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         size_t slen;
  61
  62         if (!old->d_inode->i_op->getxattr ||
  63             !new->d_inode->i_op->getxattr)
  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 = vfs_setxattr(new, name, value, size, 0);
 119                 if (error)
 120                         break;
 121         }
 122         kfree(value);
 123 out:
 124         kfree(buf);
 125         return error;
 126 }
 127
 128 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 129 {
 130         struct file *old_file;
 131         struct file *new_file;
 132         loff_t old_pos = 0;
 133         loff_t new_pos = 0;
 134         int error = 0;
 135
 136         if (len == 0)
 137                 return 0;
 138
 139         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 140         if (IS_ERR(old_file))
 141                 return PTR_ERR(old_file);
 142
 143         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 144         if (IS_ERR(new_file)) {
 145                 error = PTR_ERR(new_file);
 146                 goto out_fput;
 147         }
 148
 149         /* FIXME: copy up sparse files efficiently */
 150         while (len) {
 151                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 152                 long bytes;
 153
 154                 if (len < this_len)
 155                         this_len = len;
 156
 157                 if (signal_pending_state(TASK_KILLABLE, current)) {
 158                         error = -EINTR;
 159                         break;
 160                 }
 161
 162                 bytes = do_splice_direct(old_file, &old_pos,
 163                                          new_file, &new_pos,
 164                                          this_len, SPLICE_F_MOVE);
 165                 if (bytes <= 0) {
 166                         error = bytes;
 167                         break;
 168                 }
 169                 WARN_ON(old_pos != new_pos);
 170
 171                 len -= bytes;
 172         }
 173
 174         fput(new_file);
 175 out_fput:
 176         fput(old_file);
 177         return error;
 178 }
 179
 180 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 181 {
 182         struct iattr attr = {
 183                 .ia_valid =
 184                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 185                 .ia_atime = stat->atime,
 186                 .ia_mtime = stat->mtime,
 187         };
 188
 189         return notify_change(upperdentry, &attr, NULL);
 190 }
 191
 192 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 193 {
 194         int err = 0;
 195
 196         if (!S_ISLNK(stat->mode)) {
 197                 struct iattr attr = {
 198                         .ia_valid = ATTR_MODE,
 199                         .ia_mode = stat->mode,
 200                 };
 201                 err = notify_change(upperdentry, &attr, NULL);
 202         }
 203         if (!err) {
 204                 struct iattr attr = {
 205                         .ia_valid = ATTR_UID | ATTR_GID,
 206                         .ia_uid = stat->uid,
 207                         .ia_gid = stat->gid,
 208                 };
 209                 err = notify_change(upperdentry, &attr, NULL);
 210         }
 211         if (!err)
 212                 ovl_set_timestamps(upperdentry, stat);
 213
 214         return err;
 215 }
 216
 217 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 218                               struct dentry *dentry, struct path *lowerpath,
 219                               struct kstat *stat, const char *link)
 220 {
 221         struct inode *wdir = workdir->d_inode;
 222         struct inode *udir = upperdir->d_inode;
 223         struct dentry *newdentry = NULL;
 224         struct dentry *upper = NULL;
 225         umode_t mode = stat->mode;
 226         int err;
 227
 228         newdentry = ovl_lookup_temp(workdir, dentry);
 229         err = PTR_ERR(newdentry);
 230         if (IS_ERR(newdentry))
 231                 goto out;
 232
 233         upper = lookup_one_len(dentry->d_name.name, upperdir,
 234                                dentry->d_name.len);
 235         err = PTR_ERR(upper);
 236         if (IS_ERR(upper))
 237                 goto out1;
 238
 239         /* Can't properly set mode on creation because of the umask */
 240         stat->mode &= S_IFMT;
 241         err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
 242         stat->mode = mode;
 243         if (err)
 244                 goto out2;
 245
 246         if (S_ISREG(stat->mode)) {
 247                 struct path upperpath;
 248
 249                 ovl_path_upper(dentry, &upperpath);
 250                 BUG_ON(upperpath.dentry != NULL);
 251                 upperpath.dentry = newdentry;
 252
 253                 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
 254                 if (err)
 255                         goto out_cleanup;
 256         }
 257
 258         err = ovl_copy_xattr(lowerpath->dentry, newdentry);
 259         if (err)
 260                 goto out_cleanup;
 261
 262         inode_lock(newdentry->d_inode);
 263         err = ovl_set_attr(newdentry, stat);
 264         inode_unlock(newdentry->d_inode);
 265         if (err)
 266                 goto out_cleanup;
 267
 268         err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 269         if (err)
 270                 goto out_cleanup;
 271
 272         ovl_dentry_update(dentry, newdentry);
 273         ovl_inode_update(d_inode(dentry), d_inode(newdentry));
 274         newdentry = NULL;
 275
 276         /*
 277          * Non-directores become opaque when copied up.
 278          */
 279         if (!S_ISDIR(stat->mode))
 280                 ovl_dentry_set_opaque(dentry, true);
 281 out2:
 282         dput(upper);
 283 out1:
 284         dput(newdentry);
 285 out:
 286         return err;
 287
 288 out_cleanup:
 289         ovl_cleanup(wdir, newdentry);
 290         goto out2;
 291 }
 292
 293 /*
 294  * Copy up a single dentry
 295  *
 296  * Directory renames only allowed on "pure upper" (already created on
 297  * upper filesystem, never copied up).  Directories which are on lower or
 298  * are merged may not be renamed.  For these -EXDEV is returned and
 299  * userspace has to deal with it.  This means, when copying up a
 300  * directory we can rely on it and ancestors being stable.
 301  *
 302  * Non-directory renames start with copy up of source if necessary.  The
 303  * actual rename will only proceed once the copy up was successful.  Copy
 304  * up uses upper parent i_mutex for exclusion.  Since rename can change
 305  * d_parent it is possible that the copy up will lock the old parent.  At
 306  * that point the file will have already been copied up anyway.
 307  */
 308 int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 309                     struct path *lowerpath, struct kstat *stat)
 310 {
 311         DEFINE_DELAYED_CALL(done);
 312         struct dentry *workdir = ovl_workdir(dentry);
 313         int err;
 314         struct kstat pstat;
 315         struct path parentpath;
 316         struct dentry *lowerdentry = lowerpath->dentry;
 317         struct dentry *upperdir;
 318         struct dentry *upperdentry;
 319         const char *link = NULL;
 320
 321         if (WARN_ON(!workdir))
 322                 return -EROFS;
 323
 324         ovl_do_check_copy_up(lowerdentry);
 325
 326         ovl_path_upper(parent, &parentpath);
 327         upperdir = parentpath.dentry;
 328
 329         err = vfs_getattr(&parentpath, &pstat);
 330         if (err)
 331                 return err;
 332
 333         if (S_ISLNK(stat->mode)) {
 334                 link = vfs_get_link(lowerdentry, &done);
 335                 if (IS_ERR(link))
 336                         return PTR_ERR(link);
 337         }
 338
 339         err = -EIO;
 340         if (lock_rename(workdir, upperdir) != NULL) {
 341                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 342                 goto out_unlock;
 343         }
 344         upperdentry = ovl_dentry_upper(dentry);
 345         if (upperdentry) {
 346                 /* Raced with another copy-up?  Nothing to do, then... */
 347                 err = 0;
 348                 goto out_unlock;
 349         }
 350
 351         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 352                                  stat, link);
 353         if (!err) {
 354                 /* Restore timestamps on parent (best effort) */
 355                 ovl_set_timestamps(upperdir, &pstat);
 356         }
 357 out_unlock:
 358         unlock_rename(workdir, upperdir);
 359         do_delayed_call(&done);
 360
 361         return err;
 362 }
 363
 364 int ovl_copy_up(struct dentry *dentry)
 365 {
 366         int err = 0;
 367         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 368
 369         while (!err) {
 370                 struct dentry *next;
 371                 struct dentry *parent;
 372                 struct path lowerpath;
 373                 struct kstat stat;
 374                 enum ovl_path_type type = ovl_path_type(dentry);
 375
 376                 if (OVL_TYPE_UPPER(type))
 377                         break;
 378
 379                 next = dget(dentry);
 380                 /* find the topmost dentry not yet copied up */
 381                 for (;;) {
 382                         parent = dget_parent(next);
 383
 384                         type = ovl_path_type(parent);
 385                         if (OVL_TYPE_UPPER(type))
 386                                 break;
 387
 388                         dput(next);
 389                         next = parent;
 390                 }
 391
 392                 ovl_path_lower(next, &lowerpath);
 393                 err = vfs_getattr(&lowerpath, &stat);
 394                 if (!err)
 395                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 396
 397                 dput(parent);
 398                 dput(next);
 399         }
 400         revert_creds(old_cred);
 401
 402         return err;
 403 }