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         bool is_whiteout;
  25         struct list_head l_node;
  26         struct rb_node node;
  27         char name[];
  28 };
  29
  30 struct ovl_dir_cache {
  31         long refcount;
  32         u64 version;
  33         struct list_head entries;
  34 };
  35
  36 struct ovl_readdir_data {
  37         struct dir_context ctx;
  38         bool is_merge;
  39         struct rb_root *root;
  40         struct list_head *list;
  41         struct list_head *middle;
  42         int count;
  43         int err;
  44 };
  45
  46 struct ovl_dir_file {
  47         bool is_real;
  48         bool is_upper;
  49         struct ovl_dir_cache *cache;
  50         struct ovl_cache_entry cursor;
  51         struct file *realfile;
  52         struct file *upperfile;
  53 };
  54
  55 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
  56 {
  57         return container_of(n, struct ovl_cache_entry, node);
  58 }
  59
  60 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
  61                                                     const char *name, int len)
  62 {
  63         struct rb_node *node = root->rb_node;
  64         int cmp;
  65
  66         while (node) {
  67                 struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
  68
  69                 cmp = strncmp(name, p->name, len);
  70                 if (cmp > 0)
  71                         node = p->node.rb_right;
  72                 else if (cmp < 0 || len < p->len)
  73                         node = p->node.rb_left;
  74                 else
  75                         return p;
  76         }
  77
  78         return NULL;
  79 }
  80
  81 static struct ovl_cache_entry *ovl_cache_entry_new(const char *name, int len,
  82                                                    u64 ino, unsigned int d_type)
  83 {
  84         struct ovl_cache_entry *p;
  85         size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
  86
  87         p = kmalloc(size, GFP_KERNEL);
  88         if (p) {
  89                 memcpy(p->name, name, len);
  90                 p->name[len] = '\0';
  91                 p->len = len;
  92                 p->type = d_type;
  93                 p->ino = ino;
  94                 p->is_whiteout = false;
  95         }
  96
  97         return p;
  98 }
  99
 100 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
 101                                   const char *name, int len, u64 ino,
 102                                   unsigned int d_type)
 103 {
 104         struct rb_node **newp = &rdd->root->rb_node;
 105         struct rb_node *parent = NULL;
 106         struct ovl_cache_entry *p;
 107
 108         while (*newp) {
 109                 int cmp;
 110                 struct ovl_cache_entry *tmp;
 111
 112                 parent = *newp;
 113                 tmp = ovl_cache_entry_from_node(*newp);
 114                 cmp = strncmp(name, tmp->name, len);
 115                 if (cmp > 0)
 116                         newp = &tmp->node.rb_right;
 117                 else if (cmp < 0 || len < tmp->len)
 118                         newp = &tmp->node.rb_left;
 119                 else
 120                         return 0;
 121         }
 122
 123         p = ovl_cache_entry_new(name, len, ino, d_type);
 124         if (p == NULL)
 125                 return -ENOMEM;
 126
 127         list_add_tail(&p->l_node, rdd->list);
 128         rb_link_node(&p->node, parent, newp);
 129         rb_insert_color(&p->node, rdd->root);
 130
 131         return 0;
 132 }
 133
 134 static int ovl_fill_lower(struct ovl_readdir_data *rdd,
 135                           const char *name, int namelen,
 136                           loff_t offset, u64 ino, unsigned int d_type)
 137 {
 138         struct ovl_cache_entry *p;
 139
 140         p = ovl_cache_entry_find(rdd->root, name, namelen);
 141         if (p) {
 142                 list_move_tail(&p->l_node, rdd->middle);
 143         } else {
 144                 p = ovl_cache_entry_new(name, namelen, ino, d_type);
 145                 if (p == NULL)
 146                         rdd->err = -ENOMEM;
 147                 else
 148                         list_add_tail(&p->l_node, rdd->middle);
 149         }
 150
 151         return rdd->err;
 152 }
 153
 154 void ovl_cache_free(struct list_head *list)
 155 {
 156         struct ovl_cache_entry *p;
 157         struct ovl_cache_entry *n;
 158
 159         list_for_each_entry_safe(p, n, list, l_node)
 160                 kfree(p);
 161
 162         INIT_LIST_HEAD(list);
 163 }
 164
 165 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
 166 {
 167         struct ovl_dir_cache *cache = od->cache;
 168
 169         list_del(&od->cursor.l_node);
 170         WARN_ON(cache->refcount <= 0);
 171         cache->refcount--;
 172         if (!cache->refcount) {
 173                 if (ovl_dir_cache(dentry) == cache)
 174                         ovl_set_dir_cache(dentry, NULL);
 175
 176                 ovl_cache_free(&cache->entries);
 177                 kfree(cache);
 178         }
 179 }
 180
 181 static int ovl_fill_merge(void *buf, const char *name, int namelen,
 182                           loff_t offset, u64 ino, unsigned int d_type)
 183 {
 184         struct ovl_readdir_data *rdd = buf;
 185
 186         rdd->count++;
 187         if (!rdd->is_merge)
 188                 return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
 189         else
 190                 return ovl_fill_lower(rdd, name, namelen, offset, ino, d_type);
 191 }
 192
 193 static inline int ovl_dir_read(struct path *realpath,
 194                                struct ovl_readdir_data *rdd)
 195 {
 196         struct file *realfile;
 197         int err;
 198
 199         realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
 200         if (IS_ERR(realfile))
 201                 return PTR_ERR(realfile);
 202
 203         rdd->ctx.pos = 0;
 204         do {
 205                 rdd->count = 0;
 206                 rdd->err = 0;
 207                 err = iterate_dir(realfile, &rdd->ctx);
 208                 if (err >= 0)
 209                         err = rdd->err;
 210         } while (!err && rdd->count);
 211         fput(realfile);
 212
 213         return err;
 214 }
 215
 216 static void ovl_dir_reset(struct file *file)
 217 {
 218         struct ovl_dir_file *od = file->private_data;
 219         struct ovl_dir_cache *cache = od->cache;
 220         struct dentry *dentry = file->f_path.dentry;
 221         enum ovl_path_type type = ovl_path_type(dentry);
 222
 223         if (cache && ovl_dentry_version_get(dentry) != cache->version) {
 224                 ovl_cache_put(od, dentry);
 225                 od->cache = NULL;
 226         }
 227         WARN_ON(!od->is_real && type != OVL_PATH_MERGE);
 228         if (od->is_real && type == OVL_PATH_MERGE)
 229                 od->is_real = false;
 230 }
 231
 232 static int ovl_dir_mark_whiteouts(struct dentry *dir,
 233                                   struct ovl_readdir_data *rdd)
 234 {
 235         struct ovl_cache_entry *p;
 236         struct dentry *dentry;
 237         const struct cred *old_cred;
 238         struct cred *override_cred;
 239
 240         override_cred = prepare_creds();
 241         if (!override_cred) {
 242                 ovl_cache_free(rdd->list);
 243                 return -ENOMEM;
 244         }
 245
 246         /*
 247          * CAP_DAC_OVERRIDE for lookup
 248          */
 249         cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
 250         old_cred = override_creds(override_cred);
 251
 252         mutex_lock(&dir->d_inode->i_mutex);
 253         list_for_each_entry(p, rdd->list, l_node) {
 254                 if (!p->name)
 255                         continue;
 256
 257                 if (p->type != DT_CHR)
 258                         continue;
 259
 260                 dentry = lookup_one_len(p->name, dir, p->len);
 261                 if (IS_ERR(dentry))
 262                         continue;
 263
 264                 p->is_whiteout = ovl_is_whiteout(dentry);
 265                 dput(dentry);
 266         }
 267         mutex_unlock(&dir->d_inode->i_mutex);
 268
 269         revert_creds(old_cred);
 270         put_cred(override_cred);
 271
 272         return 0;
 273 }
 274
 275 static inline int ovl_dir_read_merged(struct path *upperpath,
 276                                       struct path *lowerpath,
 277                                       struct list_head *list)
 278 {
 279         int err;
 280         struct rb_root root = RB_ROOT;
 281         struct list_head middle;
 282         struct ovl_readdir_data rdd = {
 283                 .ctx.actor = ovl_fill_merge,
 284                 .list = list,
 285                 .root = &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(&middle, rdd.list);
 306                 rdd.middle = &middle;
 307                 rdd.is_merge = true;
 308                 err = ovl_dir_read(lowerpath, &rdd);
 309                 list_del(&middle);
 310         }
 311 out:
 312         return err;
 313
 314 }
 315
 316 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
 317 {
 318         struct ovl_cache_entry *p;
 319         loff_t off = 0;
 320
 321         list_for_each_entry(p, &od->cache->entries, l_node) {
 322                 if (!p->name)
 323                         continue;
 324                 if (off >= pos)
 325                         break;
 326                 off++;
 327         }
 328         list_move_tail(&od->cursor.l_node, &p->l_node);
 329 }
 330
 331 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 332 {
 333         int res;
 334         struct path lowerpath;
 335         struct path upperpath;
 336         struct ovl_dir_cache *cache;
 337
 338         cache = ovl_dir_cache(dentry);
 339         if (cache && ovl_dentry_version_get(dentry) == cache->version) {
 340                 cache->refcount++;
 341                 return cache;
 342         }
 343         ovl_set_dir_cache(dentry, NULL);
 344
 345         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 346         if (!cache)
 347                 return ERR_PTR(-ENOMEM);
 348
 349         cache->refcount = 1;
 350         INIT_LIST_HEAD(&cache->entries);
 351
 352         ovl_path_lower(dentry, &lowerpath);
 353         ovl_path_upper(dentry, &upperpath);
 354
 355         res = ovl_dir_read_merged(&upperpath, &lowerpath, &cache->entries);
 356         if (res) {
 357                 ovl_cache_free(&cache->entries);
 358                 kfree(cache);
 359                 return ERR_PTR(res);
 360         }
 361
 362         cache->version = ovl_dentry_version_get(dentry);
 363         ovl_set_dir_cache(dentry, cache);
 364
 365         return cache;
 366 }
 367
 368 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 369 {
 370         struct ovl_dir_file *od = file->private_data;
 371         struct dentry *dentry = file->f_path.dentry;
 372
 373         if (!ctx->pos)
 374                 ovl_dir_reset(file);
 375
 376         if (od->is_real)
 377                 return iterate_dir(od->realfile, ctx);
 378
 379         if (!od->cache) {
 380                 struct ovl_dir_cache *cache;
 381
 382                 cache = ovl_cache_get(dentry);
 383                 if (IS_ERR(cache))
 384                         return PTR_ERR(cache);
 385
 386                 od->cache = cache;
 387                 ovl_seek_cursor(od, ctx->pos);
 388         }
 389
 390         while (od->cursor.l_node.next != &od->cache->entries) {
 391                 struct ovl_cache_entry *p;
 392
 393                 p = list_entry(od->cursor.l_node.next, struct ovl_cache_entry, l_node);
 394                 /* Skip cursors */
 395                 if (p->name) {
 396                         if (!p->is_whiteout) {
 397                                 if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 398                                         break;
 399                         }
 400                         ctx->pos++;
 401                 }
 402                 list_move(&od->cursor.l_node, &p->l_node);
 403         }
 404         return 0;
 405 }
 406
 407 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 408 {
 409         loff_t res;
 410         struct ovl_dir_file *od = file->private_data;
 411
 412         mutex_lock(&file_inode(file)->i_mutex);
 413         if (!file->f_pos)
 414                 ovl_dir_reset(file);
 415
 416         if (od->is_real) {
 417                 res = vfs_llseek(od->realfile, offset, origin);
 418                 file->f_pos = od->realfile->f_pos;
 419         } else {
 420                 res = -EINVAL;
 421
 422                 switch (origin) {
 423                 case SEEK_CUR:
 424                         offset += file->f_pos;
 425                         break;
 426                 case SEEK_SET:
 427                         break;
 428                 default:
 429                         goto out_unlock;
 430                 }
 431                 if (offset < 0)
 432                         goto out_unlock;
 433
 434                 if (offset != file->f_pos) {
 435                         file->f_pos = offset;
 436                         if (od->cache)
 437                                 ovl_seek_cursor(od, offset);
 438                 }
 439                 res = offset;
 440         }
 441 out_unlock:
 442         mutex_unlock(&file_inode(file)->i_mutex);
 443
 444         return res;
 445 }
 446
 447 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 448                          int datasync)
 449 {
 450         struct ovl_dir_file *od = file->private_data;
 451         struct dentry *dentry = file->f_path.dentry;
 452         struct file *realfile = od->realfile;
 453
 454         /*
 455          * Need to check if we started out being a lower dir, but got copied up
 456          */
 457         if (!od->is_upper && ovl_path_type(dentry) == OVL_PATH_MERGE) {
 458                 struct inode *inode = file_inode(file);
 459
 460                 realfile = od->upperfile;
 461                 if (!realfile) {
 462                         struct path upperpath;
 463
 464                         ovl_path_upper(dentry, &upperpath);
 465                         realfile = ovl_path_open(&upperpath, O_RDONLY);
 466                         mutex_lock(&inode->i_mutex);
 467                         if (!od->upperfile) {
 468                                 if (IS_ERR(realfile)) {
 469                                         mutex_unlock(&inode->i_mutex);
 470                                         return PTR_ERR(realfile);
 471                                 }
 472                                 od->upperfile = realfile;
 473                         } else {
 474                                 /* somebody has beaten us to it */
 475                                 if (!IS_ERR(realfile))
 476                                         fput(realfile);
 477                                 realfile = od->upperfile;
 478                         }
 479                         mutex_unlock(&inode->i_mutex);
 480                 }
 481         }
 482
 483         return vfs_fsync_range(realfile, start, end, datasync);
 484 }
 485
 486 static int ovl_dir_release(struct inode *inode, struct file *file)
 487 {
 488         struct ovl_dir_file *od = file->private_data;
 489
 490         if (od->cache) {
 491                 mutex_lock(&inode->i_mutex);
 492                 ovl_cache_put(od, file->f_path.dentry);
 493                 mutex_unlock(&inode->i_mutex);
 494         }
 495         fput(od->realfile);
 496         if (od->upperfile)
 497                 fput(od->upperfile);
 498         kfree(od);
 499
 500         return 0;
 501 }
 502
 503 static int ovl_dir_open(struct inode *inode, struct file *file)
 504 {
 505         struct path realpath;
 506         struct file *realfile;
 507         struct ovl_dir_file *od;
 508         enum ovl_path_type type;
 509
 510         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 511         if (!od)
 512                 return -ENOMEM;
 513
 514         type = ovl_path_real(file->f_path.dentry, &realpath);
 515         realfile = ovl_path_open(&realpath, file->f_flags);
 516         if (IS_ERR(realfile)) {
 517                 kfree(od);
 518                 return PTR_ERR(realfile);
 519         }
 520         INIT_LIST_HEAD(&od->cursor.l_node);
 521         od->realfile = realfile;
 522         od->is_real = (type != OVL_PATH_MERGE);
 523         od->is_upper = (type != OVL_PATH_LOWER);
 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_PARENT);
 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 }