blob: e795987c187ac473e1c36633b86d07251cae2fb2 [file] [log] [blame]
Theodore Ts'o48364592009-07-09 14:29:12 -04001/*
2 * e4defrag.c - ext4 filesystem defragmenter
3 *
4 * Copyright (C) 2009 NEC Software Tohoku, Ltd.
5 *
6 * Author: Akira Fujita <a-fujita@rs.jp.nec.com>
7 * Takashi Sato <t-sato@yk.jp.nec.com>
8 */
9
10#ifndef _LARGEFILE_SOURCE
11#define _LARGEFILE_SOURCE
12#endif
13
14#ifndef _LARGEFILE64_SOURCE
15#define _LARGEFILE64_SOURCE
16#endif
17
Kazuya Mioae507462009-07-10 16:45:16 +090018#ifndef _GNU_SOURCE
Theodore Ts'o48364592009-07-09 14:29:12 -040019#define _GNU_SOURCE
Kazuya Mioae507462009-07-10 16:45:16 +090020#endif
21
Theodore Ts'o48364592009-07-09 14:29:12 -040022#include <ctype.h>
23#include <dirent.h>
24#include <endian.h>
25#include <errno.h>
26#include <fcntl.h>
27#include <ftw.h>
28#include <limits.h>
29#include <mntent.h>
30#include <stdio.h>
31#include <stdlib.h>
32#include <string.h>
33#include <unistd.h>
34#include <ext2fs/ext2_types.h>
35#include <ext2fs/ext2fs.h>
36#include <linux/fs.h>
37#include <sys/ioctl.h>
Theodore Ts'o88fca202009-08-22 13:16:14 -040038#include <ext2fs/fiemap.h>
Theodore Ts'o48364592009-07-09 14:29:12 -040039#include <sys/mman.h>
40#include <sys/stat.h>
41#include <sys/statfs.h>
42#include <sys/syscall.h>
43#include <sys/vfs.h>
44
Eric Sandeen4f1d0e82009-07-27 10:53:46 -040045/* A relatively new ioctl interface ... */
46#ifndef EXT4_IOC_MOVE_EXT
47#define EXT4_IOC_MOVE_EXT _IOWR('f', 15, struct move_extent)
48#endif
Theodore Ts'o48364592009-07-09 14:29:12 -040049
50/* Macro functions */
51#define PRINT_ERR_MSG(msg) fprintf(stderr, "%s\n", (msg))
52#define IN_FTW_PRINT_ERR_MSG(msg) \
53 fprintf(stderr, "\t%s\t\t[ NG ]\n", (msg))
54#define PRINT_FILE_NAME(file) fprintf(stderr, " \"%s\"\n", (file))
55#define PRINT_ERR_MSG_WITH_ERRNO(msg) \
56 fprintf(stderr, "\t%s:%s\t[ NG ]\n", (msg), strerror(errno))
57#define STATISTIC_ERR_MSG(msg) \
58 fprintf(stderr, "\t%s\n", (msg))
59#define STATISTIC_ERR_MSG_WITH_ERRNO(msg) \
60 fprintf(stderr, "\t%s:%s\n", (msg), strerror(errno))
61#define min(x, y) (((x) > (y)) ? (y) : (x))
Theodore Ts'o48364592009-07-09 14:29:12 -040062#define CALC_SCORE(ratio) \
63 ((ratio) > 10 ? (80 + 20 * (ratio) / 100) : (8 * (ratio)))
64/* Wrap up the free function */
65#define FREE(tmp) \
66 do { \
67 if ((tmp) != NULL) \
68 free(tmp); \
69 } while (0) \
70/* Insert list2 after list1 */
71#define insert(list1, list2) \
72 do { \
73 list2->next = list1->next; \
74 list1->next->prev = list2; \
75 list2->prev = list1; \
76 list1->next = list2; \
77 } while (0)
78
79/* To delete unused warning */
80#ifdef __GNUC__
81#define EXT2FS_ATTR(x) __attribute__(x)
82#else
83#define EXT2FS_ATTR(x)
84#endif
85
Theodore Ts'o48364592009-07-09 14:29:12 -040086/* The mode of defrag */
87#define DETAIL 0x01
88#define STATISTIC 0x02
89
90#define DEVNAME 0
91#define DIRNAME 1
92#define FILENAME 2
93
94#define FTW_OPEN_FD 2000
95
96#define FS_EXT4 "ext4"
97#define ROOT_UID 0
98
99#define BOUND_SCORE 55
100#define SHOW_FRAG_FILES 5
101
102/* Magic number for ext4 */
103#define EXT4_SUPER_MAGIC 0xEF53
104
105/* Definition of flex_bg */
106#define EXT4_FEATURE_INCOMPAT_FLEX_BG 0x0200
107
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400108/* The following macro is used for ioctl FS_IOC_FIEMAP
Theodore Ts'o48364592009-07-09 14:29:12 -0400109 * EXTENT_MAX_COUNT: the maximum number of extents for exchanging between
110 * kernel-space and user-space per ioctl
111 */
Theodore Ts'o48364592009-07-09 14:29:12 -0400112#define EXTENT_MAX_COUNT 512
113
114/* The following macros are error message */
115#define MSG_USAGE \
116"Usage : e4defrag [-v] file...| directory...| device...\n\
117 : e4defrag -c file...| directory...| device...\n"
118
119#define NGMSG_EXT4 "Filesystem is not ext4 filesystem"
120#define NGMSG_FILE_EXTENT "Failed to get file extents"
121#define NGMSG_FILE_INFO "Failed to get file information"
122#define NGMSG_FILE_OPEN "Failed to open"
123#define NGMSG_FILE_UNREG "File is not regular file"
124#define NGMSG_LOST_FOUND "Can not process \"lost+found\""
125
126/* Data type for filesystem-wide blocks number */
127typedef unsigned long long ext4_fsblk_t;
128
129struct fiemap_extent_data {
130 __u64 len; /* blocks count */
131 __u64 logical; /* start logical block number */
132 ext4_fsblk_t physical; /* start physical block number */
133};
134
135struct fiemap_extent_list {
136 struct fiemap_extent_list *prev;
137 struct fiemap_extent_list *next;
138 struct fiemap_extent_data data; /* extent belong to file */
139};
140
141struct fiemap_extent_group {
142 struct fiemap_extent_group *prev;
143 struct fiemap_extent_group *next;
144 __u64 len; /* length of this continuous region */
145 struct fiemap_extent_list *start; /* start ext */
146 struct fiemap_extent_list *end; /* end ext */
147};
148
149struct move_extent {
150 __s32 reserved; /* original file descriptor */
151 __u32 donor_fd; /* donor file descriptor */
152 __u64 orig_start; /* logical start offset in block for orig */
153 __u64 donor_start; /* logical start offset in block for donor */
154 __u64 len; /* block length to be moved */
155 __u64 moved_len; /* moved block length */
156};
157
Theodore Ts'o48364592009-07-09 14:29:12 -0400158struct frag_statistic_ino {
159 int now_count; /* the file's extents count of before defrag */
160 int best_count; /* the best file's extents count */
Kazuya Mio94d26c22010-12-13 09:57:26 -0500161 __u64 size_per_ext; /* size(KB) per extent */
Theodore Ts'o48364592009-07-09 14:29:12 -0400162 float ratio; /* the ratio of fragmentation */
163 char msg_buffer[PATH_MAX + 1]; /* pathname of the file */
164};
165
166typedef __u16 __le16;
167typedef __u32 __le32;
168typedef __u64 __le64;
169
170/*
171 * Structure of the super block
172 */
173struct ext4_super_block {
174/*00*/ __le32 s_inodes_count; /* Inodes count */
175 __le32 s_blocks_count_lo; /* Blocks count */
176 __le32 s_r_blocks_count_lo; /* Reserved blocks count */
177 __le32 s_free_blocks_count_lo; /* Free blocks count */
178/*10*/ __le32 s_free_inodes_count; /* Free inodes count */
179 __le32 s_first_data_block; /* First Data Block */
180 __le32 s_log_block_size; /* Block size */
181 __le32 s_obso_log_frag_size; /* Obsoleted fragment size */
182/*20*/ __le32 s_blocks_per_group; /* # Blocks per group */
183 __le32 s_obso_frags_per_group; /* Obsoleted fragments per group */
184 __le32 s_inodes_per_group; /* # Inodes per group */
185 __le32 s_mtime; /* Mount time */
186/*30*/ __le32 s_wtime; /* Write time */
187 __le16 s_mnt_count; /* Mount count */
188 __le16 s_max_mnt_count; /* Maximal mount count */
189 __le16 s_magic; /* Magic signature */
190 __le16 s_state; /* File system state */
191 __le16 s_errors; /* Behaviour when detecting errors */
192 __le16 s_minor_rev_level; /* minor revision level */
193/*40*/ __le32 s_lastcheck; /* time of last check */
194 __le32 s_checkinterval; /* max. time between checks */
195 __le32 s_creator_os; /* OS */
196 __le32 s_rev_level; /* Revision level */
197/*50*/ __le16 s_def_resuid; /* Default uid for reserved blocks */
198 __le16 s_def_resgid; /* Default gid for reserved blocks */
199 /*
200 * These fields are for EXT4_DYNAMIC_REV superblocks only.
201 *
202 * Note: the difference between the compatible feature set and
203 * the incompatible feature set is that if there is a bit set
204 * in the incompatible feature set that the kernel doesn't
205 * know about, it should refuse to mount the filesystem.
206 *
207 * e2fsck's requirements are more strict; if it doesn't know
208 * about a feature in either the compatible or incompatible
209 * feature set, it must abort and not try to meddle with
210 * things it doesn't understand...
211 */
212 __le32 s_first_ino; /* First non-reserved inode */
213 __le16 s_inode_size; /* size of inode structure */
214 __le16 s_block_group_nr; /* block group # of this superblock */
215 __le32 s_feature_compat; /* compatible feature set */
216/*60*/ __le32 s_feature_incompat; /* incompatible feature set */
217 __le32 s_feature_ro_compat; /* readonly-compatible feature set */
218/*68*/ __u8 s_uuid[16]; /* 128-bit uuid for volume */
219/*78*/ char s_volume_name[16]; /* volume name */
220/*88*/ char s_last_mounted[64]; /* directory where last mounted */
221/*C8*/ __le32 s_algorithm_usage_bitmap; /* For compression */
222 /*
223 * Performance hints. Directory preallocation should only
224 * happen if the EXT4_FEATURE_COMPAT_DIR_PREALLOC flag is on.
225 */
226 __u8 s_prealloc_blocks; /* Nr of blocks to try to preallocate*/
227 __u8 s_prealloc_dir_blocks; /* Nr to preallocate for dirs */
228 __le16 s_reserved_gdt_blocks; /* Per group desc for online growth */
229 /*
230 * Journaling support valid if EXT4_FEATURE_COMPAT_HAS_JOURNAL set.
231 */
232/*D0*/ __u8 s_journal_uuid[16]; /* uuid of journal superblock */
233/*E0*/ __le32 s_journal_inum; /* inode number of journal file */
234 __le32 s_journal_dev; /* device number of journal file */
235 __le32 s_last_orphan; /* start of list of inodes to delete */
236 __le32 s_hash_seed[4]; /* HTREE hash seed */
237 __u8 s_def_hash_version; /* Default hash version to use */
238 __u8 s_reserved_char_pad;
239 __le16 s_desc_size; /* size of group descriptor */
240/*100*/ __le32 s_default_mount_opts;
241 __le32 s_first_meta_bg; /* First metablock block group */
242 __le32 s_mkfs_time; /* When the filesystem was created */
243 __le32 s_jnl_blocks[17]; /* Backup of the journal inode */
244 /* 64bit support valid if EXT4_FEATURE_COMPAT_64BIT */
245/*150*/ __le32 s_blocks_count_hi; /* Blocks count */
246 __le32 s_r_blocks_count_hi; /* Reserved blocks count */
247 __le32 s_free_blocks_count_hi; /* Free blocks count */
248 __le16 s_min_extra_isize; /* All inodes have at least # bytes */
249 __le16 s_want_extra_isize; /* New inodes should reserve # bytes */
250 __le32 s_flags; /* Miscellaneous flags */
251 __le16 s_raid_stride; /* RAID stride */
252 __le16 s_mmp_interval; /* # seconds to wait in MMP checking */
253 __le64 s_mmp_block; /* Block for multi-mount protection */
254 __le32 s_raid_stripe_width; /* blocks on all data disks (N*stride)*/
255 __u8 s_log_groups_per_flex; /* FLEX_BG group size */
256 __u8 s_reserved_char_pad2;
257 __le16 s_reserved_pad;
258 __u32 s_reserved[162]; /* Padding to the end of the block */
259};
260
261char lost_found_dir[PATH_MAX + 1];
262int block_size;
263int extents_before_defrag;
264int extents_after_defrag;
265int mode_flag;
266unsigned int current_uid;
267unsigned int defraged_file_count;
268unsigned int frag_files_before_defrag;
269unsigned int frag_files_after_defrag;
270unsigned int regular_count;
271unsigned int succeed_cnt;
272unsigned int total_count;
273__u8 log_groups_per_flex;
274__le32 blocks_per_group;
275__le32 feature_incompat;
276ext4_fsblk_t files_block_count;
277struct frag_statistic_ino frag_rank[SHOW_FRAG_FILES];
278
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400279
280/* Local definitions of some syscalls glibc may not yet have */
281
282#ifndef HAVE_POSIX_FADVISE
283#warning Using locally defined posix_fadvise interface.
284
285#ifndef __NR_fadvise64_64
286#error Your kernel headers dont define __NR_fadvise64_64
Theodore Ts'o48364592009-07-09 14:29:12 -0400287#endif
Theodore Ts'o48364592009-07-09 14:29:12 -0400288
289/*
290 * fadvise() - Give advice about file access.
291 *
292 * @fd: defrag target file's descriptor.
293 * @offset: file offset.
294 * @len: area length.
295 * @advise: process flag.
296 */
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400297static int posix_fadvise(int fd, loff_t offset, size_t len, int advise)
Theodore Ts'o48364592009-07-09 14:29:12 -0400298{
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400299 return syscall(__NR_fadvise64_64, fd, offset, len, advise);
Theodore Ts'o48364592009-07-09 14:29:12 -0400300}
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400301#endif /* ! HAVE_FADVISE64_64 */
302
303#ifndef HAVE_SYNC_FILE_RANGE
304#warning Using locally defined sync_file_range interface.
305
306#ifndef __NR_sync_file_range
Kyle McMartina9e55a12009-08-23 22:09:58 -0700307#ifndef __NR_sync_file_range2 /* ppc */
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400308#error Your kernel headers dont define __NR_sync_file_range
309#endif
Kyle McMartina9e55a12009-08-23 22:09:58 -0700310#endif
Theodore Ts'o48364592009-07-09 14:29:12 -0400311
312/*
313 * sync_file_range() - Sync file region.
314 *
315 * @fd: defrag target file's descriptor.
316 * @offset: file offset.
317 * @length: area length.
318 * @flag: process flag.
319 */
320int sync_file_range(int fd, loff_t offset, loff_t length, unsigned int flag)
321{
Kyle McMartina9e55a12009-08-23 22:09:58 -0700322#ifdef __NR_sync_file_range
Theodore Ts'o48364592009-07-09 14:29:12 -0400323 return syscall(__NR_sync_file_range, fd, offset, length, flag);
Kyle McMartina9e55a12009-08-23 22:09:58 -0700324#else
325 return syscall(__NR_sync_file_range2, fd, flag, offset, length);
326#endif
Theodore Ts'o48364592009-07-09 14:29:12 -0400327}
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400328#endif /* ! HAVE_SYNC_FILE_RANGE */
329
Theodore Ts'o30c05292010-12-16 22:53:34 -0500330#ifndef HAVE_FALLOCATE64
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400331#warning Using locally defined fallocate syscall interface.
332
333#ifndef __NR_fallocate
334#error Your kernel headers dont define __NR_fallocate
335#endif
Theodore Ts'o48364592009-07-09 14:29:12 -0400336
337/*
Theodore Ts'o30c05292010-12-16 22:53:34 -0500338 * fallocate64() - Manipulate file space.
Theodore Ts'o48364592009-07-09 14:29:12 -0400339 *
340 * @fd: defrag target file's descriptor.
341 * @mode: process flag.
342 * @offset: file offset.
343 * @len: file size.
344 */
Theodore Ts'o30c05292010-12-16 22:53:34 -0500345static int fallocate64(int fd, int mode, loff_t offset, loff_t len)
Theodore Ts'o48364592009-07-09 14:29:12 -0400346{
347 return syscall(__NR_fallocate, fd, mode, offset, len);
348}
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400349#endif /* ! HAVE_FALLOCATE */
Theodore Ts'o48364592009-07-09 14:29:12 -0400350
351/*
352 * get_mount_point() - Get device's mount point.
353 *
354 * @devname: the device's name.
355 * @mount_point: the mount point.
356 * @dir_path_len: the length of directory.
357 */
Kazuya Mioae507462009-07-10 16:45:16 +0900358static int get_mount_point(const char *devname, char *mount_point,
359 int dir_path_len)
Theodore Ts'o48364592009-07-09 14:29:12 -0400360{
361 /* Refer to /etc/mtab */
Kazuya Mioae507462009-07-10 16:45:16 +0900362 const char *mtab = MOUNTED;
Theodore Ts'o48364592009-07-09 14:29:12 -0400363 FILE *fp = NULL;
364 struct mntent *mnt = NULL;
365
366 fp = setmntent(mtab, "r");
367 if (fp == NULL) {
368 perror("Couldn't access /etc/mtab");
369 return -1;
370 }
371
372 while ((mnt = getmntent(fp)) != NULL) {
373 if (strcmp(devname, mnt->mnt_fsname) != 0)
374 continue;
375
376 endmntent(fp);
377 if (strcmp(mnt->mnt_type, FS_EXT4) == 0) {
378 strncpy(mount_point, mnt->mnt_dir,
379 dir_path_len);
380 return 0;
381 }
382 PRINT_ERR_MSG(NGMSG_EXT4);
383 return -1;
384 }
385 endmntent(fp);
386 PRINT_ERR_MSG("Filesystem is not mounted");
387 return -1;
388}
389
390/*
391 * is_ext4() - Whether on an ext4 filesystem.
392 *
393 * @file: the file's name.
394 */
Kazuya Mioae507462009-07-10 16:45:16 +0900395static int is_ext4(const char *file)
Theodore Ts'o48364592009-07-09 14:29:12 -0400396{
397 int maxlen = 0;
398 int len, ret;
399 FILE *fp = NULL;
400 char *mnt_type = NULL;
401 /* Refer to /etc/mtab */
Kazuya Mioae507462009-07-10 16:45:16 +0900402 const char *mtab = MOUNTED;
Theodore Ts'o48364592009-07-09 14:29:12 -0400403 char file_path[PATH_MAX + 1];
404 struct mntent *mnt = NULL;
405 struct statfs64 fsbuf;
406
407 /* Get full path */
408 if (realpath(file, file_path) == NULL) {
409 perror("Couldn't get full path");
410 PRINT_FILE_NAME(file);
411 return -1;
412 }
413
414 if (statfs64(file_path, &fsbuf) < 0) {
415 perror("Failed to get filesystem information");
416 PRINT_FILE_NAME(file);
417 return -1;
418 }
419
420 if (fsbuf.f_type != EXT4_SUPER_MAGIC) {
421 PRINT_ERR_MSG(NGMSG_EXT4);
422 return -1;
423 }
424
425 fp = setmntent(mtab, "r");
426 if (fp == NULL) {
427 perror("Couldn't access /etc/mtab");
428 return -1;
429 }
430
431 while ((mnt = getmntent(fp)) != NULL) {
Eric Sandeen43eb2ad2009-09-08 06:13:50 +0000432 if (mnt->mnt_fsname[0] != '/')
433 continue;
Theodore Ts'o48364592009-07-09 14:29:12 -0400434 len = strlen(mnt->mnt_dir);
435 ret = memcmp(file_path, mnt->mnt_dir, len);
436 if (ret != 0)
437 continue;
438
439 if (maxlen >= len)
440 continue;
441
442 maxlen = len;
443
444 mnt_type = realloc(mnt_type, strlen(mnt->mnt_type) + 1);
445 if (mnt_type == NULL) {
446 endmntent(fp);
447 return -1;
448 }
449 memset(mnt_type, 0, strlen(mnt->mnt_type) + 1);
450 strncpy(mnt_type, mnt->mnt_type, strlen(mnt->mnt_type));
451 strncpy(lost_found_dir, mnt->mnt_dir, PATH_MAX);
452 }
453
454 endmntent(fp);
455 if (strcmp(mnt_type, FS_EXT4) == 0) {
456 FREE(mnt_type);
457 return 0;
458 } else {
459 FREE(mnt_type);
460 PRINT_ERR_MSG(NGMSG_EXT4);
461 return -1;
462 }
463}
464
465/*
466 * calc_entry_counts() - Calculate file counts.
467 *
468 * @file: file name.
469 * @buf: file info.
470 * @flag: file type.
471 * @ftwbuf: the pointer of a struct FTW.
472 */
Kazuya Mioae507462009-07-10 16:45:16 +0900473static int calc_entry_counts(const char *file EXT2FS_ATTR((unused)),
Theodore Ts'o48364592009-07-09 14:29:12 -0400474 const struct stat64 *buf, int flag EXT2FS_ATTR((unused)),
475 struct FTW *ftwbuf EXT2FS_ATTR((unused)))
476{
477 if (S_ISREG(buf->st_mode))
478 regular_count++;
479
480 total_count++;
481
482 return 0;
483}
484
485/*
486 * page_in_core() - Get information on whether pages are in core.
487 *
488 * @fd: defrag target file's descriptor.
489 * @defrag_data: data used for defrag.
490 * @vec: page state array.
491 * @page_num: page number.
492 */
Kazuya Mioae507462009-07-10 16:45:16 +0900493static int page_in_core(int fd, struct move_extent defrag_data,
Theodore Ts'o48364592009-07-09 14:29:12 -0400494 unsigned char **vec, unsigned int *page_num)
495{
496 long pagesize = sysconf(_SC_PAGESIZE);
497 void *page = NULL;
498 loff_t offset, end_offset, length;
499
500 if (vec == NULL || *vec != NULL)
501 return -1;
502
503 /* In mmap, offset should be a multiple of the page size */
504 offset = (loff_t)defrag_data.orig_start * block_size;
505 length = (loff_t)defrag_data.len * block_size;
506 end_offset = offset + length;
507 /* Round the offset down to the nearest multiple of pagesize */
508 offset = (offset / pagesize) * pagesize;
509 length = end_offset - offset;
510
511 page = mmap(NULL, length, PROT_READ, MAP_SHARED, fd, offset);
512 if (page == MAP_FAILED)
513 return -1;
514
515 *page_num = 0;
516 *page_num = (length + pagesize - 1) / pagesize;
517 *vec = (unsigned char *)calloc(*page_num, 1);
518 if (*vec == NULL)
519 return -1;
520
521 /* Get information on whether pages are in core */
522 if (mincore(page, (size_t)length, *vec) == -1 ||
523 munmap(page, length) == -1) {
524 FREE(*vec);
525 return -1;
526 }
527
528 return 0;
529}
530
531/*
532 * defrag_fadvise() - Predeclare an access pattern for file data.
533 *
534 * @fd: defrag target file's descriptor.
535 * @defrag_data: data used for defrag.
536 * @vec: page state array.
537 * @page_num: page number.
538 */
Kazuya Mioae507462009-07-10 16:45:16 +0900539static int defrag_fadvise(int fd, struct move_extent defrag_data,
Theodore Ts'o48364592009-07-09 14:29:12 -0400540 unsigned char *vec, unsigned int page_num)
541{
542 int flag = 1;
543 long pagesize = sysconf(_SC_PAGESIZE);
544 int fadvise_flag = POSIX_FADV_DONTNEED;
545 int sync_flag = SYNC_FILE_RANGE_WAIT_BEFORE |
546 SYNC_FILE_RANGE_WRITE |
547 SYNC_FILE_RANGE_WAIT_AFTER;
548 unsigned int i;
549 loff_t offset;
550
551 offset = (loff_t)defrag_data.orig_start * block_size;
552 offset = (offset / pagesize) * pagesize;
553
554 /* Sync file for fadvise process */
555 if (sync_file_range(fd, offset,
556 (loff_t)pagesize * page_num, sync_flag) < 0)
557 return -1;
558
559 /* Try to release buffer cache which this process used,
560 * then other process can use the released buffer
561 */
562 for (i = 0; i < page_num; i++) {
563 if ((vec[i] & 0x1) == 0) {
564 offset += pagesize;
565 continue;
566 }
Eric Sandeen4f1d0e82009-07-27 10:53:46 -0400567 if (posix_fadvise(fd, offset, pagesize, fadvise_flag) < 0) {
Theodore Ts'o48364592009-07-09 14:29:12 -0400568 if ((mode_flag & DETAIL) && flag) {
569 perror("\tFailed to fadvise");
570 flag = 0;
571 }
572 }
573 offset += pagesize;
574 }
575
576 return 0;
577}
578
579/*
580 * check_free_size() - Check if there's enough disk space.
581 *
582 * @fd: defrag target file's descriptor.
583 * @file: file name.
Kazuya Mio02808f7d2010-12-13 09:59:06 -0500584 * @blk_count: file blocks.
Theodore Ts'o48364592009-07-09 14:29:12 -0400585 */
Kazuya Mio02808f7d2010-12-13 09:59:06 -0500586static int check_free_size(int fd, const char *file, ext4_fsblk_t blk_count)
Theodore Ts'o48364592009-07-09 14:29:12 -0400587{
Theodore Ts'o48364592009-07-09 14:29:12 -0400588 ext4_fsblk_t free_blk_count;
589 struct statfs64 fsbuf;
590
591 if (fstatfs64(fd, &fsbuf) < 0) {
592 if (mode_flag & DETAIL) {
593 PRINT_FILE_NAME(file);
594 PRINT_ERR_MSG_WITH_ERRNO(
595 "Failed to get filesystem information");
596 }
597 return -1;
598 }
599
Theodore Ts'o48364592009-07-09 14:29:12 -0400600 /* Compute free space for root and normal user separately */
601 if (current_uid == ROOT_UID)
602 free_blk_count = fsbuf.f_bfree;
603 else
604 free_blk_count = fsbuf.f_bavail;
605
606 if (free_blk_count >= blk_count)
607 return 0;
608
609 return -ENOSPC;
610}
611
612/*
613 * file_frag_count() - Get file fragment count.
614 *
615 * @fd: defrag target file's descriptor.
616 */
Kazuya Mioae507462009-07-10 16:45:16 +0900617static int file_frag_count(int fd)
Theodore Ts'o48364592009-07-09 14:29:12 -0400618{
619 int ret;
620 struct fiemap fiemap_buf;
621
622 /* When fm_extent_count is 0,
623 * ioctl just get file fragment count.
624 */
625 memset(&fiemap_buf, 0, sizeof(struct fiemap));
626 fiemap_buf.fm_start = 0;
627 fiemap_buf.fm_length = FIEMAP_MAX_OFFSET;
628 fiemap_buf.fm_flags |= FIEMAP_FLAG_SYNC;
629
630 ret = ioctl(fd, FS_IOC_FIEMAP, &fiemap_buf);
631 if (ret < 0)
632 return ret;
633
634 return fiemap_buf.fm_mapped_extents;
635}
636
637/*
638 * file_check() - Check file's attributes.
639 *
640 * @fd: defrag target file's descriptor.
641 * @buf: a pointer of the struct stat64.
Kazuya Mio02808f7d2010-12-13 09:59:06 -0500642 * @file: file name.
643 * @extents: file extents.
644 * @blk_count: file blocks.
Theodore Ts'o48364592009-07-09 14:29:12 -0400645 */
Kazuya Mioae507462009-07-10 16:45:16 +0900646static int file_check(int fd, const struct stat64 *buf, const char *file,
Kazuya Mio02808f7d2010-12-13 09:59:06 -0500647 int extents, ext4_fsblk_t blk_count)
Theodore Ts'o48364592009-07-09 14:29:12 -0400648{
649 int ret;
650 struct flock lock;
651
652 /* Write-lock check is more reliable */
653 lock.l_type = F_WRLCK;
654 lock.l_start = 0;
655 lock.l_whence = SEEK_SET;
656 lock.l_len = 0;
657
658 /* Free space */
Kazuya Mio02808f7d2010-12-13 09:59:06 -0500659 ret = check_free_size(fd, file, blk_count);
Theodore Ts'o48364592009-07-09 14:29:12 -0400660 if (ret < 0) {
661 if ((mode_flag & DETAIL) && ret == -ENOSPC) {
662 printf("\033[79;0H\033[K[%u/%u] \"%s\"\t\t"
663 " extents: %d -> %d\n", defraged_file_count,
664 total_count, file, extents, extents);
665 IN_FTW_PRINT_ERR_MSG(
666 "Defrag size is larger than filesystem's free space");
667 }
668 return -1;
669 }
670
671 /* Access authority */
672 if (current_uid != ROOT_UID &&
673 buf->st_uid != current_uid) {
674 if (mode_flag & DETAIL) {
675 printf("\033[79;0H\033[K[%u/%u] \"%s\"\t\t"
676 " extents: %d -> %d\n", defraged_file_count,
677 total_count, file, extents, extents);
678 IN_FTW_PRINT_ERR_MSG(
679 "File is not current user's file"
680 " or current user is not root");
681 }
682 return -1;
683 }
684
685 /* Lock status */
686 if (fcntl(fd, F_GETLK, &lock) < 0) {
687 if (mode_flag & DETAIL) {
688 PRINT_FILE_NAME(file);
689 PRINT_ERR_MSG_WITH_ERRNO(
690 "Failed to get lock information");
691 }
692 return -1;
693 } else if (lock.l_type != F_UNLCK) {
694 if (mode_flag & DETAIL) {
695 PRINT_FILE_NAME(file);
696 IN_FTW_PRINT_ERR_MSG("File has been locked");
697 }
698 return -1;
699 }
700
701 return 0;
702}
703
704/*
705 * insert_extent_by_logical() - Sequentially insert extent by logical.
706 *
707 * @ext_list_head: the head of logical extent list.
708 * @ext: the extent element which will be inserted.
709 */
Kazuya Mioae507462009-07-10 16:45:16 +0900710static int insert_extent_by_logical(struct fiemap_extent_list **ext_list_head,
Theodore Ts'o48364592009-07-09 14:29:12 -0400711 struct fiemap_extent_list *ext)
712{
713 struct fiemap_extent_list *ext_list_tmp = *ext_list_head;
714
715 if (ext == NULL)
716 goto out;
717
718 /* First element */
719 if (*ext_list_head == NULL) {
720 (*ext_list_head) = ext;
721 (*ext_list_head)->prev = *ext_list_head;
722 (*ext_list_head)->next = *ext_list_head;
723 return 0;
724 }
725
726 if (ext->data.logical <= ext_list_tmp->data.logical) {
727 /* Insert before head */
728 if (ext_list_tmp->data.logical <
729 ext->data.logical + ext->data.len)
730 /* Overlap */
731 goto out;
732 /* Adjust head */
733 *ext_list_head = ext;
734 } else {
735 /* Insert into the middle or last of the list */
736 do {
737 if (ext->data.logical < ext_list_tmp->data.logical)
738 break;
739 ext_list_tmp = ext_list_tmp->next;
740 } while (ext_list_tmp != (*ext_list_head));
741 if (ext->data.logical <
742 ext_list_tmp->prev->data.logical +
743 ext_list_tmp->prev->data.len)
744 /* Overlap */
745 goto out;
746
747 if (ext_list_tmp != *ext_list_head &&
748 ext_list_tmp->data.logical <
749 ext->data.logical + ext->data.len)
750 /* Overlap */
751 goto out;
752 }
753 ext_list_tmp = ext_list_tmp->prev;
754 /* Insert "ext" after "ext_list_tmp" */
755 insert(ext_list_tmp, ext);
756 return 0;
757out:
758 errno = EINVAL;
759 return -1;
760}
761
762/*
763 * insert_extent_by_physical() - Sequentially insert extent by physical.
764 *
765 * @ext_list_head: the head of physical extent list.
766 * @ext: the extent element which will be inserted.
767 */
Kazuya Mioae507462009-07-10 16:45:16 +0900768static int insert_extent_by_physical(struct fiemap_extent_list **ext_list_head,
Theodore Ts'o48364592009-07-09 14:29:12 -0400769 struct fiemap_extent_list *ext)
770{
771 struct fiemap_extent_list *ext_list_tmp = *ext_list_head;
772
773 if (ext == NULL)
774 goto out;
775
776 /* First element */
777 if (*ext_list_head == NULL) {
778 (*ext_list_head) = ext;
779 (*ext_list_head)->prev = *ext_list_head;
780 (*ext_list_head)->next = *ext_list_head;
781 return 0;
782 }
783
784 if (ext->data.physical <= ext_list_tmp->data.physical) {
785 /* Insert before head */
786 if (ext_list_tmp->data.physical <
787 ext->data.physical + ext->data.len)
788 /* Overlap */
789 goto out;
790 /* Adjust head */
791 *ext_list_head = ext;
792 } else {
793 /* Insert into the middle or last of the list */
794 do {
795 if (ext->data.physical < ext_list_tmp->data.physical)
796 break;
797 ext_list_tmp = ext_list_tmp->next;
798 } while (ext_list_tmp != (*ext_list_head));
799 if (ext->data.physical <
800 ext_list_tmp->prev->data.physical +
801 ext_list_tmp->prev->data.len)
802 /* Overlap */
803 goto out;
804
805 if (ext_list_tmp != *ext_list_head &&
806 ext_list_tmp->data.physical <
807 ext->data.physical + ext->data.len)
808 /* Overlap */
809 goto out;
810 }
811 ext_list_tmp = ext_list_tmp->prev;
812 /* Insert "ext" after "ext_list_tmp" */
813 insert(ext_list_tmp, ext);
814 return 0;
815out:
816 errno = EINVAL;
817 return -1;
818}
819
820/*
821 * insert_exts_group() - Insert a exts_group.
822 *
823 * @ext_group_head: the head of a exts_group list.
824 * @exts_group: the exts_group element which will be inserted.
825 */
Kazuya Mioae507462009-07-10 16:45:16 +0900826static int insert_exts_group(struct fiemap_extent_group **ext_group_head,
Theodore Ts'o48364592009-07-09 14:29:12 -0400827 struct fiemap_extent_group *exts_group)
828{
829 struct fiemap_extent_group *ext_group_tmp = NULL;
830
831 if (exts_group == NULL) {
832 errno = EINVAL;
833 return -1;
834 }
835
836 /* Initialize list */
837 if (*ext_group_head == NULL) {
838 (*ext_group_head) = exts_group;
839 (*ext_group_head)->prev = *ext_group_head;
840 (*ext_group_head)->next = *ext_group_head;
841 return 0;
842 }
843
844 ext_group_tmp = (*ext_group_head)->prev;
845 insert(ext_group_tmp, exts_group);
846
847 return 0;
848}
849
850/*
851 * join_extents() - Find continuous region(exts_group).
852 *
853 * @ext_list_head: the head of the extent list.
854 * @ext_group_head: the head of the target exts_group list.
855 */
Kazuya Mioae507462009-07-10 16:45:16 +0900856static int join_extents(struct fiemap_extent_list *ext_list_head,
Theodore Ts'o48364592009-07-09 14:29:12 -0400857 struct fiemap_extent_group **ext_group_head)
858{
859 __u64 len = ext_list_head->data.len;
860 struct fiemap_extent_list *ext_list_start = ext_list_head;
861 struct fiemap_extent_list *ext_list_tmp = ext_list_head->next;
862
863 do {
864 struct fiemap_extent_group *ext_group_tmp = NULL;
865
866 /* This extent and previous extent are not continuous,
867 * so, all previous extents are treated as an extent group.
868 */
869 if ((ext_list_tmp->prev->data.logical +
870 ext_list_tmp->prev->data.len)
871 != ext_list_tmp->data.logical) {
872 ext_group_tmp =
873 malloc(sizeof(struct fiemap_extent_group));
874 if (ext_group_tmp == NULL)
875 return -1;
876
877 memset(ext_group_tmp, 0,
878 sizeof(struct fiemap_extent_group));
879 ext_group_tmp->len = len;
880 ext_group_tmp->start = ext_list_start;
881 ext_group_tmp->end = ext_list_tmp->prev;
882
883 if (insert_exts_group(ext_group_head,
884 ext_group_tmp) < 0) {
885 FREE(ext_group_tmp);
886 return -1;
887 }
888 ext_list_start = ext_list_tmp;
889 len = ext_list_tmp->data.len;
890 ext_list_tmp = ext_list_tmp->next;
891 continue;
892 }
893
894 /* This extent and previous extent are continuous,
895 * so, they belong to the same extent group, and we check
896 * if the next extent belongs to the same extent group.
897 */
898 len += ext_list_tmp->data.len;
899 ext_list_tmp = ext_list_tmp->next;
900 } while (ext_list_tmp != ext_list_head->next);
901
902 return 0;
903}
904
905/*
906 * get_file_extents() - Get file's extent list.
907 *
908 * @fd: defrag target file's descriptor.
909 * @ext_list_head: the head of the extent list.
910 */
Kazuya Mioae507462009-07-10 16:45:16 +0900911static int get_file_extents(int fd, struct fiemap_extent_list **ext_list_head)
Theodore Ts'o48364592009-07-09 14:29:12 -0400912{
913 __u32 i;
914 int ret;
915 int ext_buf_size, fie_buf_size;
916 __u64 pos = 0;
917 struct fiemap *fiemap_buf = NULL;
918 struct fiemap_extent *ext_buf = NULL;
919 struct fiemap_extent_list *ext_list = NULL;
920
921 /* Convert units, in bytes.
922 * Be careful : now, physical block number in extent is 48bit,
923 * and the maximum blocksize for ext4 is 4K(12bit),
924 * so there is no overflow, but in future it may be changed.
925 */
926
927 /* Alloc space for fiemap */
928 ext_buf_size = EXTENT_MAX_COUNT * sizeof(struct fiemap_extent);
929 fie_buf_size = sizeof(struct fiemap) + ext_buf_size;
930
931 fiemap_buf = malloc(fie_buf_size);
932 if (fiemap_buf == NULL)
933 return -1;
934
935 ext_buf = fiemap_buf->fm_extents;
936 memset(fiemap_buf, 0, fie_buf_size);
937 fiemap_buf->fm_length = FIEMAP_MAX_OFFSET;
938 fiemap_buf->fm_flags |= FIEMAP_FLAG_SYNC;
939 fiemap_buf->fm_extent_count = EXTENT_MAX_COUNT;
940
941 do {
942 fiemap_buf->fm_start = pos;
943 memset(ext_buf, 0, ext_buf_size);
944 ret = ioctl(fd, FS_IOC_FIEMAP, fiemap_buf);
945 if (ret < 0)
946 goto out;
947 for (i = 0; i < fiemap_buf->fm_mapped_extents; i++) {
948 ext_list = NULL;
949 ext_list = malloc(sizeof(struct fiemap_extent_list));
950 if (ext_list == NULL)
951 goto out;
952
953 ext_list->data.physical = ext_buf[i].fe_physical
954 / block_size;
955 ext_list->data.logical = ext_buf[i].fe_logical
956 / block_size;
957 ext_list->data.len = ext_buf[i].fe_length
958 / block_size;
959
960 ret = insert_extent_by_physical(
961 ext_list_head, ext_list);
962 if (ret < 0) {
963 FREE(ext_list);
964 goto out;
965 }
966 }
967 /* Record file's logical offset this time */
968 pos = ext_buf[EXTENT_MAX_COUNT-1].fe_logical +
969 ext_buf[EXTENT_MAX_COUNT-1].fe_length;
970 /*
971 * If fm_extents array has been filled and
972 * there are extents left, continue to cycle.
973 */
974 } while (fiemap_buf->fm_mapped_extents
975 == EXTENT_MAX_COUNT &&
976 !(ext_buf[EXTENT_MAX_COUNT-1].fe_flags
977 & FIEMAP_EXTENT_LAST));
978
979 FREE(fiemap_buf);
980 return 0;
981out:
982 FREE(fiemap_buf);
983 return -1;
984}
985
986/*
987 * get_logical_count() - Get the file logical extents count.
988 *
989 * @logical_list_head: the head of the logical extent list.
990 */
Kazuya Mioae507462009-07-10 16:45:16 +0900991static int get_logical_count(struct fiemap_extent_list *logical_list_head)
Theodore Ts'o48364592009-07-09 14:29:12 -0400992{
993 int ret = 0;
994 struct fiemap_extent_list *ext_list_tmp = logical_list_head;
995
996 do {
997 ret++;
998 ext_list_tmp = ext_list_tmp->next;
999 } while (ext_list_tmp != logical_list_head);
1000
1001 return ret;
1002}
1003
1004/*
1005 * get_physical_count() - Get the file physical extents count.
1006 *
1007 * @physical_list_head: the head of the physical extent list.
1008 */
Kazuya Mioae507462009-07-10 16:45:16 +09001009static int get_physical_count(struct fiemap_extent_list *physical_list_head)
Theodore Ts'o48364592009-07-09 14:29:12 -04001010{
1011 int ret = 0;
1012 struct fiemap_extent_list *ext_list_tmp = physical_list_head;
1013
1014 do {
1015 if ((ext_list_tmp->data.physical + ext_list_tmp->data.len)
1016 != ext_list_tmp->next->data.physical) {
1017 /* This extent and next extent are not continuous. */
1018 ret++;
1019 }
1020
1021 ext_list_tmp = ext_list_tmp->next;
1022 } while (ext_list_tmp != physical_list_head);
1023
1024 return ret;
1025}
1026
1027/*
1028 * change_physical_to_logical() - Change list from physical to logical.
1029 *
1030 * @physical_list_head: the head of physical extent list.
1031 * @logical_list_head: the head of logical extent list.
1032 */
Kazuya Mioae507462009-07-10 16:45:16 +09001033static int change_physical_to_logical(
1034 struct fiemap_extent_list **physical_list_head,
1035 struct fiemap_extent_list **logical_list_head)
Theodore Ts'o48364592009-07-09 14:29:12 -04001036{
1037 int ret;
1038 struct fiemap_extent_list *ext_list_tmp = *physical_list_head;
1039 struct fiemap_extent_list *ext_list_next = ext_list_tmp->next;
1040
1041 while (1) {
1042 if (ext_list_tmp == ext_list_next) {
1043 ret = insert_extent_by_logical(
1044 logical_list_head, ext_list_tmp);
1045 if (ret < 0)
1046 return -1;
1047
1048 *physical_list_head = NULL;
1049 break;
1050 }
1051
1052 ext_list_tmp->prev->next = ext_list_tmp->next;
1053 ext_list_tmp->next->prev = ext_list_tmp->prev;
1054 *physical_list_head = ext_list_next;
1055
1056 ret = insert_extent_by_logical(
1057 logical_list_head, ext_list_tmp);
1058 if (ret < 0) {
1059 FREE(ext_list_tmp);
1060 return -1;
1061 }
1062 ext_list_tmp = ext_list_next;
1063 ext_list_next = ext_list_next->next;
1064 }
1065
1066 return 0;
1067}
1068
Kazuya Mio02808f7d2010-12-13 09:59:06 -05001069/* get_file_blocks() - Get total file blocks.
1070 *
1071 * @ext_list_head: the extent list head of the target file
1072 */
1073static ext4_fsblk_t get_file_blocks(struct fiemap_extent_list *ext_list_head)
1074{
1075 ext4_fsblk_t blk_count = 0;
1076 struct fiemap_extent_list *ext_list_tmp = ext_list_head;
1077
1078 do {
1079 blk_count += ext_list_tmp->data.len;
1080 ext_list_tmp = ext_list_tmp->next;
1081 } while (ext_list_tmp != ext_list_head);
1082
1083 return blk_count;
1084}
1085
Theodore Ts'o48364592009-07-09 14:29:12 -04001086/*
1087 * free_ext() - Free the extent list.
1088 *
1089 * @ext_list_head: the extent list head of which will be free.
1090 */
Kazuya Mioae507462009-07-10 16:45:16 +09001091static void free_ext(struct fiemap_extent_list *ext_list_head)
Theodore Ts'o48364592009-07-09 14:29:12 -04001092{
1093 struct fiemap_extent_list *ext_list_tmp = NULL;
1094
1095 if (ext_list_head == NULL)
1096 return;
1097
1098 while (ext_list_head->next != ext_list_head) {
1099 ext_list_tmp = ext_list_head;
1100 ext_list_head->prev->next = ext_list_head->next;
1101 ext_list_head->next->prev = ext_list_head->prev;
1102 ext_list_head = ext_list_head->next;
1103 free(ext_list_tmp);
1104 }
1105 free(ext_list_head);
1106}
1107
1108/*
1109 * free_exts_group() - Free the exts_group.
1110 *
1111 * @*ext_group_head: the exts_group list head which will be free.
1112 */
Kazuya Mioae507462009-07-10 16:45:16 +09001113static void free_exts_group(struct fiemap_extent_group *ext_group_head)
Theodore Ts'o48364592009-07-09 14:29:12 -04001114{
1115 struct fiemap_extent_group *ext_group_tmp = NULL;
1116
1117 if (ext_group_head == NULL)
1118 return;
1119
1120 while (ext_group_head->next != ext_group_head) {
1121 ext_group_tmp = ext_group_head;
1122 ext_group_head->prev->next = ext_group_head->next;
1123 ext_group_head->next->prev = ext_group_head->prev;
1124 ext_group_head = ext_group_head->next;
1125 free(ext_group_tmp);
1126 }
1127 free(ext_group_head);
1128}
1129
1130/*
1131 * get_superblock_info() - Get superblock info by the file name.
1132 *
1133 * @file: the file's name.
1134 * @sb: the pointer of the struct ext4_super_block.
1135 */
Kazuya Mioae507462009-07-10 16:45:16 +09001136static int get_superblock_info(const char *file, struct ext4_super_block *sb)
Theodore Ts'o48364592009-07-09 14:29:12 -04001137{
1138 /* Refer to /etc/mtab */
Kazuya Mioae507462009-07-10 16:45:16 +09001139 const char *mtab = MOUNTED;
Theodore Ts'o48364592009-07-09 14:29:12 -04001140 FILE *fp = NULL;
1141
1142 int fd = -1;
1143 int ret;
1144 size_t maxlen = 0;
1145 size_t len;
1146 char dev_name[PATH_MAX + 1];
1147 struct mntent *mnt = NULL;
1148
1149 fp = setmntent(mtab, "r");
1150 if (fp == NULL)
1151 return -1;
1152
1153 while ((mnt = getmntent(fp)) != NULL) {
1154 len = strlen(mnt->mnt_dir);
1155 ret = memcmp(file, mnt->mnt_dir, len);
1156 if (ret != 0)
1157 continue;
1158
1159 if (len < maxlen)
1160 continue;
1161
1162 maxlen = len;
1163
1164 memset(dev_name, 0, PATH_MAX + 1);
1165 strncpy(dev_name, mnt->mnt_fsname,
1166 strnlen(mnt->mnt_fsname, PATH_MAX));
1167 }
1168
1169 fd = open64(dev_name, O_RDONLY);
1170 if (fd < 0) {
1171 ret = -1;
1172 goto out;
1173 }
1174
1175 /* Set offset to read superblock */
1176 ret = lseek64(fd, SUPERBLOCK_OFFSET, SEEK_SET);
1177 if (ret < 0)
1178 goto out;
1179
1180 ret = read(fd, sb, sizeof(struct ext4_super_block));
1181 if (ret < 0)
1182 goto out;
1183
1184out:
1185 if (fd != -1)
1186 close(fd);
1187 endmntent(fp);
1188 return ret;
1189}
1190
1191/*
1192 * get_best_count() - Get the file best extents count.
1193 *
1194 * @block_count: the file's physical block count.
1195 */
Kazuya Mioae507462009-07-10 16:45:16 +09001196static int get_best_count(ext4_fsblk_t block_count)
Theodore Ts'o48364592009-07-09 14:29:12 -04001197{
1198 int ret;
1199 unsigned int flex_bg_num;
1200
1201 /* Calcuate best extents count */
1202 if (feature_incompat & EXT4_FEATURE_INCOMPAT_FLEX_BG) {
1203 flex_bg_num = 1 << log_groups_per_flex;
1204 ret = ((block_count - 1) /
1205 ((ext4_fsblk_t)blocks_per_group *
1206 flex_bg_num)) + 1;
1207 } else
1208 ret = ((block_count - 1) / blocks_per_group) + 1;
1209
1210 return ret;
1211}
1212
1213
1214/*
1215 * file_statistic() - Get statistic info of the file's fragments.
1216 *
1217 * @file: the file's name.
1218 * @buf: the pointer of the struct stat64.
1219 * @flag: file type.
1220 * @ftwbuf: the pointer of a struct FTW.
1221 */
Kazuya Mioae507462009-07-10 16:45:16 +09001222static int file_statistic(const char *file, const struct stat64 *buf,
Theodore Ts'o48364592009-07-09 14:29:12 -04001223 int flag EXT2FS_ATTR((unused)),
1224 struct FTW *ftwbuf EXT2FS_ATTR((unused)))
1225{
1226 int fd;
1227 int ret;
Kazuya Mioae507462009-07-10 16:45:16 +09001228 int now_ext_count, best_ext_count = 0, physical_ext_count;
Theodore Ts'o48364592009-07-09 14:29:12 -04001229 int i, j;
Kazuya Mio94d26c22010-12-13 09:57:26 -05001230 __u64 size_per_ext = 0;
Theodore Ts'o48364592009-07-09 14:29:12 -04001231 float ratio = 0.0;
1232 ext4_fsblk_t blk_count = 0;
1233 char msg_buffer[PATH_MAX + 24];
1234 struct fiemap_extent_list *physical_list_head = NULL;
1235 struct fiemap_extent_list *logical_list_head = NULL;
1236
1237 defraged_file_count++;
1238
1239 if (mode_flag & DETAIL) {
1240 if (total_count == 1 && regular_count == 1)
1241 printf("<File>\n");
1242 else {
1243 printf("[%u/%u]", defraged_file_count, total_count);
1244 fflush(stdout);
1245 }
1246 }
1247 if (lost_found_dir[0] != '\0' &&
1248 !memcmp(file, lost_found_dir, strnlen(lost_found_dir, PATH_MAX))) {
1249 if (mode_flag & DETAIL) {
1250 PRINT_FILE_NAME(file);
1251 STATISTIC_ERR_MSG(NGMSG_LOST_FOUND);
1252 }
1253 return 0;
1254 }
1255
1256 if (!S_ISREG(buf->st_mode)) {
1257 if (mode_flag & DETAIL) {
1258 PRINT_FILE_NAME(file);
1259 STATISTIC_ERR_MSG(NGMSG_FILE_UNREG);
1260 }
1261 return 0;
1262 }
1263
1264 /* Access authority */
1265 if (current_uid != ROOT_UID &&
1266 buf->st_uid != current_uid) {
1267 if (mode_flag & DETAIL) {
1268 PRINT_FILE_NAME(file);
1269 STATISTIC_ERR_MSG(
1270 "File is not current user's file"
1271 " or current user is not root");
1272 }
1273 return 0;
1274 }
1275
1276 /* Empty file */
1277 if (buf->st_size == 0) {
1278 if (mode_flag & DETAIL) {
1279 PRINT_FILE_NAME(file);
1280 STATISTIC_ERR_MSG("File size is 0");
1281 }
1282 return 0;
1283 }
1284
Theodore Ts'o1d21f4b2009-07-09 15:03:39 -04001285 /* Has no blocks */
1286 if (buf->st_blocks == 0) {
1287 if (mode_flag & DETAIL) {
1288 PRINT_FILE_NAME(file);
1289 STATISTIC_ERR_MSG("File has no blocks");
1290 }
1291 return 0;
1292 }
1293
Theodore Ts'o48364592009-07-09 14:29:12 -04001294 fd = open64(file, O_RDONLY);
1295 if (fd < 0) {
1296 if (mode_flag & DETAIL) {
1297 PRINT_FILE_NAME(file);
1298 STATISTIC_ERR_MSG_WITH_ERRNO(NGMSG_FILE_OPEN);
1299 }
1300 return 0;
1301 }
1302
1303 /* Get file's physical extents */
1304 ret = get_file_extents(fd, &physical_list_head);
1305 if (ret < 0) {
1306 if (mode_flag & DETAIL) {
1307 PRINT_FILE_NAME(file);
1308 STATISTIC_ERR_MSG_WITH_ERRNO(NGMSG_FILE_EXTENT);
1309 }
1310 goto out;
1311 }
1312
1313 /* Get the count of file's continuous physical region */
1314 physical_ext_count = get_physical_count(physical_list_head);
1315
1316 /* Change list from physical to logical */
1317 ret = change_physical_to_logical(&physical_list_head,
1318 &logical_list_head);
1319 if (ret < 0) {
1320 if (mode_flag & DETAIL) {
1321 PRINT_FILE_NAME(file);
1322 STATISTIC_ERR_MSG_WITH_ERRNO(NGMSG_FILE_EXTENT);
1323 }
1324 goto out;
1325 }
1326
1327 /* Count file fragments before defrag */
1328 now_ext_count = get_logical_count(logical_list_head);
1329
1330 if (current_uid == ROOT_UID) {
Kazuya Mio94d26c22010-12-13 09:57:26 -05001331 /* Calculate the size per extent */
Kazuya Mio02808f7d2010-12-13 09:59:06 -05001332 blk_count = get_file_blocks(logical_list_head);
Theodore Ts'o48364592009-07-09 14:29:12 -04001333
1334 best_ext_count = get_best_count(blk_count);
1335
Kazuya Mio94d26c22010-12-13 09:57:26 -05001336 /* e4defrag rounds size_per_ext up to a block size boundary */
1337 size_per_ext = blk_count * (buf->st_blksize / 1024) /
1338 now_ext_count;
1339
Theodore Ts'o48364592009-07-09 14:29:12 -04001340 ratio = (float)(physical_ext_count - best_ext_count) * 100 /
1341 blk_count;
1342
1343 extents_before_defrag += now_ext_count;
1344 extents_after_defrag += best_ext_count;
1345 files_block_count += blk_count;
1346 }
1347
1348 if (total_count == 1 && regular_count == 1) {
1349 /* File only */
1350 if (mode_flag & DETAIL) {
1351 int count = 0;
1352 struct fiemap_extent_list *ext_list_tmp =
1353 logical_list_head;
1354
1355 /* Print extents info */
1356 do {
1357 count++;
1358 printf("[ext %d]:\tstart %llu:\tlogical "
1359 "%llu:\tlen %llu\n", count,
1360 ext_list_tmp->data.physical,
1361 ext_list_tmp->data.logical,
1362 ext_list_tmp->data.len);
1363 ext_list_tmp = ext_list_tmp->next;
1364 } while (ext_list_tmp != logical_list_head);
1365
1366 } else {
1367 printf("%-40s%10s/%-10s%9s\n",
Kazuya Mio94d26c22010-12-13 09:57:26 -05001368 "<File>", "now", "best", "size/ext");
Theodore Ts'o48364592009-07-09 14:29:12 -04001369 if (current_uid == ROOT_UID) {
1370 if (strlen(file) > 40)
Kazuya Mio94d26c22010-12-13 09:57:26 -05001371 printf("%s\n%50d/%-10d%6llu KB\n",
Theodore Ts'o48364592009-07-09 14:29:12 -04001372 file, now_ext_count,
Kazuya Mio94d26c22010-12-13 09:57:26 -05001373 best_ext_count, size_per_ext);
Theodore Ts'o48364592009-07-09 14:29:12 -04001374 else
Kazuya Mio94d26c22010-12-13 09:57:26 -05001375 printf("%-40s%10d/%-10d%6llu KB\n",
Theodore Ts'o48364592009-07-09 14:29:12 -04001376 file, now_ext_count,
Kazuya Mio94d26c22010-12-13 09:57:26 -05001377 best_ext_count, size_per_ext);
Theodore Ts'o48364592009-07-09 14:29:12 -04001378 } else {
1379 if (strlen(file) > 40)
1380 printf("%s\n%50d/%-10s%7s\n",
1381 file, now_ext_count,
1382 "-", "-");
1383 else
1384 printf("%-40s%10d/%-10s%7s\n",
1385 file, now_ext_count,
1386 "-", "-");
1387 }
1388 }
1389 succeed_cnt++;
1390 goto out;
1391 }
1392
1393 if (mode_flag & DETAIL) {
1394 /* Print statistic info */
1395 sprintf(msg_buffer, "[%u/%u]%s",
1396 defraged_file_count, total_count, file);
1397 if (current_uid == ROOT_UID) {
1398 if (strlen(msg_buffer) > 40)
1399 printf("\033[79;0H\033[K%s\n"
Kazuya Mio94d26c22010-12-13 09:57:26 -05001400 "%50d/%-10d%6llu KB\n",
Theodore Ts'o48364592009-07-09 14:29:12 -04001401 msg_buffer, now_ext_count,
Kazuya Mio94d26c22010-12-13 09:57:26 -05001402 best_ext_count, size_per_ext);
Theodore Ts'o48364592009-07-09 14:29:12 -04001403 else
1404 printf("\033[79;0H\033[K%-40s"
Kazuya Mio94d26c22010-12-13 09:57:26 -05001405 "%10d/%-10d%6llu KB\n",
Theodore Ts'o48364592009-07-09 14:29:12 -04001406 msg_buffer, now_ext_count,
Kazuya Mio94d26c22010-12-13 09:57:26 -05001407 best_ext_count, size_per_ext);
Theodore Ts'o48364592009-07-09 14:29:12 -04001408 } else {
1409 if (strlen(msg_buffer) > 40)
1410 printf("\033[79;0H\033[K%s\n%50d/%-10s%7s\n",
1411 msg_buffer, now_ext_count,
1412 "-", "-");
1413 else
1414 printf("\033[79;0H\033[K%-40s%10d/%-10s%7s\n",
1415 msg_buffer, now_ext_count,
1416 "-", "-");
1417 }
1418 }
1419
1420 for (i = 0; i < SHOW_FRAG_FILES; i++) {
1421 if (ratio >= frag_rank[i].ratio) {
1422 for (j = SHOW_FRAG_FILES - 1; j > i; j--) {
Kazuya Mio94d26c22010-12-13 09:57:26 -05001423 memset(&frag_rank[j], 0,
Theodore Ts'o48364592009-07-09 14:29:12 -04001424 sizeof(struct frag_statistic_ino));
Kazuya Mio94d26c22010-12-13 09:57:26 -05001425 strncpy(frag_rank[j].msg_buffer,
1426 frag_rank[j - 1].msg_buffer,
1427 strnlen(frag_rank[j - 1].msg_buffer,
1428 PATH_MAX));
1429 frag_rank[j].now_count =
1430 frag_rank[j - 1].now_count;
1431 frag_rank[j].best_count =
1432 frag_rank[j - 1].best_count;
1433 frag_rank[j].size_per_ext =
1434 frag_rank[j - 1].size_per_ext;
1435 frag_rank[j].ratio =
1436 frag_rank[j - 1].ratio;
Theodore Ts'o48364592009-07-09 14:29:12 -04001437 }
1438 memset(&frag_rank[i], 0,
1439 sizeof(struct frag_statistic_ino));
1440 strncpy(frag_rank[i].msg_buffer, file,
1441 strnlen(file, PATH_MAX));
1442 frag_rank[i].now_count = now_ext_count;
1443 frag_rank[i].best_count = best_ext_count;
Kazuya Mio94d26c22010-12-13 09:57:26 -05001444 frag_rank[i].size_per_ext = size_per_ext;
Theodore Ts'o48364592009-07-09 14:29:12 -04001445 frag_rank[i].ratio = ratio;
1446 break;
1447 }
1448 }
1449
1450 succeed_cnt++;
1451
1452out:
1453 close(fd);
1454 free_ext(physical_list_head);
1455 free_ext(logical_list_head);
1456 return 0;
1457}
1458
1459/*
1460 * print_progress - Print defrag progress
1461 *
1462 * @file: file name.
1463 * @start: logical offset for defrag target file
1464 * @file_size: defrag target filesize
1465 */
Kazuya Mioae507462009-07-10 16:45:16 +09001466static void print_progress(const char *file, loff_t start, loff_t file_size)
Theodore Ts'o48364592009-07-09 14:29:12 -04001467{
1468 int percent = (start * 100) / file_size;
1469 printf("\033[79;0H\033[K[%u/%u]%s:\t%3d%%",
1470 defraged_file_count, total_count, file, min(percent, 100));
1471 fflush(stdout);
1472
1473 return;
1474}
1475
1476/*
1477 * call_defrag() - Execute the defrag program.
1478 *
1479 * @fd: target file descriptor.
1480 * @donor_fd: donor file descriptor.
1481 * @file: target file name.
1482 * @buf: pointer of the struct stat64.
1483 * @ext_list_head: head of the extent list.
1484 */
Kazuya Mioae507462009-07-10 16:45:16 +09001485static int call_defrag(int fd, int donor_fd, const char *file,
Theodore Ts'o48364592009-07-09 14:29:12 -04001486 const struct stat64 *buf, struct fiemap_extent_list *ext_list_head)
1487{
1488 loff_t start = 0;
1489 unsigned int page_num;
1490 unsigned char *vec = NULL;
1491 int defraged_ret = 0;
1492 int ret;
1493 struct move_extent move_data;
1494 struct fiemap_extent_list *ext_list_tmp = NULL;
1495
1496 memset(&move_data, 0, sizeof(struct move_extent));
1497 move_data.donor_fd = donor_fd;
1498
1499 /* Print defrag progress */
1500 print_progress(file, start, buf->st_size);
1501
1502 ext_list_tmp = ext_list_head;
1503 do {
1504 move_data.orig_start = ext_list_tmp->data.logical;
1505 /* Logical offset of orig and donor should be same */
1506 move_data.donor_start = move_data.orig_start;
1507 move_data.len = ext_list_tmp->data.len;
1508 move_data.moved_len = 0;
1509
1510 ret = page_in_core(fd, move_data, &vec, &page_num);
1511 if (ret < 0) {
1512 if (mode_flag & DETAIL) {
1513 printf("\n");
1514 PRINT_ERR_MSG_WITH_ERRNO(
1515 "Failed to get file map");
1516 } else {
1517 printf("\t[ NG ]\n");
1518 }
1519 return -1;
1520 }
1521
1522 /* EXT4_IOC_MOVE_EXT */
1523 defraged_ret =
1524 ioctl(fd, EXT4_IOC_MOVE_EXT, &move_data);
1525
1526 /* Free pages */
1527 ret = defrag_fadvise(fd, move_data, vec, page_num);
1528 if (vec) {
1529 free(vec);
1530 vec = NULL;
1531 }
1532 if (ret < 0) {
1533 if (mode_flag & DETAIL) {
1534 printf("\n");
1535 PRINT_ERR_MSG_WITH_ERRNO(
1536 "Failed to free page");
1537 } else {
1538 printf("\t[ NG ]\n");
1539 }
1540 return -1;
1541 }
1542
1543 if (defraged_ret < 0) {
1544 if (mode_flag & DETAIL) {
1545 printf("\n");
1546 PRINT_ERR_MSG_WITH_ERRNO(
Peng Taoae09b932010-12-13 09:44:57 -05001547 "Failed to defrag with "
1548 "EXT4_IOC_MOVE_EXT ioctl");
1549 if (errno == ENOTTY)
1550 printf("\tAt least 2.6.31-rc1 of "
1551 "vanilla kernel is required\n");
Theodore Ts'o48364592009-07-09 14:29:12 -04001552 } else {
1553 printf("\t[ NG ]\n");
1554 }
1555 return -1;
1556 }
1557 /* Adjust logical offset for next ioctl */
1558 move_data.orig_start += move_data.moved_len;
1559 move_data.donor_start = move_data.orig_start;
1560
1561 start = move_data.orig_start * buf->st_blksize;
1562
1563 /* Print defrag progress */
1564 print_progress(file, start, buf->st_size);
1565
1566 /* End of file */
1567 if (start >= buf->st_size)
1568 break;
1569
1570 ext_list_tmp = ext_list_tmp->next;
1571 } while (ext_list_tmp != ext_list_head);
1572
1573 return 0;
1574}
1575
1576/*
1577 * file_defrag() - Check file attributes and call ioctl to defrag.
1578 *
1579 * @file: the file's name.
1580 * @buf: the pointer of the struct stat64.
1581 * @flag: file type.
1582 * @ftwbuf: the pointer of a struct FTW.
1583 */
Kazuya Mioae507462009-07-10 16:45:16 +09001584static int file_defrag(const char *file, const struct stat64 *buf,
Theodore Ts'o48364592009-07-09 14:29:12 -04001585 int flag EXT2FS_ATTR((unused)),
1586 struct FTW *ftwbuf EXT2FS_ATTR((unused)))
1587{
1588 int fd;
1589 int donor_fd = -1;
1590 int ret;
1591 int best;
1592 int file_frags_start, file_frags_end;
Kazuya Mioae507462009-07-10 16:45:16 +09001593 int orig_physical_cnt, donor_physical_cnt = 0;
Theodore Ts'o48364592009-07-09 14:29:12 -04001594 char tmp_inode_name[PATH_MAX + 8];
Kazuya Mio02808f7d2010-12-13 09:59:06 -05001595 ext4_fsblk_t blk_count = 0;
Theodore Ts'o48364592009-07-09 14:29:12 -04001596 struct fiemap_extent_list *orig_list_physical = NULL;
1597 struct fiemap_extent_list *orig_list_logical = NULL;
1598 struct fiemap_extent_list *donor_list_physical = NULL;
1599 struct fiemap_extent_list *donor_list_logical = NULL;
1600 struct fiemap_extent_group *orig_group_head = NULL;
1601 struct fiemap_extent_group *orig_group_tmp = NULL;
1602
1603 defraged_file_count++;
1604
1605 if (mode_flag & DETAIL) {
1606 printf("[%u/%u]", defraged_file_count, total_count);
1607 fflush(stdout);
1608 }
1609
1610 if (lost_found_dir[0] != '\0' &&
1611 !memcmp(file, lost_found_dir, strnlen(lost_found_dir, PATH_MAX))) {
1612 if (mode_flag & DETAIL) {
1613 PRINT_FILE_NAME(file);
1614 IN_FTW_PRINT_ERR_MSG(NGMSG_LOST_FOUND);
1615 }
1616 return 0;
1617 }
1618
1619 if (!S_ISREG(buf->st_mode)) {
1620 if (mode_flag & DETAIL) {
1621 PRINT_FILE_NAME(file);
1622 IN_FTW_PRINT_ERR_MSG(NGMSG_FILE_UNREG);
1623 }
1624 return 0;
1625 }
1626
1627 /* Empty file */
1628 if (buf->st_size == 0) {
1629 if (mode_flag & DETAIL) {
1630 PRINT_FILE_NAME(file);
1631 IN_FTW_PRINT_ERR_MSG("File size is 0");
1632 }
1633 return 0;
1634 }
1635
Theodore Ts'o1d21f4b2009-07-09 15:03:39 -04001636 /* Has no blocks */
1637 if (buf->st_blocks == 0) {
1638 if (mode_flag & DETAIL) {
1639 PRINT_FILE_NAME(file);
1640 STATISTIC_ERR_MSG("File has no blocks");
1641 }
1642 return 0;
1643 }
1644
Darrick J. Wong90b9dce2010-12-13 09:23:45 -05001645 fd = open64(file, O_RDWR);
Theodore Ts'o48364592009-07-09 14:29:12 -04001646 if (fd < 0) {
1647 if (mode_flag & DETAIL) {
1648 PRINT_FILE_NAME(file);
1649 PRINT_ERR_MSG_WITH_ERRNO(NGMSG_FILE_OPEN);
1650 }
1651 return 0;
1652 }
1653
1654 /* Get file's extents */
1655 ret = get_file_extents(fd, &orig_list_physical);
1656 if (ret < 0) {
1657 if (mode_flag & DETAIL) {
1658 PRINT_FILE_NAME(file);
1659 PRINT_ERR_MSG_WITH_ERRNO(NGMSG_FILE_EXTENT);
1660 }
1661 goto out;
1662 }
1663
1664 /* Get the count of file's continuous physical region */
1665 orig_physical_cnt = get_physical_count(orig_list_physical);
1666
1667 /* Change list from physical to logical */
1668 ret = change_physical_to_logical(&orig_list_physical,
1669 &orig_list_logical);
1670 if (ret < 0) {
1671 if (mode_flag & DETAIL) {
1672 PRINT_FILE_NAME(file);
1673 PRINT_ERR_MSG_WITH_ERRNO(NGMSG_FILE_EXTENT);
1674 }
1675 goto out;
1676 }
1677
1678 /* Count file fragments before defrag */
1679 file_frags_start = get_logical_count(orig_list_logical);
1680
Kazuya Mio02808f7d2010-12-13 09:59:06 -05001681 blk_count = get_file_blocks(orig_list_logical);
1682 if (file_check(fd, buf, file, file_frags_start, blk_count) < 0)
Theodore Ts'o48364592009-07-09 14:29:12 -04001683 goto out;
1684
1685 if (fsync(fd) < 0) {
1686 if (mode_flag & DETAIL) {
1687 PRINT_FILE_NAME(file);
1688 PRINT_ERR_MSG_WITH_ERRNO("Failed to sync(fsync)");
1689 }
1690 goto out;
1691 }
1692
1693 if (current_uid == ROOT_UID)
Kazuya Mio02808f7d2010-12-13 09:59:06 -05001694 best = get_best_count(blk_count);
Theodore Ts'o48364592009-07-09 14:29:12 -04001695 else
1696 best = 1;
1697
1698 if (file_frags_start <= best)
1699 goto check_improvement;
1700
1701 /* Combine extents to group */
1702 ret = join_extents(orig_list_logical, &orig_group_head);
1703 if (ret < 0) {
1704 if (mode_flag & DETAIL) {
1705 PRINT_FILE_NAME(file);
1706 PRINT_ERR_MSG_WITH_ERRNO(NGMSG_FILE_EXTENT);
1707 }
1708 goto out;
1709 }
1710
1711 /* Create donor inode */
1712 memset(tmp_inode_name, 0, PATH_MAX + 8);
Kazuya Mioae507462009-07-10 16:45:16 +09001713 sprintf(tmp_inode_name, "%.*s.defrag",
1714 (int)strnlen(file, PATH_MAX), file);
Theodore Ts'o48364592009-07-09 14:29:12 -04001715 donor_fd = open64(tmp_inode_name, O_WRONLY | O_CREAT | O_EXCL, S_IRUSR);
1716 if (donor_fd < 0) {
1717 if (mode_flag & DETAIL) {
1718 PRINT_FILE_NAME(file);
1719 if (errno == EEXIST)
1720 PRINT_ERR_MSG_WITH_ERRNO(
1721 "File is being defraged by other program");
1722 else
1723 PRINT_ERR_MSG_WITH_ERRNO(NGMSG_FILE_OPEN);
1724 }
1725 goto out;
1726 }
1727
1728 /* Unlink donor inode */
1729 ret = unlink(tmp_inode_name);
1730 if (ret < 0) {
1731 if (mode_flag & DETAIL) {
1732 PRINT_FILE_NAME(file);
1733 PRINT_ERR_MSG_WITH_ERRNO("Failed to unlink");
1734 }
1735 goto out;
1736 }
1737
1738 /* Allocate space for donor inode */
1739 orig_group_tmp = orig_group_head;
1740 do {
Theodore Ts'o30c05292010-12-16 22:53:34 -05001741 ret = fallocate64(donor_fd, 0,
Theodore Ts'o48364592009-07-09 14:29:12 -04001742 (loff_t)orig_group_tmp->start->data.logical * block_size,
1743 (loff_t)orig_group_tmp->len * block_size);
1744 if (ret < 0) {
1745 if (mode_flag & DETAIL) {
1746 PRINT_FILE_NAME(file);
1747 PRINT_ERR_MSG_WITH_ERRNO("Failed to fallocate");
1748 }
1749 goto out;
1750 }
1751
1752 orig_group_tmp = orig_group_tmp->next;
1753 } while (orig_group_tmp != orig_group_head);
1754
1755 /* Get donor inode's extents */
1756 ret = get_file_extents(donor_fd, &donor_list_physical);
1757 if (ret < 0) {
1758 if (mode_flag & DETAIL) {
1759 PRINT_FILE_NAME(file);
1760 PRINT_ERR_MSG_WITH_ERRNO(NGMSG_FILE_EXTENT);
1761 }
1762 goto out;
1763 }
1764
1765 /* Calcuate donor inode's continuous physical region */
1766 donor_physical_cnt = get_physical_count(donor_list_physical);
1767
1768 /* Change donor extent list from physical to logical */
1769 ret = change_physical_to_logical(&donor_list_physical,
1770 &donor_list_logical);
1771 if (ret < 0) {
1772 if (mode_flag & DETAIL) {
1773 PRINT_FILE_NAME(file);
1774 PRINT_ERR_MSG_WITH_ERRNO(NGMSG_FILE_EXTENT);
1775 }
1776 goto out;
1777 }
1778
1779check_improvement:
1780 if (mode_flag & DETAIL) {
1781 if (file_frags_start != 1)
1782 frag_files_before_defrag++;
1783
1784 extents_before_defrag += file_frags_start;
1785 }
1786
1787 if (file_frags_start <= best ||
1788 orig_physical_cnt <= donor_physical_cnt) {
1789 printf("\033[79;0H\033[K[%u/%u]%s:\t%3d%%",
1790 defraged_file_count, total_count, file, 100);
1791 if (mode_flag & DETAIL)
1792 printf(" extents: %d -> %d",
1793 file_frags_start, file_frags_start);
1794
1795 printf("\t[ OK ]\n");
1796 succeed_cnt++;
1797
1798 if (file_frags_start != 1)
1799 frag_files_after_defrag++;
1800
1801 extents_after_defrag += file_frags_start;
1802 goto out;
1803 }
1804
1805 /* Defrag the file */
1806 ret = call_defrag(fd, donor_fd, file, buf, donor_list_logical);
1807
1808 /* Count file fragments after defrag and print extents info */
1809 if (mode_flag & DETAIL) {
1810 file_frags_end = file_frag_count(fd);
1811 if (file_frags_end < 0) {
1812 printf("\n");
1813 PRINT_ERR_MSG_WITH_ERRNO(NGMSG_FILE_INFO);
1814 goto out;
1815 }
1816
1817 if (file_frags_end != 1)
1818 frag_files_after_defrag++;
1819
1820 extents_after_defrag += file_frags_end;
1821
1822 if (ret < 0)
1823 goto out;
1824
1825 printf(" extents: %d -> %d",
1826 file_frags_start, file_frags_end);
1827 fflush(stdout);
1828 }
1829
1830 if (ret < 0)
1831 goto out;
1832
1833 printf("\t[ OK ]\n");
1834 fflush(stdout);
1835 succeed_cnt++;
1836
1837out:
1838 close(fd);
1839 if (donor_fd != -1)
1840 close(donor_fd);
1841 free_ext(orig_list_physical);
1842 free_ext(orig_list_logical);
1843 free_ext(donor_list_physical);
1844 free_exts_group(orig_group_head);
1845 return 0;
1846}
1847
1848/*
1849 * main() - Ext4 online defrag.
1850 *
1851 * @argc: the number of parameter.
1852 * @argv[]: the pointer array of parameter.
1853 */
1854int main(int argc, char *argv[])
1855{
1856 int opt;
1857 int i, j;
1858 int flags = FTW_PHYS | FTW_MOUNT;
1859 int arg_type = -1;
1860 int success_flag = 0;
1861 char dir_name[PATH_MAX + 1];
1862 struct stat64 buf;
1863 struct ext4_super_block sb;
1864
1865 /* Parse arguments */
1866 if (argc == 1)
1867 goto out;
1868
1869 while ((opt = getopt(argc, argv, "vc")) != EOF) {
1870 switch (opt) {
1871 case 'v':
1872 mode_flag |= DETAIL;
1873 break;
1874 case 'c':
1875 mode_flag |= STATISTIC;
1876 break;
1877 default:
1878 goto out;
1879 }
1880 }
1881
1882 if (argc == optind)
1883 goto out;
1884
1885 current_uid = getuid();
1886
1887 /* Main process */
1888 for (i = optind; i < argc; i++) {
1889 succeed_cnt = 0;
1890 regular_count = 0;
1891 total_count = 0;
1892 frag_files_before_defrag = 0;
1893 frag_files_after_defrag = 0;
1894 extents_before_defrag = 0;
1895 extents_after_defrag = 0;
1896 defraged_file_count = 0;
1897 files_block_count = 0;
1898 blocks_per_group = 0;
1899 feature_incompat = 0;
1900 log_groups_per_flex = 0;
1901
1902 memset(dir_name, 0, PATH_MAX + 1);
1903 memset(lost_found_dir, 0, PATH_MAX + 1);
1904 memset(frag_rank, 0,
1905 sizeof(struct frag_statistic_ino) * SHOW_FRAG_FILES);
1906
1907 if ((mode_flag & STATISTIC) && i > optind)
1908 printf("\n");
1909
1910#if BYTE_ORDER != BIG_ENDIAN && BYTE_ORDER != LITTLE_ENDIAN
1911 PRINT_ERR_MSG("Endian's type is not big/little endian");
1912 PRINT_FILE_NAME(argv[i]);
1913 continue;
1914#endif
1915
1916 if (lstat64(argv[i], &buf) < 0) {
1917 perror(NGMSG_FILE_INFO);
1918 PRINT_FILE_NAME(argv[i]);
1919 continue;
1920 }
1921
1922 if (S_ISBLK(buf.st_mode)) {
1923 /* Block device */
1924 if (get_mount_point(argv[i], dir_name, PATH_MAX) < 0)
1925 continue;
1926 if (lstat64(dir_name, &buf) < 0) {
1927 perror(NGMSG_FILE_INFO);
1928 PRINT_FILE_NAME(argv[i]);
1929 continue;
1930 }
1931 arg_type = DEVNAME;
1932 if (!(mode_flag & STATISTIC))
1933 printf("ext4 defragmentation for device(%s)\n",
1934 argv[i]);
1935 } else if (S_ISDIR(buf.st_mode)) {
1936 /* Directory */
1937 if (access(argv[i], R_OK) < 0) {
1938 perror(argv[i]);
1939 continue;
1940 }
1941 arg_type = DIRNAME;
1942 strncpy(dir_name, argv[i], strnlen(argv[i], PATH_MAX));
1943 } else if (S_ISREG(buf.st_mode)) {
1944 /* Regular file */
1945 arg_type = FILENAME;
1946 } else {
1947 /* Irregular file */
1948 PRINT_ERR_MSG(NGMSG_FILE_UNREG);
1949 PRINT_FILE_NAME(argv[i]);
1950 continue;
1951 }
1952
1953 /* Set blocksize */
1954 block_size = buf.st_blksize;
1955
1956 /* For device case,
1957 * filesystem type checked in get_mount_point()
1958 */
1959 if (arg_type == FILENAME || arg_type == DIRNAME) {
1960 if (is_ext4(argv[i]) < 0)
1961 continue;
1962 if (realpath(argv[i], dir_name) == NULL) {
1963 perror("Couldn't get full path");
1964 PRINT_FILE_NAME(argv[i]);
1965 continue;
1966 }
1967 }
1968
1969 if (current_uid == ROOT_UID) {
1970 /* Get super block info */
1971 memset(&sb, 0, sizeof(struct ext4_super_block));
1972 if (get_superblock_info(dir_name, &sb) < 0) {
1973 if (mode_flag & DETAIL) {
1974 perror("Can't get super block info");
1975 PRINT_FILE_NAME(argv[i]);
1976 }
1977 continue;
1978 }
1979
1980 blocks_per_group = ext2fs_swab32(sb.s_blocks_per_group);
1981 feature_incompat = ext2fs_swab32(sb.s_feature_incompat);
1982 log_groups_per_flex = sb.s_log_groups_per_flex;
1983 }
1984
1985 switch (arg_type) {
1986 case DIRNAME:
1987 if (!(mode_flag & STATISTIC))
1988 printf("ext4 defragmentation "
1989 "for directory(%s)\n", argv[i]);
1990
1991 int mount_dir_len = 0;
1992 mount_dir_len = strnlen(lost_found_dir, PATH_MAX);
1993
1994 strncat(lost_found_dir, "/lost+found",
1995 PATH_MAX - strnlen(lost_found_dir, PATH_MAX));
1996
1997 /* Not the case("e4defrag mount_piont_dir") */
1998 if (dir_name[mount_dir_len] != '\0') {
1999 /*
2000 * "e4defrag mount_piont_dir/lost+found"
2001 * or "e4defrag mount_piont_dir/lost+found/"
2002 */
2003 if (strncmp(lost_found_dir, dir_name,
2004 strnlen(lost_found_dir,
2005 PATH_MAX)) == 0 &&
2006 (dir_name[strnlen(lost_found_dir,
2007 PATH_MAX)] == '\0' ||
2008 dir_name[strnlen(lost_found_dir,
2009 PATH_MAX)] == '/')) {
2010 PRINT_ERR_MSG(NGMSG_LOST_FOUND);
2011 PRINT_FILE_NAME(argv[i]);
2012 continue;
2013 }
2014
2015 /* "e4defrag mount_piont_dir/else_dir" */
2016 memset(lost_found_dir, 0, PATH_MAX + 1);
2017 }
2018 case DEVNAME:
2019 if (arg_type == DEVNAME) {
2020 strncpy(lost_found_dir, dir_name,
2021 strnlen(dir_name, PATH_MAX));
2022 strncat(lost_found_dir, "/lost+found/",
2023 PATH_MAX - strnlen(lost_found_dir,
2024 PATH_MAX));
2025 }
2026
2027 nftw64(dir_name, calc_entry_counts, FTW_OPEN_FD, flags);
2028
2029 if (mode_flag & STATISTIC) {
2030 if (mode_flag & DETAIL)
2031 printf("%-40s%10s/%-10s%9s\n",
Kazuya Mio94d26c22010-12-13 09:57:26 -05002032 "<File>", "now", "best", "size/ext");
Theodore Ts'o48364592009-07-09 14:29:12 -04002033
2034 if (!(mode_flag & DETAIL) &&
2035 current_uid != ROOT_UID) {
2036 printf(" Done.\n");
Kazuya Mio4baacad2010-12-13 09:59:07 -05002037 success_flag = 1;
Theodore Ts'o48364592009-07-09 14:29:12 -04002038 continue;
2039 }
2040
2041 nftw64(dir_name, file_statistic,
2042 FTW_OPEN_FD, flags);
2043
2044 if (succeed_cnt != 0 &&
2045 current_uid == ROOT_UID) {
2046 if (mode_flag & DETAIL)
2047 printf("\n");
2048 printf("%-40s%10s/%-10s%9s\n",
2049 "<Fragmented files>", "now",
Kazuya Mio94d26c22010-12-13 09:57:26 -05002050 "best", "size/ext");
Theodore Ts'o48364592009-07-09 14:29:12 -04002051 for (j = 0; j < SHOW_FRAG_FILES; j++) {
2052 if (strlen(frag_rank[j].
2053 msg_buffer) > 37) {
2054 printf("%d. %s\n%50d/"
Kazuya Mio94d26c22010-12-13 09:57:26 -05002055 "%-10d%6llu KB\n",
2056 j + 1,
Theodore Ts'o48364592009-07-09 14:29:12 -04002057 frag_rank[j].msg_buffer,
2058 frag_rank[j].now_count,
2059 frag_rank[j].best_count,
Kazuya Mio94d26c22010-12-13 09:57:26 -05002060 frag_rank[j].
2061 size_per_ext);
Theodore Ts'o48364592009-07-09 14:29:12 -04002062 } else if (strlen(frag_rank[j].
2063 msg_buffer) > 0) {
2064 printf("%d. %-37s%10d/"
Kazuya Mio94d26c22010-12-13 09:57:26 -05002065 "%-10d%6llu KB\n",
2066 j + 1,
Theodore Ts'o48364592009-07-09 14:29:12 -04002067 frag_rank[j].msg_buffer,
2068 frag_rank[j].now_count,
2069 frag_rank[j].best_count,
Kazuya Mio94d26c22010-12-13 09:57:26 -05002070 frag_rank[j].
2071 size_per_ext);
Theodore Ts'o48364592009-07-09 14:29:12 -04002072 } else
2073 break;
2074 }
2075 }
2076 break;
2077 }
2078 /* File tree walk */
2079 nftw64(dir_name, file_defrag, FTW_OPEN_FD, flags);
2080 printf("\n\tSuccess:\t\t\t[ %u/%u ]\n", succeed_cnt,
2081 total_count);
2082 printf("\tFailure:\t\t\t[ %u/%u ]\n",
2083 total_count - succeed_cnt, total_count);
2084 if (mode_flag & DETAIL) {
2085 printf("\tTotal extents:\t\t\t%4d->%d\n",
2086 extents_before_defrag,
2087 extents_after_defrag);
2088 printf("\tFragmented percentage:\t\t"
2089 "%3llu%%->%llu%%\n",
2090 !regular_count ? 0 :
2091 ((unsigned long long)
2092 frag_files_before_defrag * 100) /
2093 regular_count,
2094 !regular_count ? 0 :
2095 ((unsigned long long)
2096 frag_files_after_defrag * 100) /
2097 regular_count);
2098 }
2099 break;
2100 case FILENAME:
2101 total_count = 1;
2102 regular_count = 1;
2103 strncat(lost_found_dir, "/lost+found/",
2104 PATH_MAX - strnlen(lost_found_dir,
2105 PATH_MAX));
2106 if (strncmp(lost_found_dir, dir_name,
2107 strnlen(lost_found_dir,
2108 PATH_MAX)) == 0) {
2109 PRINT_ERR_MSG(NGMSG_LOST_FOUND);
2110 PRINT_FILE_NAME(argv[i]);
2111 continue;
2112 }
2113
2114 if (mode_flag & STATISTIC) {
2115 file_statistic(argv[i], &buf, FTW_F, NULL);
2116 break;
2117 } else
2118 printf("ext4 defragmentation for %s\n",
2119 argv[i]);
2120 /* Defrag single file process */
2121 file_defrag(argv[i], &buf, FTW_F, NULL);
2122 if (succeed_cnt != 0)
2123 printf(" Success:\t\t\t[1/1]\n");
2124 else
2125 printf(" Success:\t\t\t[0/1]\n");
2126
2127 break;
2128 }
2129
2130 if (succeed_cnt != 0)
2131 success_flag = 1;
2132 if (mode_flag & STATISTIC) {
2133 if (current_uid != ROOT_UID) {
2134 printf(" Done.\n");
2135 continue;
2136 }
2137
2138 if (!succeed_cnt) {
2139 if (mode_flag & DETAIL)
2140 printf("\n");
2141
2142 if (arg_type == DEVNAME)
2143 printf(" In this device(%s), "
2144 "none can be defragmented.\n", argv[i]);
2145 else if (arg_type == DIRNAME)
2146 printf(" In this directory(%s), "
2147 "none can be defragmented.\n", argv[i]);
2148 else
2149 printf(" This file(%s) "
2150 "can't be defragmented.\n", argv[i]);
2151 } else {
2152 float files_ratio = 0.0;
2153 float score = 0.0;
Kazuya Mio94d26c22010-12-13 09:57:26 -05002154 __u64 size_per_ext = files_block_count *
2155 (buf.st_blksize / 1024) /
2156 extents_before_defrag;
Theodore Ts'o48364592009-07-09 14:29:12 -04002157 files_ratio = (float)(extents_before_defrag -
2158 extents_after_defrag) *
2159 100 / files_block_count;
2160 score = CALC_SCORE(files_ratio);
2161 printf("\n Total/best extents\t\t\t\t%d/%d\n"
Kazuya Mio94d26c22010-12-13 09:57:26 -05002162 " Average size per extent"
2163 "\t\t\t%llu KB\n"
2164 " Fragmentation score\t\t\t\t%.0f\n",
Theodore Ts'o48364592009-07-09 14:29:12 -04002165 extents_before_defrag,
2166 extents_after_defrag,
Kazuya Mio94d26c22010-12-13 09:57:26 -05002167 size_per_ext, score);
Theodore Ts'o48364592009-07-09 14:29:12 -04002168 printf(" [0-30 no problem:"
2169 " 31-55 a little bit fragmented:"
Kazuya Mio94d26c22010-12-13 09:57:26 -05002170 " 56- needs defrag]\n");
Theodore Ts'o48364592009-07-09 14:29:12 -04002171
2172 if (arg_type == DEVNAME)
Kazuya Mio94d26c22010-12-13 09:57:26 -05002173 printf(" This device (%s) ", argv[i]);
Theodore Ts'o48364592009-07-09 14:29:12 -04002174 else if (arg_type == DIRNAME)
Kazuya Mio94d26c22010-12-13 09:57:26 -05002175 printf(" This directory (%s) ",
2176 argv[i]);
Theodore Ts'o48364592009-07-09 14:29:12 -04002177 else
Kazuya Mio94d26c22010-12-13 09:57:26 -05002178 printf(" This file (%s) ", argv[i]);
Theodore Ts'o48364592009-07-09 14:29:12 -04002179
2180 if (score > BOUND_SCORE)
2181 printf("needs defragmentation.\n");
2182 else
2183 printf("does not need "
2184 "defragmentation.\n");
2185 }
2186 printf(" Done.\n");
2187 }
2188
2189 }
2190
2191 if (success_flag)
2192 return 0;
2193
2194 exit(1);
2195
2196out:
2197 printf(MSG_USAGE);
2198 exit(1);
2199}
2200