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 #include "ovl_entry.h"
  25
  26 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  27
  28 static bool __read_mostly ovl_check_copy_up;
  29 module_param_named(check_copy_up, ovl_check_copy_up, bool,
  30                    S_IWUSR | S_IRUGO);
  31 MODULE_PARM_DESC(ovl_check_copy_up,
  32                  "Warn on copy-up when causing process also has a R/O fd open");
  33
  34 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
  35 {
  36         const struct dentry *dentry = data;
  37
  38         if (file_inode(f) == d_inode(dentry))
  39                 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",
  40                                     f, fd, current->pid, current->comm);
  41         return 0;
  42 }
  43
  44 /*
  45  * Check the fds open by this process and warn if something like the following
  46  * scenario is about to occur:
  47  *
  48  *      fd1 = open("foo", O_RDONLY);
  49  *      fd2 = open("foo", O_RDWR);
  50  */
  51 static void ovl_do_check_copy_up(struct dentry *dentry)
  52 {
  53         if (ovl_check_copy_up)
  54                 iterate_fd(current->files, 0, ovl_check_fd, dentry);
  55 }
  56
  57 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  58 {
  59         ssize_t list_size, size, value_size = 0;
  60         char *buf, *name, *value = NULL;
  61         int uninitialized_var(error);
  62         size_t slen;
  63
  64         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  65             !(new->d_inode->i_opflags & IOP_XATTR))
  66                 return 0;
  67
  68         list_size = vfs_listxattr(old, NULL, 0);
  69         if (list_size <= 0) {
  70                 if (list_size == -EOPNOTSUPP)
  71                         return 0;
  72                 return list_size;
  73         }
  74
  75         buf = kzalloc(list_size, GFP_KERNEL);
  76         if (!buf)
  77                 return -ENOMEM;
  78
  79         list_size = vfs_listxattr(old, buf, list_size);
  80         if (list_size <= 0) {
  81                 error = list_size;
  82                 goto out;
  83         }
  84
  85         for (name = buf; list_size; name += slen) {
  86                 slen = strnlen(name, list_size) + 1;
  87
  88                 /* underlying fs providing us with an broken xattr list? */
  89                 if (WARN_ON(slen > list_size)) {
  90                         error = -EIO;
  91                         break;
  92                 }
  93                 list_size -= slen;
  94
  95                 if (ovl_is_private_xattr(name))
  96                         continue;
  97 retry:
  98                 size = vfs_getxattr(old, name, value, value_size);
  99                 if (size == -ERANGE)
 100                         size = vfs_getxattr(old, name, NULL, 0);
 101
 102                 if (size < 0) {
 103                         error = size;
 104                         break;
 105                 }
 106
 107                 if (size > value_size) {
 108                         void *new;
 109
 110                         new = krealloc(value, size, GFP_KERNEL);
 111                         if (!new) {
 112                                 error = -ENOMEM;
 113                                 break;
 114                         }
 115                         value = new;
 116                         value_size = size;
 117                         goto retry;
 118                 }
 119
 120                 error = security_inode_copy_up_xattr(name);
 121                 if (error < 0 && error != -EOPNOTSUPP)
 122                         break;
 123                 if (error == 1) {
 124                         error = 0;
 125                         continue; /* Discard */
 126                 }
 127                 error = vfs_setxattr(new, name, value, size, 0);
 128                 if (error)
 129                         break;
 130         }
 131         kfree(value);
 132 out:
 133         kfree(buf);
 134         return error;
 135 }
 136
 137 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 138 {
 139         struct file *old_file;
 140         struct file *new_file;
 141         loff_t old_pos = 0;
 142         loff_t new_pos = 0;
 143         int error = 0;
 144
 145         if (len == 0)
 146                 return 0;
 147
 148         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 149         if (IS_ERR(old_file))
 150                 return PTR_ERR(old_file);
 151
 152         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 153         if (IS_ERR(new_file)) {
 154                 error = PTR_ERR(new_file);
 155                 goto out_fput;
 156         }
 157
 158         /* Try to use clone_file_range to clone up within the same fs */
 159         error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
 160         if (!error)
 161                 goto out;
 162         /* Couldn't clone, so now we try to copy the data */
 163         error = 0;
 164
 165         /* FIXME: copy up sparse files efficiently */
 166         while (len) {
 167                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 168                 long bytes;
 169
 170                 if (len < this_len)
 171                         this_len = len;
 172
 173                 if (signal_pending_state(TASK_KILLABLE, current)) {
 174                         error = -EINTR;
 175                         break;
 176                 }
 177
 178                 bytes = do_splice_direct(old_file, &old_pos,
 179                                          new_file, &new_pos,
 180                                          this_len, SPLICE_F_MOVE);
 181                 if (bytes <= 0) {
 182                         error = bytes;
 183                         break;
 184                 }
 185                 WARN_ON(old_pos != new_pos);
 186
 187                 len -= bytes;
 188         }
 189 out:
 190         if (!error)
 191                 error = vfs_fsync(new_file, 0);
 192         fput(new_file);
 193 out_fput:
 194         fput(old_file);
 195         return error;
 196 }
 197
 198 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 199 {
 200         struct iattr attr = {
 201                 .ia_valid =
 202                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 203                 .ia_atime = stat->atime,
 204                 .ia_mtime = stat->mtime,
 205         };
 206
 207         return notify_change(upperdentry, &attr, NULL);
 208 }
 209
 210 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 211 {
 212         int err = 0;
 213
 214         if (!S_ISLNK(stat->mode)) {
 215                 struct iattr attr = {
 216                         .ia_valid = ATTR_MODE,
 217                         .ia_mode = stat->mode,
 218                 };
 219                 err = notify_change(upperdentry, &attr, NULL);
 220         }
 221         if (!err) {
 222                 struct iattr attr = {
 223                         .ia_valid = ATTR_UID | ATTR_GID,
 224                         .ia_uid = stat->uid,
 225                         .ia_gid = stat->gid,
 226                 };
 227                 err = notify_change(upperdentry, &attr, NULL);
 228         }
 229         if (!err)
 230                 ovl_set_timestamps(upperdentry, stat);
 231
 232         return err;
 233 }
 234
 235 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 236                               struct dentry *dentry, struct path *lowerpath,
 237                               struct kstat *stat, const char *link,
 238                               struct kstat *pstat, bool tmpfile)
 239 {
 240         struct inode *wdir = workdir->d_inode;
 241         struct inode *udir = upperdir->d_inode;
 242         struct dentry *newdentry = NULL;
 243         struct dentry *upper = NULL;
 244         struct dentry *temp = NULL;
 245         int err;
 246         const struct cred *old_creds = NULL;
 247         struct cred *new_creds = NULL;
 248         struct cattr cattr = {
 249                 /* Can't properly set mode on creation because of the umask */
 250                 .mode = stat->mode & S_IFMT,
 251                 .rdev = stat->rdev,
 252                 .link = link
 253         };
 254
 255         err = security_inode_copy_up(dentry, &new_creds);
 256         if (err < 0)
 257                 goto out;
 258
 259         if (new_creds)
 260                 old_creds = override_creds(new_creds);
 261
 262         if (tmpfile)
 263                 temp = ovl_do_tmpfile(upperdir, stat->mode);
 264         else
 265                 temp = ovl_lookup_temp(workdir, dentry);
 266         err = 0;
 267         if (IS_ERR(temp)) {
 268                 err = PTR_ERR(temp);
 269                 temp = NULL;
 270         }
 271
 272         if (!err && !tmpfile)
 273                 err = ovl_create_real(wdir, temp, &cattr, NULL, true);
 274
 275         if (new_creds) {
 276                 revert_creds(old_creds);
 277                 put_cred(new_creds);
 278         }
 279
 280         if (err)
 281                 goto out;
 282
 283         if (S_ISREG(stat->mode)) {
 284                 struct path upperpath;
 285
 286                 ovl_path_upper(dentry, &upperpath);
 287                 BUG_ON(upperpath.dentry != NULL);
 288                 upperpath.dentry = temp;
 289
 290                 if (tmpfile) {
 291                         inode_unlock(udir);
 292                         err = ovl_copy_up_data(lowerpath, &upperpath,
 293                                                stat->size);
 294                         inode_lock_nested(udir, I_MUTEX_PARENT);
 295                 } else {
 296                         err = ovl_copy_up_data(lowerpath, &upperpath,
 297                                                stat->size);
 298                 }
 299
 300                 if (err)
 301                         goto out_cleanup;
 302         }
 303
 304         err = ovl_copy_xattr(lowerpath->dentry, temp);
 305         if (err)
 306                 goto out_cleanup;
 307
 308         inode_lock(temp->d_inode);
 309         err = ovl_set_attr(temp, stat);
 310         inode_unlock(temp->d_inode);
 311         if (err)
 312                 goto out_cleanup;
 313
 314         upper = lookup_one_len(dentry->d_name.name, upperdir,
 315                                dentry->d_name.len);
 316         if (IS_ERR(upper)) {
 317                 err = PTR_ERR(upper);
 318                 upper = NULL;
 319                 goto out_cleanup;
 320         }
 321
 322         if (tmpfile)
 323                 err = ovl_do_link(temp, udir, upper, true);
 324         else
 325                 err = ovl_do_rename(wdir, temp, udir, upper, 0);
 326         if (err)
 327                 goto out_cleanup;
 328
 329         newdentry = dget(tmpfile ? upper : temp);
 330         ovl_dentry_update(dentry, newdentry);
 331         ovl_inode_update(d_inode(dentry), d_inode(newdentry));
 332
 333         /* Restore timestamps on parent (best effort) */
 334         ovl_set_timestamps(upperdir, pstat);
 335 out:
 336         dput(temp);
 337         dput(upper);
 338         return err;
 339
 340 out_cleanup:
 341         if (!tmpfile)
 342                 ovl_cleanup(wdir, temp);
 343         goto out;
 344 }
 345
 346 /*
 347  * Copy up a single dentry
 348  *
 349  * All renames start with copy up of source if necessary.  The actual
 350  * rename will only proceed once the copy up was successful.  Copy up uses
 351  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 352  * is possible that the copy up will lock the old parent.  At that point
 353  * the file will have already been copied up anyway.
 354  */
 355 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 356                            struct path *lowerpath, struct kstat *stat)
 357 {
 358         DEFINE_DELAYED_CALL(done);
 359         struct dentry *workdir = ovl_workdir(dentry);
 360         int err;
 361         struct kstat pstat;
 362         struct path parentpath;
 363         struct dentry *lowerdentry = lowerpath->dentry;
 364         struct dentry *upperdir;
 365         const char *link = NULL;
 366         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 367
 368         if (WARN_ON(!workdir))
 369                 return -EROFS;
 370
 371         ovl_do_check_copy_up(lowerdentry);
 372
 373         ovl_path_upper(parent, &parentpath);
 374         upperdir = parentpath.dentry;
 375
 376         err = vfs_getattr(&parentpath, &pstat,
 377                           STATX_ATIME | STATX_MTIME, AT_STATX_SYNC_AS_STAT);
 378         if (err)
 379                 return err;
 380
 381         if (S_ISLNK(stat->mode)) {
 382                 link = vfs_get_link(lowerdentry, &done);
 383                 if (IS_ERR(link))
 384                         return PTR_ERR(link);
 385         }
 386
 387         /* Should we copyup with O_TMPFILE or with workdir? */
 388         if (S_ISREG(stat->mode) && ofs->tmpfile) {
 389                 err = ovl_copy_up_start(dentry);
 390                 /* err < 0: interrupted, err > 0: raced with another copy-up */
 391                 if (unlikely(err)) {
 392                         pr_debug("ovl_copy_up_start(%pd2) = %i\n", dentry, err);
 393                         if (err > 0)
 394                                 err = 0;
 395                         goto out_done;
 396                 }
 397
 398                 inode_lock_nested(upperdir->d_inode, I_MUTEX_PARENT);
 399                 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 400                                          stat, link, &pstat, true);
 401                 inode_unlock(upperdir->d_inode);
 402                 ovl_copy_up_end(dentry);
 403                 goto out_done;
 404         }
 405
 406         err = -EIO;
 407         if (lock_rename(workdir, upperdir) != NULL) {
 408                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 409                 goto out_unlock;
 410         }
 411         if (ovl_dentry_upper(dentry)) {
 412                 /* Raced with another copy-up?  Nothing to do, then... */
 413                 err = 0;
 414                 goto out_unlock;
 415         }
 416
 417         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 418                                  stat, link, &pstat, false);
 419 out_unlock:
 420         unlock_rename(workdir, upperdir);
 421 out_done:
 422         do_delayed_call(&done);
 423
 424         return err;
 425 }
 426
 427 int ovl_copy_up_flags(struct dentry *dentry, int flags)
 428 {
 429         int err = 0;
 430         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 431
 432         while (!err) {
 433                 struct dentry *next;
 434                 struct dentry *parent;
 435                 struct path lowerpath;
 436                 struct kstat stat;
 437                 enum ovl_path_type type = ovl_path_type(dentry);
 438
 439                 if (OVL_TYPE_UPPER(type))
 440                         break;
 441
 442                 next = dget(dentry);
 443                 /* find the topmost dentry not yet copied up */
 444                 for (;;) {
 445                         parent = dget_parent(next);
 446
 447                         type = ovl_path_type(parent);
 448                         if (OVL_TYPE_UPPER(type))
 449                                 break;
 450
 451                         dput(next);
 452                         next = parent;
 453                 }
 454
 455                 ovl_path_lower(next, &lowerpath);
 456                 err = vfs_getattr(&lowerpath, &stat,
 457                                   STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 458                 /* maybe truncate regular file. this has no effect on dirs */
 459                 if (flags & O_TRUNC)
 460                         stat.size = 0;
 461                 if (!err)
 462                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 463
 464                 dput(parent);
 465                 dput(next);
 466         }
 467         revert_creds(old_cred);
 468
 469         return err;
 470 }
 471
 472 int ovl_copy_up(struct dentry *dentry)
 473 {
 474         return ovl_copy_up_flags(dentry, 0);
 475 }