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                 if (ovl_is_private_xattr(name))
  84                         continue;
  85 retry:
  86                 size = vfs_getxattr(old, name, value, value_size);
  87                 if (size == -ERANGE)
  88                         size = vfs_getxattr(old, name, NULL, 0);
  89
  90                 if (size < 0) {
  91                         error = size;
  92                         break;
  93                 }
  94
  95                 if (size > value_size) {
  96                         void *new;
  97
  98                         new = krealloc(value, size, GFP_KERNEL);
  99                         if (!new) {
 100                                 error = -ENOMEM;
 101                                 break;
 102                         }
 103                         value = new;
 104                         value_size = size;
 105                         goto retry;
 106                 }
 107
 108                 error = vfs_setxattr(new, name, value, size, 0);
 109                 if (error)
 110                         break;
 111         }
 112         kfree(value);
 113 out:
 114         kfree(buf);
 115         return error;
 116 }
 117
 118 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 119 {
 120         struct file *old_file;
 121         struct file *new_file;
 122         loff_t old_pos = 0;
 123         loff_t new_pos = 0;
 124         int error = 0;
 125
 126         if (len == 0)
 127                 return 0;
 128
 129         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 130         if (IS_ERR(old_file))
 131                 return PTR_ERR(old_file);
 132
 133         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 134         if (IS_ERR(new_file)) {
 135                 error = PTR_ERR(new_file);
 136                 goto out_fput;
 137         }
 138
 139         /* FIXME: copy up sparse files efficiently */
 140         while (len) {
 141                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 142                 long bytes;
 143
 144                 if (len < this_len)
 145                         this_len = len;
 146
 147                 if (signal_pending_state(TASK_KILLABLE, current)) {
 148                         error = -EINTR;
 149                         break;
 150                 }
 151
 152                 bytes = do_splice_direct(old_file, &old_pos,
 153                                          new_file, &new_pos,
 154                                          this_len, SPLICE_F_MOVE);
 155                 if (bytes <= 0) {
 156                         error = bytes;
 157                         break;
 158                 }
 159                 WARN_ON(old_pos != new_pos);
 160
 161                 len -= bytes;
 162         }
 163
 164         fput(new_file);
 165 out_fput:
 166         fput(old_file);
 167         return error;
 168 }
 169
 170 static char *ovl_read_symlink(struct dentry *realdentry)
 171 {
 172         int res;
 173         char *buf;
 174         struct inode *inode = realdentry->d_inode;
 175         mm_segment_t old_fs;
 176
 177         res = -EINVAL;
 178         if (!inode->i_op->readlink)
 179                 goto err;
 180
 181         res = -ENOMEM;
 182         buf = (char *) __get_free_page(GFP_KERNEL);
 183         if (!buf)
 184                 goto err;
 185
 186         old_fs = get_fs();
 187         set_fs(get_ds());
 188         /* The cast to a user pointer is valid due to the set_fs() */
 189         res = inode->i_op->readlink(realdentry,
 190                                     (char __user *)buf, PAGE_SIZE - 1);
 191         set_fs(old_fs);
 192         if (res < 0) {
 193                 free_page((unsigned long) buf);
 194                 goto err;
 195         }
 196         buf[res] = '\0';
 197
 198         return buf;
 199
 200 err:
 201         return ERR_PTR(res);
 202 }
 203
 204 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 205 {
 206         struct iattr attr = {
 207                 .ia_valid =
 208                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 209                 .ia_atime = stat->atime,
 210                 .ia_mtime = stat->mtime,
 211         };
 212
 213         return notify_change(upperdentry, &attr, NULL);
 214 }
 215
 216 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 217 {
 218         int err = 0;
 219
 220         if (!S_ISLNK(stat->mode)) {
 221                 struct iattr attr = {
 222                         .ia_valid = ATTR_MODE,
 223                         .ia_mode = stat->mode,
 224                 };
 225                 err = notify_change(upperdentry, &attr, NULL);
 226         }
 227         if (!err) {
 228                 struct iattr attr = {
 229                         .ia_valid = ATTR_UID | ATTR_GID,
 230                         .ia_uid = stat->uid,
 231                         .ia_gid = stat->gid,
 232                 };
 233                 err = notify_change(upperdentry, &attr, NULL);
 234         }
 235         if (!err)
 236                 ovl_set_timestamps(upperdentry, stat);
 237
 238         return err;
 239 }
 240
 241 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 242                               struct dentry *dentry, struct path *lowerpath,
 243                               struct kstat *stat, const char *link)
 244 {
 245         struct inode *wdir = workdir->d_inode;
 246         struct inode *udir = upperdir->d_inode;
 247         struct dentry *newdentry = NULL;
 248         struct dentry *upper = NULL;
 249         umode_t mode = stat->mode;
 250         int err;
 251
 252         newdentry = ovl_lookup_temp(workdir, dentry);
 253         err = PTR_ERR(newdentry);
 254         if (IS_ERR(newdentry))
 255                 goto out;
 256
 257         upper = lookup_one_len(dentry->d_name.name, upperdir,
 258                                dentry->d_name.len);
 259         err = PTR_ERR(upper);
 260         if (IS_ERR(upper))
 261                 goto out1;
 262
 263         /* Can't properly set mode on creation because of the umask */
 264         stat->mode &= S_IFMT;
 265         err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
 266         stat->mode = mode;
 267         if (err)
 268                 goto out2;
 269
 270         if (S_ISREG(stat->mode)) {
 271                 struct path upperpath;
 272
 273                 ovl_path_upper(dentry, &upperpath);
 274                 BUG_ON(upperpath.dentry != NULL);
 275                 upperpath.dentry = newdentry;
 276
 277                 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
 278                 if (err)
 279                         goto out_cleanup;
 280         }
 281
 282         err = ovl_copy_xattr(lowerpath->dentry, newdentry);
 283         if (err)
 284                 goto out_cleanup;
 285
 286         inode_lock(newdentry->d_inode);
 287         err = ovl_set_attr(newdentry, stat);
 288         inode_unlock(newdentry->d_inode);
 289         if (err)
 290                 goto out_cleanup;
 291
 292         err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 293         if (err)
 294                 goto out_cleanup;
 295
 296         ovl_dentry_update(dentry, newdentry);
 297         ovl_inode_update(d_inode(dentry), d_inode(newdentry));
 298         newdentry = NULL;
 299
 300         /*
 301          * Non-directores become opaque when copied up.
 302          */
 303         if (!S_ISDIR(stat->mode))
 304                 ovl_dentry_set_opaque(dentry, true);
 305 out2:
 306         dput(upper);
 307 out1:
 308         dput(newdentry);
 309 out:
 310         return err;
 311
 312 out_cleanup:
 313         ovl_cleanup(wdir, newdentry);
 314         goto out2;
 315 }
 316
 317 /*
 318  * Copy up a single dentry
 319  *
 320  * Directory renames only allowed on "pure upper" (already created on
 321  * upper filesystem, never copied up).  Directories which are on lower or
 322  * are merged may not be renamed.  For these -EXDEV is returned and
 323  * userspace has to deal with it.  This means, when copying up a
 324  * directory we can rely on it and ancestors being stable.
 325  *
 326  * Non-directory renames start with copy up of source if necessary.  The
 327  * actual rename will only proceed once the copy up was successful.  Copy
 328  * up uses upper parent i_mutex for exclusion.  Since rename can change
 329  * d_parent it is possible that the copy up will lock the old parent.  At
 330  * that point the file will have already been copied up anyway.
 331  */
 332 int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 333                     struct path *lowerpath, struct kstat *stat)
 334 {
 335         struct dentry *workdir = ovl_workdir(dentry);
 336         int err;
 337         struct kstat pstat;
 338         struct path parentpath;
 339         struct dentry *upperdir;
 340         struct dentry *upperdentry;
 341         const struct cred *old_cred;
 342         char *link = NULL;
 343
 344         if (WARN_ON(!workdir))
 345                 return -EROFS;
 346
 347         ovl_do_check_copy_up(lowerpath->dentry);
 348
 349         ovl_path_upper(parent, &parentpath);
 350         upperdir = parentpath.dentry;
 351
 352         err = vfs_getattr(&parentpath, &pstat);
 353         if (err)
 354                 return err;
 355
 356         if (S_ISLNK(stat->mode)) {
 357                 link = ovl_read_symlink(lowerpath->dentry);
 358                 if (IS_ERR(link))
 359                         return PTR_ERR(link);
 360         }
 361
 362         old_cred = ovl_override_creds(dentry->d_sb);
 363
 364         err = -EIO;
 365         if (lock_rename(workdir, upperdir) != NULL) {
 366                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 367                 goto out_unlock;
 368         }
 369         upperdentry = ovl_dentry_upper(dentry);
 370         if (upperdentry) {
 371                 /* Raced with another copy-up?  Nothing to do, then... */
 372                 err = 0;
 373                 goto out_unlock;
 374         }
 375
 376         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 377                                  stat, link);
 378         if (!err) {
 379                 /* Restore timestamps on parent (best effort) */
 380                 ovl_set_timestamps(upperdir, &pstat);
 381         }
 382 out_unlock:
 383         unlock_rename(workdir, upperdir);
 384         revert_creds(old_cred);
 385
 386         if (link)
 387                 free_page((unsigned long) link);
 388
 389         return err;
 390 }
 391
 392 int ovl_copy_up(struct dentry *dentry)
 393 {
 394         int err;
 395
 396         err = 0;
 397         while (!err) {
 398                 struct dentry *next;
 399                 struct dentry *parent;
 400                 struct path lowerpath;
 401                 struct kstat stat;
 402                 enum ovl_path_type type = ovl_path_type(dentry);
 403
 404                 if (OVL_TYPE_UPPER(type))
 405                         break;
 406
 407                 next = dget(dentry);
 408                 /* find the topmost dentry not yet copied up */
 409                 for (;;) {
 410                         parent = dget_parent(next);
 411
 412                         type = ovl_path_type(parent);
 413                         if (OVL_TYPE_UPPER(type))
 414                                 break;
 415
 416                         dput(next);
 417                         next = parent;
 418                 }
 419
 420                 ovl_path_lower(next, &lowerpath);
 421                 err = vfs_getattr(&lowerpath, &stat);
 422                 if (!err)
 423                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 424
 425                 dput(parent);
 426                 dput(next);
 427         }
 428
 429         return err;
 430 }