• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /**
2  * libf2fs_zoned.c
3  *
4  * Copyright (c) 2016 Western Digital Corporation.
5  * Written by: Damien Le Moal <damien.lemoal@wdc.com>
6  *
7  * Dual licensed under the GPL or LGPL version 2 licenses.
8  */
9 #ifndef _LARGEFILE64_SOURCE
10 #define _LARGEFILE64_SOURCE
11 #endif
12 
13 #include <f2fs_fs.h>
14 #include <stdio.h>
15 #include <stdlib.h>
16 #include <string.h>
17 #include <errno.h>
18 #include <unistd.h>
19 #include <fcntl.h>
20 #include <sys/stat.h>
21 #ifdef HAVE_SYS_SYSMACROS_H
22 #include <sys/sysmacros.h>
23 #endif
24 #ifdef HAVE_LINUX_LIMITS_H
25 #include <linux/limits.h>
26 #endif
27 #ifndef ANDROID_WINDOWS_HOST
28 #include <sys/ioctl.h>
29 #endif
30 #include <libgen.h>
31 
32 #include <f2fs_fs.h>
33 
34 #ifdef HAVE_LINUX_BLKZONED_H
35 
get_sysfs_path(struct device_info * dev,const char * attr,char * buf,size_t buflen)36 int get_sysfs_path(struct device_info *dev, const char *attr,
37 		   char *buf, size_t buflen)
38 {
39 	struct stat statbuf;
40 	char str[PATH_MAX];
41 	char sysfs_path[PATH_MAX];
42 	ssize_t len;
43 	char *delim;
44 	int ret;
45 
46 	if (stat(dev->path, &statbuf) < 0)
47 		return -1;
48 
49 	snprintf(str, sizeof(str), "/sys/dev/block/%d:%d",
50 		 major(statbuf.st_rdev), minor(statbuf.st_rdev));
51 	len = readlink(str, buf, buflen - 1);
52 	if (len < 0)
53 		return -1;
54 	buf[len] = '\0';
55 
56 	ret = snprintf(sysfs_path, sizeof(sysfs_path),
57 		       "/sys/dev/block/%s", buf);
58 	if (ret >= sizeof(sysfs_path))
59 		return -1;
60 
61 	/* Test if the device is a partition */
62 	ret = snprintf(str, sizeof(str), "%s/partition", sysfs_path);
63 	if (ret >= sizeof(str))
64 		return -1;
65 	ret = stat(str, &statbuf);
66 	if (ret) {
67 		if (errno == ENOENT) {
68 			/* Not a partition */
69 			goto out;
70 		}
71 		return -1;
72 	}
73 
74 	/*
75 	 * The device is a partition: remove the device name from the
76 	 * attribute file path to obtain the sysfs path of the holder device.
77 	 *   e.g.:  /sys/dev/block/.../sda/sda1 -> /sys/dev/block/.../sda
78 	 */
79 	delim = strrchr(sysfs_path, '/');
80 	if (!delim)
81 		return -1;
82 	*delim = '\0';
83 
84 out:
85 	ret = snprintf(buf, buflen, "%s/%s", sysfs_path, attr);
86 	if (ret >= buflen)
87 		return -1;
88 
89 	return 0;
90 }
91 
f2fs_get_zoned_model(int i)92 int f2fs_get_zoned_model(int i)
93 {
94 	struct device_info *dev = c.devices + i;
95 	char str[PATH_MAX];
96 	FILE *file;
97 	int res;
98 
99 	/* Check that this is a zoned block device */
100 	res = get_sysfs_path(dev, "queue/zoned", str, sizeof(str));
101 	if (res != 0) {
102 		MSG(0, "\tInfo: can't find /sys, assuming normal block device\n");
103 		dev->zoned_model = F2FS_ZONED_NONE;
104 		return 0;
105 	}
106 
107 	file = fopen(str, "r");
108 	if (!file) {
109 		/*
110 		 * The kernel does not support zoned block devices, but we have
111 		 * a block device file. This means that if the zoned file is
112 		 * not found, then the device is not zoned or is zoned but can
113 		 * be randomly written (i.e. host-aware zoned model).
114 		 * Treat the device as a regular block device. Otherwise, signal
115 		 * the failure to verify the disk zone model.
116 		 */
117 		if (errno == ENOENT) {
118 			dev->zoned_model = F2FS_ZONED_NONE;
119 			return 0;
120 		}
121 		MSG(0, "\tError: Failed to check the device zoned model\n");
122 		return -1;
123 	}
124 
125 	memset(str, 0, sizeof(str));
126 	res = fscanf(file, "%s", str);
127 	fclose(file);
128 
129 	if (res != 1) {
130 		MSG(0, "\tError: Failed to parse the device zoned model\n");
131 		return -1;
132 	}
133 
134 	if (strcmp(str, "none") == 0) {
135 		/* Regular block device */
136 		dev->zoned_model = F2FS_ZONED_NONE;
137 	} else if (strcmp(str, "host-aware") == 0) {
138 		/* Host-aware zoned block device: can be randomly written */
139 		dev->zoned_model = F2FS_ZONED_HA;
140 	} else if (strcmp(str, "host-managed") == 0) {
141 		/* Host-managed zoned block device: sequential writes needed */
142 		dev->zoned_model = F2FS_ZONED_HM;
143 	} else {
144 		MSG(0, "\tError: Unsupported device zoned model\n");
145 		return -1;
146 	}
147 
148 	return 0;
149 }
150 
f2fs_get_zone_blocks(int i)151 int f2fs_get_zone_blocks(int i)
152 {
153 	struct device_info *dev = c.devices + i;
154 	uint64_t sectors;
155 	char str[PATH_MAX];
156 	FILE *file;
157 	int res;
158 
159 	/* Get zone size */
160 	dev->zone_blocks = 0;
161 
162 	res = get_sysfs_path(dev, "queue/chunk_sectors", str, sizeof(str));
163 	if (res != 0) {
164 		MSG(0, "\tError: Failed to get device sysfs attribute path\n");
165 		return -1;
166 	}
167 
168 	file = fopen(str, "r");
169 	if (!file)
170 		return -1;
171 
172 	memset(str, 0, sizeof(str));
173 	res = fscanf(file, "%s", str);
174 	fclose(file);
175 
176 	if (res != 1)
177 		return -1;
178 
179 	sectors = atol(str);
180 	if (!sectors)
181 		return -1;
182 
183 	dev->zone_blocks = sectors >> (F2FS_BLKSIZE_BITS - 9);
184 	sectors = (sectors << 9) / c.sector_size;
185 
186 	/*
187 	 * Total number of zones: there may
188 	 * be a last smaller runt zone.
189 	 */
190 	dev->nr_zones = dev->total_sectors / sectors;
191 	if (dev->total_sectors % sectors)
192 		dev->nr_zones++;
193 
194 	return 0;
195 }
196 
f2fs_report_zone(int i,u_int64_t sector,void * blkzone)197 int f2fs_report_zone(int i, u_int64_t sector, void *blkzone)
198 {
199 	struct blk_zone *blkz = (struct blk_zone *)blkzone;
200 	struct blk_zone_report *rep;
201 	int ret = -1;
202 
203 	rep = malloc(sizeof(struct blk_zone_report) + sizeof(struct blk_zone));
204 	if (!rep) {
205 		ERR_MSG("No memory for report zones\n");
206 		return -ENOMEM;
207 	}
208 
209 	rep->sector = sector;
210 	rep->nr_zones = 1;
211 	ret = ioctl(c.devices[i].fd, BLKREPORTZONE, rep);
212 	if (ret != 0) {
213 		ret = -errno;
214 		ERR_MSG("ioctl BLKREPORTZONE failed: errno=%d\n", errno);
215 		goto out;
216 	}
217 
218 	*blkz = *(struct blk_zone *)(rep + 1);
219 out:
220 	free(rep);
221 	return ret;
222 }
223 
224 #define F2FS_REPORT_ZONES_BUFSZ	524288
225 
f2fs_report_zones(int j,report_zones_cb_t * report_zones_cb,void * opaque)226 int f2fs_report_zones(int j, report_zones_cb_t *report_zones_cb, void *opaque)
227 {
228 	struct device_info *dev = c.devices + j;
229 	struct blk_zone_report *rep;
230 	struct blk_zone *blkz;
231 	unsigned int i, n = 0;
232 	u_int64_t total_sectors = (dev->total_sectors * c.sector_size)
233 		>> SECTOR_SHIFT;
234 	u_int64_t sector = 0;
235 	int ret = -1;
236 
237 	rep = malloc(F2FS_REPORT_ZONES_BUFSZ);
238 	if (!rep) {
239 		ERR_MSG("No memory for report zones\n");
240 		return -ENOMEM;
241 	}
242 
243 	while (sector < total_sectors) {
244 
245 		/* Get zone info */
246 		rep->sector = sector;
247 		rep->nr_zones = (F2FS_REPORT_ZONES_BUFSZ - sizeof(struct blk_zone_report))
248 			/ sizeof(struct blk_zone);
249 
250 		ret = ioctl(dev->fd, BLKREPORTZONE, rep);
251 		if (ret != 0) {
252 			ret = -errno;
253 			ERR_MSG("ioctl BLKREPORTZONE failed: errno=%d\n",
254 				errno);
255 			goto out;
256 		}
257 
258 		if (!rep->nr_zones) {
259 			ret = -EIO;
260 			ERR_MSG("Unexpected ioctl BLKREPORTZONE result\n");
261 			goto out;
262 		}
263 
264 		blkz = (struct blk_zone *)(rep + 1);
265 		for (i = 0; i < rep->nr_zones; i++) {
266 			ret = report_zones_cb(n, blkz, opaque);
267 			if (ret)
268 				goto out;
269 			sector = blk_zone_sector(blkz) + blk_zone_length(blkz);
270 			n++;
271 			blkz++;
272 		}
273 	}
274 out:
275 	free(rep);
276 	return ret;
277 }
278 
f2fs_check_zones(int j)279 int f2fs_check_zones(int j)
280 {
281 	struct device_info *dev = c.devices + j;
282 	struct blk_zone_report *rep;
283 	struct blk_zone *blkz;
284 	unsigned int i, n = 0;
285 	u_int64_t total_sectors;
286 	u_int64_t sector;
287 	int last_is_conv = 1;
288 	int ret = -1;
289 
290 	rep = malloc(F2FS_REPORT_ZONES_BUFSZ);
291 	if (!rep) {
292 		ERR_MSG("No memory for report zones\n");
293 		return -ENOMEM;
294 	}
295 
296 	dev->zone_cap_blocks = malloc(dev->nr_zones * sizeof(size_t));
297 	if (!dev->zone_cap_blocks) {
298 		ERR_MSG("No memory for zone capacity list.\n");
299 		return -ENOMEM;
300 	}
301 	memset(dev->zone_cap_blocks, 0, (dev->nr_zones * sizeof(size_t)));
302 
303 	dev->nr_rnd_zones = 0;
304 	sector = 0;
305 	total_sectors = (dev->total_sectors * c.sector_size) >> 9;
306 
307 	while (sector < total_sectors) {
308 
309 		/* Get zone info */
310 		memset(rep, 0, F2FS_REPORT_ZONES_BUFSZ);
311 		rep->sector = sector;
312 		rep->nr_zones = (F2FS_REPORT_ZONES_BUFSZ - sizeof(struct blk_zone_report))
313 			/ sizeof(struct blk_zone);
314 
315 		ret = ioctl(dev->fd, BLKREPORTZONE, rep);
316 		if (ret != 0) {
317 			ret = -errno;
318 			ERR_MSG("ioctl BLKREPORTZONE failed\n");
319 			goto out;
320 		}
321 
322 		if (!rep->nr_zones)
323 			break;
324 
325 		blkz = (struct blk_zone *)(rep + 1);
326 		for (i = 0; i < rep->nr_zones && sector < total_sectors; i++) {
327 
328 			if (blk_zone_cond(blkz) == BLK_ZONE_COND_READONLY ||
329 			    blk_zone_cond(blkz) == BLK_ZONE_COND_OFFLINE)
330 				last_is_conv = 0;
331 			if (blk_zone_conv(blkz) ||
332 			    blk_zone_seq_pref(blkz)) {
333 				if (last_is_conv)
334 					dev->nr_rnd_zones++;
335 			} else {
336 				last_is_conv = 0;
337 			}
338 
339 			if (blk_zone_conv(blkz)) {
340 				DBG(2,
341 				    "Zone %05u: Conventional, cond 0x%x (%s), sector %llu, %llu sectors\n",
342 				    n,
343 				    blk_zone_cond(blkz),
344 				    blk_zone_cond_str(blkz),
345 				    blk_zone_sector(blkz),
346 				    blk_zone_length(blkz));
347 				dev->zone_cap_blocks[n] =
348 					blk_zone_length(blkz) >>
349 					(F2FS_BLKSIZE_BITS - SECTOR_SHIFT);
350 			} else {
351 				DBG(2,
352 				    "Zone %05u: type 0x%x (%s), cond 0x%x (%s),"
353 				    " need_reset %d, non_seq %d, sector %llu,"
354 				    " %llu sectors, capacity %llu,"
355 				    " wp sector %llu\n",
356 				    n,
357 				    blk_zone_type(blkz),
358 				    blk_zone_type_str(blkz),
359 				    blk_zone_cond(blkz),
360 				    blk_zone_cond_str(blkz),
361 				    blk_zone_need_reset(blkz),
362 				    blk_zone_non_seq(blkz),
363 				    blk_zone_sector(blkz),
364 				    blk_zone_length(blkz),
365 				    blk_zone_capacity(blkz, rep->flags),
366 				    blk_zone_wp_sector(blkz));
367 				dev->zone_cap_blocks[n] =
368 					blk_zone_capacity(blkz, rep->flags) >>
369 					(F2FS_BLKSIZE_BITS - SECTOR_SHIFT);
370 			}
371 
372 			sector = blk_zone_sector(blkz) + blk_zone_length(blkz);
373 			n++;
374 			blkz++;
375 		}
376 	}
377 
378 	if (sector != total_sectors) {
379 		ERR_MSG("Invalid zones: last sector reported is %llu, expected %llu\n",
380 			(unsigned long long)(sector << 9) / c.sector_size,
381 			(unsigned long long)dev->total_sectors);
382 		ret = -1;
383 		goto out;
384 	}
385 
386 	if (n != dev->nr_zones) {
387 		ERR_MSG("Inconsistent number of zones: expected %u zones, got %u\n",
388 			dev->nr_zones, n);
389 		ret = -1;
390 		goto out;
391 	}
392 
393 	/*
394 	 * For a multi-device volume, fixed position metadata blocks are
395 	 * stored * only on the first device of the volume. Checking for the
396 	 * presence of * conventional zones (randomly writeabl zones) for
397 	 * storing these blocks * on a host-managed device is thus needed only
398 	 * for the device index 0.
399 	 */
400 	if (j == 0 && dev->zoned_model == F2FS_ZONED_HM &&
401 			!dev->nr_rnd_zones) {
402 		ERR_MSG("No conventional zone for super block\n");
403 		ret = -1;
404 	}
405 out:
406 	free(rep);
407 	return ret;
408 }
409 
f2fs_reset_zone(int i,void * blkzone)410 int f2fs_reset_zone(int i, void *blkzone)
411 {
412 	struct blk_zone *blkz = (struct blk_zone *)blkzone;
413 	struct device_info *dev = c.devices + i;
414 	struct blk_zone_range range;
415 	int ret;
416 
417 	if (!blk_zone_seq(blkz) || blk_zone_empty(blkz))
418 		return 0;
419 
420 	/* Non empty sequential zone: reset */
421 	range.sector = blk_zone_sector(blkz);
422 	range.nr_sectors = blk_zone_length(blkz);
423 	ret = ioctl(dev->fd, BLKRESETZONE, &range);
424 	if (ret != 0) {
425 		ret = -errno;
426 		ERR_MSG("ioctl BLKRESETZONE failed: errno=%d\n", errno);
427 	}
428 
429 	return ret;
430 }
431 
f2fs_reset_zones(int j)432 int f2fs_reset_zones(int j)
433 {
434 	struct device_info *dev = c.devices + j;
435 	struct blk_zone_report *rep;
436 	struct blk_zone *blkz;
437 	struct blk_zone_range range;
438 	u_int64_t total_sectors;
439 	u_int64_t sector;
440 	unsigned int i;
441 	int ret = -1;
442 
443 	rep = malloc(F2FS_REPORT_ZONES_BUFSZ);
444 	if (!rep) {
445 		ERR_MSG("No memory for report zones\n");
446 		return -1;
447 	}
448 
449 	sector = 0;
450 	total_sectors = (dev->total_sectors * c.sector_size) >> 9;
451 	while (sector < total_sectors) {
452 
453 		/* Get zone info */
454 		memset(rep, 0, F2FS_REPORT_ZONES_BUFSZ);
455 		rep->sector = sector;
456 		rep->nr_zones = (F2FS_REPORT_ZONES_BUFSZ - sizeof(struct blk_zone_report))
457 			/ sizeof(struct blk_zone);
458 
459 		ret = ioctl(dev->fd, BLKREPORTZONE, rep);
460 		if (ret != 0) {
461 			ret = -errno;
462 			ERR_MSG("ioctl BLKREPORTZONES failed\n");
463 			goto out;
464 		}
465 
466 		if (!rep->nr_zones)
467 			break;
468 
469 		blkz = (struct blk_zone *)(rep + 1);
470 		for (i = 0; i < rep->nr_zones && sector < total_sectors; i++) {
471 			if (blk_zone_seq(blkz) &&
472 			    !blk_zone_empty(blkz)) {
473 				/* Non empty sequential zone: reset */
474 				range.sector = blk_zone_sector(blkz);
475 				range.nr_sectors = blk_zone_length(blkz);
476 				ret = ioctl(dev->fd, BLKRESETZONE, &range);
477 				if (ret != 0) {
478 					ret = -errno;
479 					ERR_MSG("ioctl BLKRESETZONE failed\n");
480 					goto out;
481 				}
482 			}
483 			sector = blk_zone_sector(blkz) + blk_zone_length(blkz);
484 			blkz++;
485 		}
486 	}
487 out:
488 	free(rep);
489 	if (!ret)
490 		MSG(0, "Info: Discarded %"PRIu64" MB\n", (sector << 9) >> 20);
491 	return ret;
492 }
493 
f2fs_get_usable_segments(struct f2fs_super_block * sb)494 uint32_t f2fs_get_usable_segments(struct f2fs_super_block *sb)
495 {
496 #ifdef HAVE_BLK_ZONE_REP_V2
497 	int i, j;
498 	uint32_t usable_segs = 0, zone_segs;
499 
500 	for (i = 0; i < c.ndevs; i++) {
501 		if (c.devices[i].zoned_model != F2FS_ZONED_HM) {
502 			usable_segs += c.devices[i].total_segments;
503 			continue;
504 		}
505 		for (j = 0; j < c.devices[i].nr_zones; j++) {
506 			zone_segs = c.devices[i].zone_cap_blocks[j] >>
507 					get_sb(log_blocks_per_seg);
508 			if (c.devices[i].zone_cap_blocks[j] %
509 						DEFAULT_BLOCKS_PER_SEGMENT)
510 				usable_segs += zone_segs + 1;
511 			else
512 				usable_segs += zone_segs;
513 		}
514 	}
515 	usable_segs -= (get_sb(main_blkaddr) - get_sb(segment0_blkaddr)) >>
516 						get_sb(log_blocks_per_seg);
517 	return usable_segs;
518 #endif
519 	return get_sb(segment_count_main);
520 }
521 
522 #else
523 
f2fs_report_zone(int i,u_int64_t UNUSED (sector),void * UNUSED (blkzone))524 int f2fs_report_zone(int i, u_int64_t UNUSED(sector), void *UNUSED(blkzone))
525 {
526 	ERR_MSG("%d: Unsupported zoned block device\n", i);
527 	return -1;
528 }
529 
f2fs_report_zones(int i,report_zones_cb_t * UNUSED (report_zones_cb),void * UNUSED (opaque))530 int f2fs_report_zones(int i, report_zones_cb_t *UNUSED(report_zones_cb),
531 					void *UNUSED(opaque))
532 {
533 	ERR_MSG("%d: Unsupported zoned block device\n", i);
534 	return -1;
535 }
536 
f2fs_get_zoned_model(int i)537 int f2fs_get_zoned_model(int i)
538 {
539 	struct device_info *dev = c.devices + i;
540 
541 	c.zoned_mode = 0;
542 	dev->zoned_model = F2FS_ZONED_NONE;
543 	return 0;
544 }
545 
f2fs_get_zone_blocks(int i)546 int f2fs_get_zone_blocks(int i)
547 {
548 	struct device_info *dev = c.devices + i;
549 
550 	c.zoned_mode = 0;
551 	dev->nr_zones = 0;
552 	dev->zone_blocks = 0;
553 	dev->zoned_model = F2FS_ZONED_NONE;
554 
555 	return 0;
556 }
557 
f2fs_check_zones(int i)558 int f2fs_check_zones(int i)
559 {
560 	ERR_MSG("%d: Unsupported zoned block device\n", i);
561 	return -1;
562 }
563 
f2fs_reset_zone(int i,void * UNUSED (blkzone))564 int f2fs_reset_zone(int i, void *UNUSED(blkzone))
565 {
566 	ERR_MSG("%d: Unsupported zoned block device\n", i);
567 	return -1;
568 }
569 
f2fs_reset_zones(int i)570 int f2fs_reset_zones(int i)
571 {
572 	ERR_MSG("%d: Unsupported zoned block device\n", i);
573 	return -1;
574 }
575 
f2fs_get_usable_segments(struct f2fs_super_block * sb)576 uint32_t f2fs_get_usable_segments(struct f2fs_super_block *sb)
577 {
578 	return get_sb(segment_count_main);
579 }
580 #endif
581 
582