linux/fs/configfs/dir.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * dir.c - Operations for configfs directories.
   4 *
   5 * Based on sysfs:
   6 *      sysfs is Copyright (C) 2001, 2002, 2003 Patrick Mochel
   7 *
   8 * configfs Copyright (C) 2005 Oracle.  All rights reserved.
   9 */
  10
  11#undef DEBUG
  12
  13#include <linux/fs.h>
  14#include <linux/fsnotify.h>
  15#include <linux/mount.h>
  16#include <linux/module.h>
  17#include <linux/slab.h>
  18#include <linux/err.h>
  19
  20#include <linux/configfs.h>
  21#include "configfs_internal.h"
  22
  23/*
  24 * Protects mutations of configfs_dirent linkage together with proper i_mutex
  25 * Also protects mutations of symlinks linkage to target configfs_dirent
  26 * Mutators of configfs_dirent linkage must *both* have the proper inode locked
  27 * and configfs_dirent_lock locked, in that order.
  28 * This allows one to safely traverse configfs_dirent trees and symlinks without
  29 * having to lock inodes.
  30 *
  31 * Protects setting of CONFIGFS_USET_DROPPING: checking the flag
  32 * unlocked is not reliable unless in detach_groups() called from
  33 * rmdir()/unregister() and from configfs_attach_group()
  34 */
  35DEFINE_SPINLOCK(configfs_dirent_lock);
  36
  37static void configfs_d_iput(struct dentry * dentry,
  38                            struct inode * inode)
  39{
  40        struct configfs_dirent *sd = dentry->d_fsdata;
  41
  42        if (sd) {
  43                /* Coordinate with configfs_readdir */
  44                spin_lock(&configfs_dirent_lock);
  45                /*
  46                 * Set sd->s_dentry to null only when this dentry is the one
  47                 * that is going to be killed.  Otherwise configfs_d_iput may
  48                 * run just after configfs_attach_attr and set sd->s_dentry to
  49                 * NULL even it's still in use.
  50                 */
  51                if (sd->s_dentry == dentry)
  52                        sd->s_dentry = NULL;
  53
  54                spin_unlock(&configfs_dirent_lock);
  55                configfs_put(sd);
  56        }
  57        iput(inode);
  58}
  59
  60const struct dentry_operations configfs_dentry_ops = {
  61        .d_iput         = configfs_d_iput,
  62        .d_delete       = always_delete_dentry,
  63};
  64
  65#ifdef CONFIG_LOCKDEP
  66
  67/*
  68 * Helpers to make lockdep happy with our recursive locking of default groups'
  69 * inodes (see configfs_attach_group() and configfs_detach_group()).
  70 * We put default groups i_mutexes in separate classes according to their depth
  71 * from the youngest non-default group ancestor.
  72 *
  73 * For a non-default group A having default groups A/B, A/C, and A/C/D, default
  74 * groups A/B and A/C will have their inode's mutex in class
  75 * default_group_class[0], and default group A/C/D will be in
  76 * default_group_class[1].
  77 *
  78 * The lock classes are declared and assigned in inode.c, according to the
  79 * s_depth value.
  80 * The s_depth value is initialized to -1, adjusted to >= 0 when attaching
  81 * default groups, and reset to -1 when all default groups are attached. During
  82 * attachment, if configfs_create() sees s_depth > 0, the lock class of the new
  83 * inode's mutex is set to default_group_class[s_depth - 1].
  84 */
  85
  86static void configfs_init_dirent_depth(struct configfs_dirent *sd)
  87{
  88        sd->s_depth = -1;
  89}
  90
  91static void configfs_set_dir_dirent_depth(struct configfs_dirent *parent_sd,
  92                                          struct configfs_dirent *sd)
  93{
  94        int parent_depth = parent_sd->s_depth;
  95
  96        if (parent_depth >= 0)
  97                sd->s_depth = parent_depth + 1;
  98}
  99
 100static void
 101configfs_adjust_dir_dirent_depth_before_populate(struct configfs_dirent *sd)
 102{
 103        /*
 104         * item's i_mutex class is already setup, so s_depth is now only
 105         * used to set new sub-directories s_depth, which is always done
 106         * with item's i_mutex locked.
 107         */
 108        /*
 109         *  sd->s_depth == -1 iff we are a non default group.
 110         *  else (we are a default group) sd->s_depth > 0 (see
 111         *  create_dir()).
 112         */
 113        if (sd->s_depth == -1)
 114                /*
 115                 * We are a non default group and we are going to create
 116                 * default groups.
 117                 */
 118                sd->s_depth = 0;
 119}
 120
 121static void
 122configfs_adjust_dir_dirent_depth_after_populate(struct configfs_dirent *sd)
 123{
 124        /* We will not create default groups anymore. */
 125        sd->s_depth = -1;
 126}
 127
 128#else /* CONFIG_LOCKDEP */
 129
 130static void configfs_init_dirent_depth(struct configfs_dirent *sd)
 131{
 132}
 133
 134static void configfs_set_dir_dirent_depth(struct configfs_dirent *parent_sd,
 135                                          struct configfs_dirent *sd)
 136{
 137}
 138
 139static void
 140configfs_adjust_dir_dirent_depth_before_populate(struct configfs_dirent *sd)
 141{
 142}
 143
 144static void
 145configfs_adjust_dir_dirent_depth_after_populate(struct configfs_dirent *sd)
 146{
 147}
 148
 149#endif /* CONFIG_LOCKDEP */
 150
 151static struct configfs_fragment *new_fragment(void)
 152{
 153        struct configfs_fragment *p;
 154
 155        p = kmalloc(sizeof(struct configfs_fragment), GFP_KERNEL);
 156        if (p) {
 157                atomic_set(&p->frag_count, 1);
 158                init_rwsem(&p->frag_sem);
 159                p->frag_dead = false;
 160        }
 161        return p;
 162}
 163
 164void put_fragment(struct configfs_fragment *frag)
 165{
 166        if (frag && atomic_dec_and_test(&frag->frag_count))
 167                kfree(frag);
 168}
 169
 170struct configfs_fragment *get_fragment(struct configfs_fragment *frag)
 171{
 172        if (likely(frag))
 173                atomic_inc(&frag->frag_count);
 174        return frag;
 175}
 176
 177/*
 178 * Allocates a new configfs_dirent and links it to the parent configfs_dirent
 179 */
 180static struct configfs_dirent *configfs_new_dirent(struct configfs_dirent *parent_sd,
 181                                                   void *element, int type,
 182                                                   struct configfs_fragment *frag)
 183{
 184        struct configfs_dirent * sd;
 185
 186        sd = kmem_cache_zalloc(configfs_dir_cachep, GFP_KERNEL);
 187        if (!sd)
 188                return ERR_PTR(-ENOMEM);
 189
 190        atomic_set(&sd->s_count, 1);
 191        INIT_LIST_HEAD(&sd->s_children);
 192        sd->s_element = element;
 193        sd->s_type = type;
 194        configfs_init_dirent_depth(sd);
 195        spin_lock(&configfs_dirent_lock);
 196        if (parent_sd->s_type & CONFIGFS_USET_DROPPING) {
 197                spin_unlock(&configfs_dirent_lock);
 198                kmem_cache_free(configfs_dir_cachep, sd);
 199                return ERR_PTR(-ENOENT);
 200        }
 201        sd->s_frag = get_fragment(frag);
 202        list_add(&sd->s_sibling, &parent_sd->s_children);
 203        spin_unlock(&configfs_dirent_lock);
 204
 205        return sd;
 206}
 207
 208/*
 209 *
 210 * Return -EEXIST if there is already a configfs element with the same
 211 * name for the same parent.
 212 *
 213 * called with parent inode's i_mutex held
 214 */
 215static int configfs_dirent_exists(struct configfs_dirent *parent_sd,
 216                                  const unsigned char *new)
 217{
 218        struct configfs_dirent * sd;
 219
 220        list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
 221                if (sd->s_element) {
 222                        const unsigned char *existing = configfs_get_name(sd);
 223                        if (strcmp(existing, new))
 224                                continue;
 225                        else
 226                                return -EEXIST;
 227                }
 228        }
 229
 230        return 0;
 231}
 232
 233
 234int configfs_make_dirent(struct configfs_dirent * parent_sd,
 235                         struct dentry * dentry, void * element,
 236                         umode_t mode, int type, struct configfs_fragment *frag)
 237{
 238        struct configfs_dirent * sd;
 239
 240        sd = configfs_new_dirent(parent_sd, element, type, frag);
 241        if (IS_ERR(sd))
 242                return PTR_ERR(sd);
 243
 244        sd->s_mode = mode;
 245        sd->s_dentry = dentry;
 246        if (dentry)
 247                dentry->d_fsdata = configfs_get(sd);
 248
 249        return 0;
 250}
 251
 252static void configfs_remove_dirent(struct dentry *dentry)
 253{
 254        struct configfs_dirent *sd = dentry->d_fsdata;
 255
 256        if (!sd)
 257                return;
 258        spin_lock(&configfs_dirent_lock);
 259        list_del_init(&sd->s_sibling);
 260        spin_unlock(&configfs_dirent_lock);
 261        configfs_put(sd);
 262}
 263
 264/**
 265 *      configfs_create_dir - create a directory for an config_item.
 266 *      @item:          config_itemwe're creating directory for.
 267 *      @dentry:        config_item's dentry.
 268 *      @frag:          config_item's fragment.
 269 *
 270 *      Note: user-created entries won't be allowed under this new directory
 271 *      until it is validated by configfs_dir_set_ready()
 272 */
 273
 274static int configfs_create_dir(struct config_item *item, struct dentry *dentry,
 275                                struct configfs_fragment *frag)
 276{
 277        int error;
 278        umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
 279        struct dentry *p = dentry->d_parent;
 280        struct inode *inode;
 281
 282        BUG_ON(!item);
 283
 284        error = configfs_dirent_exists(p->d_fsdata, dentry->d_name.name);
 285        if (unlikely(error))
 286                return error;
 287
 288        error = configfs_make_dirent(p->d_fsdata, dentry, item, mode,
 289                                     CONFIGFS_DIR | CONFIGFS_USET_CREATING,
 290                                     frag);
 291        if (unlikely(error))
 292                return error;
 293
 294        configfs_set_dir_dirent_depth(p->d_fsdata, dentry->d_fsdata);
 295        inode = configfs_create(dentry, mode);
 296        if (IS_ERR(inode))
 297                goto out_remove;
 298
 299        inode->i_op = &configfs_dir_inode_operations;
 300        inode->i_fop = &configfs_dir_operations;
 301        /* directory inodes start off with i_nlink == 2 (for "." entry) */
 302        inc_nlink(inode);
 303        d_instantiate(dentry, inode);
 304        /* already hashed */
 305        dget(dentry);  /* pin directory dentries in core */
 306        inc_nlink(d_inode(p));
 307        item->ci_dentry = dentry;
 308        return 0;
 309
 310out_remove:
 311        configfs_remove_dirent(dentry);
 312        return PTR_ERR(inode);
 313}
 314
 315/*
 316 * Allow userspace to create new entries under a new directory created with
 317 * configfs_create_dir(), and under all of its chidlren directories recursively.
 318 * @sd          configfs_dirent of the new directory to validate
 319 *
 320 * Caller must hold configfs_dirent_lock.
 321 */
 322static void configfs_dir_set_ready(struct configfs_dirent *sd)
 323{
 324        struct configfs_dirent *child_sd;
 325
 326        sd->s_type &= ~CONFIGFS_USET_CREATING;
 327        list_for_each_entry(child_sd, &sd->s_children, s_sibling)
 328                if (child_sd->s_type & CONFIGFS_USET_CREATING)
 329                        configfs_dir_set_ready(child_sd);
 330}
 331
 332/*
 333 * Check that a directory does not belong to a directory hierarchy being
 334 * attached and not validated yet.
 335 * @sd          configfs_dirent of the directory to check
 336 *
 337 * @return      non-zero iff the directory was validated
 338 *
 339 * Note: takes configfs_dirent_lock, so the result may change from false to true
 340 * in two consecutive calls, but never from true to false.
 341 */
 342int configfs_dirent_is_ready(struct configfs_dirent *sd)
 343{
 344        int ret;
 345
 346        spin_lock(&configfs_dirent_lock);
 347        ret = !(sd->s_type & CONFIGFS_USET_CREATING);
 348        spin_unlock(&configfs_dirent_lock);
 349
 350        return ret;
 351}
 352
 353int configfs_create_link(struct configfs_dirent *target, struct dentry *parent,
 354                struct dentry *dentry, char *body)
 355{
 356        int err = 0;
 357        umode_t mode = S_IFLNK | S_IRWXUGO;
 358        struct configfs_dirent *p = parent->d_fsdata;
 359        struct inode *inode;
 360
 361        err = configfs_make_dirent(p, dentry, target, mode, CONFIGFS_ITEM_LINK,
 362                        p->s_frag);
 363        if (err)
 364                return err;
 365
 366        inode = configfs_create(dentry, mode);
 367        if (IS_ERR(inode))
 368                goto out_remove;
 369
 370        inode->i_link = body;
 371        inode->i_op = &configfs_symlink_inode_operations;
 372        d_instantiate(dentry, inode);
 373        dget(dentry);  /* pin link dentries in core */
 374        return 0;
 375
 376out_remove:
 377        configfs_remove_dirent(dentry);
 378        return PTR_ERR(inode);
 379}
 380
 381static void remove_dir(struct dentry * d)
 382{
 383        struct dentry * parent = dget(d->d_parent);
 384
 385        configfs_remove_dirent(d);
 386
 387        if (d_really_is_positive(d))
 388                simple_rmdir(d_inode(parent),d);
 389
 390        pr_debug(" o %pd removing done (%d)\n", d, d_count(d));
 391
 392        dput(parent);
 393}
 394
 395/**
 396 * configfs_remove_dir - remove an config_item's directory.
 397 * @item:       config_item we're removing.
 398 *
 399 * The only thing special about this is that we remove any files in
 400 * the directory before we remove the directory, and we've inlined
 401 * what used to be configfs_rmdir() below, instead of calling separately.
 402 *
 403 * Caller holds the mutex of the item's inode
 404 */
 405
 406static void configfs_remove_dir(struct config_item * item)
 407{
 408        struct dentry * dentry = dget(item->ci_dentry);
 409
 410        if (!dentry)
 411                return;
 412
 413        remove_dir(dentry);
 414        /**
 415         * Drop reference from dget() on entrance.
 416         */
 417        dput(dentry);
 418}
 419
 420
 421/* attaches attribute's configfs_dirent to the dentry corresponding to the
 422 * attribute file
 423 */
 424static int configfs_attach_attr(struct configfs_dirent * sd, struct dentry * dentry)
 425{
 426        struct configfs_attribute * attr = sd->s_element;
 427        struct inode *inode;
 428
 429        spin_lock(&configfs_dirent_lock);
 430        dentry->d_fsdata = configfs_get(sd);
 431        sd->s_dentry = dentry;
 432        spin_unlock(&configfs_dirent_lock);
 433
 434        inode = configfs_create(dentry, (attr->ca_mode & S_IALLUGO) | S_IFREG);
 435        if (IS_ERR(inode)) {
 436                configfs_put(sd);
 437                return PTR_ERR(inode);
 438        }
 439        if (sd->s_type & CONFIGFS_ITEM_BIN_ATTR) {
 440                inode->i_size = 0;
 441                inode->i_fop = &configfs_bin_file_operations;
 442        } else {
 443                inode->i_size = PAGE_SIZE;
 444                inode->i_fop = &configfs_file_operations;
 445        }
 446        d_add(dentry, inode);
 447        return 0;
 448}
 449
 450static struct dentry * configfs_lookup(struct inode *dir,
 451                                       struct dentry *dentry,
 452                                       unsigned int flags)
 453{
 454        struct configfs_dirent * parent_sd = dentry->d_parent->d_fsdata;
 455        struct configfs_dirent * sd;
 456        int found = 0;
 457        int err;
 458
 459        /*
 460         * Fake invisibility if dir belongs to a group/default groups hierarchy
 461         * being attached
 462         *
 463         * This forbids userspace to read/write attributes of items which may
 464         * not complete their initialization, since the dentries of the
 465         * attributes won't be instantiated.
 466         */
 467        err = -ENOENT;
 468        if (!configfs_dirent_is_ready(parent_sd))
 469                goto out;
 470
 471        list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
 472                if (sd->s_type & CONFIGFS_NOT_PINNED) {
 473                        const unsigned char * name = configfs_get_name(sd);
 474
 475                        if (strcmp(name, dentry->d_name.name))
 476                                continue;
 477
 478                        found = 1;
 479                        err = configfs_attach_attr(sd, dentry);
 480                        break;
 481                }
 482        }
 483
 484        if (!found) {
 485                /*
 486                 * If it doesn't exist and it isn't a NOT_PINNED item,
 487                 * it must be negative.
 488                 */
 489                if (dentry->d_name.len > NAME_MAX)
 490                        return ERR_PTR(-ENAMETOOLONG);
 491                d_add(dentry, NULL);
 492                return NULL;
 493        }
 494
 495out:
 496        return ERR_PTR(err);
 497}
 498
 499/*
 500 * Only subdirectories count here.  Files (CONFIGFS_NOT_PINNED) are
 501 * attributes and are removed by rmdir().  We recurse, setting
 502 * CONFIGFS_USET_DROPPING on all children that are candidates for
 503 * default detach.
 504 * If there is an error, the caller will reset the flags via
 505 * configfs_detach_rollback().
 506 */
 507static int configfs_detach_prep(struct dentry *dentry, struct dentry **wait)
 508{
 509        struct configfs_dirent *parent_sd = dentry->d_fsdata;
 510        struct configfs_dirent *sd;
 511        int ret;
 512
 513        /* Mark that we're trying to drop the group */
 514        parent_sd->s_type |= CONFIGFS_USET_DROPPING;
 515
 516        ret = -EBUSY;
 517        if (parent_sd->s_links)
 518                goto out;
 519
 520        ret = 0;
 521        list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
 522                if (!sd->s_element ||
 523                    (sd->s_type & CONFIGFS_NOT_PINNED))
 524                        continue;
 525                if (sd->s_type & CONFIGFS_USET_DEFAULT) {
 526                        /* Abort if racing with mkdir() */
 527                        if (sd->s_type & CONFIGFS_USET_IN_MKDIR) {
 528                                if (wait)
 529                                        *wait= dget(sd->s_dentry);
 530                                return -EAGAIN;
 531                        }
 532
 533                        /*
 534                         * Yup, recursive.  If there's a problem, blame
 535                         * deep nesting of default_groups
 536                         */
 537                        ret = configfs_detach_prep(sd->s_dentry, wait);
 538                        if (!ret)
 539                                continue;
 540                } else
 541                        ret = -ENOTEMPTY;
 542
 543                break;
 544        }
 545
 546out:
 547        return ret;
 548}
 549
 550/*
 551 * Walk the tree, resetting CONFIGFS_USET_DROPPING wherever it was
 552 * set.
 553 */
 554static void configfs_detach_rollback(struct dentry *dentry)
 555{
 556        struct configfs_dirent *parent_sd = dentry->d_fsdata;
 557        struct configfs_dirent *sd;
 558
 559        parent_sd->s_type &= ~CONFIGFS_USET_DROPPING;
 560
 561        list_for_each_entry(sd, &parent_sd->s_children, s_sibling)
 562                if (sd->s_type & CONFIGFS_USET_DEFAULT)
 563                        configfs_detach_rollback(sd->s_dentry);
 564}
 565
 566static void detach_attrs(struct config_item * item)
 567{
 568        struct dentry * dentry = dget(item->ci_dentry);
 569        struct configfs_dirent * parent_sd;
 570        struct configfs_dirent * sd, * tmp;
 571
 572        if (!dentry)
 573                return;
 574
 575        pr_debug("configfs %s: dropping attrs for  dir\n",
 576                 dentry->d_name.name);
 577
 578        parent_sd = dentry->d_fsdata;
 579        list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) {
 580                if (!sd->s_element || !(sd->s_type & CONFIGFS_NOT_PINNED))
 581                        continue;
 582                spin_lock(&configfs_dirent_lock);
 583                list_del_init(&sd->s_sibling);
 584                spin_unlock(&configfs_dirent_lock);
 585                configfs_drop_dentry(sd, dentry);
 586                configfs_put(sd);
 587        }
 588
 589        /**
 590         * Drop reference from dget() on entrance.
 591         */
 592        dput(dentry);
 593}
 594
 595static int populate_attrs(struct config_item *item)
 596{
 597        const struct config_item_type *t = item->ci_type;
 598        struct configfs_attribute *attr;
 599        struct configfs_bin_attribute *bin_attr;
 600        int error = 0;
 601        int i;
 602
 603        if (!t)
 604                return -EINVAL;
 605        if (t->ct_attrs) {
 606                for (i = 0; (attr = t->ct_attrs[i]) != NULL; i++) {
 607                        if ((error = configfs_create_file(item, attr)))
 608                                break;
 609                }
 610        }
 611        if (t->ct_bin_attrs) {
 612                for (i = 0; (bin_attr = t->ct_bin_attrs[i]) != NULL; i++) {
 613                        error = configfs_create_bin_file(item, bin_attr);
 614                        if (error)
 615                                break;
 616                }
 617        }
 618
 619        if (error)
 620                detach_attrs(item);
 621
 622        return error;
 623}
 624
 625static int configfs_attach_group(struct config_item *parent_item,
 626                                 struct config_item *item,
 627                                 struct dentry *dentry,
 628                                 struct configfs_fragment *frag);
 629static void configfs_detach_group(struct config_item *item);
 630
 631static void detach_groups(struct config_group *group)
 632{
 633        struct dentry * dentry = dget(group->cg_item.ci_dentry);
 634        struct dentry *child;
 635        struct configfs_dirent *parent_sd;
 636        struct configfs_dirent *sd, *tmp;
 637
 638        if (!dentry)
 639                return;
 640
 641        parent_sd = dentry->d_fsdata;
 642        list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) {
 643                if (!sd->s_element ||
 644                    !(sd->s_type & CONFIGFS_USET_DEFAULT))
 645                        continue;
 646
 647                child = sd->s_dentry;
 648
 649                inode_lock(d_inode(child));
 650
 651                configfs_detach_group(sd->s_element);
 652                d_inode(child)->i_flags |= S_DEAD;
 653                dont_mount(child);
 654
 655                inode_unlock(d_inode(child));
 656
 657                d_delete(child);
 658                dput(child);
 659        }
 660
 661        /**
 662         * Drop reference from dget() on entrance.
 663         */
 664        dput(dentry);
 665}
 666
 667/*
 668 * This fakes mkdir(2) on a default_groups[] entry.  It
 669 * creates a dentry, attachs it, and then does fixup
 670 * on the sd->s_type.
 671 *
 672 * We could, perhaps, tweak our parent's ->mkdir for a minute and
 673 * try using vfs_mkdir.  Just a thought.
 674 */
 675static int create_default_group(struct config_group *parent_group,
 676                                struct config_group *group,
 677                                struct configfs_fragment *frag)
 678{
 679        int ret;
 680        struct configfs_dirent *sd;
 681        /* We trust the caller holds a reference to parent */
 682        struct dentry *child, *parent = parent_group->cg_item.ci_dentry;
 683
 684        if (!group->cg_item.ci_name)
 685                group->cg_item.ci_name = group->cg_item.ci_namebuf;
 686
 687        ret = -ENOMEM;
 688        child = d_alloc_name(parent, group->cg_item.ci_name);
 689        if (child) {
 690                d_add(child, NULL);
 691
 692                ret = configfs_attach_group(&parent_group->cg_item,
 693                                            &group->cg_item, child, frag);
 694                if (!ret) {
 695                        sd = child->d_fsdata;
 696                        sd->s_type |= CONFIGFS_USET_DEFAULT;
 697                } else {
 698                        BUG_ON(d_inode(child));
 699                        d_drop(child);
 700                        dput(child);
 701                }
 702        }
 703
 704        return ret;
 705}
 706
 707static int populate_groups(struct config_group *group,
 708                           struct configfs_fragment *frag)
 709{
 710        struct config_group *new_group;
 711        int ret = 0;
 712
 713        list_for_each_entry(new_group, &group->default_groups, group_entry) {
 714                ret = create_default_group(group, new_group, frag);
 715                if (ret) {
 716                        detach_groups(group);
 717                        break;
 718                }
 719        }
 720
 721        return ret;
 722}
 723
 724void configfs_remove_default_groups(struct config_group *group)
 725{
 726        struct config_group *g, *n;
 727
 728        list_for_each_entry_safe(g, n, &group->default_groups, group_entry) {
 729                list_del(&g->group_entry);
 730                config_item_put(&g->cg_item);
 731        }
 732}
 733EXPORT_SYMBOL(configfs_remove_default_groups);
 734
 735/*
 736 * All of link_obj/unlink_obj/link_group/unlink_group require that
 737 * subsys->su_mutex is held.
 738 */
 739
 740static void unlink_obj(struct config_item *item)
 741{
 742        struct config_group *group;
 743
 744        group = item->ci_group;
 745        if (group) {
 746                list_del_init(&item->ci_entry);
 747
 748                item->ci_group = NULL;
 749                item->ci_parent = NULL;
 750
 751                /* Drop the reference for ci_entry */
 752                config_item_put(item);
 753
 754                /* Drop the reference for ci_parent */
 755                config_group_put(group);
 756        }
 757}
 758
 759static void link_obj(struct config_item *parent_item, struct config_item *item)
 760{
 761        /*
 762         * Parent seems redundant with group, but it makes certain
 763         * traversals much nicer.
 764         */
 765        item->ci_parent = parent_item;
 766
 767        /*
 768         * We hold a reference on the parent for the child's ci_parent
 769         * link.
 770         */
 771        item->ci_group = config_group_get(to_config_group(parent_item));
 772        list_add_tail(&item->ci_entry, &item->ci_group->cg_children);
 773
 774        /*
 775         * We hold a reference on the child for ci_entry on the parent's
 776         * cg_children
 777         */
 778        config_item_get(item);
 779}
 780
 781static void unlink_group(struct config_group *group)
 782{
 783        struct config_group *new_group;
 784
 785        list_for_each_entry(new_group, &group->default_groups, group_entry)
 786                unlink_group(new_group);
 787
 788        group->cg_subsys = NULL;
 789        unlink_obj(&group->cg_item);
 790}
 791
 792static void link_group(struct config_group *parent_group, struct config_group *group)
 793{
 794        struct config_group *new_group;
 795        struct configfs_subsystem *subsys = NULL; /* gcc is a turd */
 796
 797        link_obj(&parent_group->cg_item, &group->cg_item);
 798
 799        if (parent_group->cg_subsys)
 800                subsys = parent_group->cg_subsys;
 801        else if (configfs_is_root(&parent_group->cg_item))
 802                subsys = to_configfs_subsystem(group);
 803        else
 804                BUG();
 805        group->cg_subsys = subsys;
 806
 807        list_for_each_entry(new_group, &group->default_groups, group_entry)
 808                link_group(group, new_group);
 809}
 810
 811/*
 812 * The goal is that configfs_attach_item() (and
 813 * configfs_attach_group()) can be called from either the VFS or this
 814 * module.  That is, they assume that the items have been created,
 815 * the dentry allocated, and the dcache is all ready to go.
 816 *
 817 * If they fail, they must clean up after themselves as if they
 818 * had never been called.  The caller (VFS or local function) will
 819 * handle cleaning up the dcache bits.
 820 *
 821 * configfs_detach_group() and configfs_detach_item() behave similarly on
 822 * the way out.  They assume that the proper semaphores are held, they
 823 * clean up the configfs items, and they expect their callers will
 824 * handle the dcache bits.
 825 */
 826static int configfs_attach_item(struct config_item *parent_item,
 827                                struct config_item *item,
 828                                struct dentry *dentry,
 829                                struct configfs_fragment *frag)
 830{
 831        int ret;
 832
 833        ret = configfs_create_dir(item, dentry, frag);
 834        if (!ret) {
 835                ret = populate_attrs(item);
 836                if (ret) {
 837                        /*
 838                         * We are going to remove an inode and its dentry but
 839                         * the VFS may already have hit and used them. Thus,
 840                         * we must lock them as rmdir() would.
 841                         */
 842                        inode_lock(d_inode(dentry));
 843                        configfs_remove_dir(item);
 844                        d_inode(dentry)->i_flags |= S_DEAD;
 845                        dont_mount(dentry);
 846                        inode_unlock(d_inode(dentry));
 847                        d_delete(dentry);
 848                }
 849        }
 850
 851        return ret;
 852}
 853
 854/* Caller holds the mutex of the item's inode */
 855static void configfs_detach_item(struct config_item *item)
 856{
 857        detach_attrs(item);
 858        configfs_remove_dir(item);
 859}
 860
 861static int configfs_attach_group(struct config_item *parent_item,
 862                                 struct config_item *item,
 863                                 struct dentry *dentry,
 864                                 struct configfs_fragment *frag)
 865{
 866        int ret;
 867        struct configfs_dirent *sd;
 868
 869        ret = configfs_attach_item(parent_item, item, dentry, frag);
 870        if (!ret) {
 871                sd = dentry->d_fsdata;
 872                sd->s_type |= CONFIGFS_USET_DIR;
 873
 874                /*
 875                 * FYI, we're faking mkdir in populate_groups()
 876                 * We must lock the group's inode to avoid races with the VFS
 877                 * which can already hit the inode and try to add/remove entries
 878                 * under it.
 879                 *
 880                 * We must also lock the inode to remove it safely in case of
 881                 * error, as rmdir() would.
 882                 */
 883                inode_lock_nested(d_inode(dentry), I_MUTEX_CHILD);
 884                configfs_adjust_dir_dirent_depth_before_populate(sd);
 885                ret = populate_groups(to_config_group(item), frag);
 886                if (ret) {
 887                        configfs_detach_item(item);
 888                        d_inode(dentry)->i_flags |= S_DEAD;
 889                        dont_mount(dentry);
 890                }
 891                configfs_adjust_dir_dirent_depth_after_populate(sd);
 892                inode_unlock(d_inode(dentry));
 893                if (ret)
 894                        d_delete(dentry);
 895        }
 896
 897        return ret;
 898}
 899
 900/* Caller holds the mutex of the group's inode */
 901static void configfs_detach_group(struct config_item *item)
 902{
 903        detach_groups(to_config_group(item));
 904        configfs_detach_item(item);
 905}
 906
 907/*
 908 * After the item has been detached from the filesystem view, we are
 909 * ready to tear it out of the hierarchy.  Notify the client before
 910 * we do that so they can perform any cleanup that requires
 911 * navigating the hierarchy.  A client does not need to provide this
 912 * callback.  The subsystem semaphore MUST be held by the caller, and
 913 * references must be valid for both items.  It also assumes the
 914 * caller has validated ci_type.
 915 */
 916static void client_disconnect_notify(struct config_item *parent_item,
 917                                     struct config_item *item)
 918{
 919        const struct config_item_type *type;
 920
 921        type = parent_item->ci_type;
 922        BUG_ON(!type);
 923
 924        if (type->ct_group_ops && type->ct_group_ops->disconnect_notify)
 925                type->ct_group_ops->disconnect_notify(to_config_group(parent_item),
 926                                                      item);
 927}
 928
 929/*
 930 * Drop the initial reference from make_item()/make_group()
 931 * This function assumes that reference is held on item
 932 * and that item holds a valid reference to the parent.  Also, it
 933 * assumes the caller has validated ci_type.
 934 */
 935static void client_drop_item(struct config_item *parent_item,
 936                             struct config_item *item)
 937{
 938        const struct config_item_type *type;
 939
 940        type = parent_item->ci_type;
 941        BUG_ON(!type);
 942
 943        /*
 944         * If ->drop_item() exists, it is responsible for the
 945         * config_item_put().
 946         */
 947        if (type->ct_group_ops && type->ct_group_ops->drop_item)
 948                type->ct_group_ops->drop_item(to_config_group(parent_item),
 949                                              item);
 950        else
 951                config_item_put(item);
 952}
 953
 954#ifdef DEBUG
 955static void configfs_dump_one(struct configfs_dirent *sd, int level)
 956{
 957        pr_info("%*s\"%s\":\n", level, " ", configfs_get_name(sd));
 958
 959#define type_print(_type) if (sd->s_type & _type) pr_info("%*s %s\n", level, " ", #_type);
 960        type_print(CONFIGFS_ROOT);
 961        type_print(CONFIGFS_DIR);
 962        type_print(CONFIGFS_ITEM_ATTR);
 963        type_print(CONFIGFS_ITEM_LINK);
 964        type_print(CONFIGFS_USET_DIR);
 965        type_print(CONFIGFS_USET_DEFAULT);
 966        type_print(CONFIGFS_USET_DROPPING);
 967#undef type_print
 968}
 969
 970static int configfs_dump(struct configfs_dirent *sd, int level)
 971{
 972        struct configfs_dirent *child_sd;
 973        int ret = 0;
 974
 975        configfs_dump_one(sd, level);
 976
 977        if (!(sd->s_type & (CONFIGFS_DIR|CONFIGFS_ROOT)))
 978                return 0;
 979
 980        list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
 981                ret = configfs_dump(child_sd, level + 2);
 982                if (ret)
 983                        break;
 984        }
 985
 986        return ret;
 987}
 988#endif
 989
 990
 991/*
 992 * configfs_depend_item() and configfs_undepend_item()
 993 *
 994 * WARNING: Do not call these from a configfs callback!
 995 *
 996 * This describes these functions and their helpers.
 997 *
 998 * Allow another kernel system to depend on a config_item.  If this
 999 * happens, the item cannot go away until the dependent can live without
1000 * it.  The idea is to give client modules as simple an interface as
1001 * possible.  When a system asks them to depend on an item, they just
1002 * call configfs_depend_item().  If the item is live and the client
1003 * driver is in good shape, we'll happily do the work for them.
1004 *
1005 * Why is the locking complex?  Because configfs uses the VFS to handle
1006 * all locking, but this function is called outside the normal
1007 * VFS->configfs path.  So it must take VFS locks to prevent the
1008 * VFS->configfs stuff (configfs_mkdir(), configfs_rmdir(), etc).  This is
1009 * why you can't call these functions underneath configfs callbacks.
1010 *
1011 * Note, btw, that this can be called at *any* time, even when a configfs
1012 * subsystem isn't registered, or when configfs is loading or unloading.
1013 * Just like configfs_register_subsystem().  So we take the same
1014 * precautions.  We pin the filesystem.  We lock configfs_dirent_lock.
1015 * If we can find the target item in the
1016 * configfs tree, it must be part of the subsystem tree as well, so we
1017 * do not need the subsystem semaphore.  Holding configfs_dirent_lock helps
1018 * locking out mkdir() and rmdir(), who might be racing us.
1019 */
1020
1021/*
1022 * configfs_depend_prep()
1023 *
1024 * Only subdirectories count here.  Files (CONFIGFS_NOT_PINNED) are
1025 * attributes.  This is similar but not the same to configfs_detach_prep().
1026 * Note that configfs_detach_prep() expects the parent to be locked when it
1027 * is called, but we lock the parent *inside* configfs_depend_prep().  We
1028 * do that so we can unlock it if we find nothing.
1029 *
1030 * Here we do a depth-first search of the dentry hierarchy looking for
1031 * our object.
1032 * We deliberately ignore items tagged as dropping since they are virtually
1033 * dead, as well as items in the middle of attachment since they virtually
1034 * do not exist yet. This completes the locking out of racing mkdir() and
1035 * rmdir().
1036 * Note: subdirectories in the middle of attachment start with s_type =
1037 * CONFIGFS_DIR|CONFIGFS_USET_CREATING set by create_dir().  When
1038 * CONFIGFS_USET_CREATING is set, we ignore the item.  The actual set of
1039 * s_type is in configfs_new_dirent(), which has configfs_dirent_lock.
1040 *
1041 * If the target is not found, -ENOENT is bubbled up.
1042 *
1043 * This adds a requirement that all config_items be unique!
1044 *
1045 * This is recursive.  There isn't
1046 * much on the stack, though, so folks that need this function - be careful
1047 * about your stack!  Patches will be accepted to make it iterative.
1048 */
1049static int configfs_depend_prep(struct dentry *origin,
1050                                struct config_item *target)
1051{
1052        struct configfs_dirent *child_sd, *sd;
1053        int ret = 0;
1054
1055        BUG_ON(!origin || !origin->d_fsdata);
1056        sd = origin->d_fsdata;
1057
1058        if (sd->s_element == target)  /* Boo-yah */
1059                goto out;
1060
1061        list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
1062                if ((child_sd->s_type & CONFIGFS_DIR) &&
1063                    !(child_sd->s_type & CONFIGFS_USET_DROPPING) &&
1064                    !(child_sd->s_type & CONFIGFS_USET_CREATING)) {
1065                        ret = configfs_depend_prep(child_sd->s_dentry,
1066                                                   target);
1067                        if (!ret)
1068                                goto out;  /* Child path boo-yah */
1069                }
1070        }
1071
1072        /* We looped all our children and didn't find target */
1073        ret = -ENOENT;
1074
1075out:
1076        return ret;
1077}
1078
1079static int configfs_do_depend_item(struct dentry *subsys_dentry,
1080                                   struct config_item *target)
1081{
1082        struct configfs_dirent *p;
1083        int ret;
1084
1085        spin_lock(&configfs_dirent_lock);
1086        /* Scan the tree, return 0 if found */
1087        ret = configfs_depend_prep(subsys_dentry, target);
1088        if (ret)
1089                goto out_unlock_dirent_lock;
1090
1091        /*
1092         * We are sure that the item is not about to be removed by rmdir(), and
1093         * not in the middle of attachment by mkdir().
1094         */
1095        p = target->ci_dentry->d_fsdata;
1096        p->s_dependent_count += 1;
1097
1098out_unlock_dirent_lock:
1099        spin_unlock(&configfs_dirent_lock);
1100
1101        return ret;
1102}
1103
1104static inline struct configfs_dirent *
1105configfs_find_subsys_dentry(struct configfs_dirent *root_sd,
1106                            struct config_item *subsys_item)
1107{
1108        struct configfs_dirent *p;
1109        struct configfs_dirent *ret = NULL;
1110
1111        list_for_each_entry(p, &root_sd->s_children, s_sibling) {
1112                if (p->s_type & CONFIGFS_DIR &&
1113                    p->s_element == subsys_item) {
1114                        ret = p;
1115                        break;
1116                }
1117        }
1118
1119        return ret;
1120}
1121
1122
1123int configfs_depend_item(struct configfs_subsystem *subsys,
1124                         struct config_item *target)
1125{
1126        int ret;
1127        struct configfs_dirent *subsys_sd;
1128        struct config_item *s_item = &subsys->su_group.cg_item;
1129        struct dentry *root;
1130
1131        /*
1132         * Pin the configfs filesystem.  This means we can safely access
1133         * the root of the configfs filesystem.
1134         */
1135        root = configfs_pin_fs();
1136        if (IS_ERR(root))
1137                return PTR_ERR(root);
1138
1139        /*
1140         * Next, lock the root directory.  We're going to check that the
1141         * subsystem is really registered, and so we need to lock out
1142         * configfs_[un]register_subsystem().
1143         */
1144        inode_lock(d_inode(root));
1145
1146        subsys_sd = configfs_find_subsys_dentry(root->d_fsdata, s_item);
1147        if (!subsys_sd) {
1148                ret = -ENOENT;
1149                goto out_unlock_fs;
1150        }
1151
1152        /* Ok, now we can trust subsys/s_item */
1153        ret = configfs_do_depend_item(subsys_sd->s_dentry, target);
1154
1155out_unlock_fs:
1156        inode_unlock(d_inode(root));
1157
1158        /*
1159         * If we succeeded, the fs is pinned via other methods.  If not,
1160         * we're done with it anyway.  So release_fs() is always right.
1161         */
1162        configfs_release_fs();
1163
1164        return ret;
1165}
1166EXPORT_SYMBOL(configfs_depend_item);
1167
1168/*
1169 * Release the dependent linkage.  This is much simpler than
1170 * configfs_depend_item() because we know that the client driver is
1171 * pinned, thus the subsystem is pinned, and therefore configfs is pinned.
1172 */
1173void configfs_undepend_item(struct config_item *target)
1174{
1175        struct configfs_dirent *sd;
1176
1177        /*
1178         * Since we can trust everything is pinned, we just need
1179         * configfs_dirent_lock.
1180         */
1181        spin_lock(&configfs_dirent_lock);
1182
1183        sd = target->ci_dentry->d_fsdata;
1184        BUG_ON(sd->s_dependent_count < 1);
1185
1186        sd->s_dependent_count -= 1;
1187
1188        /*
1189         * After this unlock, we cannot trust the item to stay alive!
1190         * DO NOT REFERENCE item after this unlock.
1191         */
1192        spin_unlock(&configfs_dirent_lock);
1193}
1194EXPORT_SYMBOL(configfs_undepend_item);
1195
1196/*
1197 * caller_subsys is a caller's subsystem not target's. This is used to
1198 * determine if we should lock root and check subsys or not. When we are
1199 * in the same subsystem as our target there is no need to do locking as
1200 * we know that subsys is valid and is not unregistered during this function
1201 * as we are called from callback of one of his children and VFS holds a lock
1202 * on some inode. Otherwise we have to lock our root to  ensure that target's
1203 * subsystem it is not unregistered during this function.
1204 */
1205int configfs_depend_item_unlocked(struct configfs_subsystem *caller_subsys,
1206                                  struct config_item *target)
1207{
1208        struct configfs_subsystem *target_subsys;
1209        struct config_group *root, *parent;
1210        struct configfs_dirent *subsys_sd;
1211        int ret = -ENOENT;
1212
1213        /* Disallow this function for configfs root */
1214        if (configfs_is_root(target))
1215                return -EINVAL;
1216
1217        parent = target->ci_group;
1218        /*
1219         * This may happen when someone is trying to depend root
1220         * directory of some subsystem
1221         */
1222        if (configfs_is_root(&parent->cg_item)) {
1223                target_subsys = to_configfs_subsystem(to_config_group(target));
1224                root = parent;
1225        } else {
1226                target_subsys = parent->cg_subsys;
1227                /* Find a cofnigfs root as we may need it for locking */
1228                for (root = parent; !configfs_is_root(&root->cg_item);
1229                     root = root->cg_item.ci_group)
1230                        ;
1231        }
1232
1233        if (target_subsys != caller_subsys) {
1234                /*
1235                 * We are in other configfs subsystem, so we have to do
1236                 * additional locking to prevent other subsystem from being
1237                 * unregistered
1238                 */
1239                inode_lock(d_inode(root->cg_item.ci_dentry));
1240
1241                /*
1242                 * As we are trying to depend item from other subsystem
1243                 * we have to check if this subsystem is still registered
1244                 */
1245                subsys_sd = configfs_find_subsys_dentry(
1246                                root->cg_item.ci_dentry->d_fsdata,
1247                                &target_subsys->su_group.cg_item);
1248                if (!subsys_sd)
1249                        goto out_root_unlock;
1250        } else {
1251                subsys_sd = target_subsys->su_group.cg_item.ci_dentry->d_fsdata;
1252        }
1253
1254        /* Now we can execute core of depend item */
1255        ret = configfs_do_depend_item(subsys_sd->s_dentry, target);
1256
1257        if (target_subsys != caller_subsys)
1258out_root_unlock:
1259                /*
1260                 * We were called from subsystem other than our target so we
1261                 * took some locks so now it's time to release them
1262                 */
1263                inode_unlock(d_inode(root->cg_item.ci_dentry));
1264
1265        return ret;
1266}
1267EXPORT_SYMBOL(configfs_depend_item_unlocked);
1268
1269static int configfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1270                          struct dentry *dentry, umode_t mode)
1271{
1272        int ret = 0;
1273        int module_got = 0;
1274        struct config_group *group = NULL;
1275        struct config_item *item = NULL;
1276        struct config_item *parent_item;
1277        struct configfs_subsystem *subsys;
1278        struct configfs_dirent *sd;
1279        const struct config_item_type *type;
1280        struct module *subsys_owner = NULL, *new_item_owner = NULL;
1281        struct configfs_fragment *frag;
1282        char *name;
1283
1284        sd = dentry->d_parent->d_fsdata;
1285
1286        /*
1287         * Fake invisibility if dir belongs to a group/default groups hierarchy
1288         * being attached
1289         */
1290        if (!configfs_dirent_is_ready(sd)) {
1291                ret = -ENOENT;
1292                goto out;
1293        }
1294
1295        if (!(sd->s_type & CONFIGFS_USET_DIR)) {
1296                ret = -EPERM;
1297                goto out;
1298        }
1299
1300        frag = new_fragment();
1301        if (!frag) {
1302                ret = -ENOMEM;
1303                goto out;
1304        }
1305
1306        /* Get a working ref for the duration of this function */
1307        parent_item = configfs_get_config_item(dentry->d_parent);
1308        type = parent_item->ci_type;
1309        subsys = to_config_group(parent_item)->cg_subsys;
1310        BUG_ON(!subsys);
1311
1312        if (!type || !type->ct_group_ops ||
1313            (!type->ct_group_ops->make_group &&
1314             !type->ct_group_ops->make_item)) {
1315                ret = -EPERM;  /* Lack-of-mkdir returns -EPERM */
1316                goto out_put;
1317        }
1318
1319        /*
1320         * The subsystem may belong to a different module than the item
1321         * being created.  We don't want to safely pin the new item but
1322         * fail to pin the subsystem it sits under.
1323         */
1324        if (!subsys->su_group.cg_item.ci_type) {
1325                ret = -EINVAL;
1326                goto out_put;
1327        }
1328        subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1329        if (!try_module_get(subsys_owner)) {
1330                ret = -EINVAL;
1331                goto out_put;
1332        }
1333
1334        name = kmalloc(dentry->d_name.len + 1, GFP_KERNEL);
1335        if (!name) {
1336                ret = -ENOMEM;
1337                goto out_subsys_put;
1338        }
1339
1340        snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name);
1341
1342        mutex_lock(&subsys->su_mutex);
1343        if (type->ct_group_ops->make_group) {
1344                group = type->ct_group_ops->make_group(to_config_group(parent_item), name);
1345                if (!group)
1346                        group = ERR_PTR(-ENOMEM);
1347                if (!IS_ERR(group)) {
1348                        link_group(to_config_group(parent_item), group);
1349                        item = &group->cg_item;
1350                } else
1351                        ret = PTR_ERR(group);
1352        } else {
1353                item = type->ct_group_ops->make_item(to_config_group(parent_item), name);
1354                if (!item)
1355                        item = ERR_PTR(-ENOMEM);
1356                if (!IS_ERR(item))
1357                        link_obj(parent_item, item);
1358                else
1359                        ret = PTR_ERR(item);
1360        }
1361        mutex_unlock(&subsys->su_mutex);
1362
1363        kfree(name);
1364        if (ret) {
1365                /*
1366                 * If ret != 0, then link_obj() was never called.
1367                 * There are no extra references to clean up.
1368                 */
1369                goto out_subsys_put;
1370        }
1371
1372        /*
1373         * link_obj() has been called (via link_group() for groups).
1374         * From here on out, errors must clean that up.
1375         */
1376
1377        type = item->ci_type;
1378        if (!type) {
1379                ret = -EINVAL;
1380                goto out_unlink;
1381        }
1382
1383        new_item_owner = type->ct_owner;
1384        if (!try_module_get(new_item_owner)) {
1385                ret = -EINVAL;
1386                goto out_unlink;
1387        }
1388
1389        /*
1390         * I hate doing it this way, but if there is
1391         * an error,  module_put() probably should
1392         * happen after any cleanup.
1393         */
1394        module_got = 1;
1395
1396        /*
1397         * Make racing rmdir() fail if it did not tag parent with
1398         * CONFIGFS_USET_DROPPING
1399         * Note: if CONFIGFS_USET_DROPPING is already set, attach_group() will
1400         * fail and let rmdir() terminate correctly
1401         */
1402        spin_lock(&configfs_dirent_lock);
1403        /* This will make configfs_detach_prep() fail */
1404        sd->s_type |= CONFIGFS_USET_IN_MKDIR;
1405        spin_unlock(&configfs_dirent_lock);
1406
1407        if (group)
1408                ret = configfs_attach_group(parent_item, item, dentry, frag);
1409        else
1410                ret = configfs_attach_item(parent_item, item, dentry, frag);
1411
1412        spin_lock(&configfs_dirent_lock);
1413        sd->s_type &= ~CONFIGFS_USET_IN_MKDIR;
1414        if (!ret)
1415                configfs_dir_set_ready(dentry->d_fsdata);
1416        spin_unlock(&configfs_dirent_lock);
1417
1418out_unlink:
1419        if (ret) {
1420                /* Tear down everything we built up */
1421                mutex_lock(&subsys->su_mutex);
1422
1423                client_disconnect_notify(parent_item, item);
1424                if (group)
1425                        unlink_group(group);
1426                else
1427                        unlink_obj(item);
1428                client_drop_item(parent_item, item);
1429
1430                mutex_unlock(&subsys->su_mutex);
1431
1432                if (module_got)
1433                        module_put(new_item_owner);
1434        }
1435
1436out_subsys_put:
1437        if (ret)
1438                module_put(subsys_owner);
1439
1440out_put:
1441        /*
1442         * link_obj()/link_group() took a reference from child->parent,
1443         * so the parent is safely pinned.  We can drop our working
1444         * reference.
1445         */
1446        config_item_put(parent_item);
1447        put_fragment(frag);
1448
1449out:
1450        return ret;
1451}
1452
1453static int configfs_rmdir(struct inode *dir, struct dentry *dentry)
1454{
1455        struct config_item *parent_item;
1456        struct config_item *item;
1457        struct configfs_subsystem *subsys;
1458        struct configfs_dirent *sd;
1459        struct configfs_fragment *frag;
1460        struct module *subsys_owner = NULL, *dead_item_owner = NULL;
1461        int ret;
1462
1463        sd = dentry->d_fsdata;
1464        if (sd->s_type & CONFIGFS_USET_DEFAULT)
1465                return -EPERM;
1466
1467        /* Get a working ref until we have the child */
1468        parent_item = configfs_get_config_item(dentry->d_parent);
1469        subsys = to_config_group(parent_item)->cg_subsys;
1470        BUG_ON(!subsys);
1471
1472        if (!parent_item->ci_type) {
1473                config_item_put(parent_item);
1474                return -EINVAL;
1475        }
1476
1477        /* configfs_mkdir() shouldn't have allowed this */
1478        BUG_ON(!subsys->su_group.cg_item.ci_type);
1479        subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1480
1481        /*
1482         * Ensure that no racing symlink() will make detach_prep() fail while
1483         * the new link is temporarily attached
1484         */
1485        do {
1486                struct dentry *wait;
1487
1488                mutex_lock(&configfs_symlink_mutex);
1489                spin_lock(&configfs_dirent_lock);
1490                /*
1491                 * Here's where we check for dependents.  We're protected by
1492                 * configfs_dirent_lock.
1493                 * If no dependent, atomically tag the item as dropping.
1494                 */
1495                ret = sd->s_dependent_count ? -EBUSY : 0;
1496                if (!ret) {
1497                        ret = configfs_detach_prep(dentry, &wait);
1498                        if (ret)
1499                                configfs_detach_rollback(dentry);
1500                }
1501                spin_unlock(&configfs_dirent_lock);
1502                mutex_unlock(&configfs_symlink_mutex);
1503
1504                if (ret) {
1505                        if (ret != -EAGAIN) {
1506                                config_item_put(parent_item);
1507                                return ret;
1508                        }
1509
1510                        /* Wait until the racing operation terminates */
1511                        inode_lock(d_inode(wait));
1512                        inode_unlock(d_inode(wait));
1513                        dput(wait);
1514                }
1515        } while (ret == -EAGAIN);
1516
1517        frag = sd->s_frag;
1518        if (down_write_killable(&frag->frag_sem)) {
1519                spin_lock(&configfs_dirent_lock);
1520                configfs_detach_rollback(dentry);
1521                spin_unlock(&configfs_dirent_lock);
1522                config_item_put(parent_item);
1523                return -EINTR;
1524        }
1525        frag->frag_dead = true;
1526        up_write(&frag->frag_sem);
1527
1528        /* Get a working ref for the duration of this function */
1529        item = configfs_get_config_item(dentry);
1530
1531        /* Drop reference from above, item already holds one. */
1532        config_item_put(parent_item);
1533
1534        if (item->ci_type)
1535                dead_item_owner = item->ci_type->ct_owner;
1536
1537        if (sd->s_type & CONFIGFS_USET_DIR) {
1538                configfs_detach_group(item);
1539
1540                mutex_lock(&subsys->su_mutex);
1541                client_disconnect_notify(parent_item, item);
1542                unlink_group(to_config_group(item));
1543        } else {
1544                configfs_detach_item(item);
1545
1546                mutex_lock(&subsys->su_mutex);
1547                client_disconnect_notify(parent_item, item);
1548                unlink_obj(item);
1549        }
1550
1551        client_drop_item(parent_item, item);
1552        mutex_unlock(&subsys->su_mutex);
1553
1554        /* Drop our reference from above */
1555        config_item_put(item);
1556
1557        module_put(dead_item_owner);
1558        module_put(subsys_owner);
1559
1560        return 0;
1561}
1562
1563const struct inode_operations configfs_dir_inode_operations = {
1564        .mkdir          = configfs_mkdir,
1565        .rmdir          = configfs_rmdir,
1566        .symlink        = configfs_symlink,
1567        .unlink         = configfs_unlink,
1568        .lookup         = configfs_lookup,
1569        .setattr        = configfs_setattr,
1570};
1571
1572const struct inode_operations configfs_root_inode_operations = {
1573        .lookup         = configfs_lookup,
1574        .setattr        = configfs_setattr,
1575};
1576
1577static int configfs_dir_open(struct inode *inode, struct file *file)
1578{
1579        struct dentry * dentry = file->f_path.dentry;
1580        struct configfs_dirent * parent_sd = dentry->d_fsdata;
1581        int err;
1582
1583        inode_lock(d_inode(dentry));
1584        /*
1585         * Fake invisibility if dir belongs to a group/default groups hierarchy
1586         * being attached
1587         */
1588        err = -ENOENT;
1589        if (configfs_dirent_is_ready(parent_sd)) {
1590                file->private_data = configfs_new_dirent(parent_sd, NULL, 0, NULL);
1591                if (IS_ERR(file->private_data))
1592                        err = PTR_ERR(file->private_data);
1593                else
1594                        err = 0;
1595        }
1596        inode_unlock(d_inode(dentry));
1597
1598        return err;
1599}
1600
1601static int configfs_dir_close(struct inode *inode, struct file *file)
1602{
1603        struct dentry * dentry = file->f_path.dentry;
1604        struct configfs_dirent * cursor = file->private_data;
1605
1606        inode_lock(d_inode(dentry));
1607        spin_lock(&configfs_dirent_lock);
1608        list_del_init(&cursor->s_sibling);
1609        spin_unlock(&configfs_dirent_lock);
1610        inode_unlock(d_inode(dentry));
1611
1612        release_configfs_dirent(cursor);
1613
1614        return 0;
1615}
1616
1617/* Relationship between s_mode and the DT_xxx types */
1618static inline unsigned char dt_type(struct configfs_dirent *sd)
1619{
1620        return (sd->s_mode >> 12) & 15;
1621}
1622
1623static int configfs_readdir(struct file *file, struct dir_context *ctx)
1624{
1625        struct dentry *dentry = file->f_path.dentry;
1626        struct super_block *sb = dentry->d_sb;
1627        struct configfs_dirent * parent_sd = dentry->d_fsdata;
1628        struct configfs_dirent *cursor = file->private_data;
1629        struct list_head *p, *q = &cursor->s_sibling;
1630        ino_t ino = 0;
1631
1632        if (!dir_emit_dots(file, ctx))
1633                return 0;
1634        spin_lock(&configfs_dirent_lock);
1635        if (ctx->pos == 2)
1636                list_move(q, &parent_sd->s_children);
1637        for (p = q->next; p != &parent_sd->s_children; p = p->next) {
1638                struct configfs_dirent *next;
1639                const char *name;
1640                int len;
1641                struct inode *inode = NULL;
1642
1643                next = list_entry(p, struct configfs_dirent, s_sibling);
1644                if (!next->s_element)
1645                        continue;
1646
1647                /*
1648                 * We'll have a dentry and an inode for
1649                 * PINNED items and for open attribute
1650                 * files.  We lock here to prevent a race
1651                 * with configfs_d_iput() clearing
1652                 * s_dentry before calling iput().
1653                 *
1654                 * Why do we go to the trouble?  If
1655                 * someone has an attribute file open,
1656                 * the inode number should match until
1657                 * they close it.  Beyond that, we don't
1658                 * care.
1659                 */
1660                dentry = next->s_dentry;
1661                if (dentry)
1662                        inode = d_inode(dentry);
1663                if (inode)
1664                        ino = inode->i_ino;
1665                spin_unlock(&configfs_dirent_lock);
1666                if (!inode)
1667                        ino = iunique(sb, 2);
1668
1669                name = configfs_get_name(next);
1670                len = strlen(name);
1671
1672                if (!dir_emit(ctx, name, len, ino, dt_type(next)))
1673                        return 0;
1674
1675                spin_lock(&configfs_dirent_lock);
1676                list_move(q, p);
1677                p = q;
1678                ctx->pos++;
1679        }
1680        spin_unlock(&configfs_dirent_lock);
1681        return 0;
1682}
1683
1684static loff_t configfs_dir_lseek(struct file *file, loff_t offset, int whence)
1685{
1686        struct dentry * dentry = file->f_path.dentry;
1687
1688        switch (whence) {
1689                case 1:
1690                        offset += file->f_pos;
1691                        fallthrough;
1692                case 0:
1693                        if (offset >= 0)
1694                                break;
1695                        fallthrough;
1696                default:
1697                        return -EINVAL;
1698        }
1699        if (offset != file->f_pos) {
1700                file->f_pos = offset;
1701                if (file->f_pos >= 2) {
1702                        struct configfs_dirent *sd = dentry->d_fsdata;
1703                        struct configfs_dirent *cursor = file->private_data;
1704                        struct list_head *p;
1705                        loff_t n = file->f_pos - 2;
1706
1707                        spin_lock(&configfs_dirent_lock);
1708                        list_del(&cursor->s_sibling);
1709                        p = sd->s_children.next;
1710                        while (n && p != &sd->s_children) {
1711                                struct configfs_dirent *next;
1712                                next = list_entry(p, struct configfs_dirent,
1713                                                   s_sibling);
1714                                if (next->s_element)
1715                                        n--;
1716                                p = p->next;
1717                        }
1718                        list_add_tail(&cursor->s_sibling, p);
1719                        spin_unlock(&configfs_dirent_lock);
1720                }
1721        }
1722        return offset;
1723}
1724
1725const struct file_operations configfs_dir_operations = {
1726        .open           = configfs_dir_open,
1727        .release        = configfs_dir_close,
1728        .llseek         = configfs_dir_lseek,
1729        .read           = generic_read_dir,
1730        .iterate_shared = configfs_readdir,
1731};
1732
1733/**
1734 * configfs_register_group - creates a parent-child relation between two groups
1735 * @parent_group:       parent group
1736 * @group:              child group
1737 *
1738 * link groups, creates dentry for the child and attaches it to the
1739 * parent dentry.
1740 *
1741 * Return: 0 on success, negative errno code on error
1742 */
1743int configfs_register_group(struct config_group *parent_group,
1744                            struct config_group *group)
1745{
1746        struct configfs_subsystem *subsys = parent_group->cg_subsys;
1747        struct dentry *parent;
1748        struct configfs_fragment *frag;
1749        int ret;
1750
1751        frag = new_fragment();
1752        if (!frag)
1753                return -ENOMEM;
1754
1755        mutex_lock(&subsys->su_mutex);
1756        link_group(parent_group, group);
1757        mutex_unlock(&subsys->su_mutex);
1758
1759        parent = parent_group->cg_item.ci_dentry;
1760
1761        inode_lock_nested(d_inode(parent), I_MUTEX_PARENT);
1762        ret = create_default_group(parent_group, group, frag);
1763        if (ret)
1764                goto err_out;
1765
1766        spin_lock(&configfs_dirent_lock);
1767        configfs_dir_set_ready(group->cg_item.ci_dentry->d_fsdata);
1768        spin_unlock(&configfs_dirent_lock);
1769        inode_unlock(d_inode(parent));
1770        put_fragment(frag);
1771        return 0;
1772err_out:
1773        inode_unlock(d_inode(parent));
1774        mutex_lock(&subsys->su_mutex);
1775        unlink_group(group);
1776        mutex_unlock(&subsys->su_mutex);
1777        put_fragment(frag);
1778        return ret;
1779}
1780EXPORT_SYMBOL(configfs_register_group);
1781
1782/**
1783 * configfs_unregister_group() - unregisters a child group from its parent
1784 * @group: parent group to be unregistered
1785 *
1786 * Undoes configfs_register_group()
1787 */
1788void configfs_unregister_group(struct config_group *group)
1789{
1790        struct configfs_subsystem *subsys = group->cg_subsys;
1791        struct dentry *dentry = group->cg_item.ci_dentry;
1792        struct dentry *parent = group->cg_item.ci_parent->ci_dentry;
1793        struct configfs_dirent *sd = dentry->d_fsdata;
1794        struct configfs_fragment *frag = sd->s_frag;
1795
1796        down_write(&frag->frag_sem);
1797        frag->frag_dead = true;
1798        up_write(&frag->frag_sem);
1799
1800        inode_lock_nested(d_inode(parent), I_MUTEX_PARENT);
1801        spin_lock(&configfs_dirent_lock);
1802        configfs_detach_prep(dentry, NULL);
1803        spin_unlock(&configfs_dirent_lock);
1804
1805        configfs_detach_group(&group->cg_item);
1806        d_inode(dentry)->i_flags |= S_DEAD;
1807        dont_mount(dentry);
1808        fsnotify_rmdir(d_inode(parent), dentry);
1809        d_delete(dentry);
1810        inode_unlock(d_inode(parent));
1811
1812        dput(dentry);
1813
1814        mutex_lock(&subsys->su_mutex);
1815        unlink_group(group);
1816        mutex_unlock(&subsys->su_mutex);
1817}
1818EXPORT_SYMBOL(configfs_unregister_group);
1819
1820/**
1821 * configfs_register_default_group() - allocates and registers a child group
1822 * @parent_group:       parent group
1823 * @name:               child group name
1824 * @item_type:          child item type description
1825 *
1826 * boilerplate to allocate and register a child group with its parent. We need
1827 * kzalloc'ed memory because child's default_group is initially empty.
1828 *
1829 * Return: allocated config group or ERR_PTR() on error
1830 */
1831struct config_group *
1832configfs_register_default_group(struct config_group *parent_group,
1833                                const char *name,
1834                                const struct config_item_type *item_type)
1835{
1836        int ret;
1837        struct config_group *group;
1838
1839        group = kzalloc(sizeof(*group), GFP_KERNEL);
1840        if (!group)
1841                return ERR_PTR(-ENOMEM);
1842        config_group_init_type_name(group, name, item_type);
1843
1844        ret = configfs_register_group(parent_group, group);
1845        if (ret) {
1846                kfree(group);
1847                return ERR_PTR(ret);
1848        }
1849        return group;
1850}
1851EXPORT_SYMBOL(configfs_register_default_group);
1852
1853/**
1854 * configfs_unregister_default_group() - unregisters and frees a child group
1855 * @group:      the group to act on
1856 */
1857void configfs_unregister_default_group(struct config_group *group)
1858{
1859        configfs_unregister_group(group);
1860        kfree(group);
1861}
1862EXPORT_SYMBOL(configfs_unregister_default_group);
1863
1864int configfs_register_subsystem(struct configfs_subsystem *subsys)
1865{
1866        int err;
1867        struct config_group *group = &subsys->su_group;
1868        struct dentry *dentry;
1869        struct dentry *root;
1870        struct configfs_dirent *sd;
1871        struct configfs_fragment *frag;
1872
1873        frag = new_fragment();
1874        if (!frag)
1875                return -ENOMEM;
1876
1877        root = configfs_pin_fs();
1878        if (IS_ERR(root)) {
1879                put_fragment(frag);
1880                return PTR_ERR(root);
1881        }
1882
1883        if (!group->cg_item.ci_name)
1884                group->cg_item.ci_name = group->cg_item.ci_namebuf;
1885
1886        sd = root->d_fsdata;
1887        link_group(to_config_group(sd->s_element), group);
1888
1889        inode_lock_nested(d_inode(root), I_MUTEX_PARENT);
1890
1891        err = -ENOMEM;
1892        dentry = d_alloc_name(root, group->cg_item.ci_name);
1893        if (dentry) {
1894                d_add(dentry, NULL);
1895
1896                err = configfs_attach_group(sd->s_element, &group->cg_item,
1897                                            dentry, frag);
1898                if (err) {
1899                        BUG_ON(d_inode(dentry));
1900                        d_drop(dentry);
1901                        dput(dentry);
1902                } else {
1903                        spin_lock(&configfs_dirent_lock);
1904                        configfs_dir_set_ready(dentry->d_fsdata);
1905                        spin_unlock(&configfs_dirent_lock);
1906                }
1907        }
1908
1909        inode_unlock(d_inode(root));
1910
1911        if (err) {
1912                unlink_group(group);
1913                configfs_release_fs();
1914        }
1915        put_fragment(frag);
1916
1917        return err;
1918}
1919
1920void configfs_unregister_subsystem(struct configfs_subsystem *subsys)
1921{
1922        struct config_group *group = &subsys->su_group;
1923        struct dentry *dentry = group->cg_item.ci_dentry;
1924        struct dentry *root = dentry->d_sb->s_root;
1925        struct configfs_dirent *sd = dentry->d_fsdata;
1926        struct configfs_fragment *frag = sd->s_frag;
1927
1928        if (dentry->d_parent != root) {
1929                pr_err("Tried to unregister non-subsystem!\n");
1930                return;
1931        }
1932
1933        down_write(&frag->frag_sem);
1934        frag->frag_dead = true;
1935        up_write(&frag->frag_sem);
1936
1937        inode_lock_nested(d_inode(root),
1938                          I_MUTEX_PARENT);
1939        inode_lock_nested(d_inode(dentry), I_MUTEX_CHILD);
1940        mutex_lock(&configfs_symlink_mutex);
1941        spin_lock(&configfs_dirent_lock);
1942        if (configfs_detach_prep(dentry, NULL)) {
1943                pr_err("Tried to unregister non-empty subsystem!\n");
1944        }
1945        spin_unlock(&configfs_dirent_lock);
1946        mutex_unlock(&configfs_symlink_mutex);
1947        configfs_detach_group(&group->cg_item);
1948        d_inode(dentry)->i_flags |= S_DEAD;
1949        dont_mount(dentry);
1950        fsnotify_rmdir(d_inode(root), dentry);
1951        inode_unlock(d_inode(dentry));
1952
1953        d_delete(dentry);
1954
1955        inode_unlock(d_inode(root));
1956
1957        dput(dentry);
1958
1959        unlink_group(group);
1960        configfs_release_fs();
1961}
1962
1963EXPORT_SYMBOL(configfs_register_subsystem);
1964EXPORT_SYMBOL(configfs_unregister_subsystem);
1965
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.