fs/overlayfs/readdir.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/slab.h>
  12 #include <linux/namei.h>
  13 #include <linux/file.h>
  14 #include <linux/xattr.h>
  15 #include <linux/rbtree.h>
  16 #include <linux/security.h>
  17 #include <linux/cred.h>
  18 #include "overlayfs.h"
  19
  20 struct ovl_cache_entry {
  21         unsigned int len;
  22         unsigned int type;
  23         u64 ino;
  24         struct list_head l_node;
  25         struct rb_node node;
  26         bool is_whiteout;
  27         bool is_cursor;
  28         char name[];
  29 };
  30
  31 struct ovl_dir_cache {
  32         long refcount;
  33         u64 version;
  34         struct list_head entries;
  35 };
  36
  37 struct ovl_readdir_data {
  38         struct dir_context ctx;
  39         bool is_merge;
  40         struct rb_root root;
  41         struct list_head *list;
  42         struct list_head middle;
  43         int count;
  44         int err;
  45 };
  46
  47 struct ovl_dir_file {
  48         bool is_real;
  49         bool is_upper;
  50         struct ovl_dir_cache *cache;
  51         struct ovl_cache_entry cursor;
  52         struct file *realfile;
  53         struct file *upperfile;
  54 };
  55
  56 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
  57 {
  58         return container_of(n, struct ovl_cache_entry, node);
  59 }
  60
  61 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
  62                                                     const char *name, int len)
  63 {
  64         struct rb_node *node = root->rb_node;
  65         int cmp;
  66
  67         while (node) {
  68                 struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
  69
  70                 cmp = strncmp(name, p->name, len);
  71                 if (cmp > 0)
  72                         node = p->node.rb_right;
  73                 else if (cmp < 0 || len < p->len)
  74                         node = p->node.rb_left;
  75                 else
  76                         return p;
  77         }
  78
  79         return NULL;
  80 }
  81
  82 static struct ovl_cache_entry *ovl_cache_entry_new(const char *name, int len,
  83                                                    u64 ino, unsigned int d_type)
  84 {
  85         struct ovl_cache_entry *p;
  86         size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
  87
  88         p = kmalloc(size, GFP_KERNEL);
  89         if (p) {
  90                 memcpy(p->name, name, len);
  91                 p->name[len] = '\0';
  92                 p->len = len;
  93                 p->type = d_type;
  94                 p->ino = ino;
  95                 p->is_whiteout = false;
  96                 p->is_cursor = false;
  97         }
  98
  99         return p;
 100 }
 101
 102 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
 103                                   const char *name, int len, u64 ino,
 104                                   unsigned int d_type)
 105 {
 106         struct rb_node **newp = &rdd->root.rb_node;
 107         struct rb_node *parent = NULL;
 108         struct ovl_cache_entry *p;
 109
 110         while (*newp) {
 111                 int cmp;
 112                 struct ovl_cache_entry *tmp;
 113
 114                 parent = *newp;
 115                 tmp = ovl_cache_entry_from_node(*newp);
 116                 cmp = strncmp(name, tmp->name, len);
 117                 if (cmp > 0)
 118                         newp = &tmp->node.rb_right;
 119                 else if (cmp < 0 || len < tmp->len)
 120                         newp = &tmp->node.rb_left;
 121                 else
 122                         return 0;
 123         }
 124
 125         p = ovl_cache_entry_new(name, len, ino, d_type);
 126         if (p == NULL)
 127                 return -ENOMEM;
 128
 129         list_add_tail(&p->l_node, rdd->list);
 130         rb_link_node(&p->node, parent, newp);
 131         rb_insert_color(&p->node, &rdd->root);
 132
 133         return 0;
 134 }
 135
 136 static int ovl_fill_lower(struct ovl_readdir_data *rdd,
 137                           const char *name, int namelen,
 138                           loff_t offset, u64 ino, unsigned int d_type)
 139 {
 140         struct ovl_cache_entry *p;
 141
 142         p = ovl_cache_entry_find(&rdd->root, name, namelen);
 143         if (p) {
 144                 list_move_tail(&p->l_node, &rdd->middle);
 145         } else {
 146                 p = ovl_cache_entry_new(name, namelen, ino, d_type);
 147                 if (p == NULL)
 148                         rdd->err = -ENOMEM;
 149                 else
 150                         list_add_tail(&p->l_node, &rdd->middle);
 151         }
 152
 153         return rdd->err;
 154 }
 155
 156 void ovl_cache_free(struct list_head *list)
 157 {
 158         struct ovl_cache_entry *p;
 159         struct ovl_cache_entry *n;
 160
 161         list_for_each_entry_safe(p, n, list, l_node)
 162                 kfree(p);
 163
 164         INIT_LIST_HEAD(list);
 165 }
 166
 167 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
 168 {
 169         struct ovl_dir_cache *cache = od->cache;
 170
 171         list_del(&od->cursor.l_node);
 172         WARN_ON(cache->refcount <= 0);
 173         cache->refcount--;
 174         if (!cache->refcount) {
 175                 if (ovl_dir_cache(dentry) == cache)
 176                         ovl_set_dir_cache(dentry, NULL);
 177
 178                 ovl_cache_free(&cache->entries);
 179                 kfree(cache);
 180         }
 181 }
 182
 183 static int ovl_fill_merge(void *buf, const char *name, int namelen,
 184                           loff_t offset, u64 ino, unsigned int d_type)
 185 {
 186         struct ovl_readdir_data *rdd = buf;
 187
 188         rdd->count++;
 189         if (!rdd->is_merge)
 190                 return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
 191         else
 192                 return ovl_fill_lower(rdd, name, namelen, offset, ino, d_type);
 193 }
 194
 195 static inline int ovl_dir_read(struct path *realpath,
 196                                struct ovl_readdir_data *rdd)
 197 {
 198         struct file *realfile;
 199         int err;
 200
 201         realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
 202         if (IS_ERR(realfile))
 203                 return PTR_ERR(realfile);
 204
 205         rdd->ctx.pos = 0;
 206         do {
 207                 rdd->count = 0;
 208                 rdd->err = 0;
 209                 err = iterate_dir(realfile, &rdd->ctx);
 210                 if (err >= 0)
 211                         err = rdd->err;
 212         } while (!err && rdd->count);
 213         fput(realfile);
 214
 215         return err;
 216 }
 217
 218 static void ovl_dir_reset(struct file *file)
 219 {
 220         struct ovl_dir_file *od = file->private_data;
 221         struct ovl_dir_cache *cache = od->cache;
 222         struct dentry *dentry = file->f_path.dentry;
 223         enum ovl_path_type type = ovl_path_type(dentry);
 224
 225         if (cache && ovl_dentry_version_get(dentry) != cache->version) {
 226                 ovl_cache_put(od, dentry);
 227                 od->cache = NULL;
 228         }
 229         WARN_ON(!od->is_real && type != OVL_PATH_MERGE);
 230         if (od->is_real && type == OVL_PATH_MERGE)
 231                 od->is_real = false;
 232 }
 233
 234 static int ovl_dir_mark_whiteouts(struct dentry *dir,
 235                                   struct ovl_readdir_data *rdd)
 236 {
 237         struct ovl_cache_entry *p;
 238         struct dentry *dentry;
 239         const struct cred *old_cred;
 240         struct cred *override_cred;
 241
 242         override_cred = prepare_creds();
 243         if (!override_cred) {
 244                 ovl_cache_free(rdd->list);
 245                 return -ENOMEM;
 246         }
 247
 248         /*
 249          * CAP_DAC_OVERRIDE for lookup
 250          */
 251         cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
 252         old_cred = override_creds(override_cred);
 253
 254         mutex_lock(&dir->d_inode->i_mutex);
 255         list_for_each_entry(p, rdd->list, l_node) {
 256                 if (p->is_cursor)
 257                         continue;
 258
 259                 if (p->type != DT_CHR)
 260                         continue;
 261
 262                 dentry = lookup_one_len(p->name, dir, p->len);
 263                 if (IS_ERR(dentry))
 264                         continue;
 265
 266                 p->is_whiteout = ovl_is_whiteout(dentry);
 267                 dput(dentry);
 268         }
 269         mutex_unlock(&dir->d_inode->i_mutex);
 270
 271         revert_creds(old_cred);
 272         put_cred(override_cred);
 273
 274         return 0;
 275 }
 276
 277 static inline int ovl_dir_read_merged(struct path *upperpath,
 278                                       struct path *lowerpath,
 279                                       struct list_head *list)
 280 {
 281         int err;
 282         struct ovl_readdir_data rdd = {
 283                 .ctx.actor = ovl_fill_merge,
 284                 .list = list,
 285                 .root = RB_ROOT,
 286                 .is_merge = false,
 287         };
 288
 289         if (upperpath->dentry) {
 290                 err = ovl_dir_read(upperpath, &rdd);
 291                 if (err)
 292                         goto out;
 293
 294                 if (lowerpath->dentry) {
 295                         err = ovl_dir_mark_whiteouts(upperpath->dentry, &rdd);
 296                         if (err)
 297                                 goto out;
 298                 }
 299         }
 300         if (lowerpath->dentry) {
 301                 /*
 302                  * Insert lowerpath entries before upperpath ones, this allows
 303                  * offsets to be reasonably constant
 304                  */
 305                 list_add(&rdd.middle, rdd.list);
 306                 rdd.is_merge = true;
 307                 err = ovl_dir_read(lowerpath, &rdd);
 308                 list_del(&rdd.middle);
 309         }
 310 out:
 311         return err;
 312 }
 313
 314 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
 315 {
 316         struct ovl_cache_entry *p;
 317         loff_t off = 0;
 318
 319         list_for_each_entry(p, &od->cache->entries, l_node) {
 320                 if (p->is_cursor)
 321                         continue;
 322                 if (off >= pos)
 323                         break;
 324                 off++;
 325         }
 326         list_move_tail(&od->cursor.l_node, &p->l_node);
 327 }
 328
 329 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 330 {
 331         int res;
 332         struct path lowerpath;
 333         struct path upperpath;
 334         struct ovl_dir_cache *cache;
 335
 336         cache = ovl_dir_cache(dentry);
 337         if (cache && ovl_dentry_version_get(dentry) == cache->version) {
 338                 cache->refcount++;
 339                 return cache;
 340         }
 341         ovl_set_dir_cache(dentry, NULL);
 342
 343         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 344         if (!cache)
 345                 return ERR_PTR(-ENOMEM);
 346
 347         cache->refcount = 1;
 348         INIT_LIST_HEAD(&cache->entries);
 349
 350         ovl_path_lower(dentry, &lowerpath);
 351         ovl_path_upper(dentry, &upperpath);
 352
 353         res = ovl_dir_read_merged(&upperpath, &lowerpath, &cache->entries);
 354         if (res) {
 355                 ovl_cache_free(&cache->entries);
 356                 kfree(cache);
 357                 return ERR_PTR(res);
 358         }
 359
 360         cache->version = ovl_dentry_version_get(dentry);
 361         ovl_set_dir_cache(dentry, cache);
 362
 363         return cache;
 364 }
 365
 366 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 367 {
 368         struct ovl_dir_file *od = file->private_data;
 369         struct dentry *dentry = file->f_path.dentry;
 370
 371         if (!ctx->pos)
 372                 ovl_dir_reset(file);
 373
 374         if (od->is_real)
 375                 return iterate_dir(od->realfile, ctx);
 376
 377         if (!od->cache) {
 378                 struct ovl_dir_cache *cache;
 379
 380                 cache = ovl_cache_get(dentry);
 381                 if (IS_ERR(cache))
 382                         return PTR_ERR(cache);
 383
 384                 od->cache = cache;
 385                 ovl_seek_cursor(od, ctx->pos);
 386         }
 387
 388         while (od->cursor.l_node.next != &od->cache->entries) {
 389                 struct ovl_cache_entry *p;
 390
 391                 p = list_entry(od->cursor.l_node.next, struct ovl_cache_entry, l_node);
 392                 /* Skip cursors */
 393                 if (!p->is_cursor) {
 394                         if (!p->is_whiteout) {
 395                                 if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 396                                         break;
 397                         }
 398                         ctx->pos++;
 399                 }
 400                 list_move(&od->cursor.l_node, &p->l_node);
 401         }
 402         return 0;
 403 }
 404
 405 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 406 {
 407         loff_t res;
 408         struct ovl_dir_file *od = file->private_data;
 409
 410         mutex_lock(&file_inode(file)->i_mutex);
 411         if (!file->f_pos)
 412                 ovl_dir_reset(file);
 413
 414         if (od->is_real) {
 415                 res = vfs_llseek(od->realfile, offset, origin);
 416                 file->f_pos = od->realfile->f_pos;
 417         } else {
 418                 res = -EINVAL;
 419
 420                 switch (origin) {
 421                 case SEEK_CUR:
 422                         offset += file->f_pos;
 423                         break;
 424                 case SEEK_SET:
 425                         break;
 426                 default:
 427                         goto out_unlock;
 428                 }
 429                 if (offset < 0)
 430                         goto out_unlock;
 431
 432                 if (offset != file->f_pos) {
 433                         file->f_pos = offset;
 434                         if (od->cache)
 435                                 ovl_seek_cursor(od, offset);
 436                 }
 437                 res = offset;
 438         }
 439 out_unlock:
 440         mutex_unlock(&file_inode(file)->i_mutex);
 441
 442         return res;
 443 }
 444
 445 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 446                          int datasync)
 447 {
 448         struct ovl_dir_file *od = file->private_data;
 449         struct dentry *dentry = file->f_path.dentry;
 450         struct file *realfile = od->realfile;
 451
 452         /*
 453          * Need to check if we started out being a lower dir, but got copied up
 454          */
 455         if (!od->is_upper && ovl_path_type(dentry) == OVL_PATH_MERGE) {
 456                 struct inode *inode = file_inode(file);
 457
 458                 realfile =lockless_dereference(od->upperfile);
 459                 if (!realfile) {
 460                         struct path upperpath;
 461
 462                         ovl_path_upper(dentry, &upperpath);
 463                         realfile = ovl_path_open(&upperpath, O_RDONLY);
 464                         smp_mb__before_spinlock();
 465                         mutex_lock(&inode->i_mutex);
 466                         if (!od->upperfile) {
 467                                 if (IS_ERR(realfile)) {
 468                                         mutex_unlock(&inode->i_mutex);
 469                                         return PTR_ERR(realfile);
 470                                 }
 471                                 od->upperfile = realfile;
 472                         } else {
 473                                 /* somebody has beaten us to it */
 474                                 if (!IS_ERR(realfile))
 475                                         fput(realfile);
 476                                 realfile = od->upperfile;
 477                         }
 478                         mutex_unlock(&inode->i_mutex);
 479                 }
 480         }
 481
 482         return vfs_fsync_range(realfile, start, end, datasync);
 483 }
 484
 485 static int ovl_dir_release(struct inode *inode, struct file *file)
 486 {
 487         struct ovl_dir_file *od = file->private_data;
 488
 489         if (od->cache) {
 490                 mutex_lock(&inode->i_mutex);
 491                 ovl_cache_put(od, file->f_path.dentry);
 492                 mutex_unlock(&inode->i_mutex);
 493         }
 494         fput(od->realfile);
 495         if (od->upperfile)
 496                 fput(od->upperfile);
 497         kfree(od);
 498
 499         return 0;
 500 }
 501
 502 static int ovl_dir_open(struct inode *inode, struct file *file)
 503 {
 504         struct path realpath;
 505         struct file *realfile;
 506         struct ovl_dir_file *od;
 507         enum ovl_path_type type;
 508
 509         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 510         if (!od)
 511                 return -ENOMEM;
 512
 513         type = ovl_path_real(file->f_path.dentry, &realpath);
 514         realfile = ovl_path_open(&realpath, file->f_flags);
 515         if (IS_ERR(realfile)) {
 516                 kfree(od);
 517                 return PTR_ERR(realfile);
 518         }
 519         INIT_LIST_HEAD(&od->cursor.l_node);
 520         od->realfile = realfile;
 521         od->is_real = (type != OVL_PATH_MERGE);
 522         od->is_upper = (type != OVL_PATH_LOWER);
 523         od->cursor.is_cursor = true;
 524         file->private_data = od;
 525
 526         return 0;
 527 }
 528
 529 const struct file_operations ovl_dir_operations = {
 530         .read           = generic_read_dir,
 531         .open           = ovl_dir_open,
 532         .iterate        = ovl_iterate,
 533         .llseek         = ovl_dir_llseek,
 534         .fsync          = ovl_dir_fsync,
 535         .release        = ovl_dir_release,
 536 };
 537
 538 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 539 {
 540         int err;
 541         struct path lowerpath;
 542         struct path upperpath;
 543         struct ovl_cache_entry *p;
 544
 545         ovl_path_upper(dentry, &upperpath);
 546         ovl_path_lower(dentry, &lowerpath);
 547
 548         err = ovl_dir_read_merged(&upperpath, &lowerpath, list);
 549         if (err)
 550                 return err;
 551
 552         err = 0;
 553
 554         list_for_each_entry(p, list, l_node) {
 555                 if (p->is_whiteout)
 556                         continue;
 557
 558                 if (p->name[0] == '.') {
 559                         if (p->len == 1)
 560                                 continue;
 561                         if (p->len == 2 && p->name[1] == '.')
 562                                 continue;
 563                 }
 564                 err = -ENOTEMPTY;
 565                 break;
 566         }
 567
 568         return err;
 569 }
 570
 571 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
 572 {
 573         struct ovl_cache_entry *p;
 574
 575         mutex_lock_nested(&upper->d_inode->i_mutex, I_MUTEX_CHILD);
 576         list_for_each_entry(p, list, l_node) {
 577                 struct dentry *dentry;
 578
 579                 if (!p->is_whiteout)
 580                         continue;
 581
 582                 dentry = lookup_one_len(p->name, upper, p->len);
 583                 if (IS_ERR(dentry)) {
 584                         pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
 585                                upper->d_name.name, p->len, p->name,
 586                                (int) PTR_ERR(dentry));
 587                         continue;
 588                 }
 589                 ovl_cleanup(upper->d_inode, dentry);
 590                 dput(dentry);
 591         }
 592         mutex_unlock(&upper->d_inode->i_mutex);
 593 }