btrfs-progs: fi defrag: change default extent target size to 32 MiB
[platform/upstream/btrfs-progs.git] / cmds-inspect.c
1 /*
2  * This program is free software; you can redistribute it and/or
3  * modify it under the terms of the GNU General Public
4  * License v2 as published by the Free Software Foundation.
5  *
6  * This program is distributed in the hope that it will be useful,
7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
9  * General Public License for more details.
10  *
11  * You should have received a copy of the GNU General Public
12  * License along with this program; if not, write to the
13  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
14  * Boston, MA 021110-1307, USA.
15  */
16
17 #include <stdio.h>
18 #include <stdlib.h>
19 #include <unistd.h>
20 #include <stdint.h>
21 #include <sys/ioctl.h>
22 #include <errno.h>
23 #include <getopt.h>
24 #include <limits.h>
25
26 #include "kerncompat.h"
27 #include "ioctl.h"
28 #include "utils.h"
29 #include "ctree.h"
30 #include "send-utils.h"
31 #include "disk-io.h"
32 #include "commands.h"
33 #include "btrfs-list.h"
34 #include "cmds-inspect-dump-tree.h"
35 #include "cmds-inspect-dump-super.h"
36 #include "cmds-inspect-tree-stats.h"
37
38 static const char * const inspect_cmd_group_usage[] = {
39         "btrfs inspect-internal <command> <args>",
40         NULL
41 };
42
43 static int __ino_to_path_fd(u64 inum, int fd, int verbose, const char *prepend)
44 {
45         int ret;
46         int i;
47         struct btrfs_ioctl_ino_path_args ipa;
48         struct btrfs_data_container fspath[PATH_MAX];
49
50         memset(fspath, 0, sizeof(*fspath));
51         ipa.inum = inum;
52         ipa.size = PATH_MAX;
53         ipa.fspath = ptr_to_u64(fspath);
54
55         ret = ioctl(fd, BTRFS_IOC_INO_PATHS, &ipa);
56         if (ret < 0) {
57                 error("ino paths ioctl: %s", strerror(errno));
58                 goto out;
59         }
60
61         if (verbose)
62                 printf("ioctl ret=%d, bytes_left=%lu, bytes_missing=%lu, "
63                         "cnt=%d, missed=%d\n", ret,
64                         (unsigned long)fspath->bytes_left,
65                         (unsigned long)fspath->bytes_missing,
66                         fspath->elem_cnt, fspath->elem_missed);
67
68         for (i = 0; i < fspath->elem_cnt; ++i) {
69                 u64 ptr;
70                 char *str;
71                 ptr = (u64)(unsigned long)fspath->val;
72                 ptr += fspath->val[i];
73                 str = (char *)(unsigned long)ptr;
74                 if (prepend)
75                         printf("%s/%s\n", prepend, str);
76                 else
77                         printf("%s\n", str);
78         }
79
80 out:
81         return !!ret;
82 }
83
84 static const char * const cmd_inspect_inode_resolve_usage[] = {
85         "btrfs inspect-internal inode-resolve [-v] <inode> <path>",
86         "Get file system paths for the given inode",
87         "",
88         "-v   verbose mode",
89         NULL
90 };
91
92 static int cmd_inspect_inode_resolve(int argc, char **argv)
93 {
94         int fd;
95         int verbose = 0;
96         int ret;
97         DIR *dirstream = NULL;
98
99         while (1) {
100                 int c = getopt(argc, argv, "v");
101                 if (c < 0)
102                         break;
103
104                 switch (c) {
105                 case 'v':
106                         verbose = 1;
107                         break;
108                 default:
109                         usage(cmd_inspect_inode_resolve_usage);
110                 }
111         }
112
113         if (check_argc_exact(argc - optind, 2))
114                 usage(cmd_inspect_inode_resolve_usage);
115
116         fd = btrfs_open_dir(argv[optind + 1], &dirstream, 1);
117         if (fd < 0)
118                 return 1;
119
120         ret = __ino_to_path_fd(arg_strtou64(argv[optind]), fd, verbose,
121                                argv[optind+1]);
122         close_file_or_dir(fd, dirstream);
123         return !!ret;
124
125 }
126
127 static const char * const cmd_inspect_logical_resolve_usage[] = {
128         "btrfs inspect-internal logical-resolve [-Pv] [-s bufsize] <logical> <path>",
129         "Get file system paths for the given logical address",
130         "-P          skip the path resolving and print the inodes instead",
131         "-v          verbose mode",
132         "-s bufsize  set inode container's size. This is used to increase inode",
133         "            container's size in case it is not enough to read all the ",
134         "            resolved results. The max value one can set is 64k",
135         NULL
136 };
137
138 static int cmd_inspect_logical_resolve(int argc, char **argv)
139 {
140         int ret;
141         int fd;
142         int i;
143         int verbose = 0;
144         int getpath = 1;
145         int bytes_left;
146         struct btrfs_ioctl_logical_ino_args loi;
147         struct btrfs_data_container *inodes;
148         u64 size = 4096;
149         char full_path[4096];
150         char *path_ptr;
151         DIR *dirstream = NULL;
152
153         while (1) {
154                 int c = getopt(argc, argv, "Pvs:");
155                 if (c < 0)
156                         break;
157
158                 switch (c) {
159                 case 'P':
160                         getpath = 0;
161                         break;
162                 case 'v':
163                         verbose = 1;
164                         break;
165                 case 's':
166                         size = arg_strtou64(optarg);
167                         break;
168                 default:
169                         usage(cmd_inspect_logical_resolve_usage);
170                 }
171         }
172
173         if (check_argc_exact(argc - optind, 2))
174                 usage(cmd_inspect_logical_resolve_usage);
175
176         size = min(size, (u64)64 * 1024);
177         inodes = malloc(size);
178         if (!inodes)
179                 return 1;
180
181         memset(inodes, 0, sizeof(*inodes));
182         loi.logical = arg_strtou64(argv[optind]);
183         loi.size = size;
184         loi.inodes = ptr_to_u64(inodes);
185
186         fd = btrfs_open_dir(argv[optind + 1], &dirstream, 1);
187         if (fd < 0) {
188                 ret = 12;
189                 goto out;
190         }
191
192         ret = ioctl(fd, BTRFS_IOC_LOGICAL_INO, &loi);
193         if (ret < 0) {
194                 error("logical ino ioctl: %s", strerror(errno));
195                 goto out;
196         }
197
198         if (verbose)
199                 printf("ioctl ret=%d, total_size=%llu, bytes_left=%lu, "
200                         "bytes_missing=%lu, cnt=%d, missed=%d\n",
201                         ret, size,
202                         (unsigned long)inodes->bytes_left,
203                         (unsigned long)inodes->bytes_missing,
204                         inodes->elem_cnt, inodes->elem_missed);
205
206         bytes_left = sizeof(full_path);
207         ret = snprintf(full_path, bytes_left, "%s/", argv[optind+1]);
208         path_ptr = full_path + ret;
209         bytes_left -= ret + 1;
210         BUG_ON(bytes_left < 0);
211
212         for (i = 0; i < inodes->elem_cnt; i += 3) {
213                 u64 inum = inodes->val[i];
214                 u64 offset = inodes->val[i+1];
215                 u64 root = inodes->val[i+2];
216                 int path_fd;
217                 char *name;
218                 DIR *dirs = NULL;
219
220                 if (getpath) {
221                         name = btrfs_list_path_for_root(fd, root);
222                         if (IS_ERR(name)) {
223                                 ret = PTR_ERR(name);
224                                 goto out;
225                         }
226                         if (!name) {
227                                 path_ptr[-1] = '\0';
228                                 path_fd = fd;
229                         } else {
230                                 path_ptr[-1] = '/';
231                                 ret = snprintf(path_ptr, bytes_left, "%s",
232                                                 name);
233                                 BUG_ON(ret >= bytes_left);
234                                 free(name);
235                                 path_fd = btrfs_open_dir(full_path, &dirs, 1);
236                                 if (path_fd < 0) {
237                                         ret = -ENOENT;
238                                         goto out;
239                                 }
240                         }
241                         __ino_to_path_fd(inum, path_fd, verbose, full_path);
242                         if (path_fd != fd)
243                                 close_file_or_dir(path_fd, dirs);
244                 } else {
245                         printf("inode %llu offset %llu root %llu\n", inum,
246                                 offset, root);
247                 }
248         }
249
250 out:
251         close_file_or_dir(fd, dirstream);
252         free(inodes);
253         return !!ret;
254 }
255
256 static const char * const cmd_inspect_subvolid_resolve_usage[] = {
257         "btrfs inspect-internal subvolid-resolve <subvolid> <path>",
258         "Get file system paths for the given subvolume ID.",
259         NULL
260 };
261
262 static int cmd_inspect_subvolid_resolve(int argc, char **argv)
263 {
264         int ret;
265         int fd = -1;
266         u64 subvol_id;
267         char path[PATH_MAX];
268         DIR *dirstream = NULL;
269
270         clean_args_no_options(argc, argv, cmd_inspect_subvolid_resolve_usage);
271
272         if (check_argc_exact(argc - optind, 2))
273                 usage(cmd_inspect_subvolid_resolve_usage);
274
275         fd = btrfs_open_dir(argv[optind + 1], &dirstream, 1);
276         if (fd < 0) {
277                 ret = -ENOENT;
278                 goto out;
279         }
280
281         subvol_id = arg_strtou64(argv[optind]);
282         ret = btrfs_subvolid_resolve(fd, path, sizeof(path), subvol_id);
283
284         if (ret) {
285                 error("resolving subvolid %llu error %d",
286                         (unsigned long long)subvol_id, ret);
287                 goto out;
288         }
289
290         path[PATH_MAX - 1] = '\0';
291         printf("%s\n", path);
292
293 out:
294         close_file_or_dir(fd, dirstream);
295         return !!ret;
296 }
297
298 static const char* const cmd_inspect_rootid_usage[] = {
299         "btrfs inspect-internal rootid <path>",
300         "Get tree ID of the containing subvolume of path.",
301         NULL
302 };
303
304 static int cmd_inspect_rootid(int argc, char **argv)
305 {
306         int ret;
307         int fd = -1;
308         u64 rootid;
309         DIR *dirstream = NULL;
310
311         clean_args_no_options(argc, argv, cmd_inspect_rootid_usage);
312
313         if (check_argc_exact(argc - optind, 1))
314                 usage(cmd_inspect_rootid_usage);
315
316         fd = btrfs_open_dir(argv[optind], &dirstream, 1);
317         if (fd < 0) {
318                 ret = -ENOENT;
319                 goto out;
320         }
321
322         ret = lookup_ino_rootid(fd, &rootid);
323         if (ret) {
324                 error("failed to lookup root id: %s", strerror(-ret));
325                 goto out;
326         }
327
328         printf("%llu\n", (unsigned long long)rootid);
329 out:
330         close_file_or_dir(fd, dirstream);
331
332         return !!ret;
333 }
334
335 static const char* const cmd_inspect_min_dev_size_usage[] = {
336         "btrfs inspect-internal min-dev-size [options] <path>",
337         "Get the minimum size the device can be shrunk to. The",
338         "device id 1 is used by default.",
339         "--id DEVID   specify the device id to query",
340         NULL
341 };
342
343 struct dev_extent_elem {
344         u64 start;
345         /* inclusive end */
346         u64 end;
347         struct list_head list;
348 };
349
350 static int add_dev_extent(struct list_head *list,
351                           const u64 start, const u64 end,
352                           const int append)
353 {
354         struct dev_extent_elem *e;
355
356         e = malloc(sizeof(*e));
357         if (!e)
358                 return -ENOMEM;
359
360         e->start = start;
361         e->end = end;
362
363         if (append)
364                 list_add_tail(&e->list, list);
365         else
366                 list_add(&e->list, list);
367
368         return 0;
369 }
370
371 static void free_dev_extent_list(struct list_head *list)
372 {
373         while (!list_empty(list)) {
374                 struct dev_extent_elem *e;
375
376                 e = list_first_entry(list, struct dev_extent_elem, list);
377                 list_del(&e->list);
378                 free(e);
379         }
380 }
381
382 static int hole_includes_sb_mirror(const u64 start, const u64 end)
383 {
384         int i;
385         int ret = 0;
386
387         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
388                 u64 bytenr = btrfs_sb_offset(i);
389
390                 if (bytenr >= start && bytenr <= end) {
391                         ret = 1;
392                         break;
393                 }
394         }
395
396         return ret;
397 }
398
399 static void adjust_dev_min_size(struct list_head *extents,
400                                 struct list_head *holes,
401                                 u64 *min_size)
402 {
403         /*
404          * If relocation of the block group of a device extent must happen (see
405          * below) scratch space is used for the relocation. So track here the
406          * size of the largest device extent that has to be relocated. We track
407          * only the largest and not the sum of the sizes of all relocated block
408          * groups because after each block group is relocated the running
409          * transaction is committed so that pinned space is released.
410          */
411         u64 scratch_space = 0;
412
413         /*
414          * List of device extents is sorted by descending order of the extent's
415          * end offset. If some extent goes beyond the computed minimum size,
416          * which initially matches the sum of the lengths of all extents,
417          * we need to check if the extent can be relocated to an hole in the
418          * device between [0, *min_size[ (which is what the resize ioctl does).
419          */
420         while (!list_empty(extents)) {
421                 struct dev_extent_elem *e;
422                 struct dev_extent_elem *h;
423                 int found = 0;
424                 u64 extent_len;
425                 u64 hole_len = 0;
426
427                 e = list_first_entry(extents, struct dev_extent_elem, list);
428                 if (e->end <= *min_size)
429                         break;
430
431                 /*
432                  * Our extent goes beyond the computed *min_size. See if we can
433                  * find a hole large enough to relocate it to. If not we must stop
434                  * and set *min_size to the end of the extent.
435                  */
436                 extent_len = e->end - e->start + 1;
437                 list_for_each_entry(h, holes, list) {
438                         hole_len = h->end - h->start + 1;
439                         if (hole_len >= extent_len) {
440                                 found = 1;
441                                 break;
442                         }
443                 }
444
445                 if (!found) {
446                         *min_size = e->end + 1;
447                         break;
448                 }
449
450                 /*
451                  * If the hole found contains the location for a superblock
452                  * mirror, we are pessimistic and require allocating one
453                  * more extent of the same size. This is because the block
454                  * group could be in the worst case used by a single extent
455                  * with a size >= (block_group.length - superblock.size).
456                  */
457                 if (hole_includes_sb_mirror(h->start,
458                                             h->start + extent_len - 1))
459                         *min_size += extent_len;
460
461                 if (hole_len > extent_len) {
462                         h->start += extent_len;
463                 } else {
464                         list_del(&h->list);
465                         free(h);
466                 }
467
468                 list_del(&e->list);
469                 free(e);
470
471                 if (extent_len > scratch_space)
472                         scratch_space = extent_len;
473         }
474
475         if (scratch_space) {
476                 *min_size += scratch_space;
477                 /*
478                  * Chunk allocation requires inserting/updating items in the
479                  * chunk tree, so often this can lead to the need of allocating
480                  * a new system chunk too, which has a maximum size of 32Mb.
481                  */
482                 *min_size += 32 * 1024 * 1024;
483         }
484 }
485
486 static int print_min_dev_size(int fd, u64 devid)
487 {
488         int ret = 1;
489         /*
490          * Device allocations starts at 1Mb or at the value passed through the
491          * mount option alloc_start if it's bigger than 1Mb. The alloc_start
492          * option is used for debugging and testing only, and recently the
493          * possibility of deprecating/removing it has been discussed, so we
494          * ignore it here.
495          */
496         u64 min_size = 1 * 1024 * 1024ull;
497         struct btrfs_ioctl_search_args args;
498         struct btrfs_ioctl_search_key *sk = &args.key;
499         u64 last_pos = (u64)-1;
500         LIST_HEAD(extents);
501         LIST_HEAD(holes);
502
503         memset(&args, 0, sizeof(args));
504         sk->tree_id = BTRFS_DEV_TREE_OBJECTID;
505         sk->min_objectid = devid;
506         sk->max_objectid = devid;
507         sk->max_type = BTRFS_DEV_EXTENT_KEY;
508         sk->min_type = BTRFS_DEV_EXTENT_KEY;
509         sk->min_offset = 0;
510         sk->max_offset = (u64)-1;
511         sk->min_transid = 0;
512         sk->max_transid = (u64)-1;
513         sk->nr_items = 4096;
514
515         while (1) {
516                 int i;
517                 struct btrfs_ioctl_search_header *sh;
518                 unsigned long off = 0;
519
520                 ret = ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args);
521                 if (ret < 0) {
522                         error("tree search ioctl: %s", strerror(errno));
523                         ret = 1;
524                         goto out;
525                 }
526
527                 if (sk->nr_items == 0)
528                         break;
529
530                 for (i = 0; i < sk->nr_items; i++) {
531                         struct btrfs_dev_extent *extent;
532                         u64 len;
533
534                         sh = (struct btrfs_ioctl_search_header *)(args.buf +
535                                                                   off);
536                         off += sizeof(*sh);
537                         extent = (struct btrfs_dev_extent *)(args.buf + off);
538                         off += btrfs_search_header_len(sh);
539
540                         sk->min_objectid = btrfs_search_header_objectid(sh);
541                         sk->min_type = btrfs_search_header_type(sh);
542                         sk->min_offset = btrfs_search_header_offset(sh) + 1;
543
544                         if (btrfs_search_header_objectid(sh) != devid ||
545                             btrfs_search_header_type(sh) != BTRFS_DEV_EXTENT_KEY)
546                                 continue;
547
548                         len = btrfs_stack_dev_extent_length(extent);
549                         min_size += len;
550                         ret = add_dev_extent(&extents,
551                                 btrfs_search_header_offset(sh),
552                                 btrfs_search_header_offset(sh) + len - 1, 0);
553
554                         if (!ret && last_pos != (u64)-1 &&
555                             last_pos != btrfs_search_header_offset(sh))
556                                 ret = add_dev_extent(&holes, last_pos,
557                                         btrfs_search_header_offset(sh) - 1, 1);
558                         if (ret) {
559                                 error("add device extent: %s", strerror(-ret));
560                                 ret = 1;
561                                 goto out;
562                         }
563
564                         last_pos = btrfs_search_header_offset(sh) + len;
565                 }
566
567                 if (sk->min_type != BTRFS_DEV_EXTENT_KEY ||
568                     sk->min_objectid != devid)
569                         break;
570         }
571
572         adjust_dev_min_size(&extents, &holes, &min_size);
573         printf("%llu bytes (%s)\n", min_size, pretty_size(min_size));
574         ret = 0;
575 out:
576         free_dev_extent_list(&extents);
577         free_dev_extent_list(&holes);
578
579         return ret;
580 }
581
582 static int cmd_inspect_min_dev_size(int argc, char **argv)
583 {
584         int ret;
585         int fd = -1;
586         DIR *dirstream = NULL;
587         u64 devid = 1;
588
589         while (1) {
590                 int c;
591                 enum { GETOPT_VAL_DEVID = 256 };
592                 static const struct option long_options[] = {
593                         { "id", required_argument, NULL, GETOPT_VAL_DEVID },
594                         {NULL, 0, NULL, 0}
595                 };
596
597                 c = getopt_long(argc, argv, "", long_options, NULL);
598                 if (c < 0)
599                         break;
600
601                 switch (c) {
602                 case GETOPT_VAL_DEVID:
603                         devid = arg_strtou64(optarg);
604                         break;
605                 default:
606                         usage(cmd_inspect_min_dev_size_usage);
607                 }
608         }
609         if (check_argc_exact(argc - optind, 1))
610                 usage(cmd_inspect_min_dev_size_usage);
611
612         fd = btrfs_open_dir(argv[optind], &dirstream, 1);
613         if (fd < 0) {
614                 ret = -ENOENT;
615                 goto out;
616         }
617
618         ret = print_min_dev_size(fd, devid);
619         close_file_or_dir(fd, dirstream);
620 out:
621         return !!ret;
622 }
623
624 static const char inspect_cmd_group_info[] =
625 "query various internal information";
626
627 const struct cmd_group inspect_cmd_group = {
628         inspect_cmd_group_usage, inspect_cmd_group_info, {
629                 { "inode-resolve", cmd_inspect_inode_resolve,
630                         cmd_inspect_inode_resolve_usage, NULL, 0 },
631                 { "logical-resolve", cmd_inspect_logical_resolve,
632                         cmd_inspect_logical_resolve_usage, NULL, 0 },
633                 { "subvolid-resolve", cmd_inspect_subvolid_resolve,
634                         cmd_inspect_subvolid_resolve_usage, NULL, 0 },
635                 { "rootid", cmd_inspect_rootid, cmd_inspect_rootid_usage, NULL,
636                         0 },
637                 { "min-dev-size", cmd_inspect_min_dev_size,
638                         cmd_inspect_min_dev_size_usage, NULL, 0 },
639                 { "dump-tree", cmd_inspect_dump_tree,
640                                 cmd_inspect_dump_tree_usage, NULL, 0 },
641                 { "dump-super", cmd_inspect_dump_super,
642                                 cmd_inspect_dump_super_usage, NULL, 0 },
643                 { "tree-stats", cmd_inspect_tree_stats,
644                                 cmd_inspect_tree_stats_usage, NULL, 0 },
645                 NULL_CMD_STRUCT
646         }
647 };
648
649 int cmd_inspect(int argc, char **argv)
650 {
651         return handle_command_group(&inspect_cmd_group, argc, argv);
652 }