync);
2044
2045 fidp = get_fid(pdu, fid);
2046 if (fidp == NULL) {
2047 err = -ENOENT;
2048 goto out_nofid;
2049 }
2050 err = v9fs_co_fsync(pdu, fidp, datasync);
2051 if (!err) {
2052 err = offset;
2053 }
2054 put_fid(pdu, fidp);
2055 out_nofid:
2056 pdu_complete(pdu, err);
2057 }
2058
2059 static void coroutine_fn v9fs_clunk(void *opaque)
2060 {
2061 int err;
2062 int32_t fid;
2063 size_t offset = 7;
2064 V9fsFidState *fidp;
2065 V9fsPDU *pdu = opaque;
2066 V9fsState *s = pdu->s;
2067
2068 err = pdu_unmarshal(pdu, offset, "d", &fid);
2069 if (err < 0) {
2070 goto out_nofid;
2071 }
2072 trace_v9fs_clunk(pdu->tag, pdu->id, fid);
2073
2074 fidp = clunk_fid(s, fid);
2075 if (fidp == NULL) {
2076 err = -ENOENT;
2077 goto out_nofid;
2078 }
2079 /*
2080 * Bump the ref so that put_fid will
2081 * free the fid.
2082 */
2083 fidp->ref++;
2084 err = put_fid(pdu, fidp);
2085 if (!err) {
2086 err = offset;
2087 }
2088 out_nofid:
2089 pdu_complete(pdu, err);
2090 }
2091
2092 /*
2093 * Create a QEMUIOVector for a sub-region of PDU iovecs
2094 *
2095 * @qiov: uninitialized QEMUIOVector
2096 * @skip: number of bytes to skip from beginning of PDU
2097 * @size: number of bytes to include
2098 * @is_write: true - write, false - read
2099 *
2100 * The resulting QEMUIOVector has heap-allocated iovecs and must be cleaned up
2101 * with qemu_iovec_destroy().
2102 */
2103 static void v9fs_init_qiov_from_pdu(QEMUIOVector *qiov, V9fsPDU *pdu,
2104 size_t skip, size_t size,
2105 bool is_write)
2106 {
2107 QEMUIOVector elem;
2108 struct iovec *iov;
2109 unsigned int niov;
2110
2111 if (is_write) {
2112 pdu->s->transport->init_out_iov_from_pdu(pdu, &iov, &niov, size + skip);
2113 } else {
2114 pdu->s->transport->init_in_iov_from_pdu(pdu, &iov, &niov, size + skip);
2115 }
2116
2117 qemu_iovec_init_external(&elem, iov, niov);
2118 qemu_iovec_init(qiov, niov);
2119 qemu_iovec_concat(qiov, &elem, skip, size);
2120 }
2121
2122 static int v9fs_xattr_read(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
2123 uint64_t off, uint32_t max_count)
2124 {
2125 ssize_t err;
2126 size_t offset = 7;
2127 uint64_t read_count;
2128 QEMUIOVector qiov_full;
2129
2130 if (fidp->fs.xattr.len < off) {
2131 read_count = 0;
2132 } else {
2133 read_count = fidp->fs.xattr.len - off;
2134 }
2135 if (read_count > max_count) {
2136 read_count = max_count;
2137 }
2138 err = pdu_marshal(pdu, offset, "d", read_count);
2139 if (err < 0) {
2140 return err;
2141 }
2142 offset += err;
2143
2144 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, read_count, false);
2145 err = v9fs_pack(qiov_full.iov, qiov_full.niov, 0,
2146 ((char *)fidp->fs.xattr.value) + off,
2147 read_count);
2148 qemu_iovec_destroy(&qiov_full);
2149 if (err < 0) {
2150 return err;
2151 }
2152 offset += err;
2153 return offset;
2154 }
2155
2156 static int coroutine_fn v9fs_do_readdir_with_stat(V9fsPDU *pdu,
2157 V9fsFidState *fidp,
2158 uint32_t max_count)
2159 {
2160 V9fsPath path;
2161 V9fsStat v9stat;
2162 int len, err = 0;
2163 int32_t count = 0;
2164 struct stat stbuf;
2165 off_t saved_dir_pos;
2166 struct dirent *dent;
2167
2168 /* save the directory position */
2169 saved_dir_pos = v9fs_co_telldir(pdu, fidp);
2170 if (saved_dir_pos < 0) {
2171 return saved_dir_pos;
2172 }
2173
2174 while (1) {
2175 v9fs_path_init(&path);
2176
2177 v9fs_readdir_lock(&fidp->fs.dir);
2178
2179 err = v9fs_co_readdir(pdu, fidp, &dent);
2180 if (err || !dent) {
2181 break;
2182 }
2183 err = v9fs_co_name_to_path(pdu, &fidp->path, dent->d_name, &path);
2184 if (err < 0) {
2185 break;
2186 }
2187 err = v9fs_co_lstat(pdu, &path, &stbuf);
2188 if (err < 0) {
2189 break;
2190 }
2191 err = stat_to_v9stat(pdu, &path, dent->d_name, &stbuf, &v9stat);
2192 if (err < 0) {
2193 break;
2194 }
2195 if ((count + v9stat.size + 2) > max_count) {
2196 v9fs_readdir_unlock(&fidp->fs.dir);
2197
2198 /* Ran out of buffer. Set dir back to old position and return */
2199 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2200 v9fs_stat_free(&v9stat);
2201 v9fs_path_free(&path);
2202 return count;
2203 }
2204
2205 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2206 len = pdu_marshal(pdu, 11 + count, "S", &v9stat);
2207
2208 v9fs_readdir_unlock(&fidp->fs.dir);
2209
2210 if (len < 0) {
2211 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2212 v9fs_stat_free(&v9stat);
2213 v9fs_path_free(&path);
2214 return len;
2215 }
2216 count += len;
2217 v9fs_stat_free(&v9stat);
2218 v9fs_path_free(&path);
2219 saved_dir_pos = dent->d_off;
2220 }
2221
2222 v9fs_readdir_unlock(&fidp->fs.dir);
2223
2224 v9fs_path_free(&path);
2225 if (err < 0) {
2226 return err;
2227 }
2228 return count;
2229 }
2230
2231 static void coroutine_fn v9fs_read(void *opaque)
2232 {
2233 int32_t fid;
2234 uint64_t off;
2235 ssize_t err = 0;
2236 int32_t count = 0;
2237 size_t offset = 7;
2238 uint32_t max_count;
2239 V9fsFidState *fidp;
2240 V9fsPDU *pdu = opaque;
2241 V9fsState *s = pdu->s;
2242
2243 err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &max_count);
2244 if (err < 0) {
2245 goto out_nofid;
2246 }
2247 trace_v9fs_read(pdu->tag, pdu->id, fid, off, max_count);
2248
2249 fidp = get_fid(pdu, fid);
2250 if (fidp == NULL) {
2251 err = -EINVAL;
2252 goto out_nofid;
2253 }
2254 if (fidp->fid_type == P9_FID_DIR) {
2255
2256 if (off == 0) {
2257 v9fs_co_rewinddir(pdu, fidp);
2258 }
2259 count = v9fs_do_readdir_with_stat(pdu, fidp, max_count);
2260 if (count < 0) {
2261 err = count;
2262 goto out;
2263 }
2264 err = pdu_marshal(pdu, offset, "d", count);
2265 if (err < 0) {
2266 goto out;
2267 }
2268 err += offset + count;
2269 } else if (fidp->fid_type == P9_FID_FILE) {
2270 QEMUIOVector qiov_full;
2271 QEMUIOVector qiov;
2272 int32_t len;
2273
2274 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset + 4, max_count, false);
2275 qemu_iovec_init(&qiov, qiov_full.niov);
2276 do {
2277 qemu_iovec_reset(&qiov);
2278 qemu_iovec_concat(&qiov, &qiov_full, count, qiov_full.size - count);
2279 if (0) {
2280 print_sg(qiov.iov, qiov.niov);
2281 }
2282 /* Loop in case of EINTR */
2283 do {
2284 len = v9fs_co_preadv(pdu, fidp, qiov.iov, qiov.niov, off);
2285 if (len >= 0) {
2286 off += len;
2287 count += len;
2288 }
2289 } while (len == -EINTR && !pdu->cancelled);
2290 if (len < 0) {
2291 /* IO error return the error */
2292 err = len;
2293 goto out_free_iovec;
2294 }
2295 } while (count < max_count && len > 0);
2296 err = pdu_marshal(pdu, offset, "d", count);
2297 if (err < 0) {
2298 goto out_free_iovec;
2299 }
2300 err += offset + count;
2301 out_free_iovec:
2302 qemu_iovec_destroy(&qiov);
2303 qemu_iovec_destroy(&qiov_full);
2304 } else if (fidp->fid_type == P9_FID_XATTR) {
2305 err = v9fs_xattr_read(s, pdu, fidp, off, max_count);
2306 } else {
2307 err = -EINVAL;
2308 }
2309 trace_v9fs_read_return(pdu->tag, pdu->id, count, err);
2310 out:
2311 put_fid(pdu, fidp);
2312 out_nofid:
2313 pdu_complete(pdu, err);
2314 }
2315
2316 static size_t v9fs_readdir_data_size(V9fsString *name)
2317 {
2318 /*
2319 * Size of each dirent on the wire: size of qid (13) + size of offset (8)
2320 * size of type (1) + size of name.size (2) + strlen(name.data)
2321 */
2322 return 24 + v9fs_string_size(name);
2323 }
2324
2325 static int coroutine_fn v9fs_do_readdir(V9fsPDU *pdu, V9fsFidState *fidp,
2326 int32_t max_count)
2327 {
2328 size_t size;
2329 V9fsQID qid;
2330 V9fsString name;
2331 int len, err = 0;
2332 int32_t count = 0;
2333 off_t saved_dir_pos;
2334 struct dirent *dent;
2335
2336 /* save the directory position */
2337 saved_dir_pos = v9fs_co_telldir(pdu, fidp);
2338 if (saved_dir_pos < 0) {
2339 return saved_dir_pos;
2340 }
2341
2342 while (1) {
2343 v9fs_readdir_lock(&fidp->fs.dir);
2344
2345 err = v9fs_co_readdir(pdu, fidp, &dent);
2346 if (err || !dent) {
2347 break;
2348 }
2349 v9fs_string_init(&name);
2350 v9fs_string_sprintf(&name, "%s", dent->d_name);
2351 if ((count + v9fs_readdir_data_size(&name)) > max_count) {
2352 v9fs_readdir_unlock(&fidp->fs.dir);
2353
2354 /* Ran out of buffer. Set dir back to old position and return */
2355 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2356 v9fs_string_free(&name);
2357 return count;
2358 }
2359
2360 if (pdu->s->ctx.export_flags & V9FS_REMAP_INODES) {
2361 /*
2362 * dirent_to_qid() implies expensive stat call for each entry,
2363 * we must do that here though since inode remapping requires
2364 * the device id, which in turn might be different for
2365 * different entries; we cannot make any assumption to avoid
2366 * that here.
2367 */
2368 err = dirent_to_qid(pdu, fidp, dent, &qid);
2369 if (err < 0) {
2370 v9fs_readdir_unlock(&fidp->fs.dir);
2371 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2372 v9fs_string_free(&name);
2373 return err;
2374 }
2375 } else {
2376 /*
2377 * Fill up just the path field of qid because the client uses
2378 * only that. To fill the entire qid structure we will have
2379 * to stat each dirent found, which is expensive. For the
2380 * latter reason we don't call dirent_to_qid() here. Only drawback
2381 * is that no multi-device export detection of stat_to_qid()
2382 * would be done and provided as error to the user here. But
2383 * user would get that error anyway when accessing those
2384 * files/dirs through other ways.
2385 */
2386 size = MIN(sizeof(dent->d_ino), sizeof(qid.path));
2387 memcpy(&qid.path, &dent->d_ino, size);
2388 /* Fill the other fields with dummy values */
2389 qid.type = 0;
2390 qid.version = 0;
2391 }
2392
2393 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2394 len = pdu_marshal(pdu, 11 + count, "Qqbs",
2395 &qid, dent->d_off,
2396 dent->d_type, &name);
2397
2398 v9fs_readdir_unlock(&fidp->fs.dir);
2399
2400 if (len < 0) {
2401 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2402 v9fs_string_free(&name);
2403 return len;
2404 }
2405 count += len;
2406 v9fs_string_free(&name);
2407 saved_dir_pos = dent->d_off;
2408 }
2409
2410 v9fs_readdir_unlock(&fidp->fs.dir);
2411
2412 if (err < 0) {
2413 return err;
2414 }
2415 return count;
2416 }
2417
2418 static void coroutine_fn v9fs_readdir(void *opaque)
2419 {
2420 int32_t fid;
2421 V9fsFidState *fidp;
2422 ssize_t retval = 0;
2423 size_t offset = 7;
2424 uint64_t initial_offset;
2425 int32_t count;
2426 uint32_t max_count;
2427 V9fsPDU *pdu = opaque;
2428 V9fsState *s = pdu->s;
2429
2430 retval = pdu_unmarshal(pdu, offset, "dqd", &fid,
2431 &initial_offset, &max_count);
2432 if (retval < 0) {
2433 goto out_nofid;
2434 }
2435 trace_v9fs_readdir(pdu->tag, pdu->id, fid, initial_offset, max_count);
2436
2437 /* Enough space for a R_readdir header: size[4] Rreaddir tag[2] count[4] */
2438 if (max_count > s->msize - 11) {
2439 max_count = s->msize - 11;
2440 warn_report_once(
2441 "9p: bad client: T_readdir with count > msize - 11"
2442 );
2443 }
2444
2445 fidp = get_fid(pdu, fid);
2446 if (fidp == NULL) {
2447 retval = -EINVAL;
2448 goto out_nofid;
2449 }
2450 if (!fidp->fs.dir.stream) {
2451 retval = -EINVAL;
2452 goto out;
2453 }
2454 if (initial_offset == 0) {
2455 v9fs_co_rewinddir(pdu, fidp);
2456 } else {
2457 v9fs_co_seekdir(pdu, fidp, initial_offset);
2458 }
2459 count = v9fs_do_readdir(pdu, fidp, max_count);
2460 if (count < 0) {
2461 retval = count;
2462 goto out;
2463 }
2464 retval = pdu_marshal(pdu, offset, "d", count);
2465 if (retval < 0) {
2466 goto out;
2467 }
2468 retval += count + offset;
2469 trace_v9fs_readdir_return(pdu->tag, pdu->id, count, retval);
2470 out:
2471 put_fid(pdu, fidp);
2472 out_nofid:
2473 pdu_complete(pdu, retval);
2474 }
2475
2476 static int v9fs_xattr_write(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
2477 uint64_t off, uint32_t count,
2478 struct iovec *sg, int cnt)
2479 {
2480 int i, to_copy;
2481 ssize_t err = 0;
2482 uint64_t write_count;
2483 size_t offset = 7;
2484
2485
2486 if (fidp->fs.xattr.len < off) {
2487 return -ENOSPC;
2488 }
2489 write_count = fidp->fs.xattr.len - off;
2490 if (write_count > count) {
2491 write_count = count;
2492 }
2493 err = pdu_marshal(pdu, offset, "d", write_count);
2494 if (err < 0) {
2495 return err;
2496 }
2497 err += offset;
2498 fidp->fs.xattr.copied_len += write_count;
2499 /*
2500 * Now copy the content from sg list
2501 */
2502 for (i = 0; i < cnt; i++) {
2503 if (write_count > sg[i].iov_len) {
2504 to_copy = sg[i].iov_len;
2505 } else {
2506 to_copy = write_count;
2507 }
2508 memcpy((char *)fidp->fs.xattr.value + off, sg[i].iov_base, to_copy);
2509 /* updating vs->off since we are not using below */
2510 off += to_copy;
2511 write_count -= to_copy;
2512 }
2513
2514 return err;
2515 }
2516
2517 static void coroutine_fn v9fs_write(void *opaque)
2518 {
2519 ssize_t err;
2520 int32_t fid;
2521 uint64_t off;
2522 uint32_t count;
2523 int32_t len = 0;
2524 int32_t total = 0;
2525 size_t offset = 7;
2526 V9fsFidState *fidp;
2527 V9fsPDU *pdu = opaque;
2528 V9fsState *s = pdu->s;
2529 QEMUIOVector qiov_full;
2530 QEMUIOVector qiov;
2531
2532 err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &count);
2533 if (err < 0) {
2534 pdu_complete(pdu, err);
2535 return;
2536 }
2537 offset += err;
2538 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, count, true);
2539 trace_v9fs_write(pdu->tag, pdu->id, fid, off, count, qiov_full.niov);
2540
2541 fidp = get_fid(pdu, fid);
2542 if (fidp == NULL) {
2543 err = -EINVAL;
2544 goto out_nofid;
2545 }
2546 if (fidp->fid_type == P9_FID_FILE) {
2547 if (fidp->fs.fd == -1) {
2548 err = -EINVAL;
2549 goto out;
2550 }
2551 } else if (fidp->fid_type == P9_FID_XATTR) {
2552 /*
2553 * setxattr operation
2554 */
2555 err = v9fs_xattr_write(s, pdu, fidp, off, count,
2556 qiov_full.iov, qiov_full.niov);
2557 goto out;
2558 } else {
2559 err = -EINVAL;
2560 goto out;
2561 }
2562 qemu_iovec_init(&qiov, qiov_full.niov);
2563 do {
2564 qemu_iovec_reset(&qiov);
2565 qemu_iovec_concat(&qiov, &qiov_full, total, qiov_full.size - total);
2566 if (0) {
2567 print_sg(qiov.iov, qiov.niov);
2568 }
2569 /* Loop in case of EINTR */
2570 do {
2571 len = v9fs_co_pwritev(pdu, fidp, qiov.iov, qiov.niov, off);
2572 if (len >= 0) {
2573 off += len;
2574 total += len;
2575 }
2576 } while (len == -EINTR && !pdu->cancelled);
2577 if (len < 0) {
2578 /* IO error return the error */
2579 err = len;
2580 goto out_qiov;
2581 }
2582 } while (total < count && len > 0);
2583
2584 offset = 7;
2585 err = pdu_marshal(pdu, offset, "d", total);
2586 if (err < 0) {
2587 goto out_qiov;
2588 }
2589 err += offset;
2590 trace_v9fs_write_return(pdu->tag, pdu->id, total, err);
2591 out_qiov:
2592 qemu_iovec_destroy(&qiov);
2593 out:
2594 put_fid(pdu, fidp);
2595 out_nofid:
2596 qemu_iovec_destroy(&qiov_full);
2597 pdu_complete(pdu, err);
2598 }
2599
2600 static void coroutine_fn v9fs_create(void *opaque)
2601 {
2602 int32_t fid;
2603 int err = 0;
2604 size_t offset = 7;
2605 V9fsFidState *fidp;
2606 V9fsQID qid;
2607 int32_t perm;
2608 int8_t mode;
2609 V9fsPath path;
2610 struct stat stbuf;
2611 V9fsString name;
2612 V9fsString extension;
2613 int iounit;
2614 V9fsPDU *pdu = opaque;
2615 V9fsState *s = pdu->s;
2616
2617 v9fs_path_init(&path);
2618 v9fs_string_init(&name);
2619 v9fs_string_init(&extension);
2620 err = pdu_unmarshal(pdu, offset, "dsdbs", &fid, &name,
2621 &perm, &mode, &extension);
2622 if (err < 0) {
2623 goto out_nofid;
2624 }
2625 trace_v9fs_create(pdu->tag, pdu->id, fid, name.data, perm, mode);
2626
2627 if (name_is_illegal(name.data)) {
2628 err = -ENOENT;
2629 goto out_nofid;
2630 }
2631
2632 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2633 err = -EEXIST;
2634 goto out_nofid;
2635 }
2636
2637 fidp = get_fid(pdu, fid);
2638 if (fidp == NULL) {
2639 err = -EINVAL;
2640 goto out_nofid;
2641 }
2642 if (fidp->fid_type != P9_FID_NONE) {
2643 err = -EINVAL;
2644 goto out;
2645 }
2646 if (perm & P9_STAT_MODE_DIR) {
2647 err = v9fs_co_mkdir(pdu, fidp, &name, perm & 0777,
2648 fidp->uid, -1, &stbuf);
2649 if (err < 0) {
2650 goto out;
2651 }
2652 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2653 if (err < 0) {
2654 goto out;
2655 }
2656 v9fs_path_write_lock(s);
2657 v9fs_path_copy(&fidp->path, &path);
2658 v9fs_path_unlock(s);
2659 err = v9fs_co_opendir(pdu, fidp);
2660 if (err < 0) {
2661 goto out;
2662 }
2663 fidp->fid_type = P9_FID_DIR;
2664 } else if (perm & P9_STAT_MODE_SYMLINK) {
2665 err = v9fs_co_symlink(pdu, fidp, &name,
2666 extension.data, -1 , &stbuf);
2667 if (err < 0) {
2668 goto out;
2669 }
2670 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2671 if (err < 0) {
2672 goto out;
2673 }
2674 v9fs_path_write_lock(s);
2675 v9fs_path_copy(&fidp->path, &path);
2676 v9fs_path_unlock(s);
2677 } else if (perm & P9_STAT_MODE_LINK) {
2678 int32_t ofid = atoi(extension.data);
2679 V9fsFidState *ofidp = get_fid(pdu, ofid);
2680 if (ofidp == NULL) {
2681 err = -EINVAL;
2682 goto out;
2683 }
2684 err = v9fs_co_link(pdu, ofidp, fidp, &name);
2685 put_fid(pdu, ofidp);
2686 if (err < 0) {
2687 goto out;
2688 }
2689 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2690 if (err < 0) {
2691 fidp->fid_type = P9_FID_NONE;
2692 goto out;
2693 }
2694 v9fs_path_write_lock(s);
2695 v9fs_path_copy(&fidp->path, &path);
2696 v9fs_path_unlock(s);
2697 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
2698 if (err < 0) {
2699 fidp->fid_type = P9_FID_NONE;
2700 goto out;
2701 }
2702 } else if (perm & P9_STAT_MODE_DEVICE) {
2703 char ctype;
2704 uint32_t major, minor;
2705 mode_t nmode = 0;
2706
2707 if (sscanf(extension.data, "%c %u %u", &ctype, &major, &minor) != 3) {
2708 err = -errno;
2709 goto out;
2710 }
2711
2712 switch (ctype) {
2713 case 'c':
2714 nmode = S_IFCHR;
2715 break;
2716 case 'b':
2717 nmode = S_IFBLK;
2718 break;
2719 default:
2720 err = -EIO;
2721 goto out;
2722 }
2723
2724 nmode |= perm & 0777;
2725 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2726 makedev(major, minor), nmode, &stbuf);
2727 if (err < 0) {
2728 goto out;
2729 }
2730 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2731 if (err < 0) {
2732 goto out;
2733 }
2734 v9fs_path_write_lock(s);
2735 v9fs_path_copy(&fidp->path, &path);
2736 v9fs_path_unlock(s);
2737 } else if (perm & P9_STAT_MODE_NAMED_PIPE) {
2738 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2739 0, S_IFIFO | (perm & 0777), &stbuf);
2740 if (err < 0) {
2741 goto out;
2742 }
2743 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2744 if (err < 0) {
2745 goto out;
2746 }
2747 v9fs_path_write_lock(s);
2748 v9fs_path_copy(&fidp->path, &path);
2749 v9fs_path_unlock(s);
2750 } else if (perm & P9_STAT_MODE_SOCKET) {
2751 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2752 0, S_IFSOCK | (perm & 0777), &stbuf);
2753 if (err < 0) {
2754 goto out;
2755 }
2756 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2757 if (err < 0) {
2758 goto out;
2759 }
2760 v9fs_path_write_lock(s);
2761 v9fs_path_copy(&fidp->path, &path);
2762 v9fs_path_unlock(s);
2763 } else {
2764 err = v9fs_co_open2(pdu, fidp, &name, -1,
2765 omode_to_uflags(mode)|O_CREAT, perm, &stbuf);
2766 if (err < 0) {
2767 goto out;
2768 }
2769 fidp->fid_type = P9_FID_FILE;
2770 fidp->open_flags = omode_to_uflags(mode);
2771 if (fidp->open_flags & O_EXCL) {
2772 /*
2773 * We let the host file system do O_EXCL check
2774 * We should not reclaim such fd
2775 */
2776 fidp->flags |= FID_NON_RECLAIMABLE;
2777 }
2778 }
2779 iounit = get_iounit(pdu, &fidp->path);
2780 err = stat_to_qid(pdu, &stbuf, &qid);
2781 if (err < 0) {
2782 goto out;
2783 }
2784 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
2785 if (err < 0) {
2786 goto out;
2787 }
2788 err += offset;
2789 trace_v9fs_create_return(pdu->tag, pdu->id,
2790 qid.type, qid.version, qid.path, iounit);
2791 out:
2792 put_fid(pdu, fidp);
2793 out_nofid:
2794 pdu_complete(pdu, err);
2795 v9fs_string_free(&name);
2796 v9fs_string_free(&extension);
2797 v9fs_path_free(&path);
2798 }
2799
2800 static void coroutine_fn v9fs_symlink(void *opaque)
2801 {
2802 V9fsPDU *pdu = opaque;
2803 V9fsString name;
2804 V9fsString symname;
2805 V9fsFidState *dfidp;
2806 V9fsQID qid;
2807 struct stat stbuf;
2808 int32_t dfid;
2809 int err = 0;
2810 gid_t gid;
2811 size_t offset = 7;
2812
2813 v9fs_string_init(&name);
2814 v9fs_string_init(&symname);
2815 err = pdu_unmarshal(pdu, offset, "dssd", &dfid, &name, &symname, &gid);
2816 if (err < 0) {
2817 goto out_nofid;
2818 }
2819 trace_v9fs_symlink(pdu->tag, pdu->id, dfid, name.data, symname.data, gid);
2820
2821 if (name_is_illegal(name.data)) {
2822 err = -ENOENT;
2823 goto out_nofid;
2824 }
2825
2826 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2827 err = -EEXIST;
2828 goto out_nofid;
2829 }
2830
2831 dfidp = get_fid(pdu, dfid);
2832 if (dfidp == NULL) {
2833 err = -EINVAL;
2834 goto out_nofid;
2835 }
2836 err = v9fs_co_symlink(pdu, dfidp, &name, symname.data, gid, &stbuf);
2837 if (err < 0) {
2838 goto out;
2839 }
2840 err = stat_to_qid(pdu, &stbuf, &qid);
2841 if (err < 0) {
2842 goto out;
2843 }
2844 err = pdu_marshal(pdu, offset, "Q", &qid);
2845 if (err < 0) {
2846 goto out;
2847 }
2848 err += offset;
2849 trace_v9fs_symlink_return(pdu->tag, pdu->id,
2850 qid.type, qid.version, qid.path);
2851 out:
2852 put_fid(pdu, dfidp);
2853 out_nofid:
2854 pdu_complete(pdu, err);
2855 v9fs_string_free(&name);
2856 v9fs_string_free(&symname);
2857 }
2858
2859 static void coroutine_fn v9fs_flush(void *opaque)
2860 {
2861 ssize_t err;
2862 int16_t tag;
2863 size_t offset = 7;
2864 V9fsPDU *cancel_pdu = NULL;
2865 V9fsPDU *pdu = opaque;
2866 V9fsState *s = pdu->s;
2867
2868 err = pdu_unmarshal(pdu, offset, "w", &tag);
2869 if (err < 0) {
2870 pdu_complete(pdu, err);
2871 return;
2872 }
2873 trace_v9fs_flush(pdu->tag, pdu->id, tag);
2874
2875 if (pdu->tag == tag) {
2876 warn_report("the guest sent a self-referencing 9P flush request");
2877 } else {
2878 QLIST_FOREACH(cancel_pdu, &s->active_list, next) {
2879 if (cancel_pdu->tag == tag) {
2880 break;
2881 }
2882 }
2883 }
2884 if (cancel_pdu) {
2885 cancel_pdu->cancelled = 1;
2886 /*
2887 * Wait for pdu to complete.
2888 */
2889 qemu_co_queue_wait(&cancel_pdu->complete, NULL);
2890 if (!qemu_co_queue_next(&cancel_pdu->complete)) {
2891 cancel_pdu->cancelled = 0;
2892 pdu_free(cancel_pdu);
2893 }
2894 }
2895 pdu_complete(pdu, 7);
2896 }
2897
2898 static void coroutine_fn v9fs_link(void *opaque)
2899 {
2900 V9fsPDU *pdu = opaque;
2901 int32_t dfid, oldfid;
2902 V9fsFidState *dfidp, *oldfidp;
2903 V9fsString name;
2904 size_t offset = 7;
2905 int err = 0;
2906
2907 v9fs_string_init(&name);
2908 err = pdu_unmarshal(pdu, offset, "dds", &dfid, &oldfid, &name);
2909 if (err < 0) {
2910 goto out_nofid;
2911 }
2912 trace_v9fs_link(pdu->tag, pdu->id, dfid, oldfid, name.data);
2913
2914 if (name_is_illegal(name.data)) {
2915 err = -ENOENT;
2916 goto out_nofid;
2917 }
2918
2919 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2920 err = -EEXIST;
2921 goto out_nofid;
2922 }
2923
2924 dfidp = get_fid(pdu, dfid);
2925 if (dfidp == NULL) {
2926 err = -ENOENT;
2927 goto out_nofid;
2928 }
2929
2930 oldfidp = get_fid(pdu, oldfid);
2931 if (oldfidp == NULL) {
2932 err = -ENOENT;
2933 goto out;
2934 }
2935 err = v9fs_co_link(pdu, oldfidp, dfidp, &name);
2936 if (!err) {
2937 err = offset;
2938 }
2939 put_fid(pdu, oldfidp);
2940 out:
2941 put_fid(pdu, dfidp);
2942 out_nofid:
2943 v9fs_string_free(&name);
2944 pdu_complete(pdu, err);
2945 }
2946
2947 /* Only works with path name based fid */
2948 static void coroutine_fn v9fs_remove(void *opaque)
2949 {
2950 int32_t fid;
2951 int err = 0;
2952 size_t offset = 7;
2953 V9fsFidState *fidp;
2954 V9fsPDU *pdu = opaque;
2955
2956 err = pdu_unmarshal(pdu, offset, "d", &fid);
2957 if (err < 0) {
2958 goto out_nofid;
2959 }
2960 trace_v9fs_remove(pdu->tag, pdu->id, fid);
2961
2962 fidp = get_fid(pdu, fid);
2963 if (fidp == NULL) {
2964 err = -EINVAL;
2965 goto out_nofid;
2966 }
2967 /* if fs driver is not path based, return EOPNOTSUPP */
2968 if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
2969 err = -EOPNOTSUPP;
2970 goto out_err;
2971 }
2972 /*
2973 * IF the file is unlinked, we cannot reopen
2974 * the file later. So don't reclaim fd
2975 */
2976 err = v9fs_mark_fids_unreclaim(pdu, &fidp->path);
2977 if (err < 0) {
2978 goto out_err;
2979 }
2980 err = v9fs_co_remove(pdu, &fidp->path);
2981 if (!err) {
2982 err = offset;
2983 }
2984 out_err:
2985 /* For TREMOVE we need to clunk the fid even on failed remove */
2986 clunk_fid(pdu->s, fidp->fid);
2987 put_fid(pdu, fidp);
2988 out_nofid:
2989 pdu_complete(pdu, err);
2990 }
2991
2992 static void coroutine_fn v9fs_unlinkat(void *opaque)
2993 {
2994 int err = 0;
2995 V9fsString name;
2996 int32_t dfid, flags, rflags = 0;
2997 size_t offset = 7;
2998 V9fsPath path;
2999 V9fsFidState *dfidp;
3000 V9fsPDU *pdu = opaque;
3001
3002 v9fs_string_init(&name);
3003 err = pdu_unmarshal(pdu, offset, "dsd", &dfid, &name, &flags);
3004 if (err < 0) {
3005 goto out_nofid;
3006 }
3007
3008 if (name_is_illegal(name.data)) {
3009 err = -ENOENT;
3010 goto out_nofid;
3011 }
3012
3013 if (!strcmp(".", name.data)) {
3014 err = -EINVAL;
3015 goto out_nofid;
3016 }
3017
3018 if (!strcmp("..", name.data)) {
3019 err = -ENOTEMPTY;
3020 goto out_nofid;
3021 }
3022
3023 if (flags & ~P9_DOTL_AT_REMOVEDIR) {
3024 err = -EINVAL;
3025 goto out_nofid;
3026 }
3027
3028 if (flags & P9_DOTL_AT_REMOVEDIR) {
3029 rflags |= AT_REMOVEDIR;
3030 }
3031
3032 dfidp = get_fid(pdu, dfid);
3033 if (dfidp == NULL) {
3034 err = -EINVAL;
3035 goto out_nofid;
3036 }
3037 /*
3038 * IF the file is unlinked, we cannot reopen
3039 * the file later. So don't reclaim fd
3040 */
3041 v9fs_path_init(&path);
3042 err = v9fs_co_name_to_path(pdu, &dfidp->path, name.data, &path);
3043 if (err < 0) {
3044 goto out_err;
3045 }
3046 err = v9fs_mark_fids_unreclaim(pdu, &path);
3047 if (err < 0) {
3048 goto out_err;
3049 }
3050 err = v9fs_co_unlinkat(pdu, &dfidp->path, &name, rflags);
3051 if (!err) {
3052 err = offset;
3053 }
3054 out_err:
3055 put_fid(pdu, dfidp);
3056 v9fs_path_free(&path);
3057 out_nofid:
3058 pdu_complete(pdu, err);
3059 v9fs_string_free(&name);
3060 }
3061
3062
3063 /* Only works with path name based fid */
3064 static int coroutine_fn v9fs_complete_rename(V9fsPDU *pdu, V9fsFidState *fidp,
3065 int32_t newdirfid,
3066 V9fsString *name)
3067 {
3068 int err = 0;
3069 V9fsPath new_path;
3070 V9fsFidState *tfidp;
3071 V9fsState *s = pdu->s;
3072 V9fsFidState *dirfidp = NULL;
3073
3074 v9fs_path_init(&new_path);
3075 if (newdirfid != -1) {
3076 dirfidp = get_fid(pdu, newdirfid);
3077 if (dirfidp == NULL) {
3078 return -ENOENT;
3079 }
3080 if (fidp->fid_type != P9_FID_NONE) {
3081 err = -EINVAL;
3082 goto out;
3083 }
3084 err = v9fs_co_name_to_path(pdu, &dirfidp->path, name->data, &new_path);
3085 if (err < 0) {
3086 goto out;
3087 }
3088 } else {
3089 char *dir_name = g_path_get_dirname(fidp->path.data);
3090 V9fsPath dir_path;
3091
3092 v9fs_path_init(&dir_path);
3093 v9fs_path_sprintf(&dir_path, "%s", dir_name);
3094 g_free(dir_name);
3095
3096 err = v9fs_co_name_to_path(pdu, &dir_path, name->data, &new_path);
3097 v9fs_path_free(&dir_path);
3098 if (err < 0) {
3099 goto out;
3100 }
3101 }
3102 err = v9fs_co_rename(pdu, &fidp->path, &new_path);
3103 if (err < 0) {
3104 goto out;
3105 }
3106 /*
3107 * Fixup fid's pointing to the old name to
3108 * start pointing to the new name
3109 */
3110 for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
3111 if (v9fs_path_is_ancestor(&fidp->path, &tfidp->path)) {
3112 /* replace the name */
3113 v9fs_fix_path(&tfidp->path, &new_path, strlen(fidp->path.data));
3114 }
3115 }
3116 out:
3117 if (dirfidp) {
3118 put_fid(pdu, dirfidp);
3119 }
3120 v9fs_path_free(&new_path);
3121 return err;
3122 }
3123
3124 /* Only works with path name based fid */
3125 static void coroutine_fn v9fs_rename(void *opaque)
3126 {
3127 int32_t fid;
3128 ssize_t err = 0;
3129 size_t offset = 7;
3130 V9fsString name;
3131 int32_t newdirfid;
3132 V9fsFidState *fidp;
3133 V9fsPDU *pdu = opaque;
3134 V9fsState *s = pdu->s;
3135
3136 v9fs_string_init(&name);
3137 err = pdu_unmarshal(pdu, offset, "dds", &fid, &newdirfid, &name);
3138 if (err < 0) {
3139 goto out_nofid;
3140 }
3141
3142 if (name_is_illegal(name.data)) {
3143 err = -ENOENT;
3144 goto out_nofid;
3145 }
3146
3147 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3148 err = -EISDIR;
3149 goto out_nofid;
3150 }
3151
3152 fidp = get_fid(pdu, fid);
3153 if (fidp == NULL) {
3154 err = -ENOENT;
3155 goto out_nofid;
3156 }
3157 if (fidp->fid_type != P9_FID_NONE) {
3158 err = -EINVAL;
3159 goto out;
3160 }
3161 /* if fs driver is not path based, return EOPNOTSUPP */
3162 if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
3163 err = -EOPNOTSUPP;
3164 goto out;
3165 }
3166 v9fs_path_write_lock(s);
3167 err = v9fs_complete_rename(pdu, fidp, newdirfid, &name);
3168 v9fs_path_unlock(s);
3169 if (!err) {
3170 err = offset;
3171 }
3172 out:
3173 put_fid(pdu, fidp);
3174 out_nofid:
3175 pdu_complete(pdu, err);
3176 v9fs_string_free(&name);
3177 }
3178
3179 static int coroutine_fn v9fs_fix_fid_paths(V9fsPDU *pdu, V9fsPath *olddir,
3180 V9fsString *old_name,
3181 V9fsPath *newdir,
3182 V9fsString *new_name)
3183 {
3184 V9fsFidState *tfidp;
3185 V9fsPath oldpath, newpath;
3186 V9fsState *s = pdu->s;
3187 int err;
3188
3189 v9fs_path_init(&oldpath);
3190 v9fs_path_init(&newpath);
3191 err = v9fs_co_name_to_path(pdu, olddir, old_name->data, &oldpath);
3192 if (err < 0) {
3193 goto out;
3194 }
3195 err = v9fs_co_name_to_path(pdu, newdir, new_name->data, &newpath);
3196 if (err < 0) {
3197 goto out;
3198 }
3199
3200 /*
3201 * Fixup fid's pointing to the old name to
3202 * start pointing to the new name
3203 */
3204 for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
3205 if (v9fs_path_is_ancestor(&oldpath, &tfidp->path)) {
3206 /* replace the name */
3207 v9fs_fix_path(&tfidp->path, &newpath, strlen(oldpath.data));
3208 }
3209 }
3210 out:
3211 v9fs_path_free(&oldpath);
3212 v9fs_path_free(&newpath);
3213 return err;
3214 }
3215
3216 static int coroutine_fn v9fs_complete_renameat(V9fsPDU *pdu, int32_t olddirfid,
3217 V9fsString *old_name,
3218 int32_t newdirfid,
3219 V9fsString *new_name)
3220 {
3221 int err = 0;
3222 V9fsState *s = pdu->s;
3223 V9fsFidState *newdirfidp = NULL, *olddirfidp = NULL;
3224
3225 olddirfidp = get_fid(pdu, olddirfid);
3226 if (olddirfidp == NULL) {
3227 err = -ENOENT;
3228 goto out;
3229 }
3230 if (newdirfid != -1) {
3231 newdirfidp = get_fid(pdu, newdirfid);
3232 if (newdirfidp == NULL) {
3233 err = -ENOENT;
3234 goto out;
3235 }
3236 } else {
3237 newdirfidp = get_fid(pdu, olddirfid);
3238 }
3239
3240 err = v9fs_co_renameat(pdu, &olddirfidp->path, old_name,
3241 &newdirfidp->path, new_name);
3242 if (err < 0) {
3243 goto out;
3244 }
3245 if (s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT) {
3246 /* Only for path based fid we need to do the below fixup */
3247 err = v9fs_fix_fid_paths(pdu, &olddirfidp->path, old_name,
3248 &newdirfidp->path, new_name);
3249 }
3250 out:
3251 if (olddirfidp) {
3252 put_fid(pdu, olddirfidp);
3253 }
3254 if (newdirfidp) {
3255 put_fid(pdu, newdirfidp);
3256 }
3257 return err;
3258 }
3259
3260 static void coroutine_fn v9fs_renameat(void *opaque)
3261 {
3262 ssize_t err = 0;
3263 size_t offset = 7;
3264 V9fsPDU *pdu = opaque;
3265 V9fsState *s = pdu->s;
3266 int32_t olddirfid, newdirfid;
3267 V9fsString old_name, new_name;
3268
3269 v9fs_string_init(&old_name);
3270 v9fs_string_init(&new_name);
3271 err = pdu_unmarshal(pdu, offset, "dsds", &olddirfid,
3272 &old_name, &newdirfid, &new_name);
3273 if (err < 0) {
3274 goto out_err;
3275 }
3276
3277 if (name_is_illegal(old_name.data) || name_is_illegal(new_name.data)) {
3278 err = -ENOENT;
3279 goto out_err;
3280 }
3281
3282 if (!strcmp(".", old_name.data) || !strcmp("..", old_name.data) ||
3283 !strcmp(".", new_name.data) || !strcmp("..", new_name.data)) {
3284 err = -EISDIR;
3285 goto out_err;
3286 }
3287
3288 v9fs_path_write_lock(s);
3289 err = v9fs_complete_renameat(pdu, olddirfid,
3290 &old_name, newdirfid, &new_name);
3291 v9fs_path_unlock(s);
3292 if (!err) {
3293 err = offset;
3294 }
3295
3296 out_err:
3297 pdu_complete(pdu, err);
3298 v9fs_string_free(&old_name);
3299 v9fs_string_free(&new_name);
3300 }
3301
3302 static void coroutine_fn v9fs_wstat(void *opaque)
3303 {
3304 int32_t fid;
3305 int err = 0;
3306 int16_t unused;
3307 V9fsStat v9stat;
3308 size_t offset = 7;
3309 struct stat stbuf;
3310 V9fsFidState *fidp;
3311 V9fsPDU *pdu = opaque;
3312 V9fsState *s = pdu->s;
3313
3314 v9fs_stat_init(&v9stat);
3315 err = pdu_unmarshal(pdu, offset, "dwS", &fid, &unused, &v9stat);
3316 if (err < 0) {
3317 goto out_nofid;
3318 }
3319 trace_v9fs_wstat(pdu->tag, pdu->id, fid,
3320 v9stat.mode, v9stat.atime, v9stat.mtime);
3321
3322 fidp = get_fid(pdu, fid);
3323 if (fidp == NULL) {
3324 err = -EINVAL;
3325 goto out_nofid;
3326 }
3327 /* do we need to sync the file? */
3328 if (donttouch_stat(&v9stat)) {
3329 err = v9fs_co_fsync(pdu, fidp, 0);
3330 goto out;
3331 }
3332 if (v9stat.mode != -1) {
3333 uint32_t v9_mode;
3334 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
3335 if (err < 0) {
3336 goto out;
3337 }
3338 v9_mode = stat_to_v9mode(&stbuf);
3339 if ((v9stat.mode & P9_STAT_MODE_TYPE_BITS) !=
3340 (v9_mode & P9_STAT_MODE_TYPE_BITS)) {
3341 /* Attempting to change the type */
3342 err = -EIO;
3343 goto out;
3344 }
3345 err = v9fs_co_chmod(pdu, &fidp->path,
3346 v9mode_to_mode(v9stat.mode,
3347 &v9stat.extension));
3348 if (err < 0) {
3349 goto out;
3350 }
3351 }
3352 if (v9stat.mtime != -1 || v9stat.atime != -1) {
3353 struct timespec times[2];
3354 if (v9stat.atime != -1) {
3355 times[0].tv_sec = v9stat.atime;
3356 times[0].tv_nsec = 0;
3357 } else {
3358 times[0].tv_nsec = UTIME_OMIT;
3359 }
3360 if (v9stat.mtime != -1) {
3361 times[1].tv_sec = v9stat.mtime;
3362 times[1].tv_nsec = 0;
3363 } else {
3364 times[1].tv_nsec = UTIME_OMIT;
3365 }
3366 err = v9fs_co_utimensat(pdu, &fidp->path, times);
3367 if (err < 0) {
3368 goto out;
3369 }
3370 }
3371 if (v9stat.n_gid != -1 || v9stat.n_uid != -1) {
3372 err = v9fs_co_chown(pdu, &fidp->path, v9stat.n_uid, v9stat.n_gid);
3373 if (err < 0) {
3374 goto out;
3375 }
3376 }
3377 if (v9stat.name.size != 0) {
3378 v9fs_path_write_lock(s);
3379 err = v9fs_complete_rename(pdu, fidp, -1, &v9stat.name);
3380 v9fs_path_unlock(s);
3381 if (err < 0) {
3382 goto out;
3383 }
3384 }
3385 if (v9stat.length != -1) {
3386 err = v9fs_co_truncate(pdu, &fidp->path, v9stat.length);
3387 if (err < 0) {
3388 goto out;
3389 }
3390 }
3391 err = offset;
3392 out:
3393 put_fid(pdu, fidp);
3394 out_nofid:
3395 v9fs_stat_free(&v9stat);
3396 pdu_complete(pdu, err);
3397 }
3398
3399 static int v9fs_fill_statfs(V9fsState *s, V9fsPDU *pdu, struct statfs *stbuf)
3400 {
3401 uint32_t f_type;
3402 uint32_t f_bsize;
3403 uint64_t f_blocks;
3404 uint64_t f_bfree;
3405 uint64_t f_bavail;
3406 uint64_t f_files;
3407 uint64_t f_ffree;
3408 uint64_t fsid_val;
3409 uint32_t f_namelen;
3410 size_t offset = 7;
3411 int32_t bsize_factor;
3412
3413 /*
3414 * compute bsize factor based on host file system block size
3415 * and client msize
3416 */
3417 bsize_factor = (s->msize - P9_IOHDRSZ)/stbuf->f_bsize;
3418 if (!bsize_factor) {
3419 bsize_factor = 1;
3420 }
3421 f_type = stbuf->f_type;
3422 f_bsize = stbuf->f_bsize;
3423 f_bsize *= bsize_factor;
3424 /*
3425 * f_bsize is adjusted(multiplied) by bsize factor, so we need to
3426 * adjust(divide) the number of blocks, free blocks and available
3427 * blocks by bsize factor
3428 */
3429 f_blocks = stbuf->f_blocks/bsize_factor;
3430 f_bfree = stbuf->f_bfree/bsize_factor;
3431 f_bavail = stbuf->f_bavail/bsize_factor;
3432 f_files = stbuf->f_files;
3433 f_ffree = stbuf->f_ffree;
3434 fsid_val = (unsigned int) stbuf->f_fsid.__val[0] |
3435 (unsigned long long)stbuf->f_fsid.__val[1] << 32;
3436 f_namelen = stbuf->f_namelen;
3437
3438 return pdu_marshal(pdu, offset, "ddqqqqqqd",
3439 f_type, f_bsize, f_blocks, f_bfree,
3440 f_bavail, f_files, f_ffree,
3441 fsid_val, f_namelen);
3442 }
3443
3444 static void coroutine_fn v9fs_statfs(void *opaque)
3445 {
3446 int32_t fid;
3447 ssize_t retval = 0;
3448 size_t offset = 7;
3449 V9fsFidState *fidp;
3450 struct statfs stbuf;
3451 V9fsPDU *pdu = opaque;
3452 V9fsState *s = pdu->s;
3453
3454 retval = pdu_unmarshal(pdu, offset, "d", &fid);
3455 if (retval < 0) {
3456 goto out_nofid;
3457 }
3458 fidp = get_fid(pdu, fid);
3459 if (fidp == NULL) {
3460 retval = -ENOENT;
3461 goto out_nofid;
3462 }
3463 retval = v9fs_co_statfs(pdu, &fidp->path, &stbuf);
3464 if (retval < 0) {
3465 goto out;
3466 }
3467 retval = v9fs_fill_statfs(s, pdu, &stbuf);
3468 if (retval < 0) {
3469 goto out;
3470 }
3471 retval += offset;
3472 out:
3473 put_fid(pdu, fidp);
3474 out_nofid:
3475 pdu_complete(pdu, retval);
3476 }
3477
3478 static void coroutine_fn v9fs_mknod(void *opaque)
3479 {
3480
3481 int mode;
3482 gid_t gid;
3483 int32_t fid;
3484 V9fsQID qid;
3485 int err = 0;
3486 int major, minor;
3487 size_t offset = 7;
3488 V9fsString name;
3489 struct stat stbuf;
3490 V9fsFidState *fidp;
3491 V9fsPDU *pdu = opaque;
3492
3493 v9fs_string_init(&name);
3494 err = pdu_unmarshal(pdu, offset, "dsdddd", &fid, &name, &mode,
3495 &major, &minor, &gid);
3496 if (err < 0) {
3497 goto out_nofid;
3498 }
3499 trace_v9fs_mknod(pdu->tag, pdu->id, fid, mode, major, minor);
3500
3501 if (name_is_illegal(name.data)) {
3502 err = -ENOENT;
3503 goto out_nofid;
3504 }
3505
3506 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3507 err = -EEXIST;
3508 goto out_nofid;
3509 }
3510
3511 fidp = get_fid(pdu, fid);
3512 if (fidp == NULL) {
3513 err = -ENOENT;
3514 goto out_nofid;
3515 }
3516 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, gid,
3517 makedev(major, minor), mode, &stbuf);
3518 if (err < 0) {
3519 goto out;
3520 }
3521 err = stat_to_qid(pdu, &stbuf, &qid);
3522 if (err < 0) {
3523 goto out;
3524 }
3525 err = pdu_marshal(pdu, offset, "Q", &qid);
3526 if (err < 0) {
3527 goto out;
3528 }
3529 err += offset;
3530 trace_v9fs_mknod_return(pdu->tag, pdu->id,
3531 qid.type, qid.version, qid.path);
3532 out:
3533 put_fid(pdu, fidp);
3534 out_nofid:
3535 pdu_complete(pdu, err);
3536 v9fs_string_free(&name);
3537 }
3538
3539 /*
3540 * Implement posix byte range locking code
3541 * Server side handling of locking code is very simple, because 9p server in
3542 * QEMU can handle only one client. And most of the lock handling
3543 * (like conflict, merging) etc is done by the VFS layer itself, so no need to
3544 * do any thing in * qemu 9p server side lock code path.
3545 * So when a TLOCK request comes, always return success
3546 */
3547 static void coroutine_fn v9fs_lock(void *opaque)
3548 {
3549 V9fsFlock flock;
3550 size_t offset = 7;
3551 struct stat stbuf;
3552 V9fsFidState *fidp;
3553 int32_t fid, err = 0;
3554 V9fsPDU *pdu = opaque;
3555
3556 v9fs_string_init(&flock.client_id);
3557 err = pdu_unmarshal(pdu, offset, "dbdqqds", &fid, &flock.type,
3558 &flock.flags, &flock.start, &flock.length,
3559 &flock.proc_id, &flock.client_id);
3560 if (err < 0) {
3561 goto out_nofid;
3562 }
3563 trace_v9fs_lock(pdu->tag, pdu->id, fid,
3564 flock.type, flock.start, flock.length);
3565
3566
3567 /* We support only block flag now (that too ignored currently) */
3568 if (flock.flags & ~P9_LOCK_FLAGS_BLOCK) {
3569 err = -EINVAL;
3570 goto out_nofid;
3571 }
3572 fidp = get_fid(pdu, fid);
3573 if (fidp == NULL) {
3574 err = -ENOENT;
3575 goto out_nofid;
3576 }
3577 err = v9fs_co_fstat(pdu, fidp, &stbuf);
3578 if (err < 0) {
3579 goto out;
3580 }
3581 err = pdu_marshal(pdu, offset, "b", P9_LOCK_SUCCESS);
3582 if (err < 0) {
3583 goto out;
3584 }
3585 err += offset;
3586 trace_v9fs_lock_return(pdu->tag, pdu->id, P9_LOCK_SUCCESS);
3587 out:
3588 put_fid(pdu, fidp);
3589 out_nofid:
3590 pdu_complete(pdu, err);
3591 v9fs_string_free(&flock.client_id);
3592 }
3593
3594 /*
3595 * When a TGETLOCK request comes, always return success because all lock
3596 * handling is done by client's VFS layer.
3597 */
3598 static void coroutine_fn v9fs_getlock(void *opaque)
3599 {
3600 size_t offset = 7;
3601 struct stat stbuf;
3602 V9fsFidState *fidp;
3603 V9fsGetlock glock;
3604 int32_t fid, err = 0;
3605 V9fsPDU *pdu = opaque;
3606
3607 v9fs_string_init(&glock.client_id);
3608 err = pdu_unmarshal(pdu, offset, "dbqqds", &fid, &glock.type,
3609 &glock.start, &glock.length, &glock.proc_id,
3610 &glock.client_id);
3611 if (err < 0) {
3612 goto out_nofid;
3613 }
3614 trace_v9fs_getlock(pdu->tag, pdu->id, fid,
3615 glock.type, glock.start, glock.length);
3616
3617 fidp = get_fid(pdu, fid);
3618 if (fidp == NULL) {
3619 err = -ENOENT;
3620 goto out_nofid;
3621 }
3622 err = v9fs_co_fstat(pdu, fidp, &stbuf);
3623 if (err < 0) {
3624 goto out;
3625 }
3626 glock.type = P9_LOCK_TYPE_UNLCK;
3627 err = pdu_marshal(pdu, offset, "bqqds", glock.type,
3628 glock.start, glock.length, glock.proc_id,
3629 &glock.client_id);
3630 if (err < 0) {
3631 goto out;
3632 }
3633 err += offset;
3634 trace_v9fs_getlock_return(pdu->tag, pdu->id, glock.type, glock.start,
3635 glock.length, glock.proc_id);
3636 out:
3637 put_fid(pdu, fidp);
3638 out_nofid:
3639 pdu_complete(pdu, err);
3640 v9fs_string_free(&glock.client_id);
3641 }
3642
3643 static void coroutine_fn v9fs_mkdir(void *opaque)
3644 {
3645 V9fsPDU *pdu = opaque;
3646 size_t offset = 7;
3647 int32_t fid;
3648 struct stat stbuf;
3649 V9fsQID qid;
3650 V9fsString name;
3651 V9fsFidState *fidp;
3652 gid_t gid;
3653 int mode;
3654 int err = 0;
3655
3656 v9fs_string_init(&name);
3657 err = pdu_unmarshal(pdu, offset, "dsdd", &fid, &name, &mode, &gid);
3658 if (err < 0) {
3659 goto out_nofid;
3660 }
3661 trace_v9fs_mkdir(pdu->tag, pdu->id, fid, name.data, mode, gid);
3662
3663 if (name_is_illegal(name.data)) {
3664 err = -ENOENT;
3665 goto out_nofid;
3666 }
3667
3668 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3669 err = -EEXIST;
3670 goto out_nofid;
3671 }
3672
3673 fidp = get_fid(pdu, fid);
3674 if (fidp == NULL) {
3675 err = -ENOENT;
3676 goto out_nofid;
3677 }
3678 err = v9fs_co_mkdir(pdu, fidp, &name, mode, fidp->uid, gid, &stbuf);
3679 if (err < 0) {
3680 goto out;
3681 }
3682 err = stat_to_qid(pdu, &stbuf, &qid);
3683 if (err < 0) {
3684 goto out;
3685 }
3686 err = pdu_marshal(pdu, offset, "Q", &qid);
3687 if (err < 0) {
3688 goto out;
3689 }
3690 err += offset;
3691 trace_v9fs_mkdir_return(pdu->tag, pdu->id,
3692 qid.type, qid.version, qid.path, err);
3693 out:
3694 put_fid(pdu, fidp);
3695 out_nofid:
3696 pdu_complete(pdu, err);
3697 v9fs_string_free(&name);
3698 }
3699
3700 static void coroutine_fn v9fs_xattrwalk(void *opaque)
3701 {
3702 int64_t size;
3703 V9fsString name;
3704 ssize_t err = 0;
3705 size_t offset = 7;
3706 int32_t fid, newfid;
3707 V9fsFidState *file_fidp;
3708 V9fsFidState *xattr_fidp = NULL;
3709 V9fsPDU *pdu = opaque;
3710 V9fsState *s = pdu->s;
3711
3712 v9fs_string_init(&name);
3713 err = pdu_unmarshal(pdu, offset, "dds", &fid, &newfid, &name);
3714 if (err < 0) {
3715 goto out_nofid;
3716 }
3717 trace_v9fs_xattrwalk(pdu->tag, pdu->id, fid, newfid, name.data);
3718
3719 file_fidp = get_fid(pdu, fid);
3720 if (file_fidp == NULL) {
3721 err = -ENOENT;
3722 goto out_nofid;
3723 }
3724 xattr_fidp = alloc_fid(s, newfid);
3725 if (xattr_fidp == NULL) {
3726 err = -EINVAL;
3727 goto out;
3728 }
3729 v9fs_path_copy(&xattr_fidp->path, &file_fidp->path);
3730 if (!v9fs_string_size(&name)) {
3731 /*
3732 * listxattr request. Get the size first
3733 */
3734 size = v9fs_co_llistxattr(pdu, &xattr_fidp->path, NULL, 0);
3735 if (size < 0) {
3736 err = size;
3737 clunk_fid(s, xattr_fidp->fid);
3738 goto out;
3739 }
3740 /*
3741 * Read the xattr value
3742 */
3743 xattr_fidp->fs.xattr.len = size;
3744 xattr_fidp->fid_type = P9_FID_XATTR;
3745 xattr_fidp->fs.xattr.xattrwalk_fid = true;
3746 xattr_fidp->fs.xattr.value = g_malloc0(size);
3747 if (size) {
3748 err = v9fs_co_llistxattr(pdu, &xattr_fidp->path,
3749 xattr_fidp->fs.xattr.value,
3750 xattr_fidp->fs.xattr.len);
3751 if (err < 0) {
3752 clunk_fid(s, xattr_fidp->fid);
3753 goto out;
3754 }
3755 }
3756 err = pdu_marshal(pdu, offset, "q", size);
3757 if (err < 0) {
3758 goto out;
3759 }
3760 err += offset;
3761 } else {
3762 /*
3763 * specific xattr fid. We check for xattr
3764 * presence also collect the xattr size
3765 */
3766 size = v9fs_co_lgetxattr(pdu, &xattr_fidp->path,
3767 &name, NULL, 0);
3768 if (size < 0) {
3769 err = size;
3770 clunk_fid(s, xattr_fidp->fid);
3771 goto out;
3772 }
3773 /*
3774 * Read the xattr value
3775 */
3776 xattr_fidp->fs.xattr.len = size;
3777 xattr_fidp->fid_type = P9_FID_XATTR;
3778 xattr_fidp->fs.xattr.xattrwalk_fid = true;
3779 xattr_fidp->fs.xattr.value = g_malloc0(size);
3780 if (size) {
3781 err = v9fs_co_lgetxattr(pdu, &xattr_fidp->path,
3782 &name, xattr_fidp->fs.xattr.value,
3783 xattr_fidp->fs.xattr.len);
3784 if (err < 0) {
3785 clunk_fid(s, xattr_fidp->fid);
3786 goto out;
3787 }
3788 }
3789 err = pdu_marshal(pdu, offset, "q", size);
3790 if (err < 0) {
3791 goto out;
3792 }
3793 err += offset;
3794 }
3795 trace_v9fs_xattrwalk_return(pdu->tag, pdu->id, size);
3796 out:
3797 put_fid(pdu, file_fidp);
3798 if (xattr_fidp) {
3799 put_fid(pdu, xattr_fidp);
3800 }
3801 out_nofid:
3802 pdu_complete(pdu, err);
3803 v9fs_string_free(&name);
3804 }
3805
3806 static void coroutine_fn v9fs_xattrcreate(void *opaque)
3807 {
3808 int flags, rflags = 0;
3809 int32_t fid;
3810 uint64_t size;
3811 ssize_t err = 0;
3812 V9fsString name;
3813 size_t offset = 7;
3814 V9fsFidState *file_fidp;
3815 V9fsFidState *xattr_fidp;
3816 V9fsPDU *pdu = opaque;
3817
3818 v9fs_string_init(&name);
3819 err = pdu_unmarshal(pdu, offset, "dsqd", &fid, &name, &size, &flags);
3820 if (err < 0) {
3821 goto out_nofid;
3822 }
3823 trace_v9fs_xattrcreate(pdu->tag, pdu->id, fid, name.data, size, flags);
3824
3825 if (flags & ~(P9_XATTR_CREATE | P9_XATTR_REPLACE)) {
3826 err = -EINVAL;
3827 goto out_nofid;
3828 }
3829
3830 if (flags & P9_XATTR_CREATE) {
3831 rflags |= XATTR_CREATE;
3832 }
3833
3834 if (flags & P9_XATTR_REPLACE) {
3835 rflags |= XATTR_REPLACE;
3836 }
3837
3838 if (size > XATTR_SIZE_MAX) {
3839 err = -E2BIG;
3840 goto out_nofid;
3841 }
3842
3843 file_fidp = get_fid(pdu, fid);
3844 if (file_fidp == NULL) {
3845 err = -EINVAL;
3846 goto out_nofid;
3847 }
3848 if (file_fidp->fid_type != P9_FID_NONE) {
3849 err = -EINVAL;
3850 goto out_put_fid;
3851 }
3852
3853 /* Make the file fid point to xattr */
3854 xattr_fidp = file_fidp;
3855 xattr_fidp->fid_type = P9_FID_XATTR;
3856 xattr_fidp->fs.xattr.copied_len = 0;
3857 xattr_fidp->fs.xattr.xattrwalk_fid = false;
3858 xattr_fidp->fs.xattr.len = size;
3859 xattr_fidp->fs.xattr.flags = rflags;
3860 v9fs_string_init(&xattr_fidp->fs.xattr.name);
3861 v9fs_string_copy(&xattr_fidp->fs.xattr.name, &name);
3862 xattr_fidp->fs.xattr.value = g_malloc0(size);
3863 err = offset;
3864 out_put_fid:
3865 put_fid(pdu, file_fidp);
3866 out_nofid:
3867 pdu_complete(pdu, err);
3868 v9fs_string_free(&name);
3869 }
3870
3871 static void coroutine_fn v9fs_readlink(void *opaque)
3872 {
3873 V9fsPDU *pdu = opaque;
3874 size_t offset = 7;
3875 V9fsString target;
3876 int32_t fid;
3877 int err = 0;
3878 V9fsFidState *fidp;
3879
3880 err = pdu_unmarshal(pdu, offset, "d", &fid);
3881 if (err < 0) {
3882 goto out_nofid;
3883 }
3884 trace_v9fs_readlink(pdu->tag, pdu->id, fid);
3885 fidp = get_fid(pdu, fid);
3886 if (fidp == NULL) {
3887 err = -ENOENT;
3888 goto out_nofid;
3889 }
3890
3891 v9fs_string_init(&target);
3892 err = v9fs_co_readlink(pdu, &fidp->path, &target);
3893 if (err < 0) {
3894 goto out;
3895 }
3896 err = pdu_marshal(pdu, offset, "s", &target);
3897 if (err < 0) {
3898 v9fs_string_free(&target);
3899 goto out;
3900 }
3901 err += offset;
3902 trace_v9fs_readlink_return(pdu->tag, pdu->id, target.data);
3903 v9fs_string_free(&target);
3904 out:
3905 put_fid(pdu, fidp);
3906 out_nofid:
3907 pdu_complete(pdu, err);
3908 }
3909
3910 static CoroutineEntry *pdu_co_handlers[] = {
3911 [P9_TREADDIR] = v9fs_readdir,
3912 [P9_TSTATFS] = v9fs_statfs,
3913 [P9_TGETATTR] = v9fs_getattr,
3914 [P9_TSETATTR] = v9fs_setattr,
3915 [P9_TXATTRWALK] = v9fs_xattrwalk,
3916 [P9_TXATTRCREATE] = v9fs_xattrcreate,
3917 [P9_TMKNOD] = v9fs_mknod,
3918 [P9_TRENAME] = v9fs_rename,
3919 [P9_TLOCK] = v9fs_lock,
3920 [P9_TGETLOCK] = v9fs_getlock,
3921 [P9_TRENAMEAT] = v9fs_renameat,
3922 [P9_TREADLINK] = v9fs_readlink,
3923 [P9_TUNLINKAT] = v9fs_unlinkat,
3924 [P9_TMKDIR] = v9fs_mkdir,
3925 [P9_TVERSION] = v9fs_version,
3926 [P9_TLOPEN] = v9fs_open,
3927 [P9_TATTACH] = v9fs_attach,
3928 [P9_TSTAT] = v9fs_stat,
3929 [P9_TWALK] = v9fs_walk,
3930 [P9_TCLUNK] = v9fs_clunk,
3931 [P9_TFSYNC] = v9fs_fsync,
3932 [P9_TOPEN] = v9fs_open,
3933 [P9_TREAD] = v9fs_read,
3934 #if 0
3935 [P9_TAUTH] = v9fs_auth,
3936 #endif
3937 [P9_TFLUSH] = v9fs_flush,
3938 [P9_TLINK] = v9fs_link,
3939 [P9_TSYMLINK] = v9fs_symlink,
3940 [P9_TCREATE] = v9fs_create,
3941 [P9_TLCREATE] = v9fs_lcreate,
3942 [P9_TWRITE] = v9fs_write,
3943 [P9_TWSTAT] = v9fs_wstat,
3944 [P9_TREMOVE] = v9fs_remove,
3945 };
3946
3947 static void coroutine_fn v9fs_op_not_supp(void *opaque)
3948 {
3949 V9fsPDU *pdu = opaque;
3950 pdu_complete(pdu, -EOPNOTSUPP);
3951 }
3952
3953 static void coroutine_fn v9fs_fs_ro(void *opaque)
3954 {
3955 V9fsPDU *pdu = opaque;
3956 pdu_complete(pdu, -EROFS);
3957 }
3958
3959 static inline bool is_read_only_op(V9fsPDU *pdu)
3960 {
3961 switch (pdu->id) {
3962 case P9_TREADDIR:
3963 case P9_TSTATFS:
3964 case P9_TGETATTR:
3965 case P9_TXATTRWALK:
3966 case P9_TLOCK:
3967 case P9_TGETLOCK:
3968 case P9_TREADLINK:
3969 case P9_TVERSION:
3970 case P9_TLOPEN:
3971 case P9_TATTACH:
3972 case P9_TSTAT:
3973 case P9_TWALK:
3974 case P9_TCLUNK:
3975 case P9_TFSYNC:
3976 case P9_TOPEN:
3977 case P9_TREAD:
3978 case P9_TAUTH:
3979 case P9_TFLUSH:
3980 return 1;
3981 default:
3982 return 0;
3983 }
3984 }
3985
3986 void pdu_submit(V9fsPDU *pdu, P9MsgHeader *hdr)
3987 {
3988 Coroutine *co;
3989 CoroutineEntry *handler;
3990 V9fsState *s = pdu->s;
3991
3992 pdu->size = le32_to_cpu(hdr->size_le);
3993 pdu->id = hdr->id;
3994 pdu->tag = le16_to_cpu(hdr->tag_le);
3995
3996 if (pdu->id >= ARRAY_SIZE(pdu_co_handlers) ||
3997 (pdu_co_handlers[pdu->id] == NULL)) {
3998 handler = v9fs_op_not_supp;
3999 } else if (is_ro_export(&s->ctx) && !is_read_only_op(pdu)) {
4000 handler = v9fs_fs_ro;
4001 } else {
4002 handler = pdu_co_handlers[pdu->id];
4003 }
4004
4005 qemu_co_queue_init(&pdu->complete);
4006 co = qemu_coroutine_create(handler, pdu);
4007 qemu_coroutine_enter(co);
4008 }
4009
4010 /* Returns 0 on success, 1 on failure. */
4011 int v9fs_device_realize_common(V9fsState *s, const V9fsTransport *t,
4012 Error **errp)
4013 {
4014 ERRP_GUARD();
4015 int i, len;
4016 struct stat stat;
4017 FsDriverEntry *fse;
4018 V9fsPath path;
4019 int rc = 1;
4020
4021 assert(!s->transport);
4022 s->transport = t;
4023
4024 /* initialize pdu allocator */
4025 QLIST_INIT(&s->free_list);
4026 QLIST_INIT(&s->active_list);
4027 for (i = 0; i < MAX_REQ; i++) {
4028 QLIST_INSERT_HEAD(&s->free_list, &s->pdus[i], next);
4029 s->pdus[i].s = s;
4030 s->pdus[i].idx = i;
4031 }
4032
4033 v9fs_path_init(&path);
4034
4035 fse = get_fsdev_fsentry(s->fsconf.fsdev_id);
4036
4037 if (!fse) {
4038 /* We don't have a fsdev identified by fsdev_id */
4039 error_setg(errp, "9pfs device couldn't find fsdev with the "
4040 "id = %s",
4041 s->fsconf.fsdev_id ? s->fsconf.fsdev_id : "NULL");
4042 goto out;
4043 }
4044
4045 if (!s->fsconf.tag) {
4046 /* we haven't specified a mount_tag */
4047 error_setg(errp, "fsdev with id %s needs mount_tag arguments",
4048 s->fsconf.fsdev_id);
4049 goto out;
4050 }
4051
4052 s->ctx.export_flags = fse->export_flags;
4053 s->ctx.fs_root = g_strdup(fse->path);
4054 s->ctx.exops.get_st_gen = NULL;
4055 len = strlen(s->fsconf.tag);
4056 if (len > MAX_TAG_LEN - 1) {
4057 error_setg(errp, "mount tag '%s' (%d bytes) is longer than "
4058 "maximum (%d bytes)", s->fsconf.tag, len, MAX_TAG_LEN - 1);
4059 goto out;
4060 }
4061
4062 s->tag = g_strdup(s->fsconf.tag);
4063 s->ctx.uid = -1;
4064
4065 s->ops = fse->ops;
4066
4067 s->ctx.fmode = fse->fmode;
4068 s->ctx.dmode = fse->dmode;
4069
4070 s->fid_list = NULL;
4071 qemu_co_rwlock_init(&s->rename_lock);
4072
4073 if (s->ops->init(&s->ctx, errp) < 0) {
4074 error_prepend(errp, "cannot initialize fsdev '%s': ",
4075 s->fsconf.fsdev_id);
4076 goto out;
4077 }
4078
4079 /*
4080 * Check details of export path, We need to use fs driver
4081 * call back to do that. Since we are in the init path, we don't
4082 * use co-routines here.
4083 */
4084 if (s->ops->name_to_path(&s->ctx, NULL, "/", &path) < 0) {
4085 error_setg(errp,
4086 "error in converting name to path %s", strerror(errno));
4087 goto out;
4088 }
4089 if (s->ops->lstat(&s->ctx, &path, &stat)) {
4090 error_setg(errp, "share path %s does not exist", fse->path);
4091 goto out;
4092 } else if (!S_ISDIR(stat.st_mode)) {
4093 error_setg(errp, "share path %s is not a directory", fse->path);
4094 goto out;
4095 }
4096
4097 s->dev_id = stat.st_dev;
4098
4099 /* init inode remapping : */
4100 /* hash table for variable length inode suffixes */
4101 qpd_table_init(&s->qpd_table);
4102 /* hash table for slow/full inode remapping (most users won't need it) */
4103 qpf_table_init(&s->qpf_table);
4104 /* hash table for quick inode remapping */
4105 qpp_table_init(&s->qpp_table);
4106 s->qp_ndevices = 0;
4107 s->qp_affix_next = 1; /* reserve 0 to detect overflow */
4108 s->qp_fullpath_next = 1;
4109
4110 s->ctx.fst = &fse->fst;
4111 fsdev_throttle_init(s->ctx.fst);
4112
4113 rc = 0;
4114 out:
4115 if (rc) {
4116 v9fs_device_unrealize_common(s);
4117 }
4118 v9fs_path_free(&path);
4119 return rc;
4120 }
4121
4122 void v9fs_device_unrealize_common(V9fsState *s)
4123 {
4124 if (s->ops && s->ops->cleanup) {
4125 s->ops->cleanup(&s->ctx);
4126 }
4127 if (s->ctx.fst) {
4128 fsdev_throttle_cleanup(s->ctx.fst);
4129 }
4130 g_free(s->tag);
4131 qp_table_destroy(&s->qpd_table);
4132 qp_table_destroy(&s->qpp_table);
4133 qp_table_destroy(&s->qpf_table);
4134 g_free(s->ctx.fs_root);
4135 }
4136
4137 typedef struct VirtfsCoResetData {
4138 V9fsPDU pdu;
4139 bool done;
4140 } VirtfsCoResetData;
4141
4142 static void coroutine_fn virtfs_co_reset(void *opaque)
4143 {
4144 VirtfsCoResetData *data = opaque;
4145
4146 virtfs_reset(&data->pdu);
4147 data->done = true;
4148 }
4149
4150 void v9fs_reset(V9fsState *s)
4151 {
4152 VirtfsCoResetData data = { .pdu = { .s = s }, .done = false };
4153 Coroutine *co;
4154
4155 while (!QLIST_EMPTY(&s->active_list)) {
4156 aio_poll(qemu_get_aio_context(), true);
4157 }
4158
4159 co = qemu_coroutine_create(virtfs_co_reset, &data);
4160 qemu_coroutine_enter(co);
4161
4162 while (!data.done) {
4163 aio_poll(qemu_get_aio_context(), true);
4164 }
4165 }
4166
4167 static void __attribute__((__constructor__)) v9fs_set_fd_limit(void)
4168 {
4169 struct rlimit rlim;
4170 if (getrlimit(RLIMIT_NOFILE, &rlim) < 0) {
4171 error_report("Failed to get the resource limit");
4172 exit(1);
4173 }
4174 open_fd_hw = rlim.rlim_cur - MIN(400, rlim.rlim_cur/3);
4175 open_fd_rc = rlim.rlim_cur/2;
4176 }