]> git.proxmox.com Git - ceph.git/blob - ceph/src/krbd.cc
update sources to v12.1.1
[ceph.git] / ceph / src / krbd.cc
1 /*
2 * Ceph - scalable distributed file system
3 *
4 * Copyright (C) 2014 Inktank Storage, Inc.
5 *
6 * This is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License version 2.1, as published by the Free Software
9 * Foundation. See file COPYING.
10 *
11 */
12
13 #include <errno.h>
14 #include <fcntl.h>
15 #include <iostream>
16 #include <map>
17 #include <poll.h>
18 #include <sstream>
19 #include <stdio.h>
20 #include <stdlib.h>
21 #include <string.h>
22 #include <string>
23 #include <sys/stat.h>
24 #include <sys/types.h>
25 #include <unistd.h>
26
27 #include "auth/KeyRing.h"
28 #include "common/errno.h"
29 #include "common/Formatter.h"
30 #include "common/module.h"
31 #include "common/run_cmd.h"
32 #include "common/safe_io.h"
33 #include "common/secret.h"
34 #include "common/TextTable.h"
35 #include "include/assert.h"
36 #include "include/stringify.h"
37 #include "include/krbd.h"
38 #include "mon/MonMap.h"
39
40 #include <blkid/blkid.h>
41 #include <libudev.h>
42
43 using namespace std;
44
45 const static int POLL_TIMEOUT=120000;
46
47 struct krbd_ctx {
48 CephContext *cct;
49 struct udev *udev;
50 };
51
52 static string get_kernel_rbd_name(const char *id)
53 {
54 return string("/dev/rbd") + id;
55 }
56
57 static int sysfs_write_rbd(const char *which, const string& buf)
58 {
59 const string s = string("/sys/bus/rbd/") + which;
60 const string t = s + "_single_major";
61 int fd;
62 int r;
63
64 /*
65 * 'add' and 'add_single_major' interfaces are identical, but if rbd
66 * kernel module is new enough and is configured to use single-major
67 * scheme, 'add' is disabled in order to prevent old userspace from
68 * doing weird things at unmap time.
69 *
70 * Same goes for 'remove' vs 'remove_single_major'.
71 */
72 fd = open(t.c_str(), O_WRONLY);
73 if (fd < 0) {
74 if (errno == ENOENT) {
75 fd = open(s.c_str(), O_WRONLY);
76 if (fd < 0)
77 return -errno;
78 } else {
79 return -errno;
80 }
81 }
82
83 r = safe_write(fd, buf.c_str(), buf.size());
84
85 close(fd);
86 return r;
87 }
88
89 static int sysfs_write_rbd_add(const string& buf)
90 {
91 return sysfs_write_rbd("add", buf);
92 }
93
94 static int sysfs_write_rbd_remove(const string& buf)
95 {
96 return sysfs_write_rbd("remove", buf);
97 }
98
99 static int have_minor_attr(void)
100 {
101 /*
102 * 'minor' attribute was added as part of single_major merge, which
103 * exposed the 'single_major' parameter. 'minor' is always present,
104 * regardless of whether single-major scheme is turned on or not.
105 *
106 * (Something like ver >= KERNEL_VERSION(3, 14, 0) is a no-go because
107 * this has to work with rbd.ko backported to various kernels.)
108 */
109 return access("/sys/module/rbd/parameters/single_major", F_OK) == 0;
110 }
111
112 static int build_map_buf(CephContext *cct, const char *pool, const char *image,
113 const char *snap, const char *options, string *pbuf)
114 {
115 ostringstream oss;
116 int r;
117
118 MonMap monmap;
119 r = monmap.build_initial(cct, cerr);
120 if (r < 0)
121 return r;
122
123 list<entity_addr_t> mon_addr;
124 monmap.list_addrs(mon_addr);
125
126 for (const auto &p : mon_addr) {
127 if (oss.tellp() > 0) {
128 oss << ",";
129 }
130 oss << p.get_sockaddr();
131 }
132
133 oss << " name=" << cct->_conf->name.get_id();
134
135 KeyRing keyring;
136 if (cct->_conf->auth_client_required != "none") {
137 r = keyring.from_ceph_context(cct);
138 if (r == -ENOENT && !(cct->_conf->keyfile.length() ||
139 cct->_conf->key.length()))
140 r = 0;
141 if (r < 0) {
142 cerr << "rbd: failed to get secret" << std::endl;
143 return r;
144 }
145 }
146
147 CryptoKey secret;
148 string key_name = string("client.") + cct->_conf->name.get_id();
149 if (keyring.get_secret(cct->_conf->name, secret)) {
150 string secret_str;
151 secret.encode_base64(secret_str);
152
153 r = set_kernel_secret(secret_str.c_str(), key_name.c_str());
154 if (r >= 0) {
155 if (r == 0)
156 cerr << "rbd: warning: secret has length 0" << std::endl;
157 oss << ",key=" << key_name;
158 } else if (r == -ENODEV || r == -ENOSYS) {
159 // running against older kernel; fall back to secret= in options
160 oss << ",secret=" << secret_str;
161 } else {
162 cerr << "rbd: failed to add secret '" << key_name << "' to kernel"
163 << std::endl;
164 return r;
165 }
166 } else if (is_kernel_secret(key_name.c_str())) {
167 oss << ",key=" << key_name;
168 }
169
170 if (strcmp(options, "") != 0)
171 oss << "," << options;
172
173 oss << " " << pool << " " << image << " " << snap;
174
175 *pbuf = oss.str();
176 return 0;
177 }
178
179 static int wait_for_udev_add(struct udev_monitor *mon, const char *pool,
180 const char *image, const char *snap,
181 string *pname)
182 {
183 struct udev_device *bus_dev = NULL;
184
185 /*
186 * Catch /sys/devices/rbd/<id>/ and wait for the corresponding
187 * block device to show up. This is necessary because rbd devices
188 * and block devices aren't linked together in our sysfs layout.
189 */
190 for (;;) {
191 struct pollfd fds[1];
192 struct udev_device *dev;
193
194 fds[0].fd = udev_monitor_get_fd(mon);
195 fds[0].events = POLLIN;
196 if (poll(fds, 1, POLL_TIMEOUT) < 0)
197 return -errno;
198
199 dev = udev_monitor_receive_device(mon);
200 if (!dev)
201 continue;
202
203 if (strcmp(udev_device_get_action(dev), "add") != 0)
204 goto next;
205
206 if (!bus_dev) {
207 if (strcmp(udev_device_get_subsystem(dev), "rbd") == 0) {
208 const char *this_pool = udev_device_get_sysattr_value(dev, "pool");
209 const char *this_image = udev_device_get_sysattr_value(dev, "name");
210 const char *this_snap = udev_device_get_sysattr_value(dev,
211 "current_snap");
212
213 if (this_pool && strcmp(this_pool, pool) == 0 &&
214 this_image && strcmp(this_image, image) == 0 &&
215 this_snap && strcmp(this_snap, snap) == 0) {
216 bus_dev = dev;
217 continue;
218 }
219 }
220 } else {
221 if (strcmp(udev_device_get_subsystem(dev), "block") == 0) {
222 const char *major = udev_device_get_sysattr_value(bus_dev, "major");
223 const char *minor = udev_device_get_sysattr_value(bus_dev, "minor");
224 const char *this_major = udev_device_get_property_value(dev, "MAJOR");
225 const char *this_minor = udev_device_get_property_value(dev, "MINOR");
226
227 assert(!minor ^ have_minor_attr());
228
229 if (strcmp(this_major, major) == 0 &&
230 (!minor || strcmp(this_minor, minor) == 0)) {
231 string name = get_kernel_rbd_name(udev_device_get_sysname(bus_dev));
232
233 assert(strcmp(udev_device_get_devnode(dev), name.c_str()) == 0);
234 *pname = name;
235
236 udev_device_unref(dev);
237 udev_device_unref(bus_dev);
238 break;
239 }
240 }
241 }
242
243 next:
244 udev_device_unref(dev);
245 }
246
247 return 0;
248 }
249
250 static int do_map(struct udev *udev, const char *pool, const char *image,
251 const char *snap, const string& buf, string *pname)
252 {
253 struct udev_monitor *mon;
254 int r;
255
256 mon = udev_monitor_new_from_netlink(udev, "udev");
257 if (!mon)
258 return -ENOMEM;
259
260 r = udev_monitor_filter_add_match_subsystem_devtype(mon, "rbd", NULL);
261 if (r < 0)
262 goto out_mon;
263
264 r = udev_monitor_filter_add_match_subsystem_devtype(mon, "block", "disk");
265 if (r < 0)
266 goto out_mon;
267
268 r = udev_monitor_enable_receiving(mon);
269 if (r < 0)
270 goto out_mon;
271
272 r = sysfs_write_rbd_add(buf);
273 if (r < 0) {
274 cerr << "rbd: sysfs write failed" << std::endl;
275 goto out_mon;
276 }
277
278 r = wait_for_udev_add(mon, pool, image, snap, pname);
279 if (r < 0) {
280 cerr << "rbd: wait failed" << std::endl;
281 goto out_mon;
282 }
283
284 out_mon:
285 udev_monitor_unref(mon);
286 return r;
287 }
288
289 static int map_image(struct krbd_ctx *ctx, const char *pool, const char *image,
290 const char *snap, const char *options, string *pname)
291 {
292 string buf;
293 int r;
294
295 if (strcmp(snap, "") == 0)
296 snap = "-";
297
298 r = build_map_buf(ctx->cct, pool, image, snap, options, &buf);
299 if (r < 0)
300 return r;
301
302 /*
303 * Modprobe rbd kernel module. If it supports single-major device
304 * number allocation scheme, make sure it's turned on.
305 */
306 if (access("/sys/bus/rbd", F_OK) != 0) {
307 const char *module_options = NULL;
308 if (module_has_param("rbd", "single_major"))
309 module_options = "single_major=Y";
310
311 r = module_load("rbd", module_options);
312 if (r) {
313 cerr << "rbd: failed to load rbd kernel module (" << r << ")"
314 << std::endl;
315 /*
316 * Ignore the error: modprobe failing doesn't necessarily prevent
317 * from working.
318 */
319 }
320 }
321
322 return do_map(ctx->udev, pool, image, snap, buf, pname);
323 }
324
325 static int devno_to_krbd_id(struct udev *udev, dev_t devno, string *pid)
326 {
327 struct udev_enumerate *enm;
328 struct udev_list_entry *l;
329 struct udev_device *dev;
330 int r;
331
332 enm = udev_enumerate_new(udev);
333 if (!enm)
334 return -ENOMEM;
335
336 r = udev_enumerate_add_match_subsystem(enm, "rbd");
337 if (r < 0)
338 goto out_enm;
339
340 r = udev_enumerate_add_match_sysattr(enm, "major",
341 stringify(major(devno)).c_str());
342 if (r < 0)
343 goto out_enm;
344
345 if (have_minor_attr()) {
346 r = udev_enumerate_add_match_sysattr(enm, "minor",
347 stringify(minor(devno)).c_str());
348 if (r < 0)
349 goto out_enm;
350 }
351
352 r = udev_enumerate_scan_devices(enm);
353 if (r < 0)
354 goto out_enm;
355
356 l = udev_enumerate_get_list_entry(enm);
357 if (!l) {
358 r = -ENOENT;
359 goto out_enm;
360 }
361
362 /* make sure there is only one match */
363 assert(!udev_list_entry_get_next(l));
364
365 dev = udev_device_new_from_syspath(udev, udev_list_entry_get_name(l));
366 if (!dev) {
367 r = -ENOMEM;
368 goto out_enm;
369 }
370
371 *pid = udev_device_get_sysname(dev);
372
373 udev_device_unref(dev);
374 out_enm:
375 udev_enumerate_unref(enm);
376 return r;
377 }
378
379 static int spec_to_devno_and_krbd_id(struct udev *udev, const char *pool,
380 const char *image, const char *snap,
381 dev_t *pdevno, string *pid)
382 {
383 struct udev_enumerate *enm;
384 struct udev_list_entry *l;
385 struct udev_device *dev;
386 unsigned int maj, min = 0;
387 string err;
388 int r;
389
390 enm = udev_enumerate_new(udev);
391 if (!enm)
392 return -ENOMEM;
393
394 r = udev_enumerate_add_match_subsystem(enm, "rbd");
395 if (r < 0)
396 goto out_enm;
397
398 r = udev_enumerate_add_match_sysattr(enm, "pool", pool);
399 if (r < 0)
400 goto out_enm;
401
402 r = udev_enumerate_add_match_sysattr(enm, "name", image);
403 if (r < 0)
404 goto out_enm;
405
406 r = udev_enumerate_add_match_sysattr(enm, "current_snap", snap);
407 if (r < 0)
408 goto out_enm;
409
410 r = udev_enumerate_scan_devices(enm);
411 if (r < 0)
412 goto out_enm;
413
414 l = udev_enumerate_get_list_entry(enm);
415 if (!l) {
416 r = -ENOENT;
417 goto out_enm;
418 }
419
420 dev = udev_device_new_from_syspath(udev, udev_list_entry_get_name(l));
421 if (!dev) {
422 r = -ENOMEM;
423 goto out_enm;
424 }
425
426 maj = strict_strtoll(udev_device_get_sysattr_value(dev, "major"), 10, &err);
427 if (!err.empty()) {
428 cerr << "rbd: couldn't parse major: " << err << std::endl;
429 r = -EINVAL;
430 goto out_dev;
431 }
432 if (have_minor_attr()) {
433 min = strict_strtoll(udev_device_get_sysattr_value(dev, "minor"), 10, &err);
434 if (!err.empty()) {
435 cerr << "rbd: couldn't parse minor: " << err << std::endl;
436 r = -EINVAL;
437 goto out_dev;
438 }
439 }
440
441 /*
442 * If an image is mapped more than once don't bother trying to unmap
443 * all devices - let users run unmap the same number of times they
444 * ran map.
445 */
446 if (udev_list_entry_get_next(l))
447 cerr << "rbd: " << pool << "/" << image << "@" << snap
448 << ": mapped more than once, unmapping "
449 << get_kernel_rbd_name(udev_device_get_sysname(dev))
450 << " only" << std::endl;
451
452 *pdevno = makedev(maj, min);
453 *pid = udev_device_get_sysname(dev);
454
455 out_dev:
456 udev_device_unref(dev);
457 out_enm:
458 udev_enumerate_unref(enm);
459 return r;
460 }
461
462 static string build_unmap_buf(const string& id, const char *options)
463 {
464 string buf(id);
465 if (strcmp(options, "") != 0) {
466 buf += " ";
467 buf += options;
468 }
469 return buf;
470 }
471
472 static int wait_for_udev_remove(struct udev_monitor *mon, dev_t devno)
473 {
474 for (;;) {
475 struct pollfd fds[1];
476 struct udev_device *dev;
477
478 fds[0].fd = udev_monitor_get_fd(mon);
479 fds[0].events = POLLIN;
480 if (poll(fds, 1, POLL_TIMEOUT) < 0)
481 return -errno;
482
483 dev = udev_monitor_receive_device(mon);
484 if (!dev)
485 continue;
486
487 if (strcmp(udev_device_get_action(dev), "remove") == 0 &&
488 udev_device_get_devnum(dev) == devno) {
489 udev_device_unref(dev);
490 break;
491 }
492
493 udev_device_unref(dev);
494 }
495
496 return 0;
497 }
498
499 static int do_unmap(struct udev *udev, dev_t devno, const string& buf)
500 {
501 struct udev_monitor *mon;
502 int r;
503
504 mon = udev_monitor_new_from_netlink(udev, "udev");
505 if (!mon)
506 return -ENOMEM;
507
508 r = udev_monitor_filter_add_match_subsystem_devtype(mon, "block", "disk");
509 if (r < 0)
510 goto out_mon;
511
512 r = udev_monitor_enable_receiving(mon);
513 if (r < 0)
514 goto out_mon;
515
516 /*
517 * On final device close(), kernel sends a block change event, in
518 * response to which udev apparently runs blkid on the device. This
519 * makes unmap fail with EBUSY, if issued right after final close().
520 * Try to circumvent this with a retry before turning to udev.
521 */
522 for (int tries = 0; ; tries++) {
523 r = sysfs_write_rbd_remove(buf);
524 if (r >= 0) {
525 break;
526 } else if (r == -EBUSY && tries < 2) {
527 if (!tries) {
528 usleep(250 * 1000);
529 } else {
530 /*
531 * libudev does not provide the "wait until the queue is empty"
532 * API or the sufficient amount of primitives to build it from.
533 */
534 string err = run_cmd("udevadm", "settle", "--timeout", "10", NULL);
535 if (!err.empty())
536 cerr << "rbd: " << err << std::endl;
537 }
538 } else {
539 cerr << "rbd: sysfs write failed" << std::endl;
540 goto out_mon;
541 }
542 }
543
544 r = wait_for_udev_remove(mon, devno);
545 if (r < 0) {
546 cerr << "rbd: wait failed" << std::endl;
547 goto out_mon;
548 }
549
550 out_mon:
551 udev_monitor_unref(mon);
552 return r;
553 }
554
555 static int unmap_image(struct krbd_ctx *ctx, const char *devnode,
556 const char *options)
557 {
558 struct stat sb;
559 dev_t wholedevno = 0;
560 string id;
561 int r;
562
563 if (stat(devnode, &sb) < 0 || !S_ISBLK(sb.st_mode)) {
564 cerr << "rbd: '" << devnode << "' is not a block device" << std::endl;
565 return -EINVAL;
566 }
567
568 r = blkid_devno_to_wholedisk(sb.st_rdev, NULL, 0, &wholedevno);
569 if (r < 0) {
570 cerr << "rbd: couldn't compute wholedevno: " << cpp_strerror(r)
571 << std::endl;
572 /*
573 * Ignore the error: we are given whole disks most of the time, and
574 * if it turns out this is a partition we will fail later anyway.
575 */
576 wholedevno = sb.st_rdev;
577 }
578
579 r = devno_to_krbd_id(ctx->udev, wholedevno, &id);
580 if (r < 0) {
581 if (r == -ENOENT) {
582 cerr << "rbd: '" << devnode << "' is not an rbd device" << std::endl;
583 r = -EINVAL;
584 }
585 return r;
586 }
587
588 return do_unmap(ctx->udev, wholedevno, build_unmap_buf(id, options));
589 }
590
591 static int unmap_image(struct krbd_ctx *ctx, const char *pool,
592 const char *image, const char *snap,
593 const char *options)
594 {
595 dev_t devno = 0;
596 string id;
597 int r;
598
599 if (!snap)
600 snap = "-";
601
602 r = spec_to_devno_and_krbd_id(ctx->udev, pool, image, snap, &devno, &id);
603 if (r < 0) {
604 if (r == -ENOENT) {
605 cerr << "rbd: " << pool << "/" << image << "@" << snap
606 << ": not a mapped image or snapshot" << std::endl;
607 r = -EINVAL;
608 }
609 return r;
610 }
611
612 return do_unmap(ctx->udev, devno, build_unmap_buf(id, options));
613 }
614
615 static bool dump_one_image(Formatter *f, TextTable *tbl,
616 struct udev_device *dev)
617 {
618 const char *id = udev_device_get_sysname(dev);
619 const char *pool = udev_device_get_sysattr_value(dev, "pool");
620 const char *image = udev_device_get_sysattr_value(dev, "name");
621 const char *snap = udev_device_get_sysattr_value(dev, "current_snap");
622 string kname = get_kernel_rbd_name(id);
623
624 if (!pool || !image || !snap)
625 return false;
626
627 if (f) {
628 f->open_object_section(id);
629 f->dump_string("pool", pool);
630 f->dump_string("name", image);
631 f->dump_string("snap", snap);
632 f->dump_string("device", kname);
633 f->close_section();
634 } else {
635 *tbl << id << pool << image << snap << kname << TextTable::endrow;
636 }
637
638 return true;
639 }
640
641 static int do_dump(struct udev *udev, Formatter *f, TextTable *tbl)
642 {
643 struct udev_enumerate *enm;
644 struct udev_list_entry *l;
645 bool have_output = false;
646 int r;
647
648 enm = udev_enumerate_new(udev);
649 if (!enm)
650 return -ENOMEM;
651
652 r = udev_enumerate_add_match_subsystem(enm, "rbd");
653 if (r < 0)
654 goto out_enm;
655
656 r = udev_enumerate_scan_devices(enm);
657 if (r < 0)
658 goto out_enm;
659
660 udev_list_entry_foreach(l, udev_enumerate_get_list_entry(enm)) {
661 struct udev_device *dev;
662
663 dev = udev_device_new_from_syspath(udev, udev_list_entry_get_name(l));
664 if (dev) {
665 have_output |= dump_one_image(f, tbl, dev);
666 udev_device_unref(dev);
667 }
668 }
669
670 r = have_output;
671 out_enm:
672 udev_enumerate_unref(enm);
673 return r;
674 }
675
676 int dump_images(struct krbd_ctx *ctx, Formatter *f)
677 {
678 TextTable tbl;
679 int r;
680
681 if (f) {
682 f->open_object_section("devices");
683 } else {
684 tbl.define_column("id", TextTable::LEFT, TextTable::LEFT);
685 tbl.define_column("pool", TextTable::LEFT, TextTable::LEFT);
686 tbl.define_column("image", TextTable::LEFT, TextTable::LEFT);
687 tbl.define_column("snap", TextTable::LEFT, TextTable::LEFT);
688 tbl.define_column("device", TextTable::LEFT, TextTable::LEFT);
689 }
690
691 r = do_dump(ctx->udev, f, &tbl);
692
693 if (f) {
694 f->close_section();
695 f->flush(cout);
696 } else {
697 if (r > 0)
698 cout << tbl;
699 }
700
701 return r;
702 }
703
704 extern "C" int krbd_create_from_context(rados_config_t cct,
705 struct krbd_ctx **pctx)
706 {
707 struct krbd_ctx *ctx = new struct krbd_ctx();
708
709 ctx->cct = reinterpret_cast<CephContext *>(cct);
710 ctx->udev = udev_new();
711 if (!ctx->udev) {
712 delete ctx;
713 return -ENOMEM;
714 }
715
716 *pctx = ctx;
717 return 0;
718 }
719
720 extern "C" void krbd_destroy(struct krbd_ctx *ctx)
721 {
722 if (!ctx)
723 return;
724
725 udev_unref(ctx->udev);
726
727 delete ctx;
728 }
729
730 extern "C" int krbd_map(struct krbd_ctx *ctx, const char *pool,
731 const char *image, const char *snap,
732 const char *options, char **pdevnode)
733 {
734 string name;
735 char *devnode;
736 int r;
737
738 r = map_image(ctx, pool, image, snap, options, &name);
739 if (r < 0)
740 return r;
741
742 devnode = strdup(name.c_str());
743 if (!devnode)
744 return -ENOMEM;
745
746 *pdevnode = devnode;
747 return r;
748 }
749
750 extern "C" int krbd_unmap(struct krbd_ctx *ctx, const char *devnode,
751 const char *options)
752 {
753 return unmap_image(ctx, devnode, options);
754 }
755
756 extern "C" int krbd_unmap_by_spec(struct krbd_ctx *ctx, const char *pool,
757 const char *image, const char *snap,
758 const char *options)
759 {
760 return unmap_image(ctx, pool, image, snap, options);
761 }
762
763 int krbd_showmapped(struct krbd_ctx *ctx, Formatter *f)
764 {
765 return dump_images(ctx, f);
766 }