symlinks.con commit Merge branch 'lt/maint-diff-reduce-lstat' into maint (5e04a1e)
   1#include "cache.h"
   2
   3/*
   4 * Returns the length (on a path component basis) of the longest
   5 * common prefix match of 'name_a' and 'name_b'.
   6 */
   7static int longest_path_match(const char *name_a, int len_a,
   8                              const char *name_b, int len_b,
   9                              int *previous_slash)
  10{
  11        int max_len, match_len = 0, match_len_prev = 0, i = 0;
  12
  13        max_len = len_a < len_b ? len_a : len_b;
  14        while (i < max_len && name_a[i] == name_b[i]) {
  15                if (name_a[i] == '/') {
  16                        match_len_prev = match_len;
  17                        match_len = i;
  18                }
  19                i++;
  20        }
  21        /*
  22         * Is 'name_b' a substring of 'name_a', the other way around,
  23         * or is 'name_a' and 'name_b' the exact same string?
  24         */
  25        if (i >= max_len && ((len_a > len_b && name_a[len_b] == '/') ||
  26                             (len_a < len_b && name_b[len_a] == '/') ||
  27                             (len_a == len_b))) {
  28                match_len_prev = match_len;
  29                match_len = i;
  30        }
  31        *previous_slash = match_len_prev;
  32        return match_len;
  33}
  34
  35static struct cache_def {
  36        char path[PATH_MAX + 1];
  37        int len;
  38        int flags;
  39        int track_flags;
  40        int prefix_len_stat_func;
  41} cache;
  42
  43static inline void reset_lstat_cache(void)
  44{
  45        cache.path[0] = '\0';
  46        cache.len = 0;
  47        cache.flags = 0;
  48        /*
  49         * The track_flags and prefix_len_stat_func members is only
  50         * set by the safeguard rule inside lstat_cache()
  51         */
  52}
  53
  54#define FL_DIR      (1 << 0)
  55#define FL_NOENT    (1 << 1)
  56#define FL_SYMLINK  (1 << 2)
  57#define FL_LSTATERR (1 << 3)
  58#define FL_ERR      (1 << 4)
  59#define FL_FULLPATH (1 << 5)
  60
  61/*
  62 * Check if name 'name' of length 'len' has a symlink leading
  63 * component, or if the directory exists and is real, or not.
  64 *
  65 * To speed up the check, some information is allowed to be cached.
  66 * This can be indicated by the 'track_flags' argument, which also can
  67 * be used to indicate that we should check the full path.
  68 *
  69 * The 'prefix_len_stat_func' parameter can be used to set the length
  70 * of the prefix, where the cache should use the stat() function
  71 * instead of the lstat() function to test each path component.
  72 */
  73static int lstat_cache(const char *name, int len,
  74                       int track_flags, int prefix_len_stat_func)
  75{
  76        int match_len, last_slash, last_slash_dir, previous_slash;
  77        int match_flags, ret_flags, save_flags, max_len, ret;
  78        struct stat st;
  79
  80        if (cache.track_flags != track_flags ||
  81            cache.prefix_len_stat_func != prefix_len_stat_func) {
  82                /*
  83                 * As a safeguard rule we clear the cache if the
  84                 * values of track_flags and/or prefix_len_stat_func
  85                 * does not match with the last supplied values.
  86                 */
  87                reset_lstat_cache();
  88                cache.track_flags = track_flags;
  89                cache.prefix_len_stat_func = prefix_len_stat_func;
  90                match_len = last_slash = 0;
  91        } else {
  92                /*
  93                 * Check to see if we have a match from the cache for
  94                 * the 2 "excluding" path types.
  95                 */
  96                match_len = last_slash =
  97                        longest_path_match(name, len, cache.path, cache.len,
  98                                           &previous_slash);
  99                match_flags = cache.flags & track_flags & (FL_NOENT|FL_SYMLINK);
 100                if (match_flags && match_len == cache.len)
 101                        return match_flags;
 102                /*
 103                 * If we now have match_len > 0, we would know that
 104                 * the matched part will always be a directory.
 105                 *
 106                 * Also, if we are tracking directories and 'name' is
 107                 * a substring of the cache on a path component basis,
 108                 * we can return immediately.
 109                 */
 110                match_flags = track_flags & FL_DIR;
 111                if (match_flags && len == match_len)
 112                        return match_flags;
 113        }
 114
 115        /*
 116         * Okay, no match from the cache so far, so now we have to
 117         * check the rest of the path components.
 118         */
 119        ret_flags = FL_DIR;
 120        last_slash_dir = last_slash;
 121        max_len = len < PATH_MAX ? len : PATH_MAX;
 122        while (match_len < max_len) {
 123                do {
 124                        cache.path[match_len] = name[match_len];
 125                        match_len++;
 126                } while (match_len < max_len && name[match_len] != '/');
 127                if (match_len >= max_len && !(track_flags & FL_FULLPATH))
 128                        break;
 129                last_slash = match_len;
 130                cache.path[last_slash] = '\0';
 131
 132                if (last_slash <= prefix_len_stat_func)
 133                        ret = stat(cache.path, &st);
 134                else
 135                        ret = lstat(cache.path, &st);
 136
 137                if (ret) {
 138                        ret_flags = FL_LSTATERR;
 139                        if (errno == ENOENT)
 140                                ret_flags |= FL_NOENT;
 141                } else if (S_ISDIR(st.st_mode)) {
 142                        last_slash_dir = last_slash;
 143                        continue;
 144                } else if (S_ISLNK(st.st_mode)) {
 145                        ret_flags = FL_SYMLINK;
 146                } else {
 147                        ret_flags = FL_ERR;
 148                }
 149                break;
 150        }
 151
 152        /*
 153         * At the end update the cache.  Note that max 3 different
 154         * path types, FL_NOENT, FL_SYMLINK and FL_DIR, can be cached
 155         * for the moment!
 156         */
 157        save_flags = ret_flags & track_flags & (FL_NOENT|FL_SYMLINK);
 158        if (save_flags && last_slash > 0 && last_slash <= PATH_MAX) {
 159                cache.path[last_slash] = '\0';
 160                cache.len = last_slash;
 161                cache.flags = save_flags;
 162        } else if ((track_flags & FL_DIR) &&
 163                   last_slash_dir > 0 && last_slash_dir <= PATH_MAX) {
 164                /*
 165                 * We have a separate test for the directory case,
 166                 * since it could be that we have found a symlink or a
 167                 * non-existing directory and the track_flags says
 168                 * that we cannot cache this fact, so the cache would
 169                 * then have been left empty in this case.
 170                 *
 171                 * But if we are allowed to track real directories, we
 172                 * can still cache the path components before the last
 173                 * one (the found symlink or non-existing component).
 174                 */
 175                cache.path[last_slash_dir] = '\0';
 176                cache.len = last_slash_dir;
 177                cache.flags = FL_DIR;
 178        } else {
 179                reset_lstat_cache();
 180        }
 181        return ret_flags;
 182}
 183
 184/*
 185 * Invalidate the given 'name' from the cache, if 'name' matches
 186 * completely with the cache.
 187 */
 188void invalidate_lstat_cache(const char *name, int len)
 189{
 190        int match_len, previous_slash;
 191
 192        match_len = longest_path_match(name, len, cache.path, cache.len,
 193                                       &previous_slash);
 194        if (len == match_len) {
 195                if ((cache.track_flags & FL_DIR) && previous_slash > 0) {
 196                        cache.path[previous_slash] = '\0';
 197                        cache.len = previous_slash;
 198                        cache.flags = FL_DIR;
 199                } else
 200                        reset_lstat_cache();
 201        }
 202}
 203
 204/*
 205 * Completely clear the contents of the cache
 206 */
 207void clear_lstat_cache(void)
 208{
 209        reset_lstat_cache();
 210}
 211
 212#define USE_ONLY_LSTAT  0
 213
 214/*
 215 * Return non-zero if path 'name' has a leading symlink component
 216 */
 217int has_symlink_leading_path(const char *name, int len)
 218{
 219        return lstat_cache(name, len,
 220                           FL_SYMLINK|FL_DIR, USE_ONLY_LSTAT) &
 221                FL_SYMLINK;
 222}
 223
 224/*
 225 * Return non-zero if path 'name' has a leading symlink component or
 226 * if some leading path component does not exists.
 227 */
 228int has_symlink_or_noent_leading_path(const char *name, int len)
 229{
 230        return lstat_cache(name, len,
 231                           FL_SYMLINK|FL_NOENT|FL_DIR, USE_ONLY_LSTAT) &
 232                (FL_SYMLINK|FL_NOENT);
 233}
 234
 235/*
 236 * Return non-zero if all path components of 'name' exists as a
 237 * directory.  If prefix_len > 0, we will test with the stat()
 238 * function instead of the lstat() function for a prefix length of
 239 * 'prefix_len', thus we then allow for symlinks in the prefix part as
 240 * long as those points to real existing directories.
 241 */
 242int has_dirs_only_path(const char *name, int len, int prefix_len)
 243{
 244        return lstat_cache(name, len,
 245                           FL_DIR|FL_FULLPATH, prefix_len) &
 246                FL_DIR;
 247}
 248
 249static struct removal_def {
 250        char path[PATH_MAX];
 251        int len;
 252} removal;
 253
 254static void do_remove_scheduled_dirs(int new_len)
 255{
 256        while (removal.len > new_len) {
 257                removal.path[removal.len] = '\0';
 258                if (rmdir(removal.path))
 259                        break;
 260                do {
 261                        removal.len--;
 262                } while (removal.len > new_len &&
 263                         removal.path[removal.len] != '/');
 264        }
 265        removal.len = new_len;
 266        return;
 267}
 268
 269void schedule_dir_for_removal(const char *name, int len)
 270{
 271        int match_len, last_slash, i, previous_slash;
 272
 273        match_len = last_slash = i =
 274                longest_path_match(name, len, removal.path, removal.len,
 275                                   &previous_slash);
 276        /* Find last slash inside 'name' */
 277        while (i < len) {
 278                if (name[i] == '/')
 279                        last_slash = i;
 280                i++;
 281        }
 282
 283        /*
 284         * If we are about to go down the directory tree, we check if
 285         * we must first go upwards the tree, such that we then can
 286         * remove possible empty directories as we go upwards.
 287         */
 288        if (match_len < last_slash && match_len < removal.len)
 289                do_remove_scheduled_dirs(match_len);
 290        /*
 291         * If we go deeper down the directory tree, we only need to
 292         * save the new path components as we go down.
 293         */
 294        if (match_len < last_slash) {
 295                memcpy(&removal.path[match_len], &name[match_len],
 296                       last_slash - match_len);
 297                removal.len = last_slash;
 298        }
 299        return;
 300}
 301
 302void remove_scheduled_dirs(void)
 303{
 304        do_remove_scheduled_dirs(0);
 305        return;
 306}