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/signal.h>
  19 #include <linux/cred.h>
  20 #include <linux/namei.h>
  21 #include <linux/fdtable.h>
  22 #include <linux/ratelimit.h>
  23 #include "overlayfs.h"
  24
  25 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  26
  27 static bool __read_mostly ovl_check_copy_up;
  28 module_param_named(check_copy_up, ovl_check_copy_up, bool,
  29                    S_IWUSR | S_IRUGO);
  30 MODULE_PARM_DESC(ovl_check_copy_up,
  31                  "Warn on copy-up when causing process also has a R/O fd open");
  32
  33 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
  34 {
  35         const struct dentry *dentry = data;
  36
  37         if (file_inode(f) == d_inode(dentry))
  38                 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",
  39                                     f, fd, current->pid, current->comm);
  40         return 0;
  41 }
  42
  43 /*
  44  * Check the fds open by this process and warn if something like the following
  45  * scenario is about to occur:
  46  *
  47  *      fd1 = open("foo", O_RDONLY);
  48  *      fd2 = open("foo", O_RDWR);
  49  */
  50 static void ovl_do_check_copy_up(struct dentry *dentry)
  51 {
  52         if (ovl_check_copy_up)
  53                 iterate_fd(current->files, 0, ovl_check_fd, dentry);
  54 }
  55
  56 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  57 {
  58         ssize_t list_size, size, value_size = 0;
  59         char *buf, *name, *value = NULL;
  60         int uninitialized_var(error);
  61         size_t slen;
  62
  63         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  64             !(new->d_inode->i_opflags & IOP_XATTR))
  65                 return 0;
  66
  67         list_size = vfs_listxattr(old, NULL, 0);
  68         if (list_size <= 0) {
  69                 if (list_size == -EOPNOTSUPP)
  70                         return 0;
  71                 return list_size;
  72         }
  73
  74         buf = kzalloc(list_size, GFP_KERNEL);
  75         if (!buf)
  76                 return -ENOMEM;
  77
  78         list_size = vfs_listxattr(old, buf, list_size);
  79         if (list_size <= 0) {
  80                 error = list_size;
  81                 goto out;
  82         }
  83
  84         for (name = buf; list_size; name += slen) {
  85                 slen = strnlen(name, list_size) + 1;
  86
  87                 /* underlying fs providing us with an broken xattr list? */
  88                 if (WARN_ON(slen > list_size)) {
  89                         error = -EIO;
  90                         break;
  91                 }
  92                 list_size -= slen;
  93
  94                 if (ovl_is_private_xattr(name))
  95                         continue;
  96 retry:
  97                 size = vfs_getxattr(old, name, value, value_size);
  98                 if (size == -ERANGE)
  99                         size = vfs_getxattr(old, name, NULL, 0);
 100
 101                 if (size < 0) {
 102                         error = size;
 103                         break;
 104                 }
 105
 106                 if (size > value_size) {
 107                         void *new;
 108
 109                         new = krealloc(value, size, GFP_KERNEL);
 110                         if (!new) {
 111                                 error = -ENOMEM;
 112                                 break;
 113                         }
 114                         value = new;
 115                         value_size = size;
 116                         goto retry;
 117                 }
 118
 119                 error = security_inode_copy_up_xattr(name);
 120                 if (error < 0 && error != -EOPNOTSUPP)
 121                         break;
 122                 if (error == 1) {
 123                         error = 0;
 124                         continue; /* Discard */
 125                 }
 126                 error = vfs_setxattr(new, name, value, size, 0);
 127                 if (error)
 128                         break;
 129         }
 130         kfree(value);
 131 out:
 132         kfree(buf);
 133         return error;
 134 }
 135
 136 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 137 {
 138         struct file *old_file;
 139         struct file *new_file;
 140         loff_t old_pos = 0;
 141         loff_t new_pos = 0;
 142         int error = 0;
 143
 144         if (len == 0)
 145                 return 0;
 146
 147         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 148         if (IS_ERR(old_file))
 149                 return PTR_ERR(old_file);
 150
 151         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 152         if (IS_ERR(new_file)) {
 153                 error = PTR_ERR(new_file);
 154                 goto out_fput;
 155         }
 156
 157         /* Try to use clone_file_range to clone up within the same fs */
 158         error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
 159         if (!error)
 160                 goto out;
 161         /* Couldn't clone, so now we try to copy the data */
 162         error = 0;
 163
 164         /* FIXME: copy up sparse files efficiently */
 165         while (len) {
 166                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 167                 long bytes;
 168
 169                 if (len < this_len)
 170                         this_len = len;
 171
 172                 if (signal_pending_state(TASK_KILLABLE, current)) {
 173                         error = -EINTR;
 174                         break;
 175                 }
 176
 177                 bytes = do_splice_direct(old_file, &old_pos,
 178                                          new_file, &new_pos,
 179                                          this_len, SPLICE_F_MOVE);
 180                 if (bytes <= 0) {
 181                         error = bytes;
 182                         break;
 183                 }
 184                 WARN_ON(old_pos != new_pos);
 185
 186                 len -= bytes;
 187         }
 188 out:
 189         if (!error)
 190                 error = vfs_fsync(new_file, 0);
 191         fput(new_file);
 192 out_fput:
 193         fput(old_file);
 194         return error;
 195 }
 196
 197 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 198 {
 199         struct iattr attr = {
 200                 .ia_valid =
 201                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 202                 .ia_atime = stat->atime,
 203                 .ia_mtime = stat->mtime,
 204         };
 205
 206         return notify_change(upperdentry, &attr, NULL);
 207 }
 208
 209 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 210 {
 211         int err = 0;
 212
 213         if (!S_ISLNK(stat->mode)) {
 214                 struct iattr attr = {
 215                         .ia_valid = ATTR_MODE,
 216                         .ia_mode = stat->mode,
 217                 };
 218                 err = notify_change(upperdentry, &attr, NULL);
 219         }
 220         if (!err) {
 221                 struct iattr attr = {
 222                         .ia_valid = ATTR_UID | ATTR_GID,
 223                         .ia_uid = stat->uid,
 224                         .ia_gid = stat->gid,
 225                 };
 226                 err = notify_change(upperdentry, &attr, NULL);
 227         }
 228         if (!err)
 229                 ovl_set_timestamps(upperdentry, stat);
 230
 231         return err;
 232 }
 233
 234 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 235                               struct dentry *dentry, struct path *lowerpath,
 236                               struct kstat *stat, const char *link)
 237 {
 238         struct inode *wdir = workdir->d_inode;
 239         struct inode *udir = upperdir->d_inode;
 240         struct dentry *newdentry = NULL;
 241         struct dentry *upper = NULL;
 242         int err;
 243         const struct cred *old_creds = NULL;
 244         struct cred *new_creds = NULL;
 245         struct cattr cattr = {
 246                 /* Can't properly set mode on creation because of the umask */
 247                 .mode = stat->mode & S_IFMT,
 248                 .rdev = stat->rdev,
 249                 .link = link
 250         };
 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         err = security_inode_copy_up(dentry, &new_creds);
 264         if (err < 0)
 265                 goto out2;
 266
 267         if (new_creds)
 268                 old_creds = override_creds(new_creds);
 269
 270         err = ovl_create_real(wdir, newdentry, &cattr, NULL, true);
 271
 272         if (new_creds) {
 273                 revert_creds(old_creds);
 274                 put_cred(new_creds);
 275         }
 276
 277         if (err)
 278                 goto out2;
 279
 280         if (S_ISREG(stat->mode)) {
 281                 struct path upperpath;
 282
 283                 ovl_path_upper(dentry, &upperpath);
 284                 BUG_ON(upperpath.dentry != NULL);
 285                 upperpath.dentry = newdentry;
 286
 287                 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
 288                 if (err)
 289                         goto out_cleanup;
 290         }
 291
 292         err = ovl_copy_xattr(lowerpath->dentry, newdentry);
 293         if (err)
 294                 goto out_cleanup;
 295
 296         inode_lock(newdentry->d_inode);
 297         err = ovl_set_attr(newdentry, stat);
 298         inode_unlock(newdentry->d_inode);
 299         if (err)
 300                 goto out_cleanup;
 301
 302         err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 303         if (err)
 304                 goto out_cleanup;
 305
 306         ovl_dentry_update(dentry, newdentry);
 307         ovl_inode_update(d_inode(dentry), d_inode(newdentry));
 308         newdentry = NULL;
 309 out2:
 310         dput(upper);
 311 out1:
 312         dput(newdentry);
 313 out:
 314         return err;
 315
 316 out_cleanup:
 317         ovl_cleanup(wdir, newdentry);
 318         goto out2;
 319 }
 320
 321 /*
 322  * Copy up a single dentry
 323  *
 324  * All renames start with copy up of source if necessary.  The actual
 325  * rename will only proceed once the copy up was successful.  Copy up uses
 326  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 327  * is possible that the copy up will lock the old parent.  At that point
 328  * the file will have already been copied up anyway.
 329  */
 330 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 331                            struct path *lowerpath, struct kstat *stat)
 332 {
 333         DEFINE_DELAYED_CALL(done);
 334         struct dentry *workdir = ovl_workdir(dentry);
 335         int err;
 336         struct kstat pstat;
 337         struct path parentpath;
 338         struct dentry *lowerdentry = lowerpath->dentry;
 339         struct dentry *upperdir;
 340         const char *link = NULL;
 341
 342         if (WARN_ON(!workdir))
 343                 return -EROFS;
 344
 345         ovl_do_check_copy_up(lowerdentry);
 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 = vfs_get_link(lowerdentry, &done);
 356                 if (IS_ERR(link))
 357                         return PTR_ERR(link);
 358         }
 359
 360         err = -EIO;
 361         if (lock_rename(workdir, upperdir) != NULL) {
 362                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 363                 goto out_unlock;
 364         }
 365         if (ovl_dentry_upper(dentry)) {
 366                 /* Raced with another copy-up?  Nothing to do, then... */
 367                 err = 0;
 368                 goto out_unlock;
 369         }
 370
 371         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 372                                  stat, link);
 373         if (!err) {
 374                 /* Restore timestamps on parent (best effort) */
 375                 ovl_set_timestamps(upperdir, &pstat);
 376         }
 377 out_unlock:
 378         unlock_rename(workdir, upperdir);
 379         do_delayed_call(&done);
 380
 381         return err;
 382 }
 383
 384 int ovl_copy_up_flags(struct dentry *dentry, int flags)
 385 {
 386         int err = 0;
 387         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 388
 389         while (!err) {
 390                 struct dentry *next;
 391                 struct dentry *parent;
 392                 struct path lowerpath;
 393                 struct kstat stat;
 394                 enum ovl_path_type type = ovl_path_type(dentry);
 395
 396                 if (OVL_TYPE_UPPER(type))
 397                         break;
 398
 399                 next = dget(dentry);
 400                 /* find the topmost dentry not yet copied up */
 401                 for (;;) {
 402                         parent = dget_parent(next);
 403
 404                         type = ovl_path_type(parent);
 405                         if (OVL_TYPE_UPPER(type))
 406                                 break;
 407
 408                         dput(next);
 409                         next = parent;
 410                 }
 411
 412                 ovl_path_lower(next, &lowerpath);
 413                 err = vfs_getattr(&lowerpath, &stat);
 414                 /* maybe truncate regular file. this has no effect on dirs */
 415                 if (flags & O_TRUNC)
 416                         stat.size = 0;
 417                 if (!err)
 418                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 419
 420                 dput(parent);
 421                 dput(next);
 422         }
 423         revert_creds(old_cred);
 424
 425         return err;
 426 }
 427
 428 int ovl_copy_up(struct dentry *dentry)
 429 {
 430         return ovl_copy_up_flags(dentry, 0);
 431 }