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