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 (file_inode(f) == 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         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         /* Try to use clone_file_range to clone up within the same fs */
 157         error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
 158         if (!error)
 159                 goto out;
 160         /* Couldn't clone, so now we try to copy the data */
 161         error = 0;
 162
 163         /* FIXME: copy up sparse files efficiently */
 164         while (len) {
 165                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 166                 long bytes;
 167
 168                 if (len < this_len)
 169                         this_len = len;
 170
 171                 if (signal_pending_state(TASK_KILLABLE, current)) {
 172                         error = -EINTR;
 173                         break;
 174                 }
 175
 176                 bytes = do_splice_direct(old_file, &old_pos,
 177                                          new_file, &new_pos,
 178                                          this_len, SPLICE_F_MOVE);
 179                 if (bytes <= 0) {
 180                         error = bytes;
 181                         break;
 182                 }
 183                 WARN_ON(old_pos != new_pos);
 184
 185                 len -= bytes;
 186         }
 187 out:
 188         if (!error)
 189                 error = vfs_fsync(new_file, 0);
 190         fput(new_file);
 191 out_fput:
 192         fput(old_file);
 193         return error;
 194 }
 195
 196 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 197 {
 198         struct iattr attr = {
 199                 .ia_valid =
 200                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 201                 .ia_atime = stat->atime,
 202                 .ia_mtime = stat->mtime,
 203         };
 204
 205         return notify_change(upperdentry, &attr, NULL);
 206 }
 207
 208 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 209 {
 210         int err = 0;
 211
 212         if (!S_ISLNK(stat->mode)) {
 213                 struct iattr attr = {
 214                         .ia_valid = ATTR_MODE,
 215                         .ia_mode = stat->mode,
 216                 };
 217                 err = notify_change(upperdentry, &attr, NULL);
 218         }
 219         if (!err) {
 220                 struct iattr attr = {
 221                         .ia_valid = ATTR_UID | ATTR_GID,
 222                         .ia_uid = stat->uid,
 223                         .ia_gid = stat->gid,
 224                 };
 225                 err = notify_change(upperdentry, &attr, NULL);
 226         }
 227         if (!err)
 228                 ovl_set_timestamps(upperdentry, stat);
 229
 230         return err;
 231 }
 232
 233 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 234                               struct dentry *dentry, struct path *lowerpath,
 235                               struct kstat *stat, const char *link)
 236 {
 237         struct inode *wdir = workdir->d_inode;
 238         struct inode *udir = upperdir->d_inode;
 239         struct dentry *newdentry = NULL;
 240         struct dentry *upper = NULL;
 241         int err;
 242         const struct cred *old_creds = NULL;
 243         struct cred *new_creds = NULL;
 244         struct cattr cattr = {
 245                 /* Can't properly set mode on creation because of the umask */
 246                 .mode = stat->mode & S_IFMT,
 247                 .rdev = stat->rdev,
 248                 .link = link
 249         };
 250
 251         newdentry = ovl_lookup_temp(workdir, dentry);
 252         err = PTR_ERR(newdentry);
 253         if (IS_ERR(newdentry))
 254                 goto out;
 255
 256         upper = lookup_one_len(dentry->d_name.name, upperdir,
 257                                dentry->d_name.len);
 258         err = PTR_ERR(upper);
 259         if (IS_ERR(upper))
 260                 goto out1;
 261
 262         err = security_inode_copy_up(dentry, &new_creds);
 263         if (err < 0)
 264                 goto out2;
 265
 266         if (new_creds)
 267                 old_creds = override_creds(new_creds);
 268
 269         err = ovl_create_real(wdir, newdentry, &cattr, NULL, true);
 270
 271         if (new_creds) {
 272                 revert_creds(old_creds);
 273                 put_cred(new_creds);
 274         }
 275
 276         if (err)
 277                 goto out2;
 278
 279         if (S_ISREG(stat->mode)) {
 280                 struct path upperpath;
 281
 282                 ovl_path_upper(dentry, &upperpath);
 283                 BUG_ON(upperpath.dentry != NULL);
 284                 upperpath.dentry = newdentry;
 285
 286                 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
 287                 if (err)
 288                         goto out_cleanup;
 289         }
 290
 291         err = ovl_copy_xattr(lowerpath->dentry, newdentry);
 292         if (err)
 293                 goto out_cleanup;
 294
 295         inode_lock(newdentry->d_inode);
 296         err = ovl_set_attr(newdentry, stat);
 297         inode_unlock(newdentry->d_inode);
 298         if (err)
 299                 goto out_cleanup;
 300
 301         err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 302         if (err)
 303                 goto out_cleanup;
 304
 305         ovl_dentry_update(dentry, newdentry);
 306         ovl_inode_update(d_inode(dentry), d_inode(newdentry));
 307         newdentry = NULL;
 308 out2:
 309         dput(upper);
 310 out1:
 311         dput(newdentry);
 312 out:
 313         return err;
 314
 315 out_cleanup:
 316         ovl_cleanup(wdir, newdentry);
 317         goto out2;
 318 }
 319
 320 /*
 321  * Copy up a single dentry
 322  *
 323  * All renames start with copy up of source if necessary.  The actual
 324  * rename will only proceed once the copy up was successful.  Copy up uses
 325  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 326  * is possible that the copy up will lock the old parent.  At that point
 327  * the file will have already been copied up anyway.
 328  */
 329 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 330                            struct path *lowerpath, struct kstat *stat)
 331 {
 332         DEFINE_DELAYED_CALL(done);
 333         struct dentry *workdir = ovl_workdir(dentry);
 334         int err;
 335         struct kstat pstat;
 336         struct path parentpath;
 337         struct dentry *lowerdentry = lowerpath->dentry;
 338         struct dentry *upperdir;
 339         const char *link = NULL;
 340
 341         if (WARN_ON(!workdir))
 342                 return -EROFS;
 343
 344         ovl_do_check_copy_up(lowerdentry);
 345
 346         ovl_path_upper(parent, &parentpath);
 347         upperdir = parentpath.dentry;
 348
 349         err = vfs_getattr(&parentpath, &pstat);
 350         if (err)
 351                 return err;
 352
 353         if (S_ISLNK(stat->mode)) {
 354                 link = vfs_get_link(lowerdentry, &done);
 355                 if (IS_ERR(link))
 356                         return PTR_ERR(link);
 357         }
 358
 359         err = -EIO;
 360         if (lock_rename(workdir, upperdir) != NULL) {
 361                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 362                 goto out_unlock;
 363         }
 364         if (ovl_dentry_upper(dentry)) {
 365                 /* Raced with another copy-up?  Nothing to do, then... */
 366                 err = 0;
 367                 goto out_unlock;
 368         }
 369
 370         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 371                                  stat, link);
 372         if (!err) {
 373                 /* Restore timestamps on parent (best effort) */
 374                 ovl_set_timestamps(upperdir, &pstat);
 375         }
 376 out_unlock:
 377         unlock_rename(workdir, upperdir);
 378         do_delayed_call(&done);
 379
 380         return err;
 381 }
 382
 383 int ovl_copy_up_flags(struct dentry *dentry, int flags)
 384 {
 385         int err = 0;
 386         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 387
 388         while (!err) {
 389                 struct dentry *next;
 390                 struct dentry *parent;
 391                 struct path lowerpath;
 392                 struct kstat stat;
 393                 enum ovl_path_type type = ovl_path_type(dentry);
 394
 395                 if (OVL_TYPE_UPPER(type))
 396                         break;
 397
 398                 next = dget(dentry);
 399                 /* find the topmost dentry not yet copied up */
 400                 for (;;) {
 401                         parent = dget_parent(next);
 402
 403                         type = ovl_path_type(parent);
 404                         if (OVL_TYPE_UPPER(type))
 405                                 break;
 406
 407                         dput(next);
 408                         next = parent;
 409                 }
 410
 411                 ovl_path_lower(next, &lowerpath);
 412                 err = vfs_getattr(&lowerpath, &stat);
 413                 /* maybe truncate regular file. this has no effect on dirs */
 414                 if (flags & O_TRUNC)
 415                         stat.size = 0;
 416                 if (!err)
 417                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 418
 419                 dput(parent);
 420                 dput(next);
 421         }
 422         revert_creds(old_cred);
 423
 424         return err;
 425 }
 426
 427 int ovl_copy_up(struct dentry *dentry)
 428 {
 429         return ovl_copy_up_flags(dentry, 0);
 430 }