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_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         fput(new_file);
 182 out_fput:
 183         fput(old_file);
 184         return error;
 185 }
 186
 187 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 188 {
 189         struct iattr attr = {
 190                 .ia_valid =
 191                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 192                 .ia_atime = stat->atime,
 193                 .ia_mtime = stat->mtime,
 194         };
 195
 196         return notify_change(upperdentry, &attr, NULL);
 197 }
 198
 199 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 200 {
 201         int err = 0;
 202
 203         if (!S_ISLNK(stat->mode)) {
 204                 struct iattr attr = {
 205                         .ia_valid = ATTR_MODE,
 206                         .ia_mode = stat->mode,
 207                 };
 208                 err = notify_change(upperdentry, &attr, NULL);
 209         }
 210         if (!err) {
 211                 struct iattr attr = {
 212                         .ia_valid = ATTR_UID | ATTR_GID,
 213                         .ia_uid = stat->uid,
 214                         .ia_gid = stat->gid,
 215                 };
 216                 err = notify_change(upperdentry, &attr, NULL);
 217         }
 218         if (!err)
 219                 ovl_set_timestamps(upperdentry, stat);
 220
 221         return err;
 222 }
 223
 224 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 225                               struct dentry *dentry, struct path *lowerpath,
 226                               struct kstat *stat, const char *link)
 227 {
 228         struct inode *wdir = workdir->d_inode;
 229         struct inode *udir = upperdir->d_inode;
 230         struct dentry *newdentry = NULL;
 231         struct dentry *upper = NULL;
 232         umode_t mode = stat->mode;
 233         int err;
 234         const struct cred *old_creds = NULL;
 235         struct cred *new_creds = NULL;
 236
 237         newdentry = ovl_lookup_temp(workdir, dentry);
 238         err = PTR_ERR(newdentry);
 239         if (IS_ERR(newdentry))
 240                 goto out;
 241
 242         upper = lookup_one_len(dentry->d_name.name, upperdir,
 243                                dentry->d_name.len);
 244         err = PTR_ERR(upper);
 245         if (IS_ERR(upper))
 246                 goto out1;
 247
 248         err = security_inode_copy_up(dentry, &new_creds);
 249         if (err < 0)
 250                 goto out2;
 251
 252         if (new_creds)
 253                 old_creds = override_creds(new_creds);
 254
 255         /* Can't properly set mode on creation because of the umask */
 256         stat->mode &= S_IFMT;
 257         err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
 258         stat->mode = mode;
 259
 260         if (new_creds) {
 261                 revert_creds(old_creds);
 262                 put_cred(new_creds);
 263         }
 264
 265         if (err)
 266                 goto out2;
 267
 268         if (S_ISREG(stat->mode)) {
 269                 struct path upperpath;
 270
 271                 ovl_path_upper(dentry, &upperpath);
 272                 BUG_ON(upperpath.dentry != NULL);
 273                 upperpath.dentry = newdentry;
 274
 275                 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
 276                 if (err)
 277                         goto out_cleanup;
 278         }
 279
 280         err = ovl_copy_xattr(lowerpath->dentry, newdentry);
 281         if (err)
 282                 goto out_cleanup;
 283
 284         inode_lock(newdentry->d_inode);
 285         err = ovl_set_attr(newdentry, stat);
 286         inode_unlock(newdentry->d_inode);
 287         if (err)
 288                 goto out_cleanup;
 289
 290         err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 291         if (err)
 292                 goto out_cleanup;
 293
 294         ovl_dentry_update(dentry, newdentry);
 295         ovl_inode_update(d_inode(dentry), d_inode(newdentry));
 296         newdentry = NULL;
 297
 298         /*
 299          * Non-directores become opaque when copied up.
 300          */
 301         if (!S_ISDIR(stat->mode))
 302                 ovl_dentry_set_opaque(dentry, true);
 303 out2:
 304         dput(upper);
 305 out1:
 306         dput(newdentry);
 307 out:
 308         return err;
 309
 310 out_cleanup:
 311         ovl_cleanup(wdir, newdentry);
 312         goto out2;
 313 }
 314
 315 /*
 316  * Copy up a single dentry
 317  *
 318  * Directory renames only allowed on "pure upper" (already created on
 319  * upper filesystem, never copied up).  Directories which are on lower or
 320  * are merged may not be renamed.  For these -EXDEV is returned and
 321  * userspace has to deal with it.  This means, when copying up a
 322  * directory we can rely on it and ancestors being stable.
 323  *
 324  * Non-directory renames start with copy up of source if necessary.  The
 325  * actual rename will only proceed once the copy up was successful.  Copy
 326  * up uses upper parent i_mutex for exclusion.  Since rename can change
 327  * d_parent it is possible that the copy up will lock the old parent.  At
 328  * that point the file will have already been copied up anyway.
 329  */
 330 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         struct dentry *upperdentry;
 341         const char *link = NULL;
 342
 343         if (WARN_ON(!workdir))
 344                 return -EROFS;
 345
 346         ovl_do_check_copy_up(lowerdentry);
 347
 348         ovl_path_upper(parent, &parentpath);
 349         upperdir = parentpath.dentry;
 350
 351         err = vfs_getattr(&parentpath, &pstat);
 352         if (err)
 353                 return err;
 354
 355         if (S_ISLNK(stat->mode)) {
 356                 link = vfs_get_link(lowerdentry, &done);
 357                 if (IS_ERR(link))
 358                         return PTR_ERR(link);
 359         }
 360
 361         err = -EIO;
 362         if (lock_rename(workdir, upperdir) != NULL) {
 363                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 364                 goto out_unlock;
 365         }
 366         upperdentry = ovl_dentry_upper(dentry);
 367         if (upperdentry) {
 368                 /* Raced with another copy-up?  Nothing to do, then... */
 369                 err = 0;
 370                 goto out_unlock;
 371         }
 372
 373         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 374                                  stat, link);
 375         if (!err) {
 376                 /* Restore timestamps on parent (best effort) */
 377                 ovl_set_timestamps(upperdir, &pstat);
 378         }
 379 out_unlock:
 380         unlock_rename(workdir, upperdir);
 381         do_delayed_call(&done);
 382
 383         return err;
 384 }
 385
 386 int ovl_copy_up(struct dentry *dentry)
 387 {
 388         int err = 0;
 389         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 390
 391         while (!err) {
 392                 struct dentry *next;
 393                 struct dentry *parent;
 394                 struct path lowerpath;
 395                 struct kstat stat;
 396                 enum ovl_path_type type = ovl_path_type(dentry);
 397
 398                 if (OVL_TYPE_UPPER(type))
 399                         break;
 400
 401                 next = dget(dentry);
 402                 /* find the topmost dentry not yet copied up */
 403                 for (;;) {
 404                         parent = dget_parent(next);
 405
 406                         type = ovl_path_type(parent);
 407                         if (OVL_TYPE_UPPER(type))
 408                                 break;
 409
 410                         dput(next);
 411                         next = parent;
 412                 }
 413
 414                 ovl_path_lower(next, &lowerpath);
 415                 err = vfs_getattr(&lowerpath, &stat);
 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 }