xref: /freebsd/lib/geom/virstor/geom_virstor.c (revision e754909cb0aeaf759cddf79c14a04a42f8d894bc)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2005 Ivan Voras <ivoras@freebsd.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/param.h>
29 #include <errno.h>
30 #include <paths.h>
31 #include <stdio.h>
32 #include <stdlib.h>
33 #include <stdint.h>
34 #include <string.h>
35 #include <strings.h>
36 #include <fcntl.h>
37 #include <unistd.h>
38 #include <libgeom.h>
39 #include <err.h>
40 #include <assert.h>
41 
42 #include <core/geom.h>
43 #include <misc/subr.h>
44 
45 #include <geom/virstor/g_virstor_md.h>
46 #include <geom/virstor/g_virstor.h>
47 
48 uint32_t lib_version = G_LIB_VERSION;
49 uint32_t version = G_VIRSTOR_VERSION;
50 
51 #define	GVIRSTOR_CHUNK_SIZE	"4M"
52 #define	GVIRSTOR_VIR_SIZE	"2T"
53 
54 #if G_LIB_VERSION == 1
55 /* Support RELENG_6 */
56 #define G_TYPE_BOOL G_TYPE_NONE
57 #endif
58 
59 /*
60  * virstor_main gets called by the geom(8) utility
61  */
62 static void virstor_main(struct gctl_req *req, unsigned flags);
63 
64 struct g_command class_commands[] = {
65 	{ "clear", G_FLAG_VERBOSE, virstor_main, G_NULL_OPTS,
66 	    "[-v] prov ..."
67 	},
68 	{ "dump", 0, virstor_main, G_NULL_OPTS,
69 	    "prov ..."
70 	},
71 	{ "label", G_FLAG_VERBOSE | G_FLAG_LOADKLD, virstor_main,
72 	    {
73 		{ 'h', "hardcode", NULL, G_TYPE_BOOL},
74 		{ 'm', "chunk_size", GVIRSTOR_CHUNK_SIZE, G_TYPE_NUMBER},
75 		{ 's', "vir_size", GVIRSTOR_VIR_SIZE, G_TYPE_NUMBER},
76 		G_OPT_SENTINEL
77 	    },
78 	    "[-h] [-v] [-m chunk_size] [-s vir_size] name provider0 [provider1 ...]"
79 	},
80 	{ "destroy", G_FLAG_VERBOSE, NULL,
81 	    {
82 		{ 'f', "force", NULL, G_TYPE_BOOL},
83 		G_OPT_SENTINEL
84 	    },
85 	    "[-fv] name ..."
86 	},
87 	{ "stop", G_FLAG_VERBOSE, NULL,
88 	    {
89 		{ 'f', "force", NULL, G_TYPE_BOOL},
90 		G_OPT_SENTINEL
91 	    },
92 	    "[-fv] name ... (alias for \"destroy\")"
93 	},
94 	{ "add", G_FLAG_VERBOSE, NULL,
95 	    {
96 		{ 'h', "hardcode", NULL, G_TYPE_BOOL},
97 		G_OPT_SENTINEL
98 	    },
99 	    "[-vh] name prov [prov ...]"
100 	},
101 	{ "remove", G_FLAG_VERBOSE, NULL, G_NULL_OPTS,
102 	    "[-v] name ..."
103 	},
104 	G_CMD_SENTINEL
105 };
106 
107 static int verbose = 0;
108 
109 /* Helper functions' declarations */
110 static void virstor_clear(struct gctl_req *req);
111 static void virstor_dump(struct gctl_req *req);
112 static void virstor_label(struct gctl_req *req);
113 
114 /* Dispatcher function (no real work done here, only verbose flag recorder) */
115 static void
virstor_main(struct gctl_req * req,unsigned flags)116 virstor_main(struct gctl_req *req, unsigned flags)
117 {
118 	const char *name;
119 
120 	if ((flags & G_FLAG_VERBOSE) != 0)
121 		verbose = 1;
122 
123 	name = gctl_get_ascii(req, "verb");
124 	if (name == NULL) {
125 		gctl_error(req, "No '%s' argument.", "verb");
126 		return;
127 	}
128 	if (strcmp(name, "label") == 0)
129 		virstor_label(req);
130 	else if (strcmp(name, "clear") == 0)
131 		virstor_clear(req);
132 	else if (strcmp(name, "dump") == 0)
133 		virstor_dump(req);
134 	else
135 		gctl_error(req, "%s: Unknown command: %s.", __func__, name);
136 
137 	/* No CTASSERT in userland
138 	CTASSERT(VIRSTOR_MAP_BLOCK_ENTRIES*VIRSTOR_MAP_ENTRY_SIZE == MAXPHYS);
139 	*/
140 }
141 
142 /*
143  * Labels a new geom Meaning: parses and checks the parameters, calculates &
144  * writes metadata to the relevant providers so when the next round of
145  * "tasting" comes (which will be just after the provider(s) are closed) geom
146  * can be instantiated with the tasted metadata.
147  */
148 static void
virstor_label(struct gctl_req * req)149 virstor_label(struct gctl_req *req)
150 {
151 	struct g_virstor_metadata md;
152 	off_t msize;
153 	unsigned char *sect;
154 	unsigned int i;
155 	size_t ssize, secsize;
156 	const char *name;
157 	char param[32];
158 	int hardcode, nargs, error;
159 	struct virstor_map_entry *map;
160 	size_t total_chunks, write_max_map_entries;
161 	unsigned int map_chunks; /* Chunks needed by the map (map size). */
162 	size_t map_size;	/* In bytes. */
163 	ssize_t written;
164 	int fd;
165 
166 	nargs = gctl_get_int(req, "nargs");
167 	if (nargs < 2) {
168 		gctl_error(req, "Too few arguments (%d): expecting: name "
169 		    "provider0 [provider1 ...]", nargs);
170 		return;
171 	}
172 
173 	hardcode = gctl_get_int(req, "hardcode");
174 
175 	/*
176 	 * Initialize constant parts of metadata: magic signature, version,
177 	 * name.
178 	 */
179 	bzero(&md, sizeof(md));
180 	strlcpy(md.md_magic, G_VIRSTOR_MAGIC, sizeof(md.md_magic));
181 	md.md_version = G_VIRSTOR_VERSION;
182 	name = gctl_get_ascii(req, "arg0");
183 	if (name == NULL) {
184 		gctl_error(req, "No 'arg%u' argument.", 0);
185 		return;
186 	}
187 	strlcpy(md.md_name, name, sizeof(md.md_name));
188 
189 	md.md_virsize = (off_t)gctl_get_intmax(req, "vir_size");
190 	md.md_chunk_size = gctl_get_intmax(req, "chunk_size");
191 	md.md_count = nargs - 1;
192 
193 	if (md.md_virsize == 0 || md.md_chunk_size == 0) {
194 		gctl_error(req, "Virtual size and chunk size must be non-zero");
195 		return;
196 	}
197 
198 	msize = secsize = 0;
199 	for (i = 1; i < (unsigned)nargs; i++) {
200 		snprintf(param, sizeof(param), "arg%u", i);
201 		name = gctl_get_ascii(req, "%s", param);
202 		ssize = g_get_sectorsize(name);
203 		if (ssize == 0)
204 			fprintf(stderr, "%s for %s\n", strerror(errno), name);
205 		msize += g_get_mediasize(name);
206 		if (secsize == 0)
207 			secsize = ssize;
208 		else if (secsize != ssize) {
209 			gctl_error(req, "Devices need to have same sector size "
210 			    "(%u on %s needs to be %u).",
211 			    (u_int)ssize, name, (u_int)secsize);
212 			return;
213 		}
214 	}
215 
216 	if (secsize == 0) {
217 		gctl_error(req, "Device not specified");
218 		return;
219 	}
220 
221 	if (md.md_chunk_size % secsize != 0) {
222 		size_t new_size = roundup(md.md_chunk_size, secsize);
223 		fprintf(stderr, "Resizing chunk size to be a multiple of "
224 		    "sector size (%zu bytes).\n", secsize);
225 		fprintf(stderr, "New chunk size: %zu kB\n", new_size / 1024);
226 		md.md_chunk_size = new_size;
227 	}
228 
229 	if (md.md_virsize % md.md_chunk_size != 0) {
230 		off_t chunk_count = md.md_virsize / md.md_chunk_size;
231 		md.md_virsize = chunk_count * md.md_chunk_size;
232 		fprintf(stderr, "Resizing virtual size to be a multiple of "
233 		    "chunk size.\n");
234 		fprintf(stderr, "New virtual size: %zu MB\n",
235 		    (size_t)(md.md_virsize / (1024 * 1024)));
236 	}
237 
238 	total_chunks = md.md_virsize / md.md_chunk_size;
239 	map_size = total_chunks * sizeof(*map);
240 	assert(md.md_virsize % md.md_chunk_size == 0);
241 
242 	ssize = map_size % secsize;
243 	if (ssize != 0) {
244 		size_t add_chunks = (secsize - ssize) / sizeof(*map);
245 		total_chunks += add_chunks;
246 		md.md_virsize = (off_t)total_chunks * (off_t)md.md_chunk_size;
247 		map_size = total_chunks * sizeof(*map);
248 		fprintf(stderr, "Resizing virtual size to fit virstor "
249 		    "structures.\n");
250 		fprintf(stderr, "New virtual size: %ju MB (%zu new chunks)\n",
251 		    (uintmax_t)(md.md_virsize / (1024 * 1024)), add_chunks);
252 	}
253 
254 	if (verbose)
255 		printf("Total virtual chunks: %zu (%zu MB each), %ju MB total "
256 		    "virtual size.\n",
257 		    total_chunks, (size_t)(md.md_chunk_size / (1024 * 1024)),
258 		    md.md_virsize/(1024 * 1024));
259 
260 	if ((off_t)md.md_virsize < msize)
261 		fprintf(stderr, "WARNING: Virtual storage size < Physical "
262 		    "available storage (%ju < %ju)\n", md.md_virsize, msize);
263 
264 	/* Clear last sector first to spoil all components if device exists. */
265 	if (verbose)
266 		printf("Clearing metadata on");
267 
268 	for (i = 1; i < (unsigned)nargs; i++) {
269 		snprintf(param, sizeof(param), "arg%u", i);
270 		name = gctl_get_ascii(req, "%s", param);
271 
272 		if (verbose)
273 			printf(" %s", name);
274 
275 		msize = g_get_mediasize(name);
276 		ssize = g_get_sectorsize(name);
277 		if (msize == 0 || ssize == 0) {
278 			gctl_error(req, "Can't retrieve information about "
279 			    "%s: %s.", name, strerror(errno));
280 			return;
281 		}
282 		if (msize < (off_t) MAX(md.md_chunk_size*4, map_size))
283 			gctl_error(req, "Device %s is too small", name);
284 		error = g_metadata_clear(name, NULL);
285 		if (error != 0) {
286 			gctl_error(req, "Can't clear metadata on %s: %s.", name,
287 			    strerror(error));
288 			return;
289 		}
290 	}
291 
292 
293 	/* Write allocation table to the first provider - this needs to be done
294 	 * before metadata is written because when kernel tastes it it's too
295 	 * late */
296 	name = gctl_get_ascii(req, "arg1"); /* device with metadata */
297 	if (verbose)
298 		printf(".\nWriting allocation table to %s...", name);
299 
300 	/* How many chunks does the map occupy? */
301 	map_chunks = map_size/md.md_chunk_size;
302 	if (map_size % md.md_chunk_size != 0)
303 		map_chunks++;
304 	if (verbose) {
305 		printf(" (%zu MB, %d chunks) ", map_size/(1024*1024), map_chunks);
306 		fflush(stdout);
307 	}
308 
309 	if (strncmp(name, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
310 		fd = open(name, O_RDWR);
311 	else {
312 		sprintf(param, "%s%s", _PATH_DEV, name);
313 		fd = open(param, O_RDWR);
314 	}
315 	if (fd < 0) {
316 		gctl_error(req, "Cannot open provider %s to write map", name);
317 		return;
318 	}
319 
320 	/*
321 	 * Initialize and write the map.  Don't malloc the whole map at once,
322 	 * in case it's large.  Use calloc because there might be a need to set
323 	 * up chunk flags in the future.
324 	 */
325 	write_max_map_entries = 1024 * 1024 / sizeof(*map);
326 	if (write_max_map_entries > total_chunks)
327 		write_max_map_entries = total_chunks;
328 	map = calloc(write_max_map_entries, sizeof(*map));
329 	if (map == NULL) {
330 		gctl_error(req,
331 		    "Out of memory (need %zu bytes for allocation map)",
332 		    write_max_map_entries * sizeof(*map));
333 		close(fd);
334 		return;
335 	}
336 	for (size_t chunk = 0; chunk < total_chunks;
337 	    chunk += write_max_map_entries) {
338 		size_t bytes_to_write, entries_to_write;
339 
340 		entries_to_write = total_chunks - chunk;
341 		if (entries_to_write > write_max_map_entries)
342 			entries_to_write = write_max_map_entries;
343 		bytes_to_write = entries_to_write * sizeof(*map);
344 		for (size_t off = 0; off < bytes_to_write; off += written) {
345 			written = write(fd, ((char *)map) + off,
346 			    bytes_to_write - off);
347 			if (written < 0) {
348 				if (verbose) {
349 					fprintf(stderr,
350 					    "\nError writing map at offset "
351 					    "%zu of %zu: %s\n",
352 					    chunk * sizeof(*map) + off,
353 					    map_size, strerror(errno));
354 				}
355 				gctl_error(req,
356 				    "Error writing out allocation map!");
357 				free(map);
358 				close(fd);
359 				return;
360 			}
361 		}
362 	}
363 	free(map);
364 	map = NULL;
365 	close (fd);
366 
367 	if (verbose)
368 		printf("\nStoring metadata on ");
369 
370 	/*
371 	 * ID is randomly generated, unique for a geom. This is used to
372 	 * recognize all providers belonging to one geom.
373 	 */
374 	md.md_id = arc4random();
375 
376 	/* Ok, store metadata. */
377 	for (i = 1; i < (unsigned)nargs; i++) {
378 		snprintf(param, sizeof(param), "arg%u", i);
379 		name = gctl_get_ascii(req, "%s", param);
380 
381 		msize = g_get_mediasize(name);
382 		ssize = g_get_sectorsize(name);
383 
384 		if (verbose)
385 			printf("%s ", name);
386 
387 		/* this provider's position/type in geom */
388 		md.no = i - 1;
389 		/* this provider's size */
390 		md.provsize = msize;
391 		/* chunk allocation info */
392 		md.chunk_count = md.provsize / md.md_chunk_size;
393 		if (verbose)
394 			printf("(%u chunks) ", md.chunk_count);
395 		/* Check to make sure last sector is unused */
396 		if ((off_t)(md.chunk_count * md.md_chunk_size) > (off_t)(msize-ssize))
397 		    md.chunk_count--;
398 		md.chunk_next = 0;
399 		if (i != 1) {
400 			md.chunk_reserved = 0;
401 			md.flags = 0;
402 		} else {
403 			md.chunk_reserved = map_chunks * 2;
404 			md.flags = VIRSTOR_PROVIDER_ALLOCATED |
405 			    VIRSTOR_PROVIDER_CURRENT;
406 			md.chunk_next = md.chunk_reserved;
407 			if (verbose)
408 				printf("(%u reserved) ", md.chunk_reserved);
409 		}
410 
411 		if (!hardcode)
412 			bzero(md.provider, sizeof(md.provider));
413 		else {
414 			/* convert "/dev/something" to "something" */
415 			if (strncmp(name, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0) {
416 				strlcpy(md.provider, name + sizeof(_PATH_DEV) - 1,
417 				    sizeof(md.provider));
418 			} else
419 				strlcpy(md.provider, name, sizeof(md.provider));
420 		}
421 		sect = calloc(ssize, sizeof(unsigned char));
422 		if (sect == NULL)
423 			err(1, "Cannot allocate sector of %zu bytes", ssize);
424 		virstor_metadata_encode(&md, sect);
425 		error = g_metadata_store(name, sect, ssize);
426 		free(sect);
427 		if (error != 0) {
428 			if (verbose)
429 				printf("\n");
430 			fprintf(stderr, "Can't store metadata on %s: %s.\n",
431 			    name, strerror(error));
432 			gctl_error(req,
433 			    "Not fully done (error storing metadata).");
434 			return;
435 		}
436 	}
437 #if 0
438 	if (verbose)
439 		printf("\n");
440 #endif
441 }
442 
443 /* Clears metadata on given provider(s) IF it's owned by us */
444 static void
virstor_clear(struct gctl_req * req)445 virstor_clear(struct gctl_req *req)
446 {
447 	const char *name;
448 	char param[32];
449 	unsigned i;
450 	int nargs, error;
451 	int fd;
452 
453 	nargs = gctl_get_int(req, "nargs");
454 	if (nargs < 1) {
455 		gctl_error(req, "Too few arguments.");
456 		return;
457 	}
458 	for (i = 0; i < (unsigned)nargs; i++) {
459 		snprintf(param, sizeof(param), "arg%u", i);
460 		name = gctl_get_ascii(req, "%s", param);
461 
462 		error = g_metadata_clear(name, G_VIRSTOR_MAGIC);
463 		if (error != 0) {
464 			fprintf(stderr, "Can't clear metadata on %s: %s "
465 			    "(do I own it?)\n", name, strerror(error));
466 			gctl_error(req,
467 			    "Not fully done (can't clear metadata).");
468 			continue;
469 		}
470 		if (strncmp(name, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
471 			fd = open(name, O_RDWR);
472 		else {
473 			sprintf(param, "%s%s", _PATH_DEV, name);
474 			fd = open(param, O_RDWR);
475 		}
476 		if (fd < 0) {
477 			gctl_error(req, "Cannot clear header sector for %s",
478 			    name);
479 			continue;
480 		}
481 		if (verbose)
482 			printf("Metadata cleared on %s.\n", name);
483 	}
484 }
485 
486 /* Print some metadata information */
487 static void
virstor_metadata_dump(const struct g_virstor_metadata * md)488 virstor_metadata_dump(const struct g_virstor_metadata *md)
489 {
490 	printf("          Magic string: %s\n", md->md_magic);
491 	printf("      Metadata version: %u\n", (u_int) md->md_version);
492 	printf("           Device name: %s\n", md->md_name);
493 	printf("             Device ID: %u\n", (u_int) md->md_id);
494 	printf("        Provider index: %u\n", (u_int) md->no);
495 	printf("      Active providers: %u\n", (u_int) md->md_count);
496 	printf("    Hardcoded provider: %s\n",
497 	    md->provider[0] != '\0' ? md->provider : "(not hardcoded)");
498 	printf("          Virtual size: %u MB\n",
499 	    (unsigned int)(md->md_virsize/(1024 * 1024)));
500 	printf("            Chunk size: %u kB\n", md->md_chunk_size / 1024);
501 	printf("    Chunks on provider: %u\n", md->chunk_count);
502 	printf("           Chunks free: %u\n", md->chunk_count - md->chunk_next);
503 	printf("       Reserved chunks: %u\n", md->chunk_reserved);
504 }
505 
506 /* Called by geom(8) via gvirstor_main() to dump metadata information */
507 static void
virstor_dump(struct gctl_req * req)508 virstor_dump(struct gctl_req *req)
509 {
510 	struct g_virstor_metadata md;
511 	u_char tmpmd[512];	/* temporary buffer */
512 	const char *name;
513 	char param[16];
514 	int nargs, error, i;
515 
516 	assert(sizeof(tmpmd) >= sizeof(md));
517 
518 	nargs = gctl_get_int(req, "nargs");
519 	if (nargs < 1) {
520 		gctl_error(req, "Too few arguments.");
521 		return;
522 	}
523 	for (i = 0; i < nargs; i++) {
524 		snprintf(param, sizeof(param), "arg%u", i);
525 		name = gctl_get_ascii(req, "%s", param);
526 
527 		error = g_metadata_read(name, (u_char *) & tmpmd, sizeof(tmpmd),
528 		    G_VIRSTOR_MAGIC);
529 		if (error != 0) {
530 			fprintf(stderr, "Can't read metadata from %s: %s.\n",
531 			    name, strerror(error));
532 			gctl_error(req,
533 			    "Not fully done (error reading metadata).");
534 			continue;
535 		}
536 		virstor_metadata_decode((u_char *) & tmpmd, &md);
537 		printf("Metadata on %s:\n", name);
538 		virstor_metadata_dump(&md);
539 		printf("\n");
540 	}
541 }
542