| 1 | /* |
| 2 | * du.c: implementation of du.h. |
| 3 | */ |
| 4 | |
| 5 | #include "agedu.h" |
| 6 | #include "du.h" |
| 7 | #include "alloc.h" |
| 8 | |
| 9 | #if !defined __linux__ || !defined O_NOATIME || defined HAVE_FDOPENDIR |
| 10 | |
| 11 | #ifdef HAVE_DIRENT_H |
| 12 | # include <dirent.h> |
| 13 | #endif |
| 14 | #ifdef HAVE_NDIR_H |
| 15 | # include <ndir.h> |
| 16 | #endif |
| 17 | #ifdef HAVE_SYS_DIR_H |
| 18 | # include <sys/dir.h> |
| 19 | #endif |
| 20 | #ifdef HAVE_SYS_NDIR_H |
| 21 | # include <sys/ndir.h> |
| 22 | #endif |
| 23 | |
| 24 | /* |
| 25 | * Wrappers around POSIX opendir, readdir and closedir, which |
| 26 | * permit me to replace them with different wrappers in special |
| 27 | * circumstances. |
| 28 | */ |
| 29 | |
| 30 | typedef DIR *dirhandle; |
| 31 | |
| 32 | int open_dir(const char *path, dirhandle *dh) |
| 33 | { |
| 34 | #if defined O_NOATIME && defined HAVE_FDOPENDIR |
| 35 | |
| 36 | /* |
| 37 | * On Linux, we have the O_NOATIME flag. This means we can |
| 38 | * read the contents of directories without affecting their |
| 39 | * atimes, which enables us to at least try to include them in |
| 40 | * the age display rather than exempting them. |
| 41 | * |
| 42 | * Unfortunately, opendir() doesn't let us open a directory |
| 43 | * with O_NOATIME. So instead, we have to open the directory |
| 44 | * with vanilla open(), and then use fdopendir() to translate |
| 45 | * the fd into a POSIX dir handle. |
| 46 | */ |
| 47 | int fd; |
| 48 | |
| 49 | fd = open(path, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_LARGEFILE | |
| 50 | O_NOATIME | O_DIRECTORY); |
| 51 | if (fd < 0) { |
| 52 | /* |
| 53 | * Opening a file with O_NOATIME is not unconditionally |
| 54 | * permitted by the Linux kernel. As far as I can tell, |
| 55 | * it's permitted only for files on which the user would |
| 56 | * have been able to call utime(2): in other words, files |
| 57 | * for which the user could have deliberately set the |
| 58 | * atime back to its original value after finishing with |
| 59 | * it. Hence, O_NOATIME has no security implications; it's |
| 60 | * simply a cleaner, faster and more race-condition-free |
| 61 | * alternative to stat(), a normal open(), and a utimes() |
| 62 | * when finished. |
| 63 | * |
| 64 | * The upshot of all of which, for these purposes, is that |
| 65 | * we must be prepared to try again without O_NOATIME if |
| 66 | * we receive EPERM. |
| 67 | */ |
| 68 | if (errno == EPERM) |
| 69 | fd = open(path, O_RDONLY | O_NONBLOCK | O_NOCTTY | |
| 70 | O_LARGEFILE | O_DIRECTORY); |
| 71 | if (fd < 0) |
| 72 | return -1; |
| 73 | } |
| 74 | |
| 75 | *dh = fdopendir(fd); |
| 76 | #else |
| 77 | *dh = opendir(path); |
| 78 | #endif |
| 79 | |
| 80 | if (!*dh) |
| 81 | return -1; |
| 82 | return 0; |
| 83 | } |
| 84 | |
| 85 | const char *read_dir(dirhandle *dh) |
| 86 | { |
| 87 | struct dirent *de = readdir(*dh); |
| 88 | return de ? de->d_name : NULL; |
| 89 | } |
| 90 | |
| 91 | void close_dir(dirhandle *dh) |
| 92 | { |
| 93 | closedir(*dh); |
| 94 | } |
| 95 | |
| 96 | #else /* defined __linux__ && !defined HAVE_FDOPENDIR */ |
| 97 | |
| 98 | /* |
| 99 | * Earlier versions of glibc do not have fdopendir(). Therefore, |
| 100 | * if we are on Linux and still wish to make use of O_NOATIME, we |
| 101 | * have no option but to talk directly to the kernel system call |
| 102 | * interface which underlies the POSIX opendir/readdir machinery. |
| 103 | */ |
| 104 | |
| 105 | #define __KERNEL__ |
| 106 | #include <linux/types.h> |
| 107 | #include <linux/dirent.h> |
| 108 | #include <linux/unistd.h> |
| 109 | |
| 110 | _syscall3(int, getdents, uint, fd, struct dirent *, dirp, uint, count) |
| 111 | |
| 112 | typedef struct { |
| 113 | int fd; |
| 114 | struct dirent data[32]; |
| 115 | struct dirent *curr; |
| 116 | int pos, endpos; |
| 117 | } dirhandle; |
| 118 | |
| 119 | int open_dir(const char *path, dirhandle *dh) |
| 120 | { |
| 121 | /* |
| 122 | * As above, we try with O_NOATIME and then fall back to |
| 123 | * trying without it. |
| 124 | */ |
| 125 | dh->fd = open(path, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_LARGEFILE | |
| 126 | O_NOATIME | O_DIRECTORY); |
| 127 | if (dh->fd < 0) { |
| 128 | if (errno == EPERM) |
| 129 | dh->fd = open(path, O_RDONLY | O_NONBLOCK | O_NOCTTY | |
| 130 | O_LARGEFILE | O_DIRECTORY); |
| 131 | if (dh->fd < 0) |
| 132 | return -1; |
| 133 | } |
| 134 | |
| 135 | dh->pos = dh->endpos = 0; |
| 136 | |
| 137 | return 0; |
| 138 | } |
| 139 | |
| 140 | const char *read_dir(dirhandle *dh) |
| 141 | { |
| 142 | const char *ret; |
| 143 | |
| 144 | if (dh->pos >= dh->endpos) { |
| 145 | dh->curr = dh->data; |
| 146 | dh->pos = 0; |
| 147 | dh->endpos = getdents(dh->fd, dh->data, sizeof(dh->data)); |
| 148 | if (dh->endpos <= 0) |
| 149 | return NULL; |
| 150 | } |
| 151 | |
| 152 | ret = dh->curr->d_name; |
| 153 | |
| 154 | dh->pos += dh->curr->d_reclen; |
| 155 | dh->curr = (struct dirent *)((char *)dh->data + dh->pos); |
| 156 | |
| 157 | return ret; |
| 158 | } |
| 159 | |
| 160 | void close_dir(dirhandle *dh) |
| 161 | { |
| 162 | close(dh->fd); |
| 163 | } |
| 164 | |
| 165 | #endif /* !defined __linux__ || defined HAVE_FDOPENDIR */ |
| 166 | |
| 167 | static int str_cmp(const void *av, const void *bv) |
| 168 | { |
| 169 | return strcmp(*(const char **)av, *(const char **)bv); |
| 170 | } |
| 171 | |
| 172 | static void du_recurse(char **path, size_t pathlen, size_t *pathsize, |
| 173 | gotdata_fn_t gotdata, err_fn_t err, void *gotdata_ctx, |
| 174 | int toplevel) |
| 175 | { |
| 176 | const char *name; |
| 177 | dirhandle d; |
| 178 | STRUCT_STAT st; |
| 179 | char **names; |
| 180 | size_t i, nnames, namesize; |
| 181 | int statret; |
| 182 | |
| 183 | /* |
| 184 | * Special case: at the very top of the scan, we follow a |
| 185 | * symlink. |
| 186 | */ |
| 187 | if (toplevel) |
| 188 | statret = STAT(*path, &st); |
| 189 | else |
| 190 | statret = LSTAT(*path, &st); |
| 191 | if (statret < 0) { |
| 192 | err(gotdata_ctx, "%s: lstat: %s\n", *path, strerror(errno)); |
| 193 | return; |
| 194 | } |
| 195 | |
| 196 | if (!gotdata(gotdata_ctx, *path, &st)) |
| 197 | return; |
| 198 | |
| 199 | if (!S_ISDIR(st.st_mode)) |
| 200 | return; |
| 201 | |
| 202 | names = NULL; |
| 203 | nnames = namesize = 0; |
| 204 | |
| 205 | if (open_dir(*path, &d) < 0) { |
| 206 | err(gotdata_ctx, "%s: opendir: %s\n", *path, strerror(errno)); |
| 207 | return; |
| 208 | } |
| 209 | while ((name = read_dir(&d)) != NULL) { |
| 210 | if (name[0] == '.' && (!name[1] || (name[1] == '.' && !name[2]))) { |
| 211 | /* do nothing - we skip "." and ".." */ |
| 212 | } else { |
| 213 | if (nnames >= namesize) { |
| 214 | namesize = nnames * 3 / 2 + 64; |
| 215 | names = sresize(names, namesize, char *); |
| 216 | } |
| 217 | names[nnames++] = dupstr(name); |
| 218 | } |
| 219 | } |
| 220 | close_dir(&d); |
| 221 | |
| 222 | if (nnames == 0) |
| 223 | return; |
| 224 | |
| 225 | qsort(names, nnames, sizeof(*names), str_cmp); |
| 226 | |
| 227 | for (i = 0; i < nnames; i++) { |
| 228 | size_t newpathlen = pathlen + 1 + strlen(names[i]); |
| 229 | if (*pathsize <= newpathlen) { |
| 230 | *pathsize = newpathlen * 3 / 2 + 256; |
| 231 | *path = sresize(*path, *pathsize, char); |
| 232 | } |
| 233 | /* |
| 234 | * Avoid duplicating a slash if we got a trailing one to |
| 235 | * begin with (i.e. if we're starting the scan in '/' itself). |
| 236 | */ |
| 237 | if (pathlen > 0 && (*path)[pathlen-1] == '/') { |
| 238 | strcpy(*path + pathlen, names[i]); |
| 239 | newpathlen--; |
| 240 | } else { |
| 241 | sprintf(*path + pathlen, "/%s", names[i]); |
| 242 | } |
| 243 | |
| 244 | du_recurse(path, newpathlen, pathsize, gotdata, err, gotdata_ctx, 0); |
| 245 | |
| 246 | sfree(names[i]); |
| 247 | } |
| 248 | sfree(names); |
| 249 | } |
| 250 | |
| 251 | void du(const char *inpath, gotdata_fn_t gotdata, err_fn_t err, |
| 252 | void *gotdata_ctx) |
| 253 | { |
| 254 | char *path; |
| 255 | size_t pathlen, pathsize; |
| 256 | |
| 257 | pathlen = strlen(inpath); |
| 258 | pathsize = pathlen + 256; |
| 259 | path = snewn(pathsize, char); |
| 260 | strcpy(path, inpath); |
| 261 | |
| 262 | du_recurse(&path, pathlen, &pathsize, gotdata, err, gotdata_ctx, 1); |
| 263 | } |