]> git.proxmox.com Git - mirror_qemu.git/blob - block/export/fuse.c
Merge tag 'pull-target-arm-20230530-1' of https://git.linaro.org/people/pmaydell...
[mirror_qemu.git] / block / export / fuse.c
1 /*
2 * Present a block device as a raw image through FUSE
3 *
4 * Copyright (c) 2020 Max Reitz <mreitz@redhat.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; under version 2 or later of the License.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
17 */
18
19 #define FUSE_USE_VERSION 31
20
21 #include "qemu/osdep.h"
22 #include "qemu/memalign.h"
23 #include "block/aio.h"
24 #include "block/block_int-common.h"
25 #include "block/export.h"
26 #include "block/fuse.h"
27 #include "block/qapi.h"
28 #include "qapi/error.h"
29 #include "qapi/qapi-commands-block.h"
30 #include "qemu/main-loop.h"
31 #include "sysemu/block-backend.h"
32
33 #include <fuse.h>
34 #include <fuse_lowlevel.h>
35
36 #if defined(CONFIG_FALLOCATE_ZERO_RANGE)
37 #include <linux/falloc.h>
38 #endif
39
40 #ifdef __linux__
41 #include <linux/fs.h>
42 #endif
43
44 /* Prevent overly long bounce buffer allocations */
45 #define FUSE_MAX_BOUNCE_BYTES (MIN(BDRV_REQUEST_MAX_BYTES, 64 * 1024 * 1024))
46
47
48 typedef struct FuseExport {
49 BlockExport common;
50
51 struct fuse_session *fuse_session;
52 struct fuse_buf fuse_buf;
53 bool mounted, fd_handler_set_up;
54
55 char *mountpoint;
56 bool writable;
57 bool growable;
58 /* Whether allow_other was used as a mount option or not */
59 bool allow_other;
60
61 mode_t st_mode;
62 uid_t st_uid;
63 gid_t st_gid;
64 } FuseExport;
65
66 static GHashTable *exports;
67 static const struct fuse_lowlevel_ops fuse_ops;
68
69 static void fuse_export_shutdown(BlockExport *exp);
70 static void fuse_export_delete(BlockExport *exp);
71
72 static void init_exports_table(void);
73
74 static int setup_fuse_export(FuseExport *exp, const char *mountpoint,
75 bool allow_other, Error **errp);
76 static void read_from_fuse_export(void *opaque);
77
78 static bool is_regular_file(const char *path, Error **errp);
79
80
81 static int fuse_export_create(BlockExport *blk_exp,
82 BlockExportOptions *blk_exp_args,
83 Error **errp)
84 {
85 FuseExport *exp = container_of(blk_exp, FuseExport, common);
86 BlockExportOptionsFuse *args = &blk_exp_args->u.fuse;
87 int ret;
88
89 assert(blk_exp_args->type == BLOCK_EXPORT_TYPE_FUSE);
90
91 /* For growable and writable exports, take the RESIZE permission */
92 if (args->growable || blk_exp_args->writable) {
93 uint64_t blk_perm, blk_shared_perm;
94
95 blk_get_perm(exp->common.blk, &blk_perm, &blk_shared_perm);
96
97 ret = blk_set_perm(exp->common.blk, blk_perm | BLK_PERM_RESIZE,
98 blk_shared_perm, errp);
99 if (ret < 0) {
100 return ret;
101 }
102 }
103
104 init_exports_table();
105
106 /*
107 * It is important to do this check before calling is_regular_file() --
108 * that function will do a stat(), which we would have to handle if we
109 * already exported something on @mountpoint. But we cannot, because
110 * we are currently caught up here.
111 * (Note that ideally we would want to resolve relative paths here,
112 * but bdrv_make_absolute_filename() might do the wrong thing for
113 * paths that contain colons, and realpath() would resolve symlinks,
114 * which we do not want: The mount point is not going to be the
115 * symlink's destination, but the link itself.)
116 * So this will not catch all potential clashes, but hopefully at
117 * least the most common one of specifying exactly the same path
118 * string twice.
119 */
120 if (g_hash_table_contains(exports, args->mountpoint)) {
121 error_setg(errp, "There already is a FUSE export on '%s'",
122 args->mountpoint);
123 ret = -EEXIST;
124 goto fail;
125 }
126
127 if (!is_regular_file(args->mountpoint, errp)) {
128 ret = -EINVAL;
129 goto fail;
130 }
131
132 exp->mountpoint = g_strdup(args->mountpoint);
133 exp->writable = blk_exp_args->writable;
134 exp->growable = args->growable;
135
136 /* set default */
137 if (!args->has_allow_other) {
138 args->allow_other = FUSE_EXPORT_ALLOW_OTHER_AUTO;
139 }
140
141 exp->st_mode = S_IFREG | S_IRUSR;
142 if (exp->writable) {
143 exp->st_mode |= S_IWUSR;
144 }
145 exp->st_uid = getuid();
146 exp->st_gid = getgid();
147
148 if (args->allow_other == FUSE_EXPORT_ALLOW_OTHER_AUTO) {
149 /* Ignore errors on our first attempt */
150 ret = setup_fuse_export(exp, args->mountpoint, true, NULL);
151 exp->allow_other = ret == 0;
152 if (ret < 0) {
153 ret = setup_fuse_export(exp, args->mountpoint, false, errp);
154 }
155 } else {
156 exp->allow_other = args->allow_other == FUSE_EXPORT_ALLOW_OTHER_ON;
157 ret = setup_fuse_export(exp, args->mountpoint, exp->allow_other, errp);
158 }
159 if (ret < 0) {
160 goto fail;
161 }
162
163 return 0;
164
165 fail:
166 fuse_export_delete(blk_exp);
167 return ret;
168 }
169
170 /**
171 * Allocates the global @exports hash table.
172 */
173 static void init_exports_table(void)
174 {
175 if (exports) {
176 return;
177 }
178
179 exports = g_hash_table_new_full(g_str_hash, g_str_equal, g_free, NULL);
180 }
181
182 /**
183 * Create exp->fuse_session and mount it.
184 */
185 static int setup_fuse_export(FuseExport *exp, const char *mountpoint,
186 bool allow_other, Error **errp)
187 {
188 const char *fuse_argv[4];
189 char *mount_opts;
190 struct fuse_args fuse_args;
191 int ret;
192
193 /*
194 * max_read needs to match what fuse_init() sets.
195 * max_write need not be supplied.
196 */
197 mount_opts = g_strdup_printf("max_read=%zu,default_permissions%s",
198 FUSE_MAX_BOUNCE_BYTES,
199 allow_other ? ",allow_other" : "");
200
201 fuse_argv[0] = ""; /* Dummy program name */
202 fuse_argv[1] = "-o";
203 fuse_argv[2] = mount_opts;
204 fuse_argv[3] = NULL;
205 fuse_args = (struct fuse_args)FUSE_ARGS_INIT(3, (char **)fuse_argv);
206
207 exp->fuse_session = fuse_session_new(&fuse_args, &fuse_ops,
208 sizeof(fuse_ops), exp);
209 g_free(mount_opts);
210 if (!exp->fuse_session) {
211 error_setg(errp, "Failed to set up FUSE session");
212 ret = -EIO;
213 goto fail;
214 }
215
216 ret = fuse_session_mount(exp->fuse_session, mountpoint);
217 if (ret < 0) {
218 error_setg(errp, "Failed to mount FUSE session to export");
219 ret = -EIO;
220 goto fail;
221 }
222 exp->mounted = true;
223
224 g_hash_table_insert(exports, g_strdup(mountpoint), NULL);
225
226 aio_set_fd_handler(exp->common.ctx,
227 fuse_session_fd(exp->fuse_session), true,
228 read_from_fuse_export, NULL, NULL, NULL, exp);
229 exp->fd_handler_set_up = true;
230
231 return 0;
232
233 fail:
234 fuse_export_shutdown(&exp->common);
235 return ret;
236 }
237
238 /**
239 * Callback to be invoked when the FUSE session FD can be read from.
240 * (This is basically the FUSE event loop.)
241 */
242 static void read_from_fuse_export(void *opaque)
243 {
244 FuseExport *exp = opaque;
245 int ret;
246
247 blk_exp_ref(&exp->common);
248
249 do {
250 ret = fuse_session_receive_buf(exp->fuse_session, &exp->fuse_buf);
251 } while (ret == -EINTR);
252 if (ret < 0) {
253 goto out;
254 }
255
256 fuse_session_process_buf(exp->fuse_session, &exp->fuse_buf);
257
258 out:
259 blk_exp_unref(&exp->common);
260 }
261
262 static void fuse_export_shutdown(BlockExport *blk_exp)
263 {
264 FuseExport *exp = container_of(blk_exp, FuseExport, common);
265
266 if (exp->fuse_session) {
267 fuse_session_exit(exp->fuse_session);
268
269 if (exp->fd_handler_set_up) {
270 aio_set_fd_handler(exp->common.ctx,
271 fuse_session_fd(exp->fuse_session), true,
272 NULL, NULL, NULL, NULL, NULL);
273 exp->fd_handler_set_up = false;
274 }
275 }
276
277 if (exp->mountpoint) {
278 /*
279 * Safe to drop now, because we will not handle any requests
280 * for this export anymore anyway.
281 */
282 g_hash_table_remove(exports, exp->mountpoint);
283 }
284 }
285
286 static void fuse_export_delete(BlockExport *blk_exp)
287 {
288 FuseExport *exp = container_of(blk_exp, FuseExport, common);
289
290 if (exp->fuse_session) {
291 if (exp->mounted) {
292 fuse_session_unmount(exp->fuse_session);
293 }
294
295 fuse_session_destroy(exp->fuse_session);
296 }
297
298 free(exp->fuse_buf.mem);
299 g_free(exp->mountpoint);
300 }
301
302 /**
303 * Check whether @path points to a regular file. If not, put an
304 * appropriate message into *errp.
305 */
306 static bool is_regular_file(const char *path, Error **errp)
307 {
308 struct stat statbuf;
309 int ret;
310
311 ret = stat(path, &statbuf);
312 if (ret < 0) {
313 error_setg_errno(errp, errno, "Failed to stat '%s'", path);
314 return false;
315 }
316
317 if (!S_ISREG(statbuf.st_mode)) {
318 error_setg(errp, "'%s' is not a regular file", path);
319 return false;
320 }
321
322 return true;
323 }
324
325 /**
326 * A chance to set change some parameters supplied to FUSE_INIT.
327 */
328 static void fuse_init(void *userdata, struct fuse_conn_info *conn)
329 {
330 /*
331 * MIN_NON_ZERO() would not be wrong here, but what we set here
332 * must equal what has been passed to fuse_session_new().
333 * Therefore, as long as max_read must be passed as a mount option
334 * (which libfuse claims will be changed at some point), we have
335 * to set max_read to a fixed value here.
336 */
337 conn->max_read = FUSE_MAX_BOUNCE_BYTES;
338
339 conn->max_write = MIN_NON_ZERO(BDRV_REQUEST_MAX_BYTES, conn->max_write);
340 }
341
342 /**
343 * Let clients look up files. Always return ENOENT because we only
344 * care about the mountpoint itself.
345 */
346 static void fuse_lookup(fuse_req_t req, fuse_ino_t parent, const char *name)
347 {
348 fuse_reply_err(req, ENOENT);
349 }
350
351 /**
352 * Let clients get file attributes (i.e., stat() the file).
353 */
354 static void fuse_getattr(fuse_req_t req, fuse_ino_t inode,
355 struct fuse_file_info *fi)
356 {
357 struct stat statbuf;
358 int64_t length, allocated_blocks;
359 time_t now = time(NULL);
360 FuseExport *exp = fuse_req_userdata(req);
361
362 length = blk_getlength(exp->common.blk);
363 if (length < 0) {
364 fuse_reply_err(req, -length);
365 return;
366 }
367
368 allocated_blocks = bdrv_get_allocated_file_size(blk_bs(exp->common.blk));
369 if (allocated_blocks <= 0) {
370 allocated_blocks = DIV_ROUND_UP(length, 512);
371 } else {
372 allocated_blocks = DIV_ROUND_UP(allocated_blocks, 512);
373 }
374
375 statbuf = (struct stat) {
376 .st_ino = inode,
377 .st_mode = exp->st_mode,
378 .st_nlink = 1,
379 .st_uid = exp->st_uid,
380 .st_gid = exp->st_gid,
381 .st_size = length,
382 .st_blksize = blk_bs(exp->common.blk)->bl.request_alignment,
383 .st_blocks = allocated_blocks,
384 .st_atime = now,
385 .st_mtime = now,
386 .st_ctime = now,
387 };
388
389 fuse_reply_attr(req, &statbuf, 1.);
390 }
391
392 static int fuse_do_truncate(const FuseExport *exp, int64_t size,
393 bool req_zero_write, PreallocMode prealloc)
394 {
395 uint64_t blk_perm, blk_shared_perm;
396 BdrvRequestFlags truncate_flags = 0;
397 bool add_resize_perm;
398 int ret, ret_check;
399
400 /* Growable and writable exports have a permanent RESIZE permission */
401 add_resize_perm = !exp->growable && !exp->writable;
402
403 if (req_zero_write) {
404 truncate_flags |= BDRV_REQ_ZERO_WRITE;
405 }
406
407 if (add_resize_perm) {
408
409 if (!qemu_in_main_thread()) {
410 /* Changing permissions like below only works in the main thread */
411 return -EPERM;
412 }
413
414 blk_get_perm(exp->common.blk, &blk_perm, &blk_shared_perm);
415
416 ret = blk_set_perm(exp->common.blk, blk_perm | BLK_PERM_RESIZE,
417 blk_shared_perm, NULL);
418 if (ret < 0) {
419 return ret;
420 }
421 }
422
423 ret = blk_truncate(exp->common.blk, size, true, prealloc,
424 truncate_flags, NULL);
425
426 if (add_resize_perm) {
427 /* Must succeed, because we are only giving up the RESIZE permission */
428 ret_check = blk_set_perm(exp->common.blk, blk_perm,
429 blk_shared_perm, &error_abort);
430 assert(ret_check == 0);
431 }
432
433 return ret;
434 }
435
436 /**
437 * Let clients set file attributes. Only resizing and changing
438 * permissions (st_mode, st_uid, st_gid) is allowed.
439 * Changing permissions is only allowed as far as it will actually
440 * permit access: Read-only exports cannot be given +w, and exports
441 * without allow_other cannot be given a different UID or GID, and
442 * they cannot be given non-owner access.
443 */
444 static void fuse_setattr(fuse_req_t req, fuse_ino_t inode, struct stat *statbuf,
445 int to_set, struct fuse_file_info *fi)
446 {
447 FuseExport *exp = fuse_req_userdata(req);
448 int supported_attrs;
449 int ret;
450
451 supported_attrs = FUSE_SET_ATTR_SIZE | FUSE_SET_ATTR_MODE;
452 if (exp->allow_other) {
453 supported_attrs |= FUSE_SET_ATTR_UID | FUSE_SET_ATTR_GID;
454 }
455
456 if (to_set & ~supported_attrs) {
457 fuse_reply_err(req, ENOTSUP);
458 return;
459 }
460
461 /* Do some argument checks first before committing to anything */
462 if (to_set & FUSE_SET_ATTR_MODE) {
463 /*
464 * Without allow_other, non-owners can never access the export, so do
465 * not allow setting permissions for them
466 */
467 if (!exp->allow_other &&
468 (statbuf->st_mode & (S_IRWXG | S_IRWXO)) != 0)
469 {
470 fuse_reply_err(req, EPERM);
471 return;
472 }
473
474 /* +w for read-only exports makes no sense, disallow it */
475 if (!exp->writable &&
476 (statbuf->st_mode & (S_IWUSR | S_IWGRP | S_IWOTH)) != 0)
477 {
478 fuse_reply_err(req, EROFS);
479 return;
480 }
481 }
482
483 if (to_set & FUSE_SET_ATTR_SIZE) {
484 if (!exp->writable) {
485 fuse_reply_err(req, EACCES);
486 return;
487 }
488
489 ret = fuse_do_truncate(exp, statbuf->st_size, true, PREALLOC_MODE_OFF);
490 if (ret < 0) {
491 fuse_reply_err(req, -ret);
492 return;
493 }
494 }
495
496 if (to_set & FUSE_SET_ATTR_MODE) {
497 /* Ignore FUSE-supplied file type, only change the mode */
498 exp->st_mode = (statbuf->st_mode & 07777) | S_IFREG;
499 }
500
501 if (to_set & FUSE_SET_ATTR_UID) {
502 exp->st_uid = statbuf->st_uid;
503 }
504
505 if (to_set & FUSE_SET_ATTR_GID) {
506 exp->st_gid = statbuf->st_gid;
507 }
508
509 fuse_getattr(req, inode, fi);
510 }
511
512 /**
513 * Let clients open a file (i.e., the exported image).
514 */
515 static void fuse_open(fuse_req_t req, fuse_ino_t inode,
516 struct fuse_file_info *fi)
517 {
518 fuse_reply_open(req, fi);
519 }
520
521 /**
522 * Handle client reads from the exported image.
523 */
524 static void fuse_read(fuse_req_t req, fuse_ino_t inode,
525 size_t size, off_t offset, struct fuse_file_info *fi)
526 {
527 FuseExport *exp = fuse_req_userdata(req);
528 int64_t length;
529 void *buf;
530 int ret;
531
532 /* Limited by max_read, should not happen */
533 if (size > FUSE_MAX_BOUNCE_BYTES) {
534 fuse_reply_err(req, EINVAL);
535 return;
536 }
537
538 /**
539 * Clients will expect short reads at EOF, so we have to limit
540 * offset+size to the image length.
541 */
542 length = blk_getlength(exp->common.blk);
543 if (length < 0) {
544 fuse_reply_err(req, -length);
545 return;
546 }
547
548 if (offset + size > length) {
549 size = length - offset;
550 }
551
552 buf = qemu_try_blockalign(blk_bs(exp->common.blk), size);
553 if (!buf) {
554 fuse_reply_err(req, ENOMEM);
555 return;
556 }
557
558 ret = blk_pread(exp->common.blk, offset, size, buf, 0);
559 if (ret >= 0) {
560 fuse_reply_buf(req, buf, size);
561 } else {
562 fuse_reply_err(req, -ret);
563 }
564
565 qemu_vfree(buf);
566 }
567
568 /**
569 * Handle client writes to the exported image.
570 */
571 static void fuse_write(fuse_req_t req, fuse_ino_t inode, const char *buf,
572 size_t size, off_t offset, struct fuse_file_info *fi)
573 {
574 FuseExport *exp = fuse_req_userdata(req);
575 int64_t length;
576 int ret;
577
578 /* Limited by max_write, should not happen */
579 if (size > BDRV_REQUEST_MAX_BYTES) {
580 fuse_reply_err(req, EINVAL);
581 return;
582 }
583
584 if (!exp->writable) {
585 fuse_reply_err(req, EACCES);
586 return;
587 }
588
589 /**
590 * Clients will expect short writes at EOF, so we have to limit
591 * offset+size to the image length.
592 */
593 length = blk_getlength(exp->common.blk);
594 if (length < 0) {
595 fuse_reply_err(req, -length);
596 return;
597 }
598
599 if (offset + size > length) {
600 if (exp->growable) {
601 ret = fuse_do_truncate(exp, offset + size, true, PREALLOC_MODE_OFF);
602 if (ret < 0) {
603 fuse_reply_err(req, -ret);
604 return;
605 }
606 } else {
607 size = length - offset;
608 }
609 }
610
611 ret = blk_pwrite(exp->common.blk, offset, size, buf, 0);
612 if (ret >= 0) {
613 fuse_reply_write(req, size);
614 } else {
615 fuse_reply_err(req, -ret);
616 }
617 }
618
619 /**
620 * Let clients perform various fallocate() operations.
621 */
622 static void fuse_fallocate(fuse_req_t req, fuse_ino_t inode, int mode,
623 off_t offset, off_t length,
624 struct fuse_file_info *fi)
625 {
626 FuseExport *exp = fuse_req_userdata(req);
627 int64_t blk_len;
628 int ret;
629
630 if (!exp->writable) {
631 fuse_reply_err(req, EACCES);
632 return;
633 }
634
635 blk_len = blk_getlength(exp->common.blk);
636 if (blk_len < 0) {
637 fuse_reply_err(req, -blk_len);
638 return;
639 }
640
641 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
642 if (mode & FALLOC_FL_KEEP_SIZE) {
643 length = MIN(length, blk_len - offset);
644 }
645 #endif /* CONFIG_FALLOCATE_PUNCH_HOLE */
646
647 if (!mode) {
648 /* We can only fallocate at the EOF with a truncate */
649 if (offset < blk_len) {
650 fuse_reply_err(req, EOPNOTSUPP);
651 return;
652 }
653
654 if (offset > blk_len) {
655 /* No preallocation needed here */
656 ret = fuse_do_truncate(exp, offset, true, PREALLOC_MODE_OFF);
657 if (ret < 0) {
658 fuse_reply_err(req, -ret);
659 return;
660 }
661 }
662
663 ret = fuse_do_truncate(exp, offset + length, true,
664 PREALLOC_MODE_FALLOC);
665 }
666 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
667 else if (mode & FALLOC_FL_PUNCH_HOLE) {
668 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
669 fuse_reply_err(req, EINVAL);
670 return;
671 }
672
673 do {
674 int size = MIN(length, BDRV_REQUEST_MAX_BYTES);
675
676 ret = blk_pwrite_zeroes(exp->common.blk, offset, size,
677 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK);
678 if (ret == -ENOTSUP) {
679 /*
680 * fallocate() specifies to return EOPNOTSUPP for unsupported
681 * operations
682 */
683 ret = -EOPNOTSUPP;
684 }
685
686 offset += size;
687 length -= size;
688 } while (ret == 0 && length > 0);
689 }
690 #endif /* CONFIG_FALLOCATE_PUNCH_HOLE */
691 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
692 else if (mode & FALLOC_FL_ZERO_RANGE) {
693 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + length > blk_len) {
694 /* No need for zeroes, we are going to write them ourselves */
695 ret = fuse_do_truncate(exp, offset + length, false,
696 PREALLOC_MODE_OFF);
697 if (ret < 0) {
698 fuse_reply_err(req, -ret);
699 return;
700 }
701 }
702
703 do {
704 int size = MIN(length, BDRV_REQUEST_MAX_BYTES);
705
706 ret = blk_pwrite_zeroes(exp->common.blk,
707 offset, size, 0);
708 offset += size;
709 length -= size;
710 } while (ret == 0 && length > 0);
711 }
712 #endif /* CONFIG_FALLOCATE_ZERO_RANGE */
713 else {
714 ret = -EOPNOTSUPP;
715 }
716
717 fuse_reply_err(req, ret < 0 ? -ret : 0);
718 }
719
720 /**
721 * Let clients fsync the exported image.
722 */
723 static void fuse_fsync(fuse_req_t req, fuse_ino_t inode, int datasync,
724 struct fuse_file_info *fi)
725 {
726 FuseExport *exp = fuse_req_userdata(req);
727 int ret;
728
729 ret = blk_flush(exp->common.blk);
730 fuse_reply_err(req, ret < 0 ? -ret : 0);
731 }
732
733 /**
734 * Called before an FD to the exported image is closed. (libfuse
735 * notes this to be a way to return last-minute errors.)
736 */
737 static void fuse_flush(fuse_req_t req, fuse_ino_t inode,
738 struct fuse_file_info *fi)
739 {
740 fuse_fsync(req, inode, 1, fi);
741 }
742
743 #ifdef CONFIG_FUSE_LSEEK
744 /**
745 * Let clients inquire allocation status.
746 */
747 static void fuse_lseek(fuse_req_t req, fuse_ino_t inode, off_t offset,
748 int whence, struct fuse_file_info *fi)
749 {
750 FuseExport *exp = fuse_req_userdata(req);
751
752 if (whence != SEEK_HOLE && whence != SEEK_DATA) {
753 fuse_reply_err(req, EINVAL);
754 return;
755 }
756
757 while (true) {
758 int64_t pnum;
759 int ret;
760
761 ret = bdrv_block_status_above(blk_bs(exp->common.blk), NULL,
762 offset, INT64_MAX, &pnum, NULL, NULL);
763 if (ret < 0) {
764 fuse_reply_err(req, -ret);
765 return;
766 }
767
768 if (!pnum && (ret & BDRV_BLOCK_EOF)) {
769 int64_t blk_len;
770
771 /*
772 * If blk_getlength() rounds (e.g. by sectors), then the
773 * export length will be rounded, too. However,
774 * bdrv_block_status_above() may return EOF at unaligned
775 * offsets. We must not let this become visible and thus
776 * always simulate a hole between @offset (the real EOF)
777 * and @blk_len (the client-visible EOF).
778 */
779
780 blk_len = blk_getlength(exp->common.blk);
781 if (blk_len < 0) {
782 fuse_reply_err(req, -blk_len);
783 return;
784 }
785
786 if (offset > blk_len || whence == SEEK_DATA) {
787 fuse_reply_err(req, ENXIO);
788 } else {
789 fuse_reply_lseek(req, offset);
790 }
791 return;
792 }
793
794 if (ret & BDRV_BLOCK_DATA) {
795 if (whence == SEEK_DATA) {
796 fuse_reply_lseek(req, offset);
797 return;
798 }
799 } else {
800 if (whence == SEEK_HOLE) {
801 fuse_reply_lseek(req, offset);
802 return;
803 }
804 }
805
806 /* Safety check against infinite loops */
807 if (!pnum) {
808 fuse_reply_err(req, ENXIO);
809 return;
810 }
811
812 offset += pnum;
813 }
814 }
815 #endif
816
817 static const struct fuse_lowlevel_ops fuse_ops = {
818 .init = fuse_init,
819 .lookup = fuse_lookup,
820 .getattr = fuse_getattr,
821 .setattr = fuse_setattr,
822 .open = fuse_open,
823 .read = fuse_read,
824 .write = fuse_write,
825 .fallocate = fuse_fallocate,
826 .flush = fuse_flush,
827 .fsync = fuse_fsync,
828 #ifdef CONFIG_FUSE_LSEEK
829 .lseek = fuse_lseek,
830 #endif
831 };
832
833 const BlockExportDriver blk_exp_fuse = {
834 .type = BLOCK_EXPORT_TYPE_FUSE,
835 .instance_size = sizeof(FuseExport),
836 .create = fuse_export_create,
837 .delete = fuse_export_delete,
838 .request_shutdown = fuse_export_shutdown,
839 };