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/fs.h>
  11 #include <linux/slab.h>
  12 #include <linux/file.h>
  13 #include <linux/splice.h>
  14 #include <linux/xattr.h>
  15 #include <linux/security.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/sched.h>
  18 #include <linux/namei.h>
  19 #include "overlayfs.h"
  20
  21 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  22
  23 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  24 {
  25         ssize_t list_size, size;
  26         char *buf, *name, *value;
  27         int error;
  28
  29         if (!old->d_inode->i_op->getxattr ||
  30             !new->d_inode->i_op->getxattr)
  31                 return 0;
  32
  33         list_size = vfs_listxattr(old, NULL, 0);
  34         if (list_size <= 0) {
  35                 if (list_size == -EOPNOTSUPP)
  36                         return 0;
  37                 return list_size;
  38         }
  39
  40         buf = kzalloc(list_size, GFP_KERNEL);
  41         if (!buf)
  42                 return -ENOMEM;
  43
  44         error = -ENOMEM;
  45         value = kmalloc(XATTR_SIZE_MAX, GFP_KERNEL);
  46         if (!value)
  47                 goto out;
  48
  49         list_size = vfs_listxattr(old, buf, list_size);
  50         if (list_size <= 0) {
  51                 error = list_size;
  52                 goto out_free_value;
  53         }
  54
  55         for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
  56                 size = vfs_getxattr(old, name, value, XATTR_SIZE_MAX);
  57                 if (size <= 0) {
  58                         error = size;
  59                         goto out_free_value;
  60                 }
  61                 error = vfs_setxattr(new, name, value, size, 0);
  62                 if (error)
  63                         goto out_free_value;
  64         }
  65
  66 out_free_value:
  67         kfree(value);
  68 out:
  69         kfree(buf);
  70         return error;
  71 }
  72
  73 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
  74 {
  75         struct file *old_file;
  76         struct file *new_file;
  77         loff_t old_pos = 0;
  78         loff_t new_pos = 0;
  79         int error = 0;
  80
  81         if (len == 0)
  82                 return 0;
  83
  84         old_file = ovl_path_open(old, O_RDONLY);
  85         if (IS_ERR(old_file))
  86                 return PTR_ERR(old_file);
  87
  88         new_file = ovl_path_open(new, O_WRONLY);
  89         if (IS_ERR(new_file)) {
  90                 error = PTR_ERR(new_file);
  91                 goto out_fput;
  92         }
  93
  94         /* FIXME: copy up sparse files efficiently */
  95         while (len) {
  96                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
  97                 long bytes;
  98
  99                 if (len < this_len)
 100                         this_len = len;
 101
 102                 if (signal_pending_state(TASK_KILLABLE, current)) {
 103                         error = -EINTR;
 104                         break;
 105                 }
 106
 107                 bytes = do_splice_direct(old_file, &old_pos,
 108                                          new_file, &new_pos,
 109                                          this_len, SPLICE_F_MOVE);
 110                 if (bytes <= 0) {
 111                         error = bytes;
 112                         break;
 113                 }
 114                 WARN_ON(old_pos != new_pos);
 115
 116                 len -= bytes;
 117         }
 118
 119         fput(new_file);
 120 out_fput:
 121         fput(old_file);
 122         return error;
 123 }
 124
 125 static char *ovl_read_symlink(struct dentry *realdentry)
 126 {
 127         int res;
 128         char *buf;
 129         struct inode *inode = realdentry->d_inode;
 130         mm_segment_t old_fs;
 131
 132         res = -EINVAL;
 133         if (!inode->i_op->readlink)
 134                 goto err;
 135
 136         res = -ENOMEM;
 137         buf = (char *) __get_free_page(GFP_KERNEL);
 138         if (!buf)
 139                 goto err;
 140
 141         old_fs = get_fs();
 142         set_fs(get_ds());
 143         /* The cast to a user pointer is valid due to the set_fs() */
 144         res = inode->i_op->readlink(realdentry,
 145                                     (char __user *)buf, PAGE_SIZE - 1);
 146         set_fs(old_fs);
 147         if (res < 0) {
 148                 free_page((unsigned long) buf);
 149                 goto err;
 150         }
 151         buf[res] = '\0';
 152
 153         return buf;
 154
 155 err:
 156         return ERR_PTR(res);
 157 }
 158
 159 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 160 {
 161         struct iattr attr = {
 162                 .ia_valid =
 163                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 164                 .ia_atime = stat->atime,
 165                 .ia_mtime = stat->mtime,
 166         };
 167
 168         return notify_change(upperdentry, &attr, NULL);
 169 }
 170
 171 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 172 {
 173         int err = 0;
 174
 175         if (!S_ISLNK(stat->mode)) {
 176                 struct iattr attr = {
 177                         .ia_valid = ATTR_MODE,
 178                         .ia_mode = stat->mode,
 179                 };
 180                 err = notify_change(upperdentry, &attr, NULL);
 181         }
 182         if (!err) {
 183                 struct iattr attr = {
 184                         .ia_valid = ATTR_UID | ATTR_GID,
 185                         .ia_uid = stat->uid,
 186                         .ia_gid = stat->gid,
 187                 };
 188                 err = notify_change(upperdentry, &attr, NULL);
 189         }
 190         if (!err)
 191                 ovl_set_timestamps(upperdentry, stat);
 192
 193         return err;
 194 }
 195
 196 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 197                               struct dentry *dentry, struct path *lowerpath,
 198                               struct kstat *stat, struct iattr *attr,
 199                               const char *link)
 200 {
 201         struct inode *wdir = workdir->d_inode;
 202         struct inode *udir = upperdir->d_inode;
 203         struct dentry *newdentry = NULL;
 204         struct dentry *upper = NULL;
 205         umode_t mode = stat->mode;
 206         int err;
 207
 208         newdentry = ovl_lookup_temp(workdir, dentry);
 209         err = PTR_ERR(newdentry);
 210         if (IS_ERR(newdentry))
 211                 goto out;
 212
 213         upper = lookup_one_len(dentry->d_name.name, upperdir,
 214                                dentry->d_name.len);
 215         err = PTR_ERR(upper);
 216         if (IS_ERR(upper))
 217                 goto out1;
 218
 219         /* Can't properly set mode on creation because of the umask */
 220         stat->mode &= S_IFMT;
 221         err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
 222         stat->mode = mode;
 223         if (err)
 224                 goto out2;
 225
 226         if (S_ISREG(stat->mode)) {
 227                 struct path upperpath;
 228                 ovl_path_upper(dentry, &upperpath);
 229                 BUG_ON(upperpath.dentry != NULL);
 230                 upperpath.dentry = newdentry;
 231
 232                 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
 233                 if (err)
 234                         goto out_cleanup;
 235         }
 236
 237         err = ovl_copy_xattr(lowerpath->dentry, newdentry);
 238         if (err)
 239                 goto out_cleanup;
 240
 241         mutex_lock(&newdentry->d_inode->i_mutex);
 242         err = ovl_set_attr(newdentry, stat);
 243         if (!err && attr)
 244                 err = notify_change(newdentry, attr, NULL);
 245         mutex_unlock(&newdentry->d_inode->i_mutex);
 246         if (err)
 247                 goto out_cleanup;
 248
 249         err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 250         if (err)
 251                 goto out_cleanup;
 252
 253         ovl_dentry_update(dentry, newdentry);
 254         newdentry = NULL;
 255
 256         /*
 257          * Non-directores become opaque when copied up.
 258          */
 259         if (!S_ISDIR(stat->mode))
 260                 ovl_dentry_set_opaque(dentry, true);
 261 out2:
 262         dput(upper);
 263 out1:
 264         dput(newdentry);
 265 out:
 266         return err;
 267
 268 out_cleanup:
 269         ovl_cleanup(wdir, newdentry);
 270         goto out;
 271 }
 272
 273 /*
 274  * Copy up a single dentry
 275  *
 276  * Directory renames only allowed on "pure upper" (already created on
 277  * upper filesystem, never copied up).  Directories which are on lower or
 278  * are merged may not be renamed.  For these -EXDEV is returned and
 279  * userspace has to deal with it.  This means, when copying up a
 280  * directory we can rely on it and ancestors being stable.
 281  *
 282  * Non-directory renames start with copy up of source if necessary.  The
 283  * actual rename will only proceed once the copy up was successful.  Copy
 284  * up uses upper parent i_mutex for exclusion.  Since rename can change
 285  * d_parent it is possible that the copy up will lock the old parent.  At
 286  * that point the file will have already been copied up anyway.
 287  */
 288 int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 289                     struct path *lowerpath, struct kstat *stat,
 290                     struct iattr *attr)
 291 {
 292         struct dentry *workdir = ovl_workdir(dentry);
 293         int err;
 294         struct kstat pstat;
 295         struct path parentpath;
 296         struct dentry *upperdir;
 297         struct dentry *upperdentry;
 298         const struct cred *old_cred;
 299         struct cred *override_cred;
 300         char *link = NULL;
 301
 302         ovl_path_upper(parent, &parentpath);
 303         upperdir = parentpath.dentry;
 304
 305         err = vfs_getattr(&parentpath, &pstat);
 306         if (err)
 307                 return err;
 308
 309         if (S_ISLNK(stat->mode)) {
 310                 link = ovl_read_symlink(lowerpath->dentry);
 311                 if (IS_ERR(link))
 312                         return PTR_ERR(link);
 313         }
 314
 315         err = -ENOMEM;
 316         override_cred = prepare_creds();
 317         if (!override_cred)
 318                 goto out_free_link;
 319
 320         override_cred->fsuid = stat->uid;
 321         override_cred->fsgid = stat->gid;
 322         /*
 323          * CAP_SYS_ADMIN for copying up extended attributes
 324          * CAP_DAC_OVERRIDE for create
 325          * CAP_FOWNER for chmod, timestamp update
 326          * CAP_FSETID for chmod
 327          * CAP_CHOWN for chown
 328          * CAP_MKNOD for mknod
 329          */
 330         cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
 331         cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
 332         cap_raise(override_cred->cap_effective, CAP_FOWNER);
 333         cap_raise(override_cred->cap_effective, CAP_FSETID);
 334         cap_raise(override_cred->cap_effective, CAP_CHOWN);
 335         cap_raise(override_cred->cap_effective, CAP_MKNOD);
 336         old_cred = override_creds(override_cred);
 337
 338         err = -EIO;
 339         if (lock_rename(workdir, upperdir) != NULL) {
 340                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 341                 goto out_unlock;
 342         }
 343         upperdentry = ovl_dentry_upper(dentry);
 344         if (upperdentry) {
 345                 unlock_rename(workdir, upperdir);
 346                 err = 0;
 347                 /* Raced with another copy-up?  Do the setattr here */
 348                 if (attr) {
 349                         mutex_lock(&upperdentry->d_inode->i_mutex);
 350                         err = notify_change(upperdentry, attr, NULL);
 351                         mutex_unlock(&upperdentry->d_inode->i_mutex);
 352                 }
 353                 goto out_put_cred;
 354         }
 355
 356         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 357                                  stat, attr, link);
 358         if (!err) {
 359                 /* Restore timestamps on parent (best effort) */
 360                 ovl_set_timestamps(upperdir, &pstat);
 361         }
 362 out_unlock:
 363         unlock_rename(workdir, upperdir);
 364 out_put_cred:
 365         revert_creds(old_cred);
 366         put_cred(override_cred);
 367
 368 out_free_link:
 369         if (link)
 370                 free_page((unsigned long) link);
 371
 372         return err;
 373 }
 374
 375 int ovl_copy_up(struct dentry *dentry)
 376 {
 377         int err;
 378
 379         err = 0;
 380         while (!err) {
 381                 struct dentry *next;
 382                 struct dentry *parent;
 383                 struct path lowerpath;
 384                 struct kstat stat;
 385                 enum ovl_path_type type = ovl_path_type(dentry);
 386
 387                 if (OVL_TYPE_UPPER(type))
 388                         break;
 389
 390                 next = dget(dentry);
 391                 /* find the topmost dentry not yet copied up */
 392                 for (;;) {
 393                         parent = dget_parent(next);
 394
 395                         type = ovl_path_type(parent);
 396                         if (OVL_TYPE_UPPER(type))
 397                                 break;
 398
 399                         dput(next);
 400                         next = parent;
 401                 }
 402
 403                 ovl_path_lower(next, &lowerpath);
 404                 err = vfs_getattr(&lowerpath, &stat);
 405                 if (!err)
 406                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat, NULL);
 407
 408                 dput(parent);
 409                 dput(next);
 410         }
 411
 412         return err;
 413 }