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 error = 0;
  60         size_t slen;
  61
  62         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  63             !(new->d_inode->i_opflags & IOP_XATTR))
  64                 return 0;
  65
  66         list_size = vfs_listxattr(old, NULL, 0);
  67         if (list_size <= 0) {
  68                 if (list_size == -EOPNOTSUPP)
  69                         return 0;
  70                 return list_size;
  71         }
  72
  73         buf = kzalloc(list_size, GFP_KERNEL);
  74         if (!buf)
  75                 return -ENOMEM;
  76
  77         list_size = vfs_listxattr(old, buf, list_size);
  78         if (list_size <= 0) {
  79                 error = list_size;
  80                 goto out;
  81         }
  82
  83         for (name = buf; list_size; name += slen) {
  84                 slen = strnlen(name, list_size) + 1;
  85
  86                 /* underlying fs providing us with an broken xattr list? */
  87                 if (WARN_ON(slen > list_size)) {
  88                         error = -EIO;
  89                         break;
  90                 }
  91                 list_size -= slen;
  92
  93                 if (ovl_is_private_xattr(name))
  94                         continue;
  95 retry:
  96                 size = vfs_getxattr(old, name, value, value_size);
  97                 if (size == -ERANGE)
  98                         size = vfs_getxattr(old, name, NULL, 0);
  99
 100                 if (size < 0) {
 101                         error = size;
 102                         break;
 103                 }
 104
 105                 if (size > value_size) {
 106                         void *new;
 107
 108                         new = krealloc(value, size, GFP_KERNEL);
 109                         if (!new) {
 110                                 error = -ENOMEM;
 111                                 break;
 112                         }
 113                         value = new;
 114                         value_size = size;
 115                         goto retry;
 116                 }
 117
 118                 error = security_inode_copy_up_xattr(name);
 119                 if (error < 0 && error != -EOPNOTSUPP)
 120                         break;
 121                 if (error == 1) {
 122                         error = 0;
 123                         continue; /* Discard */
 124                 }
 125                 error = vfs_setxattr(new, name, value, size, 0);
 126                 if (error)
 127                         break;
 128         }
 129         kfree(value);
 130 out:
 131         kfree(buf);
 132         return error;
 133 }
 134
 135 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 136 {
 137         struct file *old_file;
 138         struct file *new_file;
 139         loff_t old_pos = 0;
 140         loff_t new_pos = 0;
 141         int error = 0;
 142
 143         if (len == 0)
 144                 return 0;
 145
 146         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 147         if (IS_ERR(old_file))
 148                 return PTR_ERR(old_file);
 149
 150         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 151         if (IS_ERR(new_file)) {
 152                 error = PTR_ERR(new_file);
 153                 goto out_fput;
 154         }
 155
 156         /* FIXME: copy up sparse files efficiently */
 157         while (len) {
 158                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 159                 long bytes;
 160
 161                 if (len < this_len)
 162                         this_len = len;
 163
 164                 if (signal_pending_state(TASK_KILLABLE, current)) {
 165                         error = -EINTR;
 166                         break;
 167                 }
 168
 169                 bytes = do_splice_direct(old_file, &old_pos,
 170                                          new_file, &new_pos,
 171                                          this_len, SPLICE_F_MOVE);
 172                 if (bytes <= 0) {
 173                         error = bytes;
 174                         break;
 175                 }
 176                 WARN_ON(old_pos != new_pos);
 177
 178                 len -= bytes;
 179         }
 180
 181         if (!error)
 182                 error = vfs_fsync(new_file, 0);
 183         fput(new_file);
 184 out_fput:
 185         fput(old_file);
 186         return error;
 187 }
 188
 189 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 190 {
 191         struct iattr attr = {
 192                 .ia_valid =
 193                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 194                 .ia_atime = stat->atime,
 195                 .ia_mtime = stat->mtime,
 196         };
 197
 198         return notify_change(upperdentry, &attr, NULL);
 199 }
 200
 201 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 202 {
 203         int err = 0;
 204
 205         if (!S_ISLNK(stat->mode)) {
 206                 struct iattr attr = {
 207                         .ia_valid = ATTR_MODE,
 208                         .ia_mode = stat->mode,
 209                 };
 210                 err = notify_change(upperdentry, &attr, NULL);
 211         }
 212         if (!err) {
 213                 struct iattr attr = {
 214                         .ia_valid = ATTR_UID | ATTR_GID,
 215                         .ia_uid = stat->uid,
 216                         .ia_gid = stat->gid,
 217                 };
 218                 err = notify_change(upperdentry, &attr, NULL);
 219         }
 220         if (!err)
 221                 ovl_set_timestamps(upperdentry, stat);
 222
 223         return err;
 224 }
 225
 226 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 227                               struct dentry *dentry, struct path *lowerpath,
 228                               struct kstat *stat, const char *link)
 229 {
 230         struct inode *wdir = workdir->d_inode;
 231         struct inode *udir = upperdir->d_inode;
 232         struct dentry *newdentry = NULL;
 233         struct dentry *upper = NULL;
 234         umode_t mode = stat->mode;
 235         int err;
 236         const struct cred *old_creds = NULL;
 237         struct cred *new_creds = NULL;
 238
 239         newdentry = ovl_lookup_temp(workdir, dentry);
 240         err = PTR_ERR(newdentry);
 241         if (IS_ERR(newdentry))
 242                 goto out;
 243
 244         upper = lookup_one_len(dentry->d_name.name, upperdir,
 245                                dentry->d_name.len);
 246         err = PTR_ERR(upper);
 247         if (IS_ERR(upper))
 248                 goto out1;
 249
 250         err = security_inode_copy_up(dentry, &new_creds);
 251         if (err < 0)
 252                 goto out2;
 253
 254         if (new_creds)
 255                 old_creds = override_creds(new_creds);
 256
 257         /* Can't properly set mode on creation because of the umask */
 258         stat->mode &= S_IFMT;
 259         err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
 260         stat->mode = mode;
 261
 262         if (new_creds) {
 263                 revert_creds(old_creds);
 264                 put_cred(new_creds);
 265         }
 266
 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         DEFINE_DELAYED_CALL(done);
 336         struct dentry *workdir = ovl_workdir(dentry);
 337         int err;
 338         struct kstat pstat;
 339         struct path parentpath;
 340         struct dentry *lowerdentry = lowerpath->dentry;
 341         struct dentry *upperdir;
 342         struct dentry *upperdentry;
 343         const char *link = NULL;
 344
 345         if (WARN_ON(!workdir))
 346                 return -EROFS;
 347
 348         ovl_do_check_copy_up(lowerdentry);
 349
 350         ovl_path_upper(parent, &parentpath);
 351         upperdir = parentpath.dentry;
 352
 353         err = vfs_getattr(&parentpath, &pstat);
 354         if (err)
 355                 return err;
 356
 357         if (S_ISLNK(stat->mode)) {
 358                 link = vfs_get_link(lowerdentry, &done);
 359                 if (IS_ERR(link))
 360                         return PTR_ERR(link);
 361         }
 362
 363         err = -EIO;
 364         if (lock_rename(workdir, upperdir) != NULL) {
 365                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 366                 goto out_unlock;
 367         }
 368         upperdentry = ovl_dentry_upper(dentry);
 369         if (upperdentry) {
 370                 /* Raced with another copy-up?  Nothing to do, then... */
 371                 err = 0;
 372                 goto out_unlock;
 373         }
 374
 375         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 376                                  stat, link);
 377         if (!err) {
 378                 /* Restore timestamps on parent (best effort) */
 379                 ovl_set_timestamps(upperdir, &pstat);
 380         }
 381 out_unlock:
 382         unlock_rename(workdir, upperdir);
 383         do_delayed_call(&done);
 384
 385         return err;
 386 }
 387
 388 int ovl_copy_up(struct dentry *dentry)
 389 {
 390         int err = 0;
 391         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 392
 393         while (!err) {
 394                 struct dentry *next;
 395                 struct dentry *parent;
 396                 struct path lowerpath;
 397                 struct kstat stat;
 398                 enum ovl_path_type type = ovl_path_type(dentry);
 399
 400                 if (OVL_TYPE_UPPER(type))
 401                         break;
 402
 403                 next = dget(dentry);
 404                 /* find the topmost dentry not yet copied up */
 405                 for (;;) {
 406                         parent = dget_parent(next);
 407
 408                         type = ovl_path_type(parent);
 409                         if (OVL_TYPE_UPPER(type))
 410                                 break;
 411
 412                         dput(next);
 413                         next = parent;
 414                 }
 415
 416                 ovl_path_lower(next, &lowerpath);
 417                 err = vfs_getattr(&lowerpath, &stat);
 418                 if (!err)
 419                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 420
 421                 dput(parent);
 422                 dput(next);
 423         }
 424         revert_creds(old_cred);
 425
 426         return err;
 427 }