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.
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.
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.
17 #include <sys/types.h>
28 #include <sys/ioctl.h>
30 #include <linux/fiemap.h>
34 #include "kerncompat.h"
37 #include "interval_tree_generic.h"
39 static int summarize = 0;
40 static unsigned unit_mode = UNITS_RAW;
41 static char path[PATH_MAX] = { 0, };
42 static char *pathp = path;
43 static char *path_max = &path[PATH_MAX - 1];
45 struct shared_extent {
47 u64 start; /* Start of interval */
48 u64 last; /* Last location _in_ interval */
53 * extent_tree_* functions are defined in the massive interval tree
54 * macro below. This serves to illustrate the api in human-readable
58 extent_tree_insert(struct shared_extent *node, struct rb_root *root);
61 extent_tree_remove(struct shared_extent *node, struct rb_root *root);
63 static struct shared_extent *
64 extent_tree_iter_first(struct rb_root *root,
67 static struct shared_extent *
68 extent_tree_iter_next(struct shared_extent *node,
71 #define START(node) ((node)->start)
72 #define LAST(node) ((node)->last)
74 INTERVAL_TREE_DEFINE(struct shared_extent, rb,
76 START, LAST, static, extent_tree)
78 static int add_shared_extent(u64 start, u64 len, struct rb_root *root)
80 struct shared_extent *sh;
84 sh = calloc(1, sizeof(*sh));
89 sh->last = (start + len - 1);
91 extent_tree_insert(sh, root);
96 static void cleanup_shared_extents(struct rb_root *root)
98 struct shared_extent *s;
99 struct shared_extent *tmp;
104 s = extent_tree_iter_first(root, 0, -1ULL);
106 tmp = extent_tree_iter_next(s, 0, -1ULL);
107 extent_tree_remove(s, root);
117 * Find all extents which overlap 'n', calculate the space
118 * covered by them and remove those nodes from the tree.
120 static u64 count_unique_bytes(struct rb_root *root, struct shared_extent *n)
122 struct shared_extent *tmp;
123 u64 wstart = n->start;
126 dprintf("Count overlaps:");
130 * Expand our search window based on the lastest
131 * overlapping extent. Doing this will allow us to
132 * find all possible overlaps
134 if (wstart > n->start)
139 dprintf(" (%llu, %llu)", n->start, n->last);
142 n = extent_tree_iter_next(n, wstart, wlast);
144 extent_tree_remove(tmp, root);
148 dprintf("; wstart: %llu wlast: %llu total: %llu\n", wstart,
149 wlast, wlast - wstart + 1);
151 return wlast - wstart + 1;
155 * What we want to do here is get a count of shared bytes within the
156 * set of extents we have collected. Specifcally, we don't want to
157 * count any byte more than once, so just adding them up doesn't
160 * For each set of overlapping extents we find the lowest start and
161 * highest end. From there we have the actual number of bytes which is
162 * shared across all of the extents in our set. A sum of each sets
163 * extent length is returned.
165 static void count_shared_bytes(struct rb_root *root, u64 *ret_cnt)
168 struct shared_extent *s = extent_tree_iter_first(root,
176 * Find all extents which overlap 's', calculate the space
177 * covered by them and remove those nodes from the tree.
179 count += count_unique_bytes(root, s);
182 * Since count_unique_bytes will be emptying the tree,
183 * we can grab the first node here
185 s = extent_tree_iter_first(root, 0, -1ULL);
188 BUG_ON(!RB_EMPTY_ROOT(root));
193 /* Track which inodes we've seen for the purposes of hardlink detection. */
195 struct rb_node i_node;
199 static struct rb_root seen_inodes = RB_ROOT;
201 static int cmp_si(struct seen_inode *si, u64 ino, u64 subvol)
205 else if (ino > si->i_ino)
207 if (subvol < si->i_subvol)
209 else if (subvol > si->i_subvol)
214 static int mark_inode_seen(u64 ino, u64 subvol)
217 struct rb_node **p = &seen_inodes.rb_node;
218 struct rb_node *parent = NULL;
219 struct seen_inode *si;
224 si = rb_entry(parent, struct seen_inode, i_node);
225 cmp = cmp_si(si, ino, subvol);
234 si = calloc(1, sizeof(*si));
239 si->i_subvol = subvol;
241 rb_link_node(&si->i_node, parent, p);
242 rb_insert_color(&si->i_node, &seen_inodes);
247 static int inode_seen(u64 ino, u64 subvol)
250 struct rb_node *n = seen_inodes.rb_node;
251 struct seen_inode *si;
254 si = rb_entry(n, struct seen_inode, i_node);
256 cmp = cmp_si(si, ino, subvol);
267 static void clear_seen_inodes(void)
269 struct rb_node *n = rb_first(&seen_inodes);
270 struct seen_inode *si;
273 si = rb_entry(n, struct seen_inode, i_node);
275 rb_erase(&si->i_node, &seen_inodes);
278 n = rb_first(&seen_inodes);
282 const char * const cmd_filesystem_du_usage[] = {
283 "btrfs filesystem du [options] <path> [<path>..]",
284 "Summarize disk usage of each file.",
285 "-h|--human-readable",
286 " human friendly numbers, base 1024 (default)",
287 "-s display only a total for each argument",
292 * Inline extents are skipped because they do not take data space,
293 * delalloc and unknown are skipped because we do not know how much
294 * space they will use yet.
296 #define SKIP_FLAGS (FIEMAP_EXTENT_UNKNOWN|FIEMAP_EXTENT_DELALLOC|FIEMAP_EXTENT_DATA_INLINE)
297 static int du_calc_file_space(int fd, struct rb_root *shared_extents,
298 uint64_t *ret_total, uint64_t *ret_shared)
301 struct fiemap *fiemap = (struct fiemap *)buf;
302 struct fiemap_extent *fm_ext = &fiemap->fm_extents[0];
303 int count = (sizeof(buf) - sizeof(*fiemap)) /
304 sizeof(struct fiemap_extent);
313 memset(fiemap, 0, sizeof(struct fiemap));
316 fiemap->fm_length = ~0ULL;
317 fiemap->fm_extent_count = count;
318 rc = ioctl(fd, FS_IOC_FIEMAP, (unsigned long) fiemap);
324 /* If 0 extents are returned, then more ioctls are not needed */
325 if (fiemap->fm_mapped_extents == 0)
328 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
329 ext_len = fm_ext[i].fe_length;
330 flags = fm_ext[i].fe_flags;
332 if (flags & FIEMAP_EXTENT_LAST)
335 if (flags & SKIP_FLAGS)
338 file_total += ext_len;
339 if (flags & FIEMAP_EXTENT_SHARED) {
340 file_shared += ext_len;
342 if (shared_extents) {
343 ret = add_shared_extent(fm_ext[i].fe_physical,
352 fiemap->fm_start = (fm_ext[i - 1].fe_logical +
353 fm_ext[i - 1].fe_length);
356 *ret_total = file_total;
357 *ret_shared = file_shared;
365 uint64_t bytes_total;
366 uint64_t bytes_shared;
368 struct rb_root shared_extents;
370 #define INIT_DU_DIR_CTXT (struct du_dir_ctxt) { 0ULL, 0ULL, NULL, RB_ROOT }
372 static int du_add_file(const char *filename, int dirfd,
373 struct rb_root *shared_extents, uint64_t *ret_total,
374 uint64_t *ret_shared, int top_level);
376 static int du_walk_dir(struct du_dir_ctxt *ctxt, struct rb_root *shared_extents)
379 struct dirent *entry;
380 DIR *dirstream = ctxt->dirstream;
387 entry = readdir(dirstream);
389 if (strcmp(entry->d_name, ".") == 0
390 || strcmp(entry->d_name, "..") == 0)
393 type = entry->d_type;
394 if (type == DT_REG || type == DT_DIR) {
397 ret = du_add_file(entry->d_name,
399 shared_extents, &tot, &shr,
404 ctxt->bytes_total += tot;
405 ctxt->bytes_shared += shr;
408 } while (entry != NULL);
413 static int du_add_file(const char *filename, int dirfd,
414 struct rb_root *shared_extents, uint64_t *ret_total,
415 uint64_t *ret_shared, int top_level)
417 int ret, len = strlen(filename);
420 struct du_dir_ctxt dir = INIT_DU_DIR_CTXT;
422 uint64_t file_total = 0;
423 uint64_t file_shared = 0;
424 u64 dir_set_shared = 0;
427 DIR *dirstream = NULL;
429 ret = fstatat(dirfd, filename, &st, 0);
435 if (!S_ISREG(st.st_mode) && !S_ISDIR(st.st_mode))
438 if (len > (path_max - pathp)) {
439 fprintf(stderr, "ERROR: Path max exceeded: %s %s\n", path,
446 ret = sprintf(pathp, "%s", filename);
448 ret = sprintf(pathp, "/%s", filename);
451 fd = open_file_or_dir(path, &dirstream);
457 ret = lookup_ino_rootid(fd, &subvol);
461 if (inode_seen(st.st_ino, subvol))
464 ret = mark_inode_seen(st.st_ino, subvol);
468 if (S_ISREG(st.st_mode)) {
469 ret = du_calc_file_space(fd, shared_extents, &file_total,
473 } else if (S_ISDIR(st.st_mode)) {
474 struct rb_root *root = shared_extents;
477 * We collect shared extents in an rb_root, the top
478 * level caller will not pass a root down, so use the
479 * one on our dir context.
482 root = &dir.shared_extents;
486 dir.dirstream = dirstream;
487 ret = du_walk_dir(&dir, root);
491 cleanup_shared_extents(root);
495 file_total = dir.bytes_total;
496 file_shared = dir.bytes_shared;
498 count_shared_bytes(root, &dir_set_shared);
501 if (!summarize || top_level) {
502 u64 excl = file_total - file_shared;
505 u64 set_shared = file_shared;
508 set_shared = dir_set_shared;
510 printf("%s\t%s\t%s\t%s\n",
511 pretty_size_mode(file_total, unit_mode),
512 pretty_size_mode(excl, unit_mode),
513 pretty_size_mode(set_shared, unit_mode),
516 printf("%s\t%s\t\t\t%s\n",
517 pretty_size_mode(file_total, unit_mode),
518 pretty_size_mode(excl, unit_mode), path);
523 *ret_total = file_total;
525 *ret_shared = file_shared;
528 close_file_or_dir(fd, dirstream);
530 /* reset path to just before this element */
536 int cmd_filesystem_du(int argc, char **argv)
538 int ret = 0, error = 0;
544 static const struct option long_options[] = {
545 { "summarize", no_argument, NULL, 's'},
546 { "human-readable", no_argument, NULL, 'h'},
549 int c = getopt_long(argc, argv, "sh", long_options,
556 unit_mode = UNITS_HUMAN;
562 usage(cmd_filesystem_du_usage);
566 if (check_argc_min(argc - optind, 1))
567 usage(cmd_filesystem_du_usage);
569 printf("total\texclusive\tset shared\tfilename\n");
571 for (i = optind; i < argc; i++) {
572 ret = du_add_file(argv[i], AT_FDCWD, NULL, NULL, NULL, 1);
574 fprintf(stderr, "ERROR: can't check space of '%s': %s\n",
575 argv[i], strerror(ret));
579 /* reset hard-link detection for each argument */