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         struct ovl_cache_entry *next_maybe_whiteout;
  27         bool is_whiteout;
  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         struct ovl_cache_entry *first_maybe_whiteout;
  44         int count;
  45         int err;
  46         bool d_type_supported;
  47 };
  48
  49 struct ovl_dir_file {
  50         bool is_real;
  51         bool is_upper;
  52         struct ovl_dir_cache *cache;
  53         struct list_head *cursor;
  54         struct file *realfile;
  55         struct file *upperfile;
  56 };
  57
  58 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
  59 {
  60         return container_of(n, struct ovl_cache_entry, node);
  61 }
  62
  63 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
  64                                                     const char *name, int len)
  65 {
  66         struct rb_node *node = root->rb_node;
  67         int cmp;
  68
  69         while (node) {
  70                 struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
  71
  72                 cmp = strncmp(name, p->name, len);
  73                 if (cmp > 0)
  74                         node = p->node.rb_right;
  75                 else if (cmp < 0 || len < p->len)
  76                         node = p->node.rb_left;
  77                 else
  78                         return p;
  79         }
  80
  81         return NULL;
  82 }
  83
  84 static struct ovl_cache_entry *ovl_cache_entry_new(struct ovl_readdir_data *rdd,
  85                                                    const char *name, int len,
  86                                                    u64 ino, unsigned int d_type)
  87 {
  88         struct ovl_cache_entry *p;
  89         size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
  90
  91         p = kmalloc(size, GFP_KERNEL);
  92         if (!p)
  93                 return NULL;
  94
  95         memcpy(p->name, name, len);
  96         p->name[len] = '\0';
  97         p->len = len;
  98         p->type = d_type;
  99         p->ino = ino;
 100         p->is_whiteout = false;
 101
 102         if (d_type == DT_CHR) {
 103                 p->next_maybe_whiteout = rdd->first_maybe_whiteout;
 104                 rdd->first_maybe_whiteout = p;
 105         }
 106         return p;
 107 }
 108
 109 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
 110                                   const char *name, int len, u64 ino,
 111                                   unsigned int d_type)
 112 {
 113         struct rb_node **newp = &rdd->root.rb_node;
 114         struct rb_node *parent = NULL;
 115         struct ovl_cache_entry *p;
 116
 117         while (*newp) {
 118                 int cmp;
 119                 struct ovl_cache_entry *tmp;
 120
 121                 parent = *newp;
 122                 tmp = ovl_cache_entry_from_node(*newp);
 123                 cmp = strncmp(name, tmp->name, len);
 124                 if (cmp > 0)
 125                         newp = &tmp->node.rb_right;
 126                 else if (cmp < 0 || len < tmp->len)
 127                         newp = &tmp->node.rb_left;
 128                 else
 129                         return 0;
 130         }
 131
 132         p = ovl_cache_entry_new(rdd, name, len, ino, d_type);
 133         if (p == NULL)
 134                 return -ENOMEM;
 135
 136         list_add_tail(&p->l_node, rdd->list);
 137         rb_link_node(&p->node, parent, newp);
 138         rb_insert_color(&p->node, &rdd->root);
 139
 140         return 0;
 141 }
 142
 143 static int ovl_fill_lower(struct ovl_readdir_data *rdd,
 144                           const char *name, int namelen,
 145                           loff_t offset, u64 ino, unsigned int d_type)
 146 {
 147         struct ovl_cache_entry *p;
 148
 149         p = ovl_cache_entry_find(&rdd->root, name, namelen);
 150         if (p) {
 151                 list_move_tail(&p->l_node, &rdd->middle);
 152         } else {
 153                 p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 154                 if (p == NULL)
 155                         rdd->err = -ENOMEM;
 156                 else
 157                         list_add_tail(&p->l_node, &rdd->middle);
 158         }
 159
 160         return rdd->err;
 161 }
 162
 163 void ovl_cache_free(struct list_head *list)
 164 {
 165         struct ovl_cache_entry *p;
 166         struct ovl_cache_entry *n;
 167
 168         list_for_each_entry_safe(p, n, list, l_node)
 169                 kfree(p);
 170
 171         INIT_LIST_HEAD(list);
 172 }
 173
 174 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
 175 {
 176         struct ovl_dir_cache *cache = od->cache;
 177
 178         WARN_ON(cache->refcount <= 0);
 179         cache->refcount--;
 180         if (!cache->refcount) {
 181                 if (ovl_dir_cache(dentry) == cache)
 182                         ovl_set_dir_cache(dentry, NULL);
 183
 184                 ovl_cache_free(&cache->entries);
 185                 kfree(cache);
 186         }
 187 }
 188
 189 static int ovl_fill_merge(struct dir_context *ctx, const char *name,
 190                           int namelen, loff_t offset, u64 ino,
 191                           unsigned int d_type)
 192 {
 193         struct ovl_readdir_data *rdd =
 194                 container_of(ctx, struct ovl_readdir_data, ctx);
 195
 196         rdd->count++;
 197         if (!rdd->is_merge)
 198                 return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
 199         else
 200                 return ovl_fill_lower(rdd, name, namelen, offset, ino, d_type);
 201 }
 202
 203 static int ovl_check_whiteouts(struct dentry *dir, struct ovl_readdir_data *rdd)
 204 {
 205         int err;
 206         struct ovl_cache_entry *p;
 207         struct dentry *dentry;
 208         const struct cred *old_cred;
 209         struct cred *override_cred;
 210
 211         override_cred = prepare_creds();
 212         if (!override_cred)
 213                 return -ENOMEM;
 214
 215         /*
 216          * CAP_DAC_OVERRIDE for lookup
 217          */
 218         cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
 219         old_cred = override_creds(override_cred);
 220
 221         err = mutex_lock_killable(&dir->d_inode->i_mutex);
 222         if (!err) {
 223                 while (rdd->first_maybe_whiteout) {
 224                         p = rdd->first_maybe_whiteout;
 225                         rdd->first_maybe_whiteout = p->next_maybe_whiteout;
 226                         dentry = lookup_one_len(p->name, dir, p->len);
 227                         if (!IS_ERR(dentry)) {
 228                                 p->is_whiteout = ovl_is_whiteout(dentry);
 229                                 dput(dentry);
 230                         }
 231                 }
 232                 inode_unlock(dir->d_inode);
 233         }
 234         revert_creds(old_cred);
 235         put_cred(override_cred);
 236
 237         return err;
 238 }
 239
 240 static inline int ovl_dir_read(struct path *realpath,
 241                                struct ovl_readdir_data *rdd)
 242 {
 243         struct file *realfile;
 244         int err;
 245
 246         realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
 247         if (IS_ERR(realfile))
 248                 return PTR_ERR(realfile);
 249
 250         rdd->first_maybe_whiteout = NULL;
 251         rdd->ctx.pos = 0;
 252         do {
 253                 rdd->count = 0;
 254                 rdd->err = 0;
 255                 err = iterate_dir(realfile, &rdd->ctx);
 256                 if (err >= 0)
 257                         err = rdd->err;
 258         } while (!err && rdd->count);
 259
 260         if (!err && rdd->first_maybe_whiteout)
 261                 err = ovl_check_whiteouts(realpath->dentry, rdd);
 262
 263         fput(realfile);
 264
 265         return err;
 266 }
 267
 268 static void ovl_dir_reset(struct file *file)
 269 {
 270         struct ovl_dir_file *od = file->private_data;
 271         struct ovl_dir_cache *cache = od->cache;
 272         struct dentry *dentry = file->f_path.dentry;
 273         enum ovl_path_type type = ovl_path_type(dentry);
 274
 275         if (cache && ovl_dentry_version_get(dentry) != cache->version) {
 276                 ovl_cache_put(od, dentry);
 277                 od->cache = NULL;
 278                 od->cursor = NULL;
 279         }
 280         WARN_ON(!od->is_real && !OVL_TYPE_MERGE(type));
 281         if (od->is_real && OVL_TYPE_MERGE(type))
 282                 od->is_real = false;
 283 }
 284
 285 static int ovl_dir_read_merged(struct dentry *dentry, struct list_head *list)
 286 {
 287         int err;
 288         struct path realpath;
 289         struct ovl_readdir_data rdd = {
 290                 .ctx.actor = ovl_fill_merge,
 291                 .list = list,
 292                 .root = RB_ROOT,
 293                 .is_merge = false,
 294         };
 295         int idx, next;
 296
 297         for (idx = 0; idx != -1; idx = next) {
 298                 next = ovl_path_next(idx, dentry, &realpath);
 299
 300                 if (next != -1) {
 301                         err = ovl_dir_read(&realpath, &rdd);
 302                         if (err)
 303                                 break;
 304                 } else {
 305                         /*
 306                          * Insert lowest layer entries before upper ones, this
 307                          * allows offsets to be reasonably constant
 308                          */
 309                         list_add(&rdd.middle, rdd.list);
 310                         rdd.is_merge = true;
 311                         err = ovl_dir_read(&realpath, &rdd);
 312                         list_del(&rdd.middle);
 313                 }
 314         }
 315         return err;
 316 }
 317
 318 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
 319 {
 320         struct list_head *p;
 321         loff_t off = 0;
 322
 323         list_for_each(p, &od->cache->entries) {
 324                 if (off >= pos)
 325                         break;
 326                 off++;
 327         }
 328         /* Cursor is safe since the cache is stable */
 329         od->cursor = p;
 330 }
 331
 332 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 333 {
 334         int res;
 335         struct ovl_dir_cache *cache;
 336
 337         cache = ovl_dir_cache(dentry);
 338         if (cache && ovl_dentry_version_get(dentry) == cache->version) {
 339                 cache->refcount++;
 340                 return cache;
 341         }
 342         ovl_set_dir_cache(dentry, NULL);
 343
 344         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 345         if (!cache)
 346                 return ERR_PTR(-ENOMEM);
 347
 348         cache->refcount = 1;
 349         INIT_LIST_HEAD(&cache->entries);
 350
 351         res = ovl_dir_read_merged(dentry, &cache->entries);
 352         if (res) {
 353                 ovl_cache_free(&cache->entries);
 354                 kfree(cache);
 355                 return ERR_PTR(res);
 356         }
 357
 358         cache->version = ovl_dentry_version_get(dentry);
 359         ovl_set_dir_cache(dentry, cache);
 360
 361         return cache;
 362 }
 363
 364 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 365 {
 366         struct ovl_dir_file *od = file->private_data;
 367         struct dentry *dentry = file->f_path.dentry;
 368         struct ovl_cache_entry *p;
 369
 370         if (!ctx->pos)
 371                 ovl_dir_reset(file);
 372
 373         if (od->is_real)
 374                 return iterate_dir(od->realfile, ctx);
 375
 376         if (!od->cache) {
 377                 struct ovl_dir_cache *cache;
 378
 379                 cache = ovl_cache_get(dentry);
 380                 if (IS_ERR(cache))
 381                         return PTR_ERR(cache);
 382
 383                 od->cache = cache;
 384                 ovl_seek_cursor(od, ctx->pos);
 385         }
 386
 387         while (od->cursor != &od->cache->entries) {
 388                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 389                 if (!p->is_whiteout)
 390                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 391                                 break;
 392                 od->cursor = p->l_node.next;
 393                 ctx->pos++;
 394         }
 395         return 0;
 396 }
 397
 398 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 399 {
 400         loff_t res;
 401         struct ovl_dir_file *od = file->private_data;
 402
 403         inode_lock(file_inode(file));
 404         if (!file->f_pos)
 405                 ovl_dir_reset(file);
 406
 407         if (od->is_real) {
 408                 res = vfs_llseek(od->realfile, offset, origin);
 409                 file->f_pos = od->realfile->f_pos;
 410         } else {
 411                 res = -EINVAL;
 412
 413                 switch (origin) {
 414                 case SEEK_CUR:
 415                         offset += file->f_pos;
 416                         break;
 417                 case SEEK_SET:
 418                         break;
 419                 default:
 420                         goto out_unlock;
 421                 }
 422                 if (offset < 0)
 423                         goto out_unlock;
 424
 425                 if (offset != file->f_pos) {
 426                         file->f_pos = offset;
 427                         if (od->cache)
 428                                 ovl_seek_cursor(od, offset);
 429                 }
 430                 res = offset;
 431         }
 432 out_unlock:
 433         inode_unlock(file_inode(file));
 434
 435         return res;
 436 }
 437
 438 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 439                          int datasync)
 440 {
 441         struct ovl_dir_file *od = file->private_data;
 442         struct dentry *dentry = file->f_path.dentry;
 443         struct file *realfile = od->realfile;
 444
 445         /*
 446          * Need to check if we started out being a lower dir, but got copied up
 447          */
 448         if (!od->is_upper && OVL_TYPE_UPPER(ovl_path_type(dentry))) {
 449                 struct inode *inode = file_inode(file);
 450
 451                 realfile = lockless_dereference(od->upperfile);
 452                 if (!realfile) {
 453                         struct path upperpath;
 454
 455                         ovl_path_upper(dentry, &upperpath);
 456                         realfile = ovl_path_open(&upperpath, O_RDONLY);
 457                         smp_mb__before_spinlock();
 458                         inode_lock(inode);
 459                         if (!od->upperfile) {
 460                                 if (IS_ERR(realfile)) {
 461                                         inode_unlock(inode);
 462                                         return PTR_ERR(realfile);
 463                                 }
 464                                 od->upperfile = realfile;
 465                         } else {
 466                                 /* somebody has beaten us to it */
 467                                 if (!IS_ERR(realfile))
 468                                         fput(realfile);
 469                                 realfile = od->upperfile;
 470                         }
 471                         inode_unlock(inode);
 472                 }
 473         }
 474
 475         return vfs_fsync_range(realfile, start, end, datasync);
 476 }
 477
 478 static int ovl_dir_release(struct inode *inode, struct file *file)
 479 {
 480         struct ovl_dir_file *od = file->private_data;
 481
 482         if (od->cache) {
 483                 inode_lock(inode);
 484                 ovl_cache_put(od, file->f_path.dentry);
 485                 inode_unlock(inode);
 486         }
 487         fput(od->realfile);
 488         if (od->upperfile)
 489                 fput(od->upperfile);
 490         kfree(od);
 491
 492         return 0;
 493 }
 494
 495 static int ovl_dir_open(struct inode *inode, struct file *file)
 496 {
 497         struct path realpath;
 498         struct file *realfile;
 499         struct ovl_dir_file *od;
 500         enum ovl_path_type type;
 501
 502         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 503         if (!od)
 504                 return -ENOMEM;
 505
 506         type = ovl_path_real(file->f_path.dentry, &realpath);
 507         realfile = ovl_path_open(&realpath, file->f_flags);
 508         if (IS_ERR(realfile)) {
 509                 kfree(od);
 510                 return PTR_ERR(realfile);
 511         }
 512         od->realfile = realfile;
 513         od->is_real = !OVL_TYPE_MERGE(type);
 514         od->is_upper = OVL_TYPE_UPPER(type);
 515         file->private_data = od;
 516
 517         return 0;
 518 }
 519
 520 const struct file_operations ovl_dir_operations = {
 521         .read           = generic_read_dir,
 522         .open           = ovl_dir_open,
 523         .iterate        = ovl_iterate,
 524         .llseek         = ovl_dir_llseek,
 525         .fsync          = ovl_dir_fsync,
 526         .release        = ovl_dir_release,
 527 };
 528
 529 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 530 {
 531         int err;
 532         struct ovl_cache_entry *p;
 533
 534         err = ovl_dir_read_merged(dentry, list);
 535         if (err)
 536                 return err;
 537
 538         err = 0;
 539
 540         list_for_each_entry(p, list, l_node) {
 541                 if (p->is_whiteout)
 542                         continue;
 543
 544                 if (p->name[0] == '.') {
 545                         if (p->len == 1)
 546                                 continue;
 547                         if (p->len == 2 && p->name[1] == '.')
 548                                 continue;
 549                 }
 550                 err = -ENOTEMPTY;
 551                 break;
 552         }
 553
 554         return err;
 555 }
 556
 557 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
 558 {
 559         struct ovl_cache_entry *p;
 560
 561         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
 562         list_for_each_entry(p, list, l_node) {
 563                 struct dentry *dentry;
 564
 565                 if (!p->is_whiteout)
 566                         continue;
 567
 568                 dentry = lookup_one_len(p->name, upper, p->len);
 569                 if (IS_ERR(dentry)) {
 570                         pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
 571                                upper->d_name.name, p->len, p->name,
 572                                (int) PTR_ERR(dentry));
 573                         continue;
 574                 }
 575                 if (dentry->d_inode)
 576                         ovl_cleanup(upper->d_inode, dentry);
 577                 dput(dentry);
 578         }
 579         inode_unlock(upper->d_inode);
 580 }
 581
 582 static int ovl_check_d_type(struct dir_context *ctx, const char *name,
 583                           int namelen, loff_t offset, u64 ino,
 584                           unsigned int d_type)
 585 {
 586         struct ovl_readdir_data *rdd =
 587                 container_of(ctx, struct ovl_readdir_data, ctx);
 588
 589         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
 590         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
 591                 return 0;
 592
 593         if (d_type != DT_UNKNOWN)
 594                 rdd->d_type_supported = true;
 595
 596         return 0;
 597 }
 598
 599 /*
 600  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
 601  * if error is encountered.
 602  */
 603 int ovl_check_d_type_supported(struct path *realpath)
 604 {
 605         int err;
 606         struct ovl_readdir_data rdd = {
 607                 .ctx.actor = ovl_check_d_type,
 608                 .d_type_supported = false,
 609         };
 610
 611         err = ovl_dir_read(realpath, &rdd);
 612         if (err)
 613                 return err;
 614
 615         return rdd.d_type_supported;
 616 }