fs/overlayfs/copy_up.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/module.h>
  11 #include <linux/fs.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/splice.h>
  15 #include <linux/xattr.h>
  16 #include <linux/security.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/sched.h>
  19 #include <linux/namei.h>
  20 #include <linux/fdtable.h>
  21 #include <linux/ratelimit.h>
  22 #include "overlayfs.h"
  23
  24 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  25
  26 static bool __read_mostly ovl_check_copy_up;
  27 module_param_named(check_copy_up, ovl_check_copy_up, bool,
  28                    S_IWUSR | S_IRUGO);
  29 MODULE_PARM_DESC(ovl_check_copy_up,
  30                  "Warn on copy-up when causing process also has a R/O fd open");
  31
  32 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
  33 {
  34         const struct dentry *dentry = data;
  35
  36         if (f->f_inode == d_inode(dentry))
  37                 pr_warn_ratelimited("overlayfs: Warning: Copying up %pD, but open R/O on fd %u which will cease to be coherent [pid=%d %s]\n",
  38                                     f, fd, current->pid, current->comm);
  39         return 0;
  40 }
  41
  42 /*
  43  * Check the fds open by this process and warn if something like the following
  44  * scenario is about to occur:
  45  *
  46  *      fd1 = open("foo", O_RDONLY);
  47  *      fd2 = open("foo", O_RDWR);
  48  */
  49 static void ovl_do_check_copy_up(struct dentry *dentry)
  50 {
  51         if (ovl_check_copy_up)
  52                 iterate_fd(current->files, 0, ovl_check_fd, dentry);
  53 }
  54
  55 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  56 {
  57         ssize_t list_size, size, value_size = 0;
  58         char *buf, *name, *value = NULL;
  59         int uninitialized_var(error);
  60
  61         if (!old->d_inode->i_op->getxattr ||
  62             !new->d_inode->i_op->getxattr)
  63                 return 0;
  64
  65         list_size = vfs_listxattr(old, NULL, 0);
  66         if (list_size <= 0) {
  67                 if (list_size == -EOPNOTSUPP)
  68                         return 0;
  69                 return list_size;
  70         }
  71
  72         buf = kzalloc(list_size, GFP_KERNEL);
  73         if (!buf)
  74                 return -ENOMEM;
  75
  76         list_size = vfs_listxattr(old, buf, list_size);
  77         if (list_size <= 0) {
  78                 error = list_size;
  79                 goto out;
  80         }
  81
  82         for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
  83 retry:
  84                 size = vfs_getxattr(old, name, value, value_size);
  85                 if (size == -ERANGE)
  86                         size = vfs_getxattr(old, name, NULL, 0);
  87
  88                 if (size < 0) {
  89                         error = size;
  90                         break;
  91                 }
  92
  93                 if (size > value_size) {
  94                         void *new;
  95
  96                         new = krealloc(value, size, GFP_KERNEL);
  97                         if (!new) {
  98                                 error = -ENOMEM;
  99                                 break;
 100                         }
 101                         value = new;
 102                         value_size = size;
 103                         goto retry;
 104                 }
 105
 106                 error = vfs_setxattr(new, name, value, size, 0);
 107                 if (error)
 108                         break;
 109         }
 110         kfree(value);
 111 out:
 112         kfree(buf);
 113         return error;
 114 }
 115
 116 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 117 {
 118         struct file *old_file;
 119         struct file *new_file;
 120         loff_t old_pos = 0;
 121         loff_t new_pos = 0;
 122         int error = 0;
 123
 124         if (len == 0)
 125                 return 0;
 126
 127         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 128         if (IS_ERR(old_file))
 129                 return PTR_ERR(old_file);
 130
 131         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 132         if (IS_ERR(new_file)) {
 133                 error = PTR_ERR(new_file);
 134                 goto out_fput;
 135         }
 136
 137         /* FIXME: copy up sparse files efficiently */
 138         while (len) {
 139                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 140                 long bytes;
 141
 142                 if (len < this_len)
 143                         this_len = len;
 144
 145                 if (signal_pending_state(TASK_KILLABLE, current)) {
 146                         error = -EINTR;
 147                         break;
 148                 }
 149
 150                 bytes = do_splice_direct(old_file, &old_pos,
 151                                          new_file, &new_pos,
 152                                          this_len, SPLICE_F_MOVE);
 153                 if (bytes <= 0) {
 154                         error = bytes;
 155                         break;
 156                 }
 157                 WARN_ON(old_pos != new_pos);
 158
 159                 len -= bytes;
 160         }
 161
 162         fput(new_file);
 163 out_fput:
 164         fput(old_file);
 165         return error;
 166 }
 167
 168 static char *ovl_read_symlink(struct dentry *realdentry)
 169 {
 170         int res;
 171         char *buf;
 172         struct inode *inode = realdentry->d_inode;
 173         mm_segment_t old_fs;
 174
 175         res = -EINVAL;
 176         if (!inode->i_op->readlink)
 177                 goto err;
 178
 179         res = -ENOMEM;
 180         buf = (char *) __get_free_page(GFP_KERNEL);
 181         if (!buf)
 182                 goto err;
 183
 184         old_fs = get_fs();
 185         set_fs(get_ds());
 186         /* The cast to a user pointer is valid due to the set_fs() */
 187         res = inode->i_op->readlink(realdentry,
 188                                     (char __user *)buf, PAGE_SIZE - 1);
 189         set_fs(old_fs);
 190         if (res < 0) {
 191                 free_page((unsigned long) buf);
 192                 goto err;
 193         }
 194         buf[res] = '\0';
 195
 196         return buf;
 197
 198 err:
 199         return ERR_PTR(res);
 200 }
 201
 202 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 203 {
 204         struct iattr attr = {
 205                 .ia_valid =
 206                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 207                 .ia_atime = stat->atime,
 208                 .ia_mtime = stat->mtime,
 209         };
 210
 211         return notify_change(upperdentry, &attr, NULL);
 212 }
 213
 214 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 215 {
 216         int err = 0;
 217
 218         if (!S_ISLNK(stat->mode)) {
 219                 struct iattr attr = {
 220                         .ia_valid = ATTR_MODE,
 221                         .ia_mode = stat->mode,
 222                 };
 223                 err = notify_change(upperdentry, &attr, NULL);
 224         }
 225         if (!err) {
 226                 struct iattr attr = {
 227                         .ia_valid = ATTR_UID | ATTR_GID,
 228                         .ia_uid = stat->uid,
 229                         .ia_gid = stat->gid,
 230                 };
 231                 err = notify_change(upperdentry, &attr, NULL);
 232         }
 233         if (!err)
 234                 ovl_set_timestamps(upperdentry, stat);
 235
 236         return err;
 237 }
 238
 239 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 240                               struct dentry *dentry, struct path *lowerpath,
 241                               struct kstat *stat, const char *link)
 242 {
 243         struct inode *wdir = workdir->d_inode;
 244         struct inode *udir = upperdir->d_inode;
 245         struct dentry *newdentry = NULL;
 246         struct dentry *upper = NULL;
 247         umode_t mode = stat->mode;
 248         int err;
 249
 250         newdentry = ovl_lookup_temp(workdir, dentry);
 251         err = PTR_ERR(newdentry);
 252         if (IS_ERR(newdentry))
 253                 goto out;
 254
 255         upper = lookup_one_len(dentry->d_name.name, upperdir,
 256                                dentry->d_name.len);
 257         err = PTR_ERR(upper);
 258         if (IS_ERR(upper))
 259                 goto out1;
 260
 261         /* Can't properly set mode on creation because of the umask */
 262         stat->mode &= S_IFMT;
 263         err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
 264         stat->mode = mode;
 265         if (err)
 266                 goto out2;
 267
 268         if (S_ISREG(stat->mode)) {
 269                 struct path upperpath;
 270
 271                 ovl_path_upper(dentry, &upperpath);
 272                 BUG_ON(upperpath.dentry != NULL);
 273                 upperpath.dentry = newdentry;
 274
 275                 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
 276                 if (err)
 277                         goto out_cleanup;
 278         }
 279
 280         err = ovl_copy_xattr(lowerpath->dentry, newdentry);
 281         if (err)
 282                 goto out_cleanup;
 283
 284         inode_lock(newdentry->d_inode);
 285         err = ovl_set_attr(newdentry, stat);
 286         inode_unlock(newdentry->d_inode);
 287         if (err)
 288                 goto out_cleanup;
 289
 290         err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 291         if (err)
 292                 goto out_cleanup;
 293
 294         ovl_dentry_update(dentry, newdentry);
 295         newdentry = NULL;
 296
 297         /*
 298          * Non-directores become opaque when copied up.
 299          */
 300         if (!S_ISDIR(stat->mode))
 301                 ovl_dentry_set_opaque(dentry, true);
 302 out2:
 303         dput(upper);
 304 out1:
 305         dput(newdentry);
 306 out:
 307         return err;
 308
 309 out_cleanup:
 310         ovl_cleanup(wdir, newdentry);
 311         goto out2;
 312 }
 313
 314 /*
 315  * Copy up a single dentry
 316  *
 317  * Directory renames only allowed on "pure upper" (already created on
 318  * upper filesystem, never copied up).  Directories which are on lower or
 319  * are merged may not be renamed.  For these -EXDEV is returned and
 320  * userspace has to deal with it.  This means, when copying up a
 321  * directory we can rely on it and ancestors being stable.
 322  *
 323  * Non-directory renames start with copy up of source if necessary.  The
 324  * actual rename will only proceed once the copy up was successful.  Copy
 325  * up uses upper parent i_mutex for exclusion.  Since rename can change
 326  * d_parent it is possible that the copy up will lock the old parent.  At
 327  * that point the file will have already been copied up anyway.
 328  */
 329 int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 330                     struct path *lowerpath, struct kstat *stat)
 331 {
 332         struct dentry *workdir = ovl_workdir(dentry);
 333         int err;
 334         struct kstat pstat;
 335         struct path parentpath;
 336         struct dentry *upperdir;
 337         struct dentry *upperdentry;
 338         const struct cred *old_cred;
 339         struct cred *override_cred;
 340         char *link = NULL;
 341
 342         if (WARN_ON(!workdir))
 343                 return -EROFS;
 344
 345         ovl_do_check_copy_up(lowerpath->dentry);
 346
 347         ovl_path_upper(parent, &parentpath);
 348         upperdir = parentpath.dentry;
 349
 350         err = vfs_getattr(&parentpath, &pstat);
 351         if (err)
 352                 return err;
 353
 354         if (S_ISLNK(stat->mode)) {
 355                 link = ovl_read_symlink(lowerpath->dentry);
 356                 if (IS_ERR(link))
 357                         return PTR_ERR(link);
 358         }
 359
 360         err = -ENOMEM;
 361         override_cred = prepare_creds();
 362         if (!override_cred)
 363                 goto out_free_link;
 364
 365         override_cred->fsuid = stat->uid;
 366         override_cred->fsgid = stat->gid;
 367         /*
 368          * CAP_SYS_ADMIN for copying up extended attributes
 369          * CAP_DAC_OVERRIDE for create
 370          * CAP_FOWNER for chmod, timestamp update
 371          * CAP_FSETID for chmod
 372          * CAP_CHOWN for chown
 373          * CAP_MKNOD for mknod
 374          */
 375         cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
 376         cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
 377         cap_raise(override_cred->cap_effective, CAP_FOWNER);
 378         cap_raise(override_cred->cap_effective, CAP_FSETID);
 379         cap_raise(override_cred->cap_effective, CAP_CHOWN);
 380         cap_raise(override_cred->cap_effective, CAP_MKNOD);
 381         old_cred = override_creds(override_cred);
 382
 383         err = -EIO;
 384         if (lock_rename(workdir, upperdir) != NULL) {
 385                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 386                 goto out_unlock;
 387         }
 388         upperdentry = ovl_dentry_upper(dentry);
 389         if (upperdentry) {
 390                 /* Raced with another copy-up?  Nothing to do, then... */
 391                 err = 0;
 392                 goto out_unlock;
 393         }
 394
 395         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 396                                  stat, link);
 397         if (!err) {
 398                 /* Restore timestamps on parent (best effort) */
 399                 ovl_set_timestamps(upperdir, &pstat);
 400         }
 401 out_unlock:
 402         unlock_rename(workdir, upperdir);
 403         revert_creds(old_cred);
 404         put_cred(override_cred);
 405
 406 out_free_link:
 407         if (link)
 408                 free_page((unsigned long) link);
 409
 410         return err;
 411 }
 412
 413 int ovl_copy_up(struct dentry *dentry)
 414 {
 415         int err;
 416
 417         err = 0;
 418         while (!err) {
 419                 struct dentry *next;
 420                 struct dentry *parent;
 421                 struct path lowerpath;
 422                 struct kstat stat;
 423                 enum ovl_path_type type = ovl_path_type(dentry);
 424
 425                 if (OVL_TYPE_UPPER(type))
 426                         break;
 427
 428                 next = dget(dentry);
 429                 /* find the topmost dentry not yet copied up */
 430                 for (;;) {
 431                         parent = dget_parent(next);
 432
 433                         type = ovl_path_type(parent);
 434                         if (OVL_TYPE_UPPER(type))
 435                                 break;
 436
 437                         dput(next);
 438                         next = parent;
 439                 }
 440
 441                 ovl_path_lower(next, &lowerpath);
 442                 err = vfs_getattr(&lowerpath, &stat);
 443                 if (!err)
 444                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 445
 446                 dput(parent);
 447                 dput(next);
 448         }
 449
 450         return err;
 451 }