tests/tcg: Do not require FE_* exception bits
[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(s->proto_version, &f->fs.dir);
318 v9fs_readdir_init(s->proto_version, &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 V9fsPDU *pdu_alloc(V9fsState *s)
976 {
977 V9fsPDU *pdu = NULL;
978
979 if (!QLIST_EMPTY(&s->free_list)) {
980 pdu = QLIST_FIRST(&s->free_list);
981 QLIST_REMOVE(pdu, next);
982 QLIST_INSERT_HEAD(&s->active_list, pdu, next);
983 }
984 return pdu;
985 }
986
987 void pdu_free(V9fsPDU *pdu)
988 {
989 V9fsState *s = pdu->s;
990
991 g_assert(!pdu->cancelled);
992 QLIST_REMOVE(pdu, next);
993 QLIST_INSERT_HEAD(&s->free_list, pdu, next);
994 }
995
996 static void coroutine_fn pdu_complete(V9fsPDU *pdu, ssize_t len)
997 {
998 int8_t id = pdu->id + 1; /* Response */
999 V9fsState *s = pdu->s;
1000 int ret;
1001
1002 /*
1003 * The 9p spec requires that successfully cancelled pdus receive no reply.
1004 * Sending a reply would confuse clients because they would
1005 * assume that any EINTR is the actual result of the operation,
1006 * rather than a consequence of the cancellation. However, if
1007 * the operation completed (succesfully or with an error other
1008 * than caused be cancellation), we do send out that reply, both
1009 * for efficiency and to avoid confusing the rest of the state machine
1010 * that assumes passing a non-error here will mean a successful
1011 * transmission of the reply.
1012 */
1013 bool discard = pdu->cancelled && len == -EINTR;
1014 if (discard) {
1015 trace_v9fs_rcancel(pdu->tag, pdu->id);
1016 pdu->size = 0;
1017 goto out_notify;
1018 }
1019
1020 if (len < 0) {
1021 int err = -len;
1022 len = 7;
1023
1024 if (s->proto_version != V9FS_PROTO_2000L) {
1025 V9fsString str;
1026
1027 str.data = strerror(err);
1028 str.size = strlen(str.data);
1029
1030 ret = pdu_marshal(pdu, len, "s", &str);
1031 if (ret < 0) {
1032 goto out_notify;
1033 }
1034 len += ret;
1035 id = P9_RERROR;
1036 }
1037
1038 ret = pdu_marshal(pdu, len, "d", err);
1039 if (ret < 0) {
1040 goto out_notify;
1041 }
1042 len += ret;
1043
1044 if (s->proto_version == V9FS_PROTO_2000L) {
1045 id = P9_RLERROR;
1046 }
1047 trace_v9fs_rerror(pdu->tag, pdu->id, err); /* Trace ERROR */
1048 }
1049
1050 /* fill out the header */
1051 if (pdu_marshal(pdu, 0, "dbw", (int32_t)len, id, pdu->tag) < 0) {
1052 goto out_notify;
1053 }
1054
1055 /* keep these in sync */
1056 pdu->size = len;
1057 pdu->id = id;
1058
1059 out_notify:
1060 pdu->s->transport->push_and_notify(pdu);
1061
1062 /* Now wakeup anybody waiting in flush for this request */
1063 if (!qemu_co_queue_next(&pdu->complete)) {
1064 pdu_free(pdu);
1065 }
1066 }
1067
1068 static mode_t v9mode_to_mode(uint32_t mode, V9fsString *extension)
1069 {
1070 mode_t ret;
1071
1072 ret = mode & 0777;
1073 if (mode & P9_STAT_MODE_DIR) {
1074 ret |= S_IFDIR;
1075 }
1076
1077 if (mode & P9_STAT_MODE_SYMLINK) {
1078 ret |= S_IFLNK;
1079 }
1080 if (mode & P9_STAT_MODE_SOCKET) {
1081 ret |= S_IFSOCK;
1082 }
1083 if (mode & P9_STAT_MODE_NAMED_PIPE) {
1084 ret |= S_IFIFO;
1085 }
1086 if (mode & P9_STAT_MODE_DEVICE) {
1087 if (extension->size && extension->data[0] == 'c') {
1088 ret |= S_IFCHR;
1089 } else {
1090 ret |= S_IFBLK;
1091 }
1092 }
1093
1094 if (!(ret&~0777)) {
1095 ret |= S_IFREG;
1096 }
1097
1098 if (mode & P9_STAT_MODE_SETUID) {
1099 ret |= S_ISUID;
1100 }
1101 if (mode & P9_STAT_MODE_SETGID) {
1102 ret |= S_ISGID;
1103 }
1104 if (mode & P9_STAT_MODE_SETVTX) {
1105 ret |= S_ISVTX;
1106 }
1107
1108 return ret;
1109 }
1110
1111 static int donttouch_stat(V9fsStat *stat)
1112 {
1113 if (stat->type == -1 &&
1114 stat->dev == -1 &&
1115 stat->qid.type == 0xff &&
1116 stat->qid.version == (uint32_t) -1 &&
1117 stat->qid.path == (uint64_t) -1 &&
1118 stat->mode == -1 &&
1119 stat->atime == -1 &&
1120 stat->mtime == -1 &&
1121 stat->length == -1 &&
1122 !stat->name.size &&
1123 !stat->uid.size &&
1124 !stat->gid.size &&
1125 !stat->muid.size &&
1126 stat->n_uid == -1 &&
1127 stat->n_gid == -1 &&
1128 stat->n_muid == -1) {
1129 return 1;
1130 }
1131
1132 return 0;
1133 }
1134
1135 static void v9fs_stat_init(V9fsStat *stat)
1136 {
1137 v9fs_string_init(&stat->name);
1138 v9fs_string_init(&stat->uid);
1139 v9fs_string_init(&stat->gid);
1140 v9fs_string_init(&stat->muid);
1141 v9fs_string_init(&stat->extension);
1142 }
1143
1144 static void v9fs_stat_free(V9fsStat *stat)
1145 {
1146 v9fs_string_free(&stat->name);
1147 v9fs_string_free(&stat->uid);
1148 v9fs_string_free(&stat->gid);
1149 v9fs_string_free(&stat->muid);
1150 v9fs_string_free(&stat->extension);
1151 }
1152
1153 static uint32_t stat_to_v9mode(const struct stat *stbuf)
1154 {
1155 uint32_t mode;
1156
1157 mode = stbuf->st_mode & 0777;
1158 if (S_ISDIR(stbuf->st_mode)) {
1159 mode |= P9_STAT_MODE_DIR;
1160 }
1161
1162 if (S_ISLNK(stbuf->st_mode)) {
1163 mode |= P9_STAT_MODE_SYMLINK;
1164 }
1165
1166 if (S_ISSOCK(stbuf->st_mode)) {
1167 mode |= P9_STAT_MODE_SOCKET;
1168 }
1169
1170 if (S_ISFIFO(stbuf->st_mode)) {
1171 mode |= P9_STAT_MODE_NAMED_PIPE;
1172 }
1173
1174 if (S_ISBLK(stbuf->st_mode) || S_ISCHR(stbuf->st_mode)) {
1175 mode |= P9_STAT_MODE_DEVICE;
1176 }
1177
1178 if (stbuf->st_mode & S_ISUID) {
1179 mode |= P9_STAT_MODE_SETUID;
1180 }
1181
1182 if (stbuf->st_mode & S_ISGID) {
1183 mode |= P9_STAT_MODE_SETGID;
1184 }
1185
1186 if (stbuf->st_mode & S_ISVTX) {
1187 mode |= P9_STAT_MODE_SETVTX;
1188 }
1189
1190 return mode;
1191 }
1192
1193 static int coroutine_fn stat_to_v9stat(V9fsPDU *pdu, V9fsPath *path,
1194 const char *basename,
1195 const struct stat *stbuf,
1196 V9fsStat *v9stat)
1197 {
1198 int err;
1199
1200 memset(v9stat, 0, sizeof(*v9stat));
1201
1202 err = stat_to_qid(pdu, stbuf, &v9stat->qid);
1203 if (err < 0) {
1204 return err;
1205 }
1206 v9stat->mode = stat_to_v9mode(stbuf);
1207 v9stat->atime = stbuf->st_atime;
1208 v9stat->mtime = stbuf->st_mtime;
1209 v9stat->length = stbuf->st_size;
1210
1211 v9fs_string_free(&v9stat->uid);
1212 v9fs_string_free(&v9stat->gid);
1213 v9fs_string_free(&v9stat->muid);
1214
1215 v9stat->n_uid = stbuf->st_uid;
1216 v9stat->n_gid = stbuf->st_gid;
1217 v9stat->n_muid = 0;
1218
1219 v9fs_string_free(&v9stat->extension);
1220
1221 if (v9stat->mode & P9_STAT_MODE_SYMLINK) {
1222 err = v9fs_co_readlink(pdu, path, &v9stat->extension);
1223 if (err < 0) {
1224 return err;
1225 }
1226 } else if (v9stat->mode & P9_STAT_MODE_DEVICE) {
1227 v9fs_string_sprintf(&v9stat->extension, "%c %u %u",
1228 S_ISCHR(stbuf->st_mode) ? 'c' : 'b',
1229 major(stbuf->st_rdev), minor(stbuf->st_rdev));
1230 } else if (S_ISDIR(stbuf->st_mode) || S_ISREG(stbuf->st_mode)) {
1231 v9fs_string_sprintf(&v9stat->extension, "%s %lu",
1232 "HARDLINKCOUNT", (unsigned long)stbuf->st_nlink);
1233 }
1234
1235 v9fs_string_sprintf(&v9stat->name, "%s", basename);
1236
1237 v9stat->size = 61 +
1238 v9fs_string_size(&v9stat->name) +
1239 v9fs_string_size(&v9stat->uid) +
1240 v9fs_string_size(&v9stat->gid) +
1241 v9fs_string_size(&v9stat->muid) +
1242 v9fs_string_size(&v9stat->extension);
1243 return 0;
1244 }
1245
1246 #define P9_STATS_MODE 0x00000001ULL
1247 #define P9_STATS_NLINK 0x00000002ULL
1248 #define P9_STATS_UID 0x00000004ULL
1249 #define P9_STATS_GID 0x00000008ULL
1250 #define P9_STATS_RDEV 0x00000010ULL
1251 #define P9_STATS_ATIME 0x00000020ULL
1252 #define P9_STATS_MTIME 0x00000040ULL
1253 #define P9_STATS_CTIME 0x00000080ULL
1254 #define P9_STATS_INO 0x00000100ULL
1255 #define P9_STATS_SIZE 0x00000200ULL
1256 #define P9_STATS_BLOCKS 0x00000400ULL
1257
1258 #define P9_STATS_BTIME 0x00000800ULL
1259 #define P9_STATS_GEN 0x00001000ULL
1260 #define P9_STATS_DATA_VERSION 0x00002000ULL
1261
1262 #define P9_STATS_BASIC 0x000007ffULL /* Mask for fields up to BLOCKS */
1263 #define P9_STATS_ALL 0x00003fffULL /* Mask for All fields above */
1264
1265
1266 static int stat_to_v9stat_dotl(V9fsPDU *pdu, const struct stat *stbuf,
1267 V9fsStatDotl *v9lstat)
1268 {
1269 memset(v9lstat, 0, sizeof(*v9lstat));
1270
1271 v9lstat->st_mode = stbuf->st_mode;
1272 v9lstat->st_nlink = stbuf->st_nlink;
1273 v9lstat->st_uid = stbuf->st_uid;
1274 v9lstat->st_gid = stbuf->st_gid;
1275 v9lstat->st_rdev = stbuf->st_rdev;
1276 v9lstat->st_size = stbuf->st_size;
1277 v9lstat->st_blksize = stbuf->st_blksize;
1278 v9lstat->st_blocks = stbuf->st_blocks;
1279 v9lstat->st_atime_sec = stbuf->st_atime;
1280 v9lstat->st_atime_nsec = stbuf->st_atim.tv_nsec;
1281 v9lstat->st_mtime_sec = stbuf->st_mtime;
1282 v9lstat->st_mtime_nsec = stbuf->st_mtim.tv_nsec;
1283 v9lstat->st_ctime_sec = stbuf->st_ctime;
1284 v9lstat->st_ctime_nsec = stbuf->st_ctim.tv_nsec;
1285 /* Currently we only support BASIC fields in stat */
1286 v9lstat->st_result_mask = P9_STATS_BASIC;
1287
1288 return stat_to_qid(pdu, stbuf, &v9lstat->qid);
1289 }
1290
1291 static void print_sg(struct iovec *sg, int cnt)
1292 {
1293 int i;
1294
1295 printf("sg[%d]: {", cnt);
1296 for (i = 0; i < cnt; i++) {
1297 if (i) {
1298 printf(", ");
1299 }
1300 printf("(%p, %zd)", sg[i].iov_base, sg[i].iov_len);
1301 }
1302 printf("}\n");
1303 }
1304
1305 /* Will call this only for path name based fid */
1306 static void v9fs_fix_path(V9fsPath *dst, V9fsPath *src, int len)
1307 {
1308 V9fsPath str;
1309 v9fs_path_init(&str);
1310 v9fs_path_copy(&str, dst);
1311 v9fs_path_sprintf(dst, "%s%s", src->data, str.data + len);
1312 v9fs_path_free(&str);
1313 }
1314
1315 static inline bool is_ro_export(FsContext *ctx)
1316 {
1317 return ctx->export_flags & V9FS_RDONLY;
1318 }
1319
1320 static void coroutine_fn v9fs_version(void *opaque)
1321 {
1322 ssize_t err;
1323 V9fsPDU *pdu = opaque;
1324 V9fsState *s = pdu->s;
1325 V9fsString version;
1326 size_t offset = 7;
1327
1328 v9fs_string_init(&version);
1329 err = pdu_unmarshal(pdu, offset, "ds", &s->msize, &version);
1330 if (err < 0) {
1331 goto out;
1332 }
1333 trace_v9fs_version(pdu->tag, pdu->id, s->msize, version.data);
1334
1335 virtfs_reset(pdu);
1336
1337 if (!strcmp(version.data, "9P2000.u")) {
1338 s->proto_version = V9FS_PROTO_2000U;
1339 } else if (!strcmp(version.data, "9P2000.L")) {
1340 s->proto_version = V9FS_PROTO_2000L;
1341 } else {
1342 v9fs_string_sprintf(&version, "unknown");
1343 /* skip min. msize check, reporting invalid version has priority */
1344 goto marshal;
1345 }
1346
1347 if (s->msize < P9_MIN_MSIZE) {
1348 err = -EMSGSIZE;
1349 error_report(
1350 "9pfs: Client requested msize < minimum msize ("
1351 stringify(P9_MIN_MSIZE) ") supported by this server."
1352 );
1353 goto out;
1354 }
1355
1356 marshal:
1357 err = pdu_marshal(pdu, offset, "ds", s->msize, &version);
1358 if (err < 0) {
1359 goto out;
1360 }
1361 err += offset;
1362 trace_v9fs_version_return(pdu->tag, pdu->id, s->msize, version.data);
1363 out:
1364 pdu_complete(pdu, err);
1365 v9fs_string_free(&version);
1366 }
1367
1368 static void coroutine_fn v9fs_attach(void *opaque)
1369 {
1370 V9fsPDU *pdu = opaque;
1371 V9fsState *s = pdu->s;
1372 int32_t fid, afid, n_uname;
1373 V9fsString uname, aname;
1374 V9fsFidState *fidp;
1375 size_t offset = 7;
1376 V9fsQID qid;
1377 ssize_t err;
1378
1379 v9fs_string_init(&uname);
1380 v9fs_string_init(&aname);
1381 err = pdu_unmarshal(pdu, offset, "ddssd", &fid,
1382 &afid, &uname, &aname, &n_uname);
1383 if (err < 0) {
1384 goto out_nofid;
1385 }
1386 trace_v9fs_attach(pdu->tag, pdu->id, fid, afid, uname.data, aname.data);
1387
1388 fidp = alloc_fid(s, fid);
1389 if (fidp == NULL) {
1390 err = -EINVAL;
1391 goto out_nofid;
1392 }
1393 fidp->uid = n_uname;
1394 err = v9fs_co_name_to_path(pdu, NULL, "/", &fidp->path);
1395 if (err < 0) {
1396 err = -EINVAL;
1397 clunk_fid(s, fid);
1398 goto out;
1399 }
1400 err = fid_to_qid(pdu, fidp, &qid);
1401 if (err < 0) {
1402 err = -EINVAL;
1403 clunk_fid(s, fid);
1404 goto out;
1405 }
1406
1407 /*
1408 * disable migration if we haven't done already.
1409 * attach could get called multiple times for the same export.
1410 */
1411 if (!s->migration_blocker) {
1412 error_setg(&s->migration_blocker,
1413 "Migration is disabled when VirtFS export path '%s' is mounted in the guest using mount_tag '%s'",
1414 s->ctx.fs_root ? s->ctx.fs_root : "NULL", s->tag);
1415 err = migrate_add_blocker(s->migration_blocker, NULL);
1416 if (err < 0) {
1417 error_free(s->migration_blocker);
1418 s->migration_blocker = NULL;
1419 clunk_fid(s, fid);
1420 goto out;
1421 }
1422 s->root_fid = fid;
1423 }
1424
1425 err = pdu_marshal(pdu, offset, "Q", &qid);
1426 if (err < 0) {
1427 clunk_fid(s, fid);
1428 goto out;
1429 }
1430 err += offset;
1431
1432 memcpy(&s->root_qid, &qid, sizeof(qid));
1433 trace_v9fs_attach_return(pdu->tag, pdu->id,
1434 qid.type, qid.version, qid.path);
1435 out:
1436 put_fid(pdu, fidp);
1437 out_nofid:
1438 pdu_complete(pdu, err);
1439 v9fs_string_free(&uname);
1440 v9fs_string_free(&aname);
1441 }
1442
1443 static void coroutine_fn v9fs_stat(void *opaque)
1444 {
1445 int32_t fid;
1446 V9fsStat v9stat;
1447 ssize_t err = 0;
1448 size_t offset = 7;
1449 struct stat stbuf;
1450 V9fsFidState *fidp;
1451 V9fsPDU *pdu = opaque;
1452 char *basename;
1453
1454 err = pdu_unmarshal(pdu, offset, "d", &fid);
1455 if (err < 0) {
1456 goto out_nofid;
1457 }
1458 trace_v9fs_stat(pdu->tag, pdu->id, fid);
1459
1460 fidp = get_fid(pdu, fid);
1461 if (fidp == NULL) {
1462 err = -ENOENT;
1463 goto out_nofid;
1464 }
1465 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1466 if (err < 0) {
1467 goto out;
1468 }
1469 basename = g_path_get_basename(fidp->path.data);
1470 err = stat_to_v9stat(pdu, &fidp->path, basename, &stbuf, &v9stat);
1471 g_free(basename);
1472 if (err < 0) {
1473 goto out;
1474 }
1475 err = pdu_marshal(pdu, offset, "wS", 0, &v9stat);
1476 if (err < 0) {
1477 v9fs_stat_free(&v9stat);
1478 goto out;
1479 }
1480 trace_v9fs_stat_return(pdu->tag, pdu->id, v9stat.mode,
1481 v9stat.atime, v9stat.mtime, v9stat.length);
1482 err += offset;
1483 v9fs_stat_free(&v9stat);
1484 out:
1485 put_fid(pdu, fidp);
1486 out_nofid:
1487 pdu_complete(pdu, err);
1488 }
1489
1490 static void coroutine_fn v9fs_getattr(void *opaque)
1491 {
1492 int32_t fid;
1493 size_t offset = 7;
1494 ssize_t retval = 0;
1495 struct stat stbuf;
1496 V9fsFidState *fidp;
1497 uint64_t request_mask;
1498 V9fsStatDotl v9stat_dotl;
1499 V9fsPDU *pdu = opaque;
1500
1501 retval = pdu_unmarshal(pdu, offset, "dq", &fid, &request_mask);
1502 if (retval < 0) {
1503 goto out_nofid;
1504 }
1505 trace_v9fs_getattr(pdu->tag, pdu->id, fid, request_mask);
1506
1507 fidp = get_fid(pdu, fid);
1508 if (fidp == NULL) {
1509 retval = -ENOENT;
1510 goto out_nofid;
1511 }
1512 /*
1513 * Currently we only support BASIC fields in stat, so there is no
1514 * need to look at request_mask.
1515 */
1516 retval = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1517 if (retval < 0) {
1518 goto out;
1519 }
1520 retval = stat_to_v9stat_dotl(pdu, &stbuf, &v9stat_dotl);
1521 if (retval < 0) {
1522 goto out;
1523 }
1524
1525 /* fill st_gen if requested and supported by underlying fs */
1526 if (request_mask & P9_STATS_GEN) {
1527 retval = v9fs_co_st_gen(pdu, &fidp->path, stbuf.st_mode, &v9stat_dotl);
1528 switch (retval) {
1529 case 0:
1530 /* we have valid st_gen: update result mask */
1531 v9stat_dotl.st_result_mask |= P9_STATS_GEN;
1532 break;
1533 case -EINTR:
1534 /* request cancelled, e.g. by Tflush */
1535 goto out;
1536 default:
1537 /* failed to get st_gen: not fatal, ignore */
1538 break;
1539 }
1540 }
1541 retval = pdu_marshal(pdu, offset, "A", &v9stat_dotl);
1542 if (retval < 0) {
1543 goto out;
1544 }
1545 retval += offset;
1546 trace_v9fs_getattr_return(pdu->tag, pdu->id, v9stat_dotl.st_result_mask,
1547 v9stat_dotl.st_mode, v9stat_dotl.st_uid,
1548 v9stat_dotl.st_gid);
1549 out:
1550 put_fid(pdu, fidp);
1551 out_nofid:
1552 pdu_complete(pdu, retval);
1553 }
1554
1555 /* Attribute flags */
1556 #define P9_ATTR_MODE (1 << 0)
1557 #define P9_ATTR_UID (1 << 1)
1558 #define P9_ATTR_GID (1 << 2)
1559 #define P9_ATTR_SIZE (1 << 3)
1560 #define P9_ATTR_ATIME (1 << 4)
1561 #define P9_ATTR_MTIME (1 << 5)
1562 #define P9_ATTR_CTIME (1 << 6)
1563 #define P9_ATTR_ATIME_SET (1 << 7)
1564 #define P9_ATTR_MTIME_SET (1 << 8)
1565
1566 #define P9_ATTR_MASK 127
1567
1568 static void coroutine_fn v9fs_setattr(void *opaque)
1569 {
1570 int err = 0;
1571 int32_t fid;
1572 V9fsFidState *fidp;
1573 size_t offset = 7;
1574 V9fsIattr v9iattr;
1575 V9fsPDU *pdu = opaque;
1576
1577 err = pdu_unmarshal(pdu, offset, "dI", &fid, &v9iattr);
1578 if (err < 0) {
1579 goto out_nofid;
1580 }
1581
1582 trace_v9fs_setattr(pdu->tag, pdu->id, fid,
1583 v9iattr.valid, v9iattr.mode, v9iattr.uid, v9iattr.gid,
1584 v9iattr.size, v9iattr.atime_sec, v9iattr.mtime_sec);
1585
1586 fidp = get_fid(pdu, fid);
1587 if (fidp == NULL) {
1588 err = -EINVAL;
1589 goto out_nofid;
1590 }
1591 if (v9iattr.valid & P9_ATTR_MODE) {
1592 err = v9fs_co_chmod(pdu, &fidp->path, v9iattr.mode);
1593 if (err < 0) {
1594 goto out;
1595 }
1596 }
1597 if (v9iattr.valid & (P9_ATTR_ATIME | P9_ATTR_MTIME)) {
1598 struct timespec times[2];
1599 if (v9iattr.valid & P9_ATTR_ATIME) {
1600 if (v9iattr.valid & P9_ATTR_ATIME_SET) {
1601 times[0].tv_sec = v9iattr.atime_sec;
1602 times[0].tv_nsec = v9iattr.atime_nsec;
1603 } else {
1604 times[0].tv_nsec = UTIME_NOW;
1605 }
1606 } else {
1607 times[0].tv_nsec = UTIME_OMIT;
1608 }
1609 if (v9iattr.valid & P9_ATTR_MTIME) {
1610 if (v9iattr.valid & P9_ATTR_MTIME_SET) {
1611 times[1].tv_sec = v9iattr.mtime_sec;
1612 times[1].tv_nsec = v9iattr.mtime_nsec;
1613 } else {
1614 times[1].tv_nsec = UTIME_NOW;
1615 }
1616 } else {
1617 times[1].tv_nsec = UTIME_OMIT;
1618 }
1619 err = v9fs_co_utimensat(pdu, &fidp->path, times);
1620 if (err < 0) {
1621 goto out;
1622 }
1623 }
1624 /*
1625 * If the only valid entry in iattr is ctime we can call
1626 * chown(-1,-1) to update the ctime of the file
1627 */
1628 if ((v9iattr.valid & (P9_ATTR_UID | P9_ATTR_GID)) ||
1629 ((v9iattr.valid & P9_ATTR_CTIME)
1630 && !((v9iattr.valid & P9_ATTR_MASK) & ~P9_ATTR_CTIME))) {
1631 if (!(v9iattr.valid & P9_ATTR_UID)) {
1632 v9iattr.uid = -1;
1633 }
1634 if (!(v9iattr.valid & P9_ATTR_GID)) {
1635 v9iattr.gid = -1;
1636 }
1637 err = v9fs_co_chown(pdu, &fidp->path, v9iattr.uid,
1638 v9iattr.gid);
1639 if (err < 0) {
1640 goto out;
1641 }
1642 }
1643 if (v9iattr.valid & (P9_ATTR_SIZE)) {
1644 err = v9fs_co_truncate(pdu, &fidp->path, v9iattr.size);
1645 if (err < 0) {
1646 goto out;
1647 }
1648 }
1649 err = offset;
1650 trace_v9fs_setattr_return(pdu->tag, pdu->id);
1651 out:
1652 put_fid(pdu, fidp);
1653 out_nofid:
1654 pdu_complete(pdu, err);
1655 }
1656
1657 static int v9fs_walk_marshal(V9fsPDU *pdu, uint16_t nwnames, V9fsQID *qids)
1658 {
1659 int i;
1660 ssize_t err;
1661 size_t offset = 7;
1662
1663 err = pdu_marshal(pdu, offset, "w", nwnames);
1664 if (err < 0) {
1665 return err;
1666 }
1667 offset += err;
1668 for (i = 0; i < nwnames; i++) {
1669 err = pdu_marshal(pdu, offset, "Q", &qids[i]);
1670 if (err < 0) {
1671 return err;
1672 }
1673 offset += err;
1674 }
1675 return offset;
1676 }
1677
1678 static bool name_is_illegal(const char *name)
1679 {
1680 return !*name || strchr(name, '/') != NULL;
1681 }
1682
1683 static bool not_same_qid(const V9fsQID *qid1, const V9fsQID *qid2)
1684 {
1685 return
1686 qid1->type != qid2->type ||
1687 qid1->version != qid2->version ||
1688 qid1->path != qid2->path;
1689 }
1690
1691 static void coroutine_fn v9fs_walk(void *opaque)
1692 {
1693 int name_idx;
1694 V9fsQID *qids = NULL;
1695 int i, err = 0;
1696 V9fsPath dpath, path;
1697 uint16_t nwnames;
1698 struct stat stbuf;
1699 size_t offset = 7;
1700 int32_t fid, newfid;
1701 V9fsString *wnames = NULL;
1702 V9fsFidState *fidp;
1703 V9fsFidState *newfidp = NULL;
1704 V9fsPDU *pdu = opaque;
1705 V9fsState *s = pdu->s;
1706 V9fsQID qid;
1707
1708 err = pdu_unmarshal(pdu, offset, "ddw", &fid, &newfid, &nwnames);
1709 if (err < 0) {
1710 pdu_complete(pdu, err);
1711 return ;
1712 }
1713 offset += err;
1714
1715 trace_v9fs_walk(pdu->tag, pdu->id, fid, newfid, nwnames);
1716
1717 if (nwnames && nwnames <= P9_MAXWELEM) {
1718 wnames = g_new0(V9fsString, nwnames);
1719 qids = g_new0(V9fsQID, nwnames);
1720 for (i = 0; i < nwnames; i++) {
1721 err = pdu_unmarshal(pdu, offset, "s", &wnames[i]);
1722 if (err < 0) {
1723 goto out_nofid;
1724 }
1725 if (name_is_illegal(wnames[i].data)) {
1726 err = -ENOENT;
1727 goto out_nofid;
1728 }
1729 offset += err;
1730 }
1731 } else if (nwnames > P9_MAXWELEM) {
1732 err = -EINVAL;
1733 goto out_nofid;
1734 }
1735 fidp = get_fid(pdu, fid);
1736 if (fidp == NULL) {
1737 err = -ENOENT;
1738 goto out_nofid;
1739 }
1740
1741 v9fs_path_init(&dpath);
1742 v9fs_path_init(&path);
1743
1744 err = fid_to_qid(pdu, fidp, &qid);
1745 if (err < 0) {
1746 goto out;
1747 }
1748
1749 /*
1750 * Both dpath and path initially poin to fidp.
1751 * Needed to handle request with nwnames == 0
1752 */
1753 v9fs_path_copy(&dpath, &fidp->path);
1754 v9fs_path_copy(&path, &fidp->path);
1755 for (name_idx = 0; name_idx < nwnames; name_idx++) {
1756 if (not_same_qid(&pdu->s->root_qid, &qid) ||
1757 strcmp("..", wnames[name_idx].data)) {
1758 err = v9fs_co_name_to_path(pdu, &dpath, wnames[name_idx].data,
1759 &path);
1760 if (err < 0) {
1761 goto out;
1762 }
1763
1764 err = v9fs_co_lstat(pdu, &path, &stbuf);
1765 if (err < 0) {
1766 goto out;
1767 }
1768 err = stat_to_qid(pdu, &stbuf, &qid);
1769 if (err < 0) {
1770 goto out;
1771 }
1772 v9fs_path_copy(&dpath, &path);
1773 }
1774 memcpy(&qids[name_idx], &qid, sizeof(qid));
1775 }
1776 if (fid == newfid) {
1777 if (fidp->fid_type != P9_FID_NONE) {
1778 err = -EINVAL;
1779 goto out;
1780 }
1781 v9fs_path_write_lock(s);
1782 v9fs_path_copy(&fidp->path, &path);
1783 v9fs_path_unlock(s);
1784 } else {
1785 newfidp = alloc_fid(s, newfid);
1786 if (newfidp == NULL) {
1787 err = -EINVAL;
1788 goto out;
1789 }
1790 newfidp->uid = fidp->uid;
1791 v9fs_path_copy(&newfidp->path, &path);
1792 }
1793 err = v9fs_walk_marshal(pdu, nwnames, qids);
1794 trace_v9fs_walk_return(pdu->tag, pdu->id, nwnames, qids);
1795 out:
1796 put_fid(pdu, fidp);
1797 if (newfidp) {
1798 put_fid(pdu, newfidp);
1799 }
1800 v9fs_path_free(&dpath);
1801 v9fs_path_free(&path);
1802 out_nofid:
1803 pdu_complete(pdu, err);
1804 if (nwnames && nwnames <= P9_MAXWELEM) {
1805 for (name_idx = 0; name_idx < nwnames; name_idx++) {
1806 v9fs_string_free(&wnames[name_idx]);
1807 }
1808 g_free(wnames);
1809 g_free(qids);
1810 }
1811 }
1812
1813 static int32_t coroutine_fn get_iounit(V9fsPDU *pdu, V9fsPath *path)
1814 {
1815 struct statfs stbuf;
1816 int32_t iounit = 0;
1817 V9fsState *s = pdu->s;
1818
1819 /*
1820 * iounit should be multiples of f_bsize (host filesystem block size
1821 * and as well as less than (client msize - P9_IOHDRSZ))
1822 */
1823 if (!v9fs_co_statfs(pdu, path, &stbuf)) {
1824 if (stbuf.f_bsize) {
1825 iounit = stbuf.f_bsize;
1826 iounit *= (s->msize - P9_IOHDRSZ) / stbuf.f_bsize;
1827 }
1828 }
1829 if (!iounit) {
1830 iounit = s->msize - P9_IOHDRSZ;
1831 }
1832 return iounit;
1833 }
1834
1835 static void coroutine_fn v9fs_open(void *opaque)
1836 {
1837 int flags;
1838 int32_t fid;
1839 int32_t mode;
1840 V9fsQID qid;
1841 int iounit = 0;
1842 ssize_t err = 0;
1843 size_t offset = 7;
1844 struct stat stbuf;
1845 V9fsFidState *fidp;
1846 V9fsPDU *pdu = opaque;
1847 V9fsState *s = pdu->s;
1848
1849 if (s->proto_version == V9FS_PROTO_2000L) {
1850 err = pdu_unmarshal(pdu, offset, "dd", &fid, &mode);
1851 } else {
1852 uint8_t modebyte;
1853 err = pdu_unmarshal(pdu, offset, "db", &fid, &modebyte);
1854 mode = modebyte;
1855 }
1856 if (err < 0) {
1857 goto out_nofid;
1858 }
1859 trace_v9fs_open(pdu->tag, pdu->id, fid, mode);
1860
1861 fidp = get_fid(pdu, fid);
1862 if (fidp == NULL) {
1863 err = -ENOENT;
1864 goto out_nofid;
1865 }
1866 if (fidp->fid_type != P9_FID_NONE) {
1867 err = -EINVAL;
1868 goto out;
1869 }
1870
1871 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
1872 if (err < 0) {
1873 goto out;
1874 }
1875 err = stat_to_qid(pdu, &stbuf, &qid);
1876 if (err < 0) {
1877 goto out;
1878 }
1879 if (S_ISDIR(stbuf.st_mode)) {
1880 err = v9fs_co_opendir(pdu, fidp);
1881 if (err < 0) {
1882 goto out;
1883 }
1884 fidp->fid_type = P9_FID_DIR;
1885 err = pdu_marshal(pdu, offset, "Qd", &qid, 0);
1886 if (err < 0) {
1887 goto out;
1888 }
1889 err += offset;
1890 } else {
1891 if (s->proto_version == V9FS_PROTO_2000L) {
1892 flags = get_dotl_openflags(s, mode);
1893 } else {
1894 flags = omode_to_uflags(mode);
1895 }
1896 if (is_ro_export(&s->ctx)) {
1897 if (mode & O_WRONLY || mode & O_RDWR ||
1898 mode & O_APPEND || mode & O_TRUNC) {
1899 err = -EROFS;
1900 goto out;
1901 }
1902 }
1903 err = v9fs_co_open(pdu, fidp, flags);
1904 if (err < 0) {
1905 goto out;
1906 }
1907 fidp->fid_type = P9_FID_FILE;
1908 fidp->open_flags = flags;
1909 if (flags & O_EXCL) {
1910 /*
1911 * We let the host file system do O_EXCL check
1912 * We should not reclaim such fd
1913 */
1914 fidp->flags |= FID_NON_RECLAIMABLE;
1915 }
1916 iounit = get_iounit(pdu, &fidp->path);
1917 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1918 if (err < 0) {
1919 goto out;
1920 }
1921 err += offset;
1922 }
1923 trace_v9fs_open_return(pdu->tag, pdu->id,
1924 qid.type, qid.version, qid.path, iounit);
1925 out:
1926 put_fid(pdu, fidp);
1927 out_nofid:
1928 pdu_complete(pdu, err);
1929 }
1930
1931 static void coroutine_fn v9fs_lcreate(void *opaque)
1932 {
1933 int32_t dfid, flags, mode;
1934 gid_t gid;
1935 ssize_t err = 0;
1936 ssize_t offset = 7;
1937 V9fsString name;
1938 V9fsFidState *fidp;
1939 struct stat stbuf;
1940 V9fsQID qid;
1941 int32_t iounit;
1942 V9fsPDU *pdu = opaque;
1943
1944 v9fs_string_init(&name);
1945 err = pdu_unmarshal(pdu, offset, "dsddd", &dfid,
1946 &name, &flags, &mode, &gid);
1947 if (err < 0) {
1948 goto out_nofid;
1949 }
1950 trace_v9fs_lcreate(pdu->tag, pdu->id, dfid, flags, mode, gid);
1951
1952 if (name_is_illegal(name.data)) {
1953 err = -ENOENT;
1954 goto out_nofid;
1955 }
1956
1957 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
1958 err = -EEXIST;
1959 goto out_nofid;
1960 }
1961
1962 fidp = get_fid(pdu, dfid);
1963 if (fidp == NULL) {
1964 err = -ENOENT;
1965 goto out_nofid;
1966 }
1967 if (fidp->fid_type != P9_FID_NONE) {
1968 err = -EINVAL;
1969 goto out;
1970 }
1971
1972 flags = get_dotl_openflags(pdu->s, flags);
1973 err = v9fs_co_open2(pdu, fidp, &name, gid,
1974 flags | O_CREAT, mode, &stbuf);
1975 if (err < 0) {
1976 goto out;
1977 }
1978 fidp->fid_type = P9_FID_FILE;
1979 fidp->open_flags = flags;
1980 if (flags & O_EXCL) {
1981 /*
1982 * We let the host file system do O_EXCL check
1983 * We should not reclaim such fd
1984 */
1985 fidp->flags |= FID_NON_RECLAIMABLE;
1986 }
1987 iounit = get_iounit(pdu, &fidp->path);
1988 err = stat_to_qid(pdu, &stbuf, &qid);
1989 if (err < 0) {
1990 goto out;
1991 }
1992 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
1993 if (err < 0) {
1994 goto out;
1995 }
1996 err += offset;
1997 trace_v9fs_lcreate_return(pdu->tag, pdu->id,
1998 qid.type, qid.version, qid.path, iounit);
1999 out:
2000 put_fid(pdu, fidp);
2001 out_nofid:
2002 pdu_complete(pdu, err);
2003 v9fs_string_free(&name);
2004 }
2005
2006 static void coroutine_fn v9fs_fsync(void *opaque)
2007 {
2008 int err;
2009 int32_t fid;
2010 int datasync;
2011 size_t offset = 7;
2012 V9fsFidState *fidp;
2013 V9fsPDU *pdu = opaque;
2014
2015 err = pdu_unmarshal(pdu, offset, "dd", &fid, &datasync);
2016 if (err < 0) {
2017 goto out_nofid;
2018 }
2019 trace_v9fs_fsync(pdu->tag, pdu->id, fid, datasync);
2020
2021 fidp = get_fid(pdu, fid);
2022 if (fidp == NULL) {
2023 err = -ENOENT;
2024 goto out_nofid;
2025 }
2026 err = v9fs_co_fsync(pdu, fidp, datasync);
2027 if (!err) {
2028 err = offset;
2029 }
2030 put_fid(pdu, fidp);
2031 out_nofid:
2032 pdu_complete(pdu, err);
2033 }
2034
2035 static void coroutine_fn v9fs_clunk(void *opaque)
2036 {
2037 int err;
2038 int32_t fid;
2039 size_t offset = 7;
2040 V9fsFidState *fidp;
2041 V9fsPDU *pdu = opaque;
2042 V9fsState *s = pdu->s;
2043
2044 err = pdu_unmarshal(pdu, offset, "d", &fid);
2045 if (err < 0) {
2046 goto out_nofid;
2047 }
2048 trace_v9fs_clunk(pdu->tag, pdu->id, fid);
2049
2050 fidp = clunk_fid(s, fid);
2051 if (fidp == NULL) {
2052 err = -ENOENT;
2053 goto out_nofid;
2054 }
2055 /*
2056 * Bump the ref so that put_fid will
2057 * free the fid.
2058 */
2059 fidp->ref++;
2060 err = put_fid(pdu, fidp);
2061 if (!err) {
2062 err = offset;
2063 }
2064 out_nofid:
2065 pdu_complete(pdu, err);
2066 }
2067
2068 /*
2069 * Create a QEMUIOVector for a sub-region of PDU iovecs
2070 *
2071 * @qiov: uninitialized QEMUIOVector
2072 * @skip: number of bytes to skip from beginning of PDU
2073 * @size: number of bytes to include
2074 * @is_write: true - write, false - read
2075 *
2076 * The resulting QEMUIOVector has heap-allocated iovecs and must be cleaned up
2077 * with qemu_iovec_destroy().
2078 */
2079 static void v9fs_init_qiov_from_pdu(QEMUIOVector *qiov, V9fsPDU *pdu,
2080 size_t skip, size_t size,
2081 bool is_write)
2082 {
2083 QEMUIOVector elem;
2084 struct iovec *iov;
2085 unsigned int niov;
2086
2087 if (is_write) {
2088 pdu->s->transport->init_out_iov_from_pdu(pdu, &iov, &niov, size + skip);
2089 } else {
2090 pdu->s->transport->init_in_iov_from_pdu(pdu, &iov, &niov, size + skip);
2091 }
2092
2093 qemu_iovec_init_external(&elem, iov, niov);
2094 qemu_iovec_init(qiov, niov);
2095 qemu_iovec_concat(qiov, &elem, skip, size);
2096 }
2097
2098 static int v9fs_xattr_read(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
2099 uint64_t off, uint32_t max_count)
2100 {
2101 ssize_t err;
2102 size_t offset = 7;
2103 uint64_t read_count;
2104 QEMUIOVector qiov_full;
2105
2106 if (fidp->fs.xattr.len < off) {
2107 read_count = 0;
2108 } else {
2109 read_count = fidp->fs.xattr.len - off;
2110 }
2111 if (read_count > max_count) {
2112 read_count = max_count;
2113 }
2114 err = pdu_marshal(pdu, offset, "d", read_count);
2115 if (err < 0) {
2116 return err;
2117 }
2118 offset += err;
2119
2120 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, read_count, false);
2121 err = v9fs_pack(qiov_full.iov, qiov_full.niov, 0,
2122 ((char *)fidp->fs.xattr.value) + off,
2123 read_count);
2124 qemu_iovec_destroy(&qiov_full);
2125 if (err < 0) {
2126 return err;
2127 }
2128 offset += err;
2129 return offset;
2130 }
2131
2132 static int coroutine_fn v9fs_do_readdir_with_stat(V9fsPDU *pdu,
2133 V9fsFidState *fidp,
2134 uint32_t max_count)
2135 {
2136 V9fsPath path;
2137 V9fsStat v9stat;
2138 int len, err = 0;
2139 int32_t count = 0;
2140 struct stat stbuf;
2141 off_t saved_dir_pos;
2142 struct dirent *dent;
2143
2144 /* save the directory position */
2145 saved_dir_pos = v9fs_co_telldir(pdu, fidp);
2146 if (saved_dir_pos < 0) {
2147 return saved_dir_pos;
2148 }
2149
2150 while (1) {
2151 v9fs_path_init(&path);
2152
2153 v9fs_readdir_lock(&fidp->fs.dir);
2154
2155 err = v9fs_co_readdir(pdu, fidp, &dent);
2156 if (err || !dent) {
2157 break;
2158 }
2159 err = v9fs_co_name_to_path(pdu, &fidp->path, dent->d_name, &path);
2160 if (err < 0) {
2161 break;
2162 }
2163 err = v9fs_co_lstat(pdu, &path, &stbuf);
2164 if (err < 0) {
2165 break;
2166 }
2167 err = stat_to_v9stat(pdu, &path, dent->d_name, &stbuf, &v9stat);
2168 if (err < 0) {
2169 break;
2170 }
2171 if ((count + v9stat.size + 2) > max_count) {
2172 v9fs_readdir_unlock(&fidp->fs.dir);
2173
2174 /* Ran out of buffer. Set dir back to old position and return */
2175 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2176 v9fs_stat_free(&v9stat);
2177 v9fs_path_free(&path);
2178 return count;
2179 }
2180
2181 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2182 len = pdu_marshal(pdu, 11 + count, "S", &v9stat);
2183
2184 v9fs_readdir_unlock(&fidp->fs.dir);
2185
2186 if (len < 0) {
2187 v9fs_co_seekdir(pdu, fidp, saved_dir_pos);
2188 v9fs_stat_free(&v9stat);
2189 v9fs_path_free(&path);
2190 return len;
2191 }
2192 count += len;
2193 v9fs_stat_free(&v9stat);
2194 v9fs_path_free(&path);
2195 saved_dir_pos = dent->d_off;
2196 }
2197
2198 v9fs_readdir_unlock(&fidp->fs.dir);
2199
2200 v9fs_path_free(&path);
2201 if (err < 0) {
2202 return err;
2203 }
2204 return count;
2205 }
2206
2207 static void coroutine_fn v9fs_read(void *opaque)
2208 {
2209 int32_t fid;
2210 uint64_t off;
2211 ssize_t err = 0;
2212 int32_t count = 0;
2213 size_t offset = 7;
2214 uint32_t max_count;
2215 V9fsFidState *fidp;
2216 V9fsPDU *pdu = opaque;
2217 V9fsState *s = pdu->s;
2218
2219 err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &max_count);
2220 if (err < 0) {
2221 goto out_nofid;
2222 }
2223 trace_v9fs_read(pdu->tag, pdu->id, fid, off, max_count);
2224
2225 fidp = get_fid(pdu, fid);
2226 if (fidp == NULL) {
2227 err = -EINVAL;
2228 goto out_nofid;
2229 }
2230 if (fidp->fid_type == P9_FID_DIR) {
2231 if (s->proto_version != V9FS_PROTO_2000U) {
2232 warn_report_once(
2233 "9p: bad client: T_read request on directory only expected "
2234 "with 9P2000.u protocol version"
2235 );
2236 err = -EOPNOTSUPP;
2237 goto out;
2238 }
2239 if (off == 0) {
2240 v9fs_co_rewinddir(pdu, fidp);
2241 }
2242 count = v9fs_do_readdir_with_stat(pdu, fidp, max_count);
2243 if (count < 0) {
2244 err = count;
2245 goto out;
2246 }
2247 err = pdu_marshal(pdu, offset, "d", count);
2248 if (err < 0) {
2249 goto out;
2250 }
2251 err += offset + count;
2252 } else if (fidp->fid_type == P9_FID_FILE) {
2253 QEMUIOVector qiov_full;
2254 QEMUIOVector qiov;
2255 int32_t len;
2256
2257 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset + 4, max_count, false);
2258 qemu_iovec_init(&qiov, qiov_full.niov);
2259 do {
2260 qemu_iovec_reset(&qiov);
2261 qemu_iovec_concat(&qiov, &qiov_full, count, qiov_full.size - count);
2262 if (0) {
2263 print_sg(qiov.iov, qiov.niov);
2264 }
2265 /* Loop in case of EINTR */
2266 do {
2267 len = v9fs_co_preadv(pdu, fidp, qiov.iov, qiov.niov, off);
2268 if (len >= 0) {
2269 off += len;
2270 count += len;
2271 }
2272 } while (len == -EINTR && !pdu->cancelled);
2273 if (len < 0) {
2274 /* IO error return the error */
2275 err = len;
2276 goto out_free_iovec;
2277 }
2278 } while (count < max_count && len > 0);
2279 err = pdu_marshal(pdu, offset, "d", count);
2280 if (err < 0) {
2281 goto out_free_iovec;
2282 }
2283 err += offset + count;
2284 out_free_iovec:
2285 qemu_iovec_destroy(&qiov);
2286 qemu_iovec_destroy(&qiov_full);
2287 } else if (fidp->fid_type == P9_FID_XATTR) {
2288 err = v9fs_xattr_read(s, pdu, fidp, off, max_count);
2289 } else {
2290 err = -EINVAL;
2291 }
2292 trace_v9fs_read_return(pdu->tag, pdu->id, count, err);
2293 out:
2294 put_fid(pdu, fidp);
2295 out_nofid:
2296 pdu_complete(pdu, err);
2297 }
2298
2299 /**
2300 * Returns size required in Rreaddir response for the passed dirent @p name.
2301 *
2302 * @param name - directory entry's name (i.e. file name, directory name)
2303 * @returns required size in bytes
2304 */
2305 size_t v9fs_readdir_response_size(V9fsString *name)
2306 {
2307 /*
2308 * Size of each dirent on the wire: size of qid (13) + size of offset (8)
2309 * size of type (1) + size of name.size (2) + strlen(name.data)
2310 */
2311 return 24 + v9fs_string_size(name);
2312 }
2313
2314 static void v9fs_free_dirents(struct V9fsDirEnt *e)
2315 {
2316 struct V9fsDirEnt *next = NULL;
2317
2318 for (; e; e = next) {
2319 next = e->next;
2320 g_free(e->dent);
2321 g_free(e->st);
2322 g_free(e);
2323 }
2324 }
2325
2326 static int coroutine_fn v9fs_do_readdir(V9fsPDU *pdu, V9fsFidState *fidp,
2327 off_t offset, int32_t max_count)
2328 {
2329 size_t size;
2330 V9fsQID qid;
2331 V9fsString name;
2332 int len, err = 0;
2333 int32_t count = 0;
2334 struct dirent *dent;
2335 struct stat *st;
2336 struct V9fsDirEnt *entries = NULL;
2337
2338 /*
2339 * inode remapping requires the device id, which in turn might be
2340 * different for different directory entries, so if inode remapping is
2341 * enabled we have to make a full stat for each directory entry
2342 */
2343 const bool dostat = pdu->s->ctx.export_flags & V9FS_REMAP_INODES;
2344
2345 /*
2346 * Fetch all required directory entries altogether on a background IO
2347 * thread from fs driver. We don't want to do that for each entry
2348 * individually, because hopping between threads (this main IO thread
2349 * and background IO driver thread) would sum up to huge latencies.
2350 */
2351 count = v9fs_co_readdir_many(pdu, fidp, &entries, offset, max_count,
2352 dostat);
2353 if (count < 0) {
2354 err = count;
2355 count = 0;
2356 goto out;
2357 }
2358 count = 0;
2359
2360 for (struct V9fsDirEnt *e = entries; e; e = e->next) {
2361 dent = e->dent;
2362
2363 if (pdu->s->ctx.export_flags & V9FS_REMAP_INODES) {
2364 st = e->st;
2365 /* e->st should never be NULL, but just to be sure */
2366 if (!st) {
2367 err = -1;
2368 break;
2369 }
2370
2371 /* remap inode */
2372 err = stat_to_qid(pdu, st, &qid);
2373 if (err < 0) {
2374 break;
2375 }
2376 } else {
2377 /*
2378 * Fill up just the path field of qid because the client uses
2379 * only that. To fill the entire qid structure we will have
2380 * to stat each dirent found, which is expensive. For the
2381 * latter reason we don't call stat_to_qid() here. Only drawback
2382 * is that no multi-device export detection of stat_to_qid()
2383 * would be done and provided as error to the user here. But
2384 * user would get that error anyway when accessing those
2385 * files/dirs through other ways.
2386 */
2387 size = MIN(sizeof(dent->d_ino), sizeof(qid.path));
2388 memcpy(&qid.path, &dent->d_ino, size);
2389 /* Fill the other fields with dummy values */
2390 qid.type = 0;
2391 qid.version = 0;
2392 }
2393
2394 v9fs_string_init(&name);
2395 v9fs_string_sprintf(&name, "%s", dent->d_name);
2396
2397 /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
2398 len = pdu_marshal(pdu, 11 + count, "Qqbs",
2399 &qid, dent->d_off,
2400 dent->d_type, &name);
2401
2402 v9fs_string_free(&name);
2403
2404 if (len < 0) {
2405 err = len;
2406 break;
2407 }
2408
2409 count += len;
2410 }
2411
2412 out:
2413 v9fs_free_dirents(entries);
2414 if (err < 0) {
2415 return err;
2416 }
2417 return count;
2418 }
2419
2420 static void coroutine_fn v9fs_readdir(void *opaque)
2421 {
2422 int32_t fid;
2423 V9fsFidState *fidp;
2424 ssize_t retval = 0;
2425 size_t offset = 7;
2426 uint64_t initial_offset;
2427 int32_t count;
2428 uint32_t max_count;
2429 V9fsPDU *pdu = opaque;
2430 V9fsState *s = pdu->s;
2431
2432 retval = pdu_unmarshal(pdu, offset, "dqd", &fid,
2433 &initial_offset, &max_count);
2434 if (retval < 0) {
2435 goto out_nofid;
2436 }
2437 trace_v9fs_readdir(pdu->tag, pdu->id, fid, initial_offset, max_count);
2438
2439 /* Enough space for a R_readdir header: size[4] Rreaddir tag[2] count[4] */
2440 if (max_count > s->msize - 11) {
2441 max_count = s->msize - 11;
2442 warn_report_once(
2443 "9p: bad client: T_readdir with count > msize - 11"
2444 );
2445 }
2446
2447 fidp = get_fid(pdu, fid);
2448 if (fidp == NULL) {
2449 retval = -EINVAL;
2450 goto out_nofid;
2451 }
2452 if (!fidp->fs.dir.stream) {
2453 retval = -EINVAL;
2454 goto out;
2455 }
2456 if (s->proto_version != V9FS_PROTO_2000L) {
2457 warn_report_once(
2458 "9p: bad client: T_readdir request only expected with 9P2000.L "
2459 "protocol version"
2460 );
2461 retval = -EOPNOTSUPP;
2462 goto out;
2463 }
2464 count = v9fs_do_readdir(pdu, fidp, (off_t) initial_offset, max_count);
2465 if (count < 0) {
2466 retval = count;
2467 goto out;
2468 }
2469 retval = pdu_marshal(pdu, offset, "d", count);
2470 if (retval < 0) {
2471 goto out;
2472 }
2473 retval += count + offset;
2474 trace_v9fs_readdir_return(pdu->tag, pdu->id, count, retval);
2475 out:
2476 put_fid(pdu, fidp);
2477 out_nofid:
2478 pdu_complete(pdu, retval);
2479 }
2480
2481 static int v9fs_xattr_write(V9fsState *s, V9fsPDU *pdu, V9fsFidState *fidp,
2482 uint64_t off, uint32_t count,
2483 struct iovec *sg, int cnt)
2484 {
2485 int i, to_copy;
2486 ssize_t err = 0;
2487 uint64_t write_count;
2488 size_t offset = 7;
2489
2490
2491 if (fidp->fs.xattr.len < off) {
2492 return -ENOSPC;
2493 }
2494 write_count = fidp->fs.xattr.len - off;
2495 if (write_count > count) {
2496 write_count = count;
2497 }
2498 err = pdu_marshal(pdu, offset, "d", write_count);
2499 if (err < 0) {
2500 return err;
2501 }
2502 err += offset;
2503 fidp->fs.xattr.copied_len += write_count;
2504 /*
2505 * Now copy the content from sg list
2506 */
2507 for (i = 0; i < cnt; i++) {
2508 if (write_count > sg[i].iov_len) {
2509 to_copy = sg[i].iov_len;
2510 } else {
2511 to_copy = write_count;
2512 }
2513 memcpy((char *)fidp->fs.xattr.value + off, sg[i].iov_base, to_copy);
2514 /* updating vs->off since we are not using below */
2515 off += to_copy;
2516 write_count -= to_copy;
2517 }
2518
2519 return err;
2520 }
2521
2522 static void coroutine_fn v9fs_write(void *opaque)
2523 {
2524 ssize_t err;
2525 int32_t fid;
2526 uint64_t off;
2527 uint32_t count;
2528 int32_t len = 0;
2529 int32_t total = 0;
2530 size_t offset = 7;
2531 V9fsFidState *fidp;
2532 V9fsPDU *pdu = opaque;
2533 V9fsState *s = pdu->s;
2534 QEMUIOVector qiov_full;
2535 QEMUIOVector qiov;
2536
2537 err = pdu_unmarshal(pdu, offset, "dqd", &fid, &off, &count);
2538 if (err < 0) {
2539 pdu_complete(pdu, err);
2540 return;
2541 }
2542 offset += err;
2543 v9fs_init_qiov_from_pdu(&qiov_full, pdu, offset, count, true);
2544 trace_v9fs_write(pdu->tag, pdu->id, fid, off, count, qiov_full.niov);
2545
2546 fidp = get_fid(pdu, fid);
2547 if (fidp == NULL) {
2548 err = -EINVAL;
2549 goto out_nofid;
2550 }
2551 if (fidp->fid_type == P9_FID_FILE) {
2552 if (fidp->fs.fd == -1) {
2553 err = -EINVAL;
2554 goto out;
2555 }
2556 } else if (fidp->fid_type == P9_FID_XATTR) {
2557 /*
2558 * setxattr operation
2559 */
2560 err = v9fs_xattr_write(s, pdu, fidp, off, count,
2561 qiov_full.iov, qiov_full.niov);
2562 goto out;
2563 } else {
2564 err = -EINVAL;
2565 goto out;
2566 }
2567 qemu_iovec_init(&qiov, qiov_full.niov);
2568 do {
2569 qemu_iovec_reset(&qiov);
2570 qemu_iovec_concat(&qiov, &qiov_full, total, qiov_full.size - total);
2571 if (0) {
2572 print_sg(qiov.iov, qiov.niov);
2573 }
2574 /* Loop in case of EINTR */
2575 do {
2576 len = v9fs_co_pwritev(pdu, fidp, qiov.iov, qiov.niov, off);
2577 if (len >= 0) {
2578 off += len;
2579 total += len;
2580 }
2581 } while (len == -EINTR && !pdu->cancelled);
2582 if (len < 0) {
2583 /* IO error return the error */
2584 err = len;
2585 goto out_qiov;
2586 }
2587 } while (total < count && len > 0);
2588
2589 offset = 7;
2590 err = pdu_marshal(pdu, offset, "d", total);
2591 if (err < 0) {
2592 goto out_qiov;
2593 }
2594 err += offset;
2595 trace_v9fs_write_return(pdu->tag, pdu->id, total, err);
2596 out_qiov:
2597 qemu_iovec_destroy(&qiov);
2598 out:
2599 put_fid(pdu, fidp);
2600 out_nofid:
2601 qemu_iovec_destroy(&qiov_full);
2602 pdu_complete(pdu, err);
2603 }
2604
2605 static void coroutine_fn v9fs_create(void *opaque)
2606 {
2607 int32_t fid;
2608 int err = 0;
2609 size_t offset = 7;
2610 V9fsFidState *fidp;
2611 V9fsQID qid;
2612 int32_t perm;
2613 int8_t mode;
2614 V9fsPath path;
2615 struct stat stbuf;
2616 V9fsString name;
2617 V9fsString extension;
2618 int iounit;
2619 V9fsPDU *pdu = opaque;
2620 V9fsState *s = pdu->s;
2621
2622 v9fs_path_init(&path);
2623 v9fs_string_init(&name);
2624 v9fs_string_init(&extension);
2625 err = pdu_unmarshal(pdu, offset, "dsdbs", &fid, &name,
2626 &perm, &mode, &extension);
2627 if (err < 0) {
2628 goto out_nofid;
2629 }
2630 trace_v9fs_create(pdu->tag, pdu->id, fid, name.data, perm, mode);
2631
2632 if (name_is_illegal(name.data)) {
2633 err = -ENOENT;
2634 goto out_nofid;
2635 }
2636
2637 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2638 err = -EEXIST;
2639 goto out_nofid;
2640 }
2641
2642 fidp = get_fid(pdu, fid);
2643 if (fidp == NULL) {
2644 err = -EINVAL;
2645 goto out_nofid;
2646 }
2647 if (fidp->fid_type != P9_FID_NONE) {
2648 err = -EINVAL;
2649 goto out;
2650 }
2651 if (perm & P9_STAT_MODE_DIR) {
2652 err = v9fs_co_mkdir(pdu, fidp, &name, perm & 0777,
2653 fidp->uid, -1, &stbuf);
2654 if (err < 0) {
2655 goto out;
2656 }
2657 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2658 if (err < 0) {
2659 goto out;
2660 }
2661 v9fs_path_write_lock(s);
2662 v9fs_path_copy(&fidp->path, &path);
2663 v9fs_path_unlock(s);
2664 err = v9fs_co_opendir(pdu, fidp);
2665 if (err < 0) {
2666 goto out;
2667 }
2668 fidp->fid_type = P9_FID_DIR;
2669 } else if (perm & P9_STAT_MODE_SYMLINK) {
2670 err = v9fs_co_symlink(pdu, fidp, &name,
2671 extension.data, -1 , &stbuf);
2672 if (err < 0) {
2673 goto out;
2674 }
2675 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2676 if (err < 0) {
2677 goto out;
2678 }
2679 v9fs_path_write_lock(s);
2680 v9fs_path_copy(&fidp->path, &path);
2681 v9fs_path_unlock(s);
2682 } else if (perm & P9_STAT_MODE_LINK) {
2683 int32_t ofid = atoi(extension.data);
2684 V9fsFidState *ofidp = get_fid(pdu, ofid);
2685 if (ofidp == NULL) {
2686 err = -EINVAL;
2687 goto out;
2688 }
2689 err = v9fs_co_link(pdu, ofidp, fidp, &name);
2690 put_fid(pdu, ofidp);
2691 if (err < 0) {
2692 goto out;
2693 }
2694 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2695 if (err < 0) {
2696 fidp->fid_type = P9_FID_NONE;
2697 goto out;
2698 }
2699 v9fs_path_write_lock(s);
2700 v9fs_path_copy(&fidp->path, &path);
2701 v9fs_path_unlock(s);
2702 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
2703 if (err < 0) {
2704 fidp->fid_type = P9_FID_NONE;
2705 goto out;
2706 }
2707 } else if (perm & P9_STAT_MODE_DEVICE) {
2708 char ctype;
2709 uint32_t major, minor;
2710 mode_t nmode = 0;
2711
2712 if (sscanf(extension.data, "%c %u %u", &ctype, &major, &minor) != 3) {
2713 err = -errno;
2714 goto out;
2715 }
2716
2717 switch (ctype) {
2718 case 'c':
2719 nmode = S_IFCHR;
2720 break;
2721 case 'b':
2722 nmode = S_IFBLK;
2723 break;
2724 default:
2725 err = -EIO;
2726 goto out;
2727 }
2728
2729 nmode |= perm & 0777;
2730 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2731 makedev(major, minor), nmode, &stbuf);
2732 if (err < 0) {
2733 goto out;
2734 }
2735 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2736 if (err < 0) {
2737 goto out;
2738 }
2739 v9fs_path_write_lock(s);
2740 v9fs_path_copy(&fidp->path, &path);
2741 v9fs_path_unlock(s);
2742 } else if (perm & P9_STAT_MODE_NAMED_PIPE) {
2743 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2744 0, S_IFIFO | (perm & 0777), &stbuf);
2745 if (err < 0) {
2746 goto out;
2747 }
2748 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2749 if (err < 0) {
2750 goto out;
2751 }
2752 v9fs_path_write_lock(s);
2753 v9fs_path_copy(&fidp->path, &path);
2754 v9fs_path_unlock(s);
2755 } else if (perm & P9_STAT_MODE_SOCKET) {
2756 err = v9fs_co_mknod(pdu, fidp, &name, fidp->uid, -1,
2757 0, S_IFSOCK | (perm & 0777), &stbuf);
2758 if (err < 0) {
2759 goto out;
2760 }
2761 err = v9fs_co_name_to_path(pdu, &fidp->path, name.data, &path);
2762 if (err < 0) {
2763 goto out;
2764 }
2765 v9fs_path_write_lock(s);
2766 v9fs_path_copy(&fidp->path, &path);
2767 v9fs_path_unlock(s);
2768 } else {
2769 err = v9fs_co_open2(pdu, fidp, &name, -1,
2770 omode_to_uflags(mode)|O_CREAT, perm, &stbuf);
2771 if (err < 0) {
2772 goto out;
2773 }
2774 fidp->fid_type = P9_FID_FILE;
2775 fidp->open_flags = omode_to_uflags(mode);
2776 if (fidp->open_flags & O_EXCL) {
2777 /*
2778 * We let the host file system do O_EXCL check
2779 * We should not reclaim such fd
2780 */
2781 fidp->flags |= FID_NON_RECLAIMABLE;
2782 }
2783 }
2784 iounit = get_iounit(pdu, &fidp->path);
2785 err = stat_to_qid(pdu, &stbuf, &qid);
2786 if (err < 0) {
2787 goto out;
2788 }
2789 err = pdu_marshal(pdu, offset, "Qd", &qid, iounit);
2790 if (err < 0) {
2791 goto out;
2792 }
2793 err += offset;
2794 trace_v9fs_create_return(pdu->tag, pdu->id,
2795 qid.type, qid.version, qid.path, iounit);
2796 out:
2797 put_fid(pdu, fidp);
2798 out_nofid:
2799 pdu_complete(pdu, err);
2800 v9fs_string_free(&name);
2801 v9fs_string_free(&extension);
2802 v9fs_path_free(&path);
2803 }
2804
2805 static void coroutine_fn v9fs_symlink(void *opaque)
2806 {
2807 V9fsPDU *pdu = opaque;
2808 V9fsString name;
2809 V9fsString symname;
2810 V9fsFidState *dfidp;
2811 V9fsQID qid;
2812 struct stat stbuf;
2813 int32_t dfid;
2814 int err = 0;
2815 gid_t gid;
2816 size_t offset = 7;
2817
2818 v9fs_string_init(&name);
2819 v9fs_string_init(&symname);
2820 err = pdu_unmarshal(pdu, offset, "dssd", &dfid, &name, &symname, &gid);
2821 if (err < 0) {
2822 goto out_nofid;
2823 }
2824 trace_v9fs_symlink(pdu->tag, pdu->id, dfid, name.data, symname.data, gid);
2825
2826 if (name_is_illegal(name.data)) {
2827 err = -ENOENT;
2828 goto out_nofid;
2829 }
2830
2831 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2832 err = -EEXIST;
2833 goto out_nofid;
2834 }
2835
2836 dfidp = get_fid(pdu, dfid);
2837 if (dfidp == NULL) {
2838 err = -EINVAL;
2839 goto out_nofid;
2840 }
2841 err = v9fs_co_symlink(pdu, dfidp, &name, symname.data, gid, &stbuf);
2842 if (err < 0) {
2843 goto out;
2844 }
2845 err = stat_to_qid(pdu, &stbuf, &qid);
2846 if (err < 0) {
2847 goto out;
2848 }
2849 err = pdu_marshal(pdu, offset, "Q", &qid);
2850 if (err < 0) {
2851 goto out;
2852 }
2853 err += offset;
2854 trace_v9fs_symlink_return(pdu->tag, pdu->id,
2855 qid.type, qid.version, qid.path);
2856 out:
2857 put_fid(pdu, dfidp);
2858 out_nofid:
2859 pdu_complete(pdu, err);
2860 v9fs_string_free(&name);
2861 v9fs_string_free(&symname);
2862 }
2863
2864 static void coroutine_fn v9fs_flush(void *opaque)
2865 {
2866 ssize_t err;
2867 int16_t tag;
2868 size_t offset = 7;
2869 V9fsPDU *cancel_pdu = NULL;
2870 V9fsPDU *pdu = opaque;
2871 V9fsState *s = pdu->s;
2872
2873 err = pdu_unmarshal(pdu, offset, "w", &tag);
2874 if (err < 0) {
2875 pdu_complete(pdu, err);
2876 return;
2877 }
2878 trace_v9fs_flush(pdu->tag, pdu->id, tag);
2879
2880 if (pdu->tag == tag) {
2881 warn_report("the guest sent a self-referencing 9P flush request");
2882 } else {
2883 QLIST_FOREACH(cancel_pdu, &s->active_list, next) {
2884 if (cancel_pdu->tag == tag) {
2885 break;
2886 }
2887 }
2888 }
2889 if (cancel_pdu) {
2890 cancel_pdu->cancelled = 1;
2891 /*
2892 * Wait for pdu to complete.
2893 */
2894 qemu_co_queue_wait(&cancel_pdu->complete, NULL);
2895 if (!qemu_co_queue_next(&cancel_pdu->complete)) {
2896 cancel_pdu->cancelled = 0;
2897 pdu_free(cancel_pdu);
2898 }
2899 }
2900 pdu_complete(pdu, 7);
2901 }
2902
2903 static void coroutine_fn v9fs_link(void *opaque)
2904 {
2905 V9fsPDU *pdu = opaque;
2906 int32_t dfid, oldfid;
2907 V9fsFidState *dfidp, *oldfidp;
2908 V9fsString name;
2909 size_t offset = 7;
2910 int err = 0;
2911
2912 v9fs_string_init(&name);
2913 err = pdu_unmarshal(pdu, offset, "dds", &dfid, &oldfid, &name);
2914 if (err < 0) {
2915 goto out_nofid;
2916 }
2917 trace_v9fs_link(pdu->tag, pdu->id, dfid, oldfid, name.data);
2918
2919 if (name_is_illegal(name.data)) {
2920 err = -ENOENT;
2921 goto out_nofid;
2922 }
2923
2924 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
2925 err = -EEXIST;
2926 goto out_nofid;
2927 }
2928
2929 dfidp = get_fid(pdu, dfid);
2930 if (dfidp == NULL) {
2931 err = -ENOENT;
2932 goto out_nofid;
2933 }
2934
2935 oldfidp = get_fid(pdu, oldfid);
2936 if (oldfidp == NULL) {
2937 err = -ENOENT;
2938 goto out;
2939 }
2940 err = v9fs_co_link(pdu, oldfidp, dfidp, &name);
2941 if (!err) {
2942 err = offset;
2943 }
2944 put_fid(pdu, oldfidp);
2945 out:
2946 put_fid(pdu, dfidp);
2947 out_nofid:
2948 v9fs_string_free(&name);
2949 pdu_complete(pdu, err);
2950 }
2951
2952 /* Only works with path name based fid */
2953 static void coroutine_fn v9fs_remove(void *opaque)
2954 {
2955 int32_t fid;
2956 int err = 0;
2957 size_t offset = 7;
2958 V9fsFidState *fidp;
2959 V9fsPDU *pdu = opaque;
2960
2961 err = pdu_unmarshal(pdu, offset, "d", &fid);
2962 if (err < 0) {
2963 goto out_nofid;
2964 }
2965 trace_v9fs_remove(pdu->tag, pdu->id, fid);
2966
2967 fidp = get_fid(pdu, fid);
2968 if (fidp == NULL) {
2969 err = -EINVAL;
2970 goto out_nofid;
2971 }
2972 /* if fs driver is not path based, return EOPNOTSUPP */
2973 if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
2974 err = -EOPNOTSUPP;
2975 goto out_err;
2976 }
2977 /*
2978 * IF the file is unlinked, we cannot reopen
2979 * the file later. So don't reclaim fd
2980 */
2981 err = v9fs_mark_fids_unreclaim(pdu, &fidp->path);
2982 if (err < 0) {
2983 goto out_err;
2984 }
2985 err = v9fs_co_remove(pdu, &fidp->path);
2986 if (!err) {
2987 err = offset;
2988 }
2989 out_err:
2990 /* For TREMOVE we need to clunk the fid even on failed remove */
2991 clunk_fid(pdu->s, fidp->fid);
2992 put_fid(pdu, fidp);
2993 out_nofid:
2994 pdu_complete(pdu, err);
2995 }
2996
2997 static void coroutine_fn v9fs_unlinkat(void *opaque)
2998 {
2999 int err = 0;
3000 V9fsString name;
3001 int32_t dfid, flags, rflags = 0;
3002 size_t offset = 7;
3003 V9fsPath path;
3004 V9fsFidState *dfidp;
3005 V9fsPDU *pdu = opaque;
3006
3007 v9fs_string_init(&name);
3008 err = pdu_unmarshal(pdu, offset, "dsd", &dfid, &name, &flags);
3009 if (err < 0) {
3010 goto out_nofid;
3011 }
3012
3013 if (name_is_illegal(name.data)) {
3014 err = -ENOENT;
3015 goto out_nofid;
3016 }
3017
3018 if (!strcmp(".", name.data)) {
3019 err = -EINVAL;
3020 goto out_nofid;
3021 }
3022
3023 if (!strcmp("..", name.data)) {
3024 err = -ENOTEMPTY;
3025 goto out_nofid;
3026 }
3027
3028 if (flags & ~P9_DOTL_AT_REMOVEDIR) {
3029 err = -EINVAL;
3030 goto out_nofid;
3031 }
3032
3033 if (flags & P9_DOTL_AT_REMOVEDIR) {
3034 rflags |= AT_REMOVEDIR;
3035 }
3036
3037 dfidp = get_fid(pdu, dfid);
3038 if (dfidp == NULL) {
3039 err = -EINVAL;
3040 goto out_nofid;
3041 }
3042 /*
3043 * IF the file is unlinked, we cannot reopen
3044 * the file later. So don't reclaim fd
3045 */
3046 v9fs_path_init(&path);
3047 err = v9fs_co_name_to_path(pdu, &dfidp->path, name.data, &path);
3048 if (err < 0) {
3049 goto out_err;
3050 }
3051 err = v9fs_mark_fids_unreclaim(pdu, &path);
3052 if (err < 0) {
3053 goto out_err;
3054 }
3055 err = v9fs_co_unlinkat(pdu, &dfidp->path, &name, rflags);
3056 if (!err) {
3057 err = offset;
3058 }
3059 out_err:
3060 put_fid(pdu, dfidp);
3061 v9fs_path_free(&path);
3062 out_nofid:
3063 pdu_complete(pdu, err);
3064 v9fs_string_free(&name);
3065 }
3066
3067
3068 /* Only works with path name based fid */
3069 static int coroutine_fn v9fs_complete_rename(V9fsPDU *pdu, V9fsFidState *fidp,
3070 int32_t newdirfid,
3071 V9fsString *name)
3072 {
3073 int err = 0;
3074 V9fsPath new_path;
3075 V9fsFidState *tfidp;
3076 V9fsState *s = pdu->s;
3077 V9fsFidState *dirfidp = NULL;
3078
3079 v9fs_path_init(&new_path);
3080 if (newdirfid != -1) {
3081 dirfidp = get_fid(pdu, newdirfid);
3082 if (dirfidp == NULL) {
3083 return -ENOENT;
3084 }
3085 if (fidp->fid_type != P9_FID_NONE) {
3086 err = -EINVAL;
3087 goto out;
3088 }
3089 err = v9fs_co_name_to_path(pdu, &dirfidp->path, name->data, &new_path);
3090 if (err < 0) {
3091 goto out;
3092 }
3093 } else {
3094 char *dir_name = g_path_get_dirname(fidp->path.data);
3095 V9fsPath dir_path;
3096
3097 v9fs_path_init(&dir_path);
3098 v9fs_path_sprintf(&dir_path, "%s", dir_name);
3099 g_free(dir_name);
3100
3101 err = v9fs_co_name_to_path(pdu, &dir_path, name->data, &new_path);
3102 v9fs_path_free(&dir_path);
3103 if (err < 0) {
3104 goto out;
3105 }
3106 }
3107 err = v9fs_co_rename(pdu, &fidp->path, &new_path);
3108 if (err < 0) {
3109 goto out;
3110 }
3111 /*
3112 * Fixup fid's pointing to the old name to
3113 * start pointing to the new name
3114 */
3115 for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
3116 if (v9fs_path_is_ancestor(&fidp->path, &tfidp->path)) {
3117 /* replace the name */
3118 v9fs_fix_path(&tfidp->path, &new_path, strlen(fidp->path.data));
3119 }
3120 }
3121 out:
3122 if (dirfidp) {
3123 put_fid(pdu, dirfidp);
3124 }
3125 v9fs_path_free(&new_path);
3126 return err;
3127 }
3128
3129 /* Only works with path name based fid */
3130 static void coroutine_fn v9fs_rename(void *opaque)
3131 {
3132 int32_t fid;
3133 ssize_t err = 0;
3134 size_t offset = 7;
3135 V9fsString name;
3136 int32_t newdirfid;
3137 V9fsFidState *fidp;
3138 V9fsPDU *pdu = opaque;
3139 V9fsState *s = pdu->s;
3140
3141 v9fs_string_init(&name);
3142 err = pdu_unmarshal(pdu, offset, "dds", &fid, &newdirfid, &name);
3143 if (err < 0) {
3144 goto out_nofid;
3145 }
3146
3147 if (name_is_illegal(name.data)) {
3148 err = -ENOENT;
3149 goto out_nofid;
3150 }
3151
3152 if (!strcmp(".", name.data) || !strcmp("..", name.data)) {
3153 err = -EISDIR;
3154 goto out_nofid;
3155 }
3156
3157 fidp = get_fid(pdu, fid);
3158 if (fidp == NULL) {
3159 err = -ENOENT;
3160 goto out_nofid;
3161 }
3162 if (fidp->fid_type != P9_FID_NONE) {
3163 err = -EINVAL;
3164 goto out;
3165 }
3166 /* if fs driver is not path based, return EOPNOTSUPP */
3167 if (!(pdu->s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT)) {
3168 err = -EOPNOTSUPP;
3169 goto out;
3170 }
3171 v9fs_path_write_lock(s);
3172 err = v9fs_complete_rename(pdu, fidp, newdirfid, &name);
3173 v9fs_path_unlock(s);
3174 if (!err) {
3175 err = offset;
3176 }
3177 out:
3178 put_fid(pdu, fidp);
3179 out_nofid:
3180 pdu_complete(pdu, err);
3181 v9fs_string_free(&name);
3182 }
3183
3184 static int coroutine_fn v9fs_fix_fid_paths(V9fsPDU *pdu, V9fsPath *olddir,
3185 V9fsString *old_name,
3186 V9fsPath *newdir,
3187 V9fsString *new_name)
3188 {
3189 V9fsFidState *tfidp;
3190 V9fsPath oldpath, newpath;
3191 V9fsState *s = pdu->s;
3192 int err;
3193
3194 v9fs_path_init(&oldpath);
3195 v9fs_path_init(&newpath);
3196 err = v9fs_co_name_to_path(pdu, olddir, old_name->data, &oldpath);
3197 if (err < 0) {
3198 goto out;
3199 }
3200 err = v9fs_co_name_to_path(pdu, newdir, new_name->data, &newpath);
3201 if (err < 0) {
3202 goto out;
3203 }
3204
3205 /*
3206 * Fixup fid's pointing to the old name to
3207 * start pointing to the new name
3208 */
3209 for (tfidp = s->fid_list; tfidp; tfidp = tfidp->next) {
3210 if (v9fs_path_is_ancestor(&oldpath, &tfidp->path)) {
3211 /* replace the name */
3212 v9fs_fix_path(&tfidp->path, &newpath, strlen(oldpath.data));
3213 }
3214 }
3215 out:
3216 v9fs_path_free(&oldpath);
3217 v9fs_path_free(&newpath);
3218 return err;
3219 }
3220
3221 static int coroutine_fn v9fs_complete_renameat(V9fsPDU *pdu, int32_t olddirfid,
3222 V9fsString *old_name,
3223 int32_t newdirfid,
3224 V9fsString *new_name)
3225 {
3226 int err = 0;
3227 V9fsState *s = pdu->s;
3228 V9fsFidState *newdirfidp = NULL, *olddirfidp = NULL;
3229
3230 olddirfidp = get_fid(pdu, olddirfid);
3231 if (olddirfidp == NULL) {
3232 err = -ENOENT;
3233 goto out;
3234 }
3235 if (newdirfid != -1) {
3236 newdirfidp = get_fid(pdu, newdirfid);
3237 if (newdirfidp == NULL) {
3238 err = -ENOENT;
3239 goto out;
3240 }
3241 } else {
3242 newdirfidp = get_fid(pdu, olddirfid);
3243 }
3244
3245 err = v9fs_co_renameat(pdu, &olddirfidp->path, old_name,
3246 &newdirfidp->path, new_name);
3247 if (err < 0) {
3248 goto out;
3249 }
3250 if (s->ctx.export_flags & V9FS_PATHNAME_FSCONTEXT) {
3251 /* Only for path based fid we need to do the below fixup */
3252 err = v9fs_fix_fid_paths(pdu, &olddirfidp->path, old_name,
3253 &newdirfidp->path, new_name);
3254 }
3255 out:
3256 if (olddirfidp) {
3257 put_fid(pdu, olddirfidp);
3258 }
3259 if (newdirfidp) {
3260 put_fid(pdu, newdirfidp);
3261 }
3262 return err;
3263 }
3264
3265 static void coroutine_fn v9fs_renameat(void *opaque)
3266 {
3267 ssize_t err = 0;
3268 size_t offset = 7;
3269 V9fsPDU *pdu = opaque;
3270 V9fsState *s = pdu->s;
3271 int32_t olddirfid, newdirfid;
3272 V9fsString old_name, new_name;
3273
3274 v9fs_string_init(&old_name);
3275 v9fs_string_init(&new_name);
3276 err = pdu_unmarshal(pdu, offset, "dsds", &olddirfid,
3277 &old_name, &newdirfid, &new_name);
3278 if (err < 0) {
3279 goto out_err;
3280 }
3281
3282 if (name_is_illegal(old_name.data) || name_is_illegal(new_name.data)) {
3283 err = -ENOENT;
3284 goto out_err;
3285 }
3286
3287 if (!strcmp(".", old_name.data) || !strcmp("..", old_name.data) ||
3288 !strcmp(".", new_name.data) || !strcmp("..", new_name.data)) {
3289 err = -EISDIR;
3290 goto out_err;
3291 }
3292
3293 v9fs_path_write_lock(s);
3294 err = v9fs_complete_renameat(pdu, olddirfid,
3295 &old_name, newdirfid, &new_name);
3296 v9fs_path_unlock(s);
3297 if (!err) {
3298 err = offset;
3299 }
3300
3301 out_err:
3302 pdu_complete(pdu, err);
3303 v9fs_string_free(&old_name);
3304 v9fs_string_free(&new_name);
3305 }
3306
3307 static void coroutine_fn v9fs_wstat(void *opaque)
3308 {
3309 int32_t fid;
3310 int err = 0;
3311 int16_t unused;
3312 V9fsStat v9stat;
3313 size_t offset = 7;
3314 struct stat stbuf;
3315 V9fsFidState *fidp;
3316 V9fsPDU *pdu = opaque;
3317 V9fsState *s = pdu->s;
3318
3319 v9fs_stat_init(&v9stat);
3320 err = pdu_unmarshal(pdu, offset, "dwS", &fid, &unused, &v9stat);
3321 if (err < 0) {
3322 goto out_nofid;
3323 }
3324 trace_v9fs_wstat(pdu->tag, pdu->id, fid,
3325 v9stat.mode, v9stat.atime, v9stat.mtime);
3326
3327 fidp = get_fid(pdu, fid);
3328 if (fidp == NULL) {
3329 err = -EINVAL;
3330 goto out_nofid;
3331 }
3332 /* do we need to sync the file? */
3333 if (donttouch_stat(&v9stat)) {
3334 err = v9fs_co_fsync(pdu, fidp, 0);
3335 goto out;
3336 }
3337 if (v9stat.mode != -1) {
3338 uint32_t v9_mode;
3339 err = v9fs_co_lstat(pdu, &fidp->path, &stbuf);
3340 if (err < 0) {
3341 goto out;
3342 }
3343 v9_mode = stat_to_v9mode(&stbuf);
3344 if ((v9stat.mode & P9_STAT_MODE_TYPE_BITS) !=
3345 (v9_mode & P9_STAT_MODE_TYPE_BITS)) {
3346 /* Attempting to change the type */
3347 err = -EIO;
3348 goto out;
3349 }
3350 err = v9fs_co_chmod(pdu, &fidp->path,
3351 v9mode_to_mode(v9stat.mode,
3352 &v9stat.extension));
3353 if (err < 0) {
3354 goto out;
3355 }
3356 }
3357 if (v9stat.mtime != -1 || v9stat.atime != -1) {
3358 struct timespec times[2];
3359 if (v9stat.atime != -1) {
3360 times[0].tv_sec = v9stat.atime;
3361 times[0].tv_nsec = 0;
3362 } else {
3363 times[0].tv_nsec = UTIME_OMIT;
3364 }
3365 if (v9stat.mtime != -1) {
3366 times[1].tv_sec = v9stat.mtime;
3367 times[1].tv_nsec = 0;
3368 } else {
3369 times[1].tv_nsec = UTIME_OMIT;
3370 }
3371 err = v9fs_co_utimensat(pdu, &fidp->path, times);
3372 if (err < 0) {
3373 goto out;
3374 }
3375 }
3376 if (v9stat.n_gid != -1 || v9stat.n_uid != -1) {
3377 err = v9fs_co_chown(pdu, &fidp->path, v9stat.n_uid, v9stat.n_gid);
3378 if (err < 0) {
3379 goto out;
3380 }
3381 }
3382 if (v9stat.name.size != 0) {
3383 v9fs_path_write_lock(s);
3384 err = v9fs_complete_rename(pdu, fidp, -1, &v9stat.name);
3385 v9fs_path_unlock(s);
3386 if (err < 0) {
3387 goto out;
3388 }
3389 }
3390 if (v9stat.length != -1) {
3391 err = v9fs_co_truncate(pdu, &fidp->path, v9stat.length);
3392 if (err < 0) {
3393 goto out;
3394 }
3395 }
3396 err = offset;
3397 out:
3398 put_fid(pdu, fidp);
3399 out_nofid:
3400 v9fs_stat_free(&v9stat);
3401 pdu_complete(pdu, err);
3402 }
3403
3404 static int v9fs_fill_statfs(V9fsState *s, V9fsPDU *pdu, struct statfs *stbuf)
3405 {
3406 uint32_t f_type;
3407 uint32_t f_bsize;
3408 uint64_t f_blocks;
3409 uint64_t f_bfree;
3410 uint64_t f_bavail;
3411 uint64_t f_files;
3412 uint64_t f_ffree;
3413 uint64_t fsid_val;
3414 uint32_t f_namelen;
3415 size_t offset = 7;
3416 int32_t bsize_factor;
3417
3418 /*
3419 * compute bsize factor based on host file system block size
3420 * and client msize
3421 */
3422 bsize_factor = (s->msize - P9_IOHDRSZ)/stbuf->f_bsize;
3423 if (!bsize_factor) {
3424 bsize_factor = 1;
3425 }