btrfs-progs: Introduce function to fix super block total bytes
[platform/upstream/btrfs-progs.git] / cmds-fi-du.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 <sys/types.h>
18 #include <sys/stat.h>
19 #include <unistd.h>
20 #include <stdio.h>
21 #include <stdlib.h>
22 #include <string.h>
23 #include <stdarg.h>
24 #include <getopt.h>
25 #include <fcntl.h>
26 #include <dirent.h>
27
28 #include <sys/ioctl.h>
29 #include <linux/fs.h>
30 #include <linux/version.h>
31 #include <linux/fiemap.h>
32
33 #if !defined(FIEMAP_EXTENT_SHARED) && (HAVE_OWN_FIEMAP_EXTENT_SHARED_DEFINE == 1)
34 #define FIEMAP_EXTENT_SHARED           0x00002000
35 #endif
36
37 #include "utils.h"
38 #include "commands.h"
39 #include "kerncompat.h"
40 #include "rbtree.h"
41
42 #include "interval_tree_generic.h"
43 #include "help.h"
44 #include "fsfeatures.h"
45
46 static int summarize = 0;
47 static unsigned unit_mode = UNITS_RAW;
48 static char path[PATH_MAX] = { 0, };
49 static char *pathp = path;
50 static char *path_max = &path[PATH_MAX - 1];
51
52 struct shared_extent {
53         struct rb_node  rb;
54         u64     start;  /* Start of interval */
55         u64     last;   /* Last location _in_ interval */
56         u64     __subtree_last;
57 };
58
59 /*
60  * extent_tree_* functions are defined in the massive interval tree
61  * macro below. This serves to illustrate the api in human-readable
62  * terms.
63  */
64 static void
65 extent_tree_insert(struct shared_extent *node, struct rb_root *root);
66
67 static void
68 extent_tree_remove(struct shared_extent *node, struct rb_root *root);
69
70 static struct shared_extent *
71 extent_tree_iter_first(struct rb_root *root,
72                        u64 start, u64 last);
73
74 static struct shared_extent *
75 extent_tree_iter_next(struct shared_extent *node,
76                         u64 start, u64 last);
77
78 #define START(node) ((node)->start)
79 #define LAST(node)  ((node)->last)
80
81 INTERVAL_TREE_DEFINE(struct shared_extent, rb,
82                      u64, __subtree_last,
83                      START, LAST, static, extent_tree)
84
85 static int add_shared_extent(u64 start, u64 len, struct rb_root *root)
86 {
87         struct shared_extent *sh;
88
89         ASSERT(len != 0);
90
91         sh = calloc(1, sizeof(*sh));
92         if (!sh)
93                 return -ENOMEM;
94
95         sh->start = start;
96         sh->last = (start + len - 1);
97
98         extent_tree_insert(sh, root);
99
100         return 0;
101 }
102
103 static void cleanup_shared_extents(struct rb_root *root)
104 {
105         struct shared_extent *s;
106         struct shared_extent *tmp;
107
108         if (!root)
109                 return;
110
111         s = extent_tree_iter_first(root, 0, -1ULL);
112         while (s) {
113                 tmp = extent_tree_iter_next(s, 0, -1ULL);
114                 extent_tree_remove(s, root);
115
116                 free(s);
117                 s = tmp;
118         }
119 }
120
121 #define dbgprintf(...)
122
123 /*
124  * Find all extents which overlap 'n', calculate the space
125  * covered by them and remove those nodes from the tree.
126  */
127 static u64 count_unique_bytes(struct rb_root *root, struct shared_extent *n)
128 {
129         struct shared_extent *tmp;
130         u64 wstart = n->start;
131         u64 wlast = n->last;
132
133         dbgprintf("Count overlaps:");
134
135         do {
136                 /*
137                  * Expand our search window based on the lastest
138                  * overlapping extent. Doing this will allow us to
139                  * find all possible overlaps
140                  */
141                 if (wstart > n->start)
142                         wstart = n->start;
143                 if (wlast < n->last)
144                         wlast = n->last;
145
146                 dbgprintf(" (%llu, %llu)", n->start, n->last);
147
148                 tmp = n;
149                 n = extent_tree_iter_next(n, wstart, wlast);
150
151                 extent_tree_remove(tmp, root);
152                 free(tmp);
153         } while (n);
154
155         dbgprintf("; wstart: %llu wlast: %llu total: %llu\n", wstart,
156                 wlast, wlast - wstart + 1);
157
158         return wlast - wstart + 1;
159 }
160
161 /*
162  * What we want to do here is get a count of shared bytes within the
163  * set of extents we have collected. Specifically, we don't want to
164  * count any byte more than once, so just adding them up doesn't
165  * work.
166  *
167  * For each set of overlapping extents we find the lowest start and
168  * highest end. From there we have the actual number of bytes which is
169  * shared across all of the extents in our set. A sum of each sets
170  * extent length is returned.
171  */
172 static void count_shared_bytes(struct rb_root *root, u64 *ret_cnt)
173 {
174         u64 count = 0;
175         struct shared_extent *s = extent_tree_iter_first(root,
176                                                          0, -1ULL);
177
178         if (!s)
179                 goto out;
180
181         while (s) {
182                 /*
183                  * Find all extents which overlap 's', calculate the space
184                  * covered by them and remove those nodes from the tree.
185                  */
186                 count += count_unique_bytes(root, s);
187
188                 /*
189                  * Since count_unique_bytes will be emptying the tree,
190                  * we can grab the first node here
191                  */
192                 s = extent_tree_iter_first(root, 0, -1ULL);
193         }
194
195         BUG_ON(!RB_EMPTY_ROOT(root));
196 out:
197         *ret_cnt = count;
198 }
199
200 /* Track which inodes we've seen for the purposes of hardlink detection. */
201 struct seen_inode {
202         struct rb_node  i_node;
203         u64             i_ino;
204         u64             i_subvol;
205 };
206 static struct rb_root seen_inodes = RB_ROOT;
207
208 static int cmp_si(struct seen_inode *si, u64 ino, u64 subvol)
209 {
210         if (ino < si->i_ino)
211                 return -1;
212         else if (ino > si->i_ino)
213                 return 1;
214         if (subvol < si->i_subvol)
215                 return -1;
216         else if (subvol > si->i_subvol)
217                 return 1;
218         return 0;
219 }
220
221 static int mark_inode_seen(u64 ino, u64 subvol)
222 {
223         int cmp;
224         struct rb_node **p = &seen_inodes.rb_node;
225         struct rb_node *parent = NULL;
226         struct seen_inode *si;
227
228         while (*p) {
229                 parent = *p;
230
231                 si = rb_entry(parent, struct seen_inode, i_node);
232                 cmp = cmp_si(si, ino, subvol);
233                 if (cmp < 0)
234                         p = &(*p)->rb_left;
235                 else if (cmp > 0)
236                         p = &(*p)->rb_right;
237                 else
238                         return -EEXIST;
239         }
240
241         si = calloc(1, sizeof(*si));
242         if (!si)
243                 return -ENOMEM;
244
245         si->i_ino = ino;
246         si->i_subvol = subvol;
247
248         rb_link_node(&si->i_node, parent, p);
249         rb_insert_color(&si->i_node, &seen_inodes);
250
251         return 0;
252 }
253
254 static int inode_seen(u64 ino, u64 subvol)
255 {
256         int cmp;
257         struct rb_node *n = seen_inodes.rb_node;
258         struct seen_inode *si;
259
260         while (n) {
261                 si = rb_entry(n, struct seen_inode, i_node);
262
263                 cmp = cmp_si(si, ino, subvol);
264                 if (cmp < 0)
265                         n = n->rb_left;
266                 else if (cmp > 0)
267                         n = n->rb_right;
268                 else
269                         return -EEXIST;
270         }
271         return 0;
272 }
273
274 static void clear_seen_inodes(void)
275 {
276         struct rb_node *n = rb_first(&seen_inodes);
277         struct seen_inode *si;
278
279         while (n) {
280                 si = rb_entry(n, struct seen_inode, i_node);
281
282                 rb_erase(&si->i_node, &seen_inodes);
283                 free(si);
284
285                 n = rb_first(&seen_inodes);
286         }
287 }
288
289 /*
290  * Inline extents are skipped because they do not take data space,
291  * delalloc and unknown are skipped because we do not know how much
292  * space they will use yet.
293  */
294 #define SKIP_FLAGS      (FIEMAP_EXTENT_UNKNOWN|FIEMAP_EXTENT_DELALLOC|FIEMAP_EXTENT_DATA_INLINE)
295 static int du_calc_file_space(int fd, struct rb_root *shared_extents,
296                               u64 *ret_total, u64 *ret_shared)
297 {
298         char buf[16384];
299         struct fiemap *fiemap = (struct fiemap *)buf;
300         struct fiemap_extent *fm_ext = &fiemap->fm_extents[0];
301         int count = (sizeof(buf) - sizeof(*fiemap)) /
302                         sizeof(struct fiemap_extent);
303         unsigned int i, ret;
304         int last = 0;
305         int rc;
306         u64 ext_len;
307         u64 file_total = 0;
308         u64 file_shared = 0;
309         u32 flags;
310
311         memset(fiemap, 0, sizeof(struct fiemap));
312
313         do {
314                 fiemap->fm_length = ~0ULL;
315                 fiemap->fm_extent_count = count;
316                 rc = ioctl(fd, FS_IOC_FIEMAP, (unsigned long) fiemap);
317                 if (rc < 0) {
318                         ret = -errno;
319                         goto out;
320                 }
321
322                 /* If 0 extents are returned, then more ioctls are not needed */
323                 if (fiemap->fm_mapped_extents == 0)
324                         break;
325
326                 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
327                         ext_len = fm_ext[i].fe_length;
328                         flags = fm_ext[i].fe_flags;
329
330                         if (flags & FIEMAP_EXTENT_LAST)
331                                 last = 1;
332
333                         if (flags & SKIP_FLAGS)
334                                 continue;
335
336                         if (ext_len == 0) {
337                                 warning("extent %llu has length 0, skipping",
338                                         (unsigned long long)fm_ext[i].fe_physical);
339                                 continue;
340                         }
341
342                         file_total += ext_len;
343                         if (flags & FIEMAP_EXTENT_SHARED) {
344                                 file_shared += ext_len;
345
346                                 if (shared_extents) {
347                                         ret = add_shared_extent(fm_ext[i].fe_physical,
348                                                                 ext_len,
349                                                                 shared_extents);
350                                         if (ret)
351                                                 goto out;
352                                 }
353                         }
354                 }
355
356                 fiemap->fm_start = (fm_ext[i - 1].fe_logical +
357                                     fm_ext[i - 1].fe_length);
358         } while (last == 0);
359
360         *ret_total = file_total;
361         *ret_shared = file_shared;
362
363         ret = 0;
364 out:
365         return ret;
366 }
367
368 struct du_dir_ctxt {
369         u64             bytes_total;
370         u64             bytes_shared;
371         DIR             *dirstream;
372         struct rb_root  shared_extents;
373 };
374 #define INIT_DU_DIR_CTXT        (struct du_dir_ctxt) { 0ULL, 0ULL, NULL, RB_ROOT }
375
376 static int du_add_file(const char *filename, int dirfd,
377                        struct rb_root *shared_extents, u64 *ret_total,
378                        u64 *ret_shared, int top_level);
379
380 static int du_walk_dir(struct du_dir_ctxt *ctxt, struct rb_root *shared_extents)
381 {
382         int ret, type;
383         struct dirent *entry;
384         DIR *dirstream = ctxt->dirstream;
385
386         ret = 0;
387         do {
388                 u64 tot, shr;
389
390                 errno = 0;
391                 entry = readdir(dirstream);
392                 if (entry) {
393                         if (strcmp(entry->d_name, ".") == 0
394                             || strcmp(entry->d_name, "..") == 0)
395                                 continue;
396
397                         type = entry->d_type;
398                         if (type == DT_REG || type == DT_DIR) {
399                                 tot = shr = 0;
400
401                                 ret = du_add_file(entry->d_name,
402                                                   dirfd(dirstream),
403                                                   shared_extents, &tot, &shr,
404                                                   0);
405                                 if (ret == -ENOTTY) {
406                                         ret = 0;
407                                         continue;
408                                 } else if (ret) {
409                                         fprintf(stderr,
410                                                 "failed to walk dir/file: %s :%s\n",
411                                                 entry->d_name, strerror(-ret));
412                                         break;
413                                 }
414
415                                 ctxt->bytes_total += tot;
416                                 ctxt->bytes_shared += shr;
417                         }
418                 }
419         } while (entry != NULL);
420
421         return ret;
422 }
423
424 static int du_add_file(const char *filename, int dirfd,
425                        struct rb_root *shared_extents, u64 *ret_total,
426                        u64 *ret_shared, int top_level)
427 {
428         int ret, len = strlen(filename);
429         char *pathtmp;
430         struct stat st;
431         struct du_dir_ctxt dir = INIT_DU_DIR_CTXT;
432         int is_dir = 0;
433         u64 file_total = 0;
434         u64 file_shared = 0;
435         u64 dir_set_shared = 0;
436         int fd;
437         DIR *dirstream = NULL;
438
439         ret = fstatat(dirfd, filename, &st, 0);
440         if (ret)
441                 return -errno;
442
443         if (!S_ISREG(st.st_mode) && !S_ISDIR(st.st_mode))
444                 return 0;
445
446         if (len > (path_max - pathp)) {
447                 error("path too long: %s %s", path, filename);
448                 return -ENAMETOOLONG;
449         }
450
451         pathtmp = pathp;
452         if (pathp == path)
453                 ret = sprintf(pathp, "%s", filename);
454         else
455                 ret = sprintf(pathp, "/%s", filename);
456         pathp += ret;
457
458         fd = open_file_or_dir3(path, &dirstream, O_RDONLY);
459         if (fd < 0) {
460                 ret = -errno;
461                 goto out;
462         }
463
464         /*
465          * If st.st_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ==2, there is no any
466          * related tree
467          */
468         if (st.st_ino != BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
469                 u64 subvol;
470
471                 ret = lookup_path_rootid(fd, &subvol);
472                 if (ret)
473                         goto out_close;
474
475                 if (inode_seen(st.st_ino, subvol))
476                         goto out_close;
477
478                 ret = mark_inode_seen(st.st_ino, subvol);
479                 if (ret)
480                         goto out_close;
481         }
482
483         if (S_ISREG(st.st_mode)) {
484                 ret = du_calc_file_space(fd, shared_extents, &file_total,
485                                          &file_shared);
486                 if (ret)
487                         goto out_close;
488         } else if (S_ISDIR(st.st_mode)) {
489                 struct rb_root *root = shared_extents;
490
491                 /*
492                  * We collect shared extents in an rb_root, the top
493                  * level caller will not pass a root down, so use the
494                  * one on our dir context.
495                  */
496                 if (top_level)
497                         root = &dir.shared_extents;
498
499                 is_dir = 1;
500
501                 dir.dirstream = dirstream;
502                 ret = du_walk_dir(&dir, root);
503                 *pathp = '\0';
504                 if (ret) {
505                         if (top_level)
506                                 cleanup_shared_extents(root);
507                         goto out_close;
508                 }
509
510                 file_total = dir.bytes_total;
511                 file_shared = dir.bytes_shared;
512                 if (top_level)
513                         count_shared_bytes(root, &dir_set_shared);
514         }
515
516         if (!summarize || top_level) {
517                 u64 excl = file_total - file_shared;
518
519                 if (top_level) {
520                         u64 set_shared = file_shared;
521
522                         if (is_dir)
523                                 set_shared = dir_set_shared;
524
525                         printf("%10s  %10s  %10s  %s\n",
526                                pretty_size_mode(file_total, unit_mode),
527                                pretty_size_mode(excl, unit_mode),
528                                pretty_size_mode(set_shared, unit_mode),
529                                path);
530                 } else {
531                         printf("%10s  %10s  %10s  %s\n",
532                                pretty_size_mode(file_total, unit_mode),
533                                pretty_size_mode(excl, unit_mode),
534                                "-", path);
535                 }
536         }
537
538         if (ret_total)
539                 *ret_total = file_total;
540         if (ret_shared)
541                 *ret_shared = file_shared;
542
543 out_close:
544         close_file_or_dir(fd, dirstream);
545 out:
546         /* reset path to just before this element */
547         pathp = pathtmp;
548
549         return ret;
550 }
551
552 const char * const cmd_filesystem_du_usage[] = {
553         "btrfs filesystem du [options] <path> [<path>..]",
554         "Summarize disk usage of each file.",
555         "-s|--summarize     display only a total for each argument",
556         HELPINFO_UNITS_LONG,
557         NULL
558 };
559
560 int cmd_filesystem_du(int argc, char **argv)
561 {
562         int ret = 0, err = 0;
563         int i;
564         u32 kernel_version;
565
566         unit_mode = get_unit_mode_from_arg(&argc, argv, 1);
567
568         while (1) {
569                 static const struct option long_options[] = {
570                         { "summarize", no_argument, NULL, 's'},
571                         { NULL, 0, NULL, 0 }
572                 };
573                 int c = getopt_long(argc, argv, "s", long_options, NULL);
574
575                 if (c < 0)
576                         break;
577                 switch (c) {
578                 case 's':
579                         summarize = 1;
580                         break;
581                 default:
582                         usage(cmd_filesystem_du_usage);
583                 }
584         }
585
586         if (check_argc_min(argc - optind, 1))
587                 usage(cmd_filesystem_du_usage);
588
589         kernel_version = get_running_kernel_version();
590
591         if (kernel_version < KERNEL_VERSION(2,6,33)) {
592                 warning(
593 "old kernel version detected, shared space will be reported as exclusive\n"
594 "due to missing support for FIEMAP_EXTENT_SHARED flag");
595         }
596
597         printf("%10s  %10s  %10s  %s\n", "Total", "Exclusive", "Set shared",
598                         "Filename");
599
600         for (i = optind; i < argc; i++) {
601                 ret = du_add_file(argv[i], AT_FDCWD, NULL, NULL, NULL, 1);
602                 if (ret) {
603                         error("cannot check space of '%s': %s", argv[i],
604                                         strerror(-ret));
605                         err = 1;
606                 }
607
608                 /* reset hard-link detection for each argument */
609                 clear_seen_inodes();
610         }
611
612         return err;
613 }