include/linux/kernfs.h

   1 /*
   2  * kernfs.h - pseudo filesystem decoupled from vfs locking
   3  *
   4  * This file is released under the GPLv2.
   5  */
   6
   7 #ifndef __LINUX_KERNFS_H
   8 #define __LINUX_KERNFS_H
   9
  10 #include <linux/kernel.h>
  11 #include <linux/err.h>
  12 #include <linux/list.h>
  13 #include <linux/mutex.h>
  14 #include <linux/idr.h>
  15 #include <linux/lockdep.h>
  16 #include <linux/rbtree.h>
  17 #include <linux/atomic.h>
  18 #include <linux/uidgid.h>
  19 #include <linux/wait.h>
  20
  21 struct file;
  22 struct dentry;
  23 struct iattr;
  24 struct seq_file;
  25 struct vm_area_struct;
  26 struct super_block;
  27 struct file_system_type;
  28
  29 struct kernfs_open_node;
  30 struct kernfs_iattrs;
  31
  32 enum kernfs_node_type {
  33         KERNFS_DIR              = 0x0001,
  34         KERNFS_FILE             = 0x0002,
  35         KERNFS_LINK             = 0x0004,
  36 };
  37
  38 #define KERNFS_TYPE_MASK        0x000f
  39 #define KERNFS_FLAG_MASK        ~KERNFS_TYPE_MASK
  40
  41 enum kernfs_node_flag {
  42         KERNFS_ACTIVATED        = 0x0010,
  43         KERNFS_NS               = 0x0020,
  44         KERNFS_HAS_SEQ_SHOW     = 0x0040,
  45         KERNFS_HAS_MMAP         = 0x0080,
  46         KERNFS_LOCKDEP          = 0x0100,
  47         KERNFS_SUICIDAL         = 0x0400,
  48         KERNFS_SUICIDED         = 0x0800,
  49         KERNFS_EMPTY_DIR        = 0x1000,
  50         KERNFS_HAS_RELEASE      = 0x2000,
  51 };
  52
  53 /* @flags for kernfs_create_root() */
  54 enum kernfs_root_flag {
  55         /*
  56          * kernfs_nodes are created in the deactivated state and invisible.
  57          * They require explicit kernfs_activate() to become visible.  This
  58          * can be used to make related nodes become visible atomically
  59          * after all nodes are created successfully.
  60          */
  61         KERNFS_ROOT_CREATE_DEACTIVATED          = 0x0001,
  62
  63         /*
  64          * For regular flies, if the opener has CAP_DAC_OVERRIDE, open(2)
  65          * succeeds regardless of the RW permissions.  sysfs had an extra
  66          * layer of enforcement where open(2) fails with -EACCES regardless
  67          * of CAP_DAC_OVERRIDE if the permission doesn't have the
  68          * respective read or write access at all (none of S_IRUGO or
  69          * S_IWUGO) or the respective operation isn't implemented.  The
  70          * following flag enables that behavior.
  71          */
  72         KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK       = 0x0002,
  73
  74         /*
  75          * The filesystem supports exportfs operation, so userspace can use
  76          * fhandle to access nodes of the fs.
  77          */
  78         KERNFS_ROOT_SUPPORT_EXPORTOP            = 0x0004,
  79 };
  80
  81 /* type-specific structures for kernfs_node union members */
  82 struct kernfs_elem_dir {
  83         unsigned long           subdirs;
  84         /* children rbtree starts here and goes through kn->rb */
  85         struct rb_root          children;
  86
  87         /*
  88          * The kernfs hierarchy this directory belongs to.  This fits
  89          * better directly in kernfs_node but is here to save space.
  90          */
  91         struct kernfs_root      *root;
  92 };
  93
  94 struct kernfs_elem_symlink {
  95         struct kernfs_node      *target_kn;
  96 };
  97
  98 struct kernfs_elem_attr {
  99         const struct kernfs_ops *ops;
 100         struct kernfs_open_node *open;
 101         loff_t                  size;
 102         struct kernfs_node      *notify_next;   /* for kernfs_notify() */
 103 };
 104
 105 /* represent a kernfs node */
 106 union kernfs_node_id {
 107         struct {
 108                 /*
 109                  * blktrace will export this struct as a simplified 'struct
 110                  * fid' (which is a big data struction), so userspace can use
 111                  * it to find kernfs node. The layout must match the first two
 112                  * fields of 'struct fid' exactly.
 113                  */
 114                 u32             ino;
 115                 u32             generation;
 116         };
 117         u64                     id;
 118 };
 119
 120 /*
 121  * kernfs_node - the building block of kernfs hierarchy.  Each and every
 122  * kernfs node is represented by single kernfs_node.  Most fields are
 123  * private to kernfs and shouldn't be accessed directly by kernfs users.
 124  *
 125  * As long as s_count reference is held, the kernfs_node itself is
 126  * accessible.  Dereferencing elem or any other outer entity requires
 127  * active reference.
 128  */
 129 struct kernfs_node {
 130         atomic_t                count;
 131         atomic_t                active;
 132 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 133         struct lockdep_map      dep_map;
 134 #endif
 135         /*
 136          * Use kernfs_get_parent() and kernfs_name/path() instead of
 137          * accessing the following two fields directly.  If the node is
 138          * never moved to a different parent, it is safe to access the
 139          * parent directly.
 140          */
 141         struct kernfs_node      *parent;
 142         const char              *name;
 143
 144         struct rb_node          rb;
 145
 146         const void              *ns;    /* namespace tag */
 147         unsigned int            hash;   /* ns + name hash */
 148         union {
 149                 struct kernfs_elem_dir          dir;
 150                 struct kernfs_elem_symlink      symlink;
 151                 struct kernfs_elem_attr         attr;
 152         };
 153
 154         void                    *priv;
 155
 156         union kernfs_node_id    id;
 157         unsigned short          flags;
 158         umode_t                 mode;
 159         struct kernfs_iattrs    *iattr;
 160 };
 161
 162 /*
 163  * kernfs_syscall_ops may be specified on kernfs_create_root() to support
 164  * syscalls.  These optional callbacks are invoked on the matching syscalls
 165  * and can perform any kernfs operations which don't necessarily have to be
 166  * the exact operation requested.  An active reference is held for each
 167  * kernfs_node parameter.
 168  */
 169 struct kernfs_syscall_ops {
 170         int (*remount_fs)(struct kernfs_root *root, int *flags, char *data);
 171         int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
 172
 173         int (*mkdir)(struct kernfs_node *parent, const char *name,
 174                      umode_t mode);
 175         int (*rmdir)(struct kernfs_node *kn);
 176         int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
 177                       const char *new_name);
 178         int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
 179                          struct kernfs_root *root);
 180 };
 181
 182 struct kernfs_root {
 183         /* published fields */
 184         struct kernfs_node      *kn;
 185         unsigned int            flags;  /* KERNFS_ROOT_* flags */
 186
 187         /* private fields, do not use outside kernfs proper */
 188         struct idr              ino_idr;
 189         u32                     last_ino;
 190         u32                     next_generation;
 191         struct kernfs_syscall_ops *syscall_ops;
 192
 193         /* list of kernfs_super_info of this root, protected by kernfs_mutex */
 194         struct list_head        supers;
 195
 196         wait_queue_head_t       deactivate_waitq;
 197 };
 198
 199 struct kernfs_open_file {
 200         /* published fields */
 201         struct kernfs_node      *kn;
 202         struct file             *file;
 203         struct seq_file         *seq_file;
 204         void                    *priv;
 205
 206         /* private fields, do not use outside kernfs proper */
 207         struct mutex            mutex;
 208         struct mutex            prealloc_mutex;
 209         int                     event;
 210         struct list_head        list;
 211         char                    *prealloc_buf;
 212
 213         size_t                  atomic_write_len;
 214         bool                    mmapped:1;
 215         bool                    released:1;
 216         const struct vm_operations_struct *vm_ops;
 217 };
 218
 219 struct kernfs_ops {
 220         /*
 221          * Optional open/release methods.  Both are called with
 222          * @of->seq_file populated.
 223          */
 224         int (*open)(struct kernfs_open_file *of);
 225         void (*release)(struct kernfs_open_file *of);
 226
 227         /*
 228          * Read is handled by either seq_file or raw_read().
 229          *
 230          * If seq_show() is present, seq_file path is active.  Other seq
 231          * operations are optional and if not implemented, the behavior is
 232          * equivalent to single_open().  @sf->private points to the
 233          * associated kernfs_open_file.
 234          *
 235          * read() is bounced through kernel buffer and a read larger than
 236          * PAGE_SIZE results in partial operation of PAGE_SIZE.
 237          */
 238         int (*seq_show)(struct seq_file *sf, void *v);
 239
 240         void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
 241         void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
 242         void (*seq_stop)(struct seq_file *sf, void *v);
 243
 244         ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
 245                         loff_t off);
 246
 247         /*
 248          * write() is bounced through kernel buffer.  If atomic_write_len
 249          * is not set, a write larger than PAGE_SIZE results in partial
 250          * operations of PAGE_SIZE chunks.  If atomic_write_len is set,
 251          * writes upto the specified size are executed atomically but
 252          * larger ones are rejected with -E2BIG.
 253          */
 254         size_t atomic_write_len;
 255         /*
 256          * "prealloc" causes a buffer to be allocated at open for
 257          * all read/write requests.  As ->seq_show uses seq_read()
 258          * which does its own allocation, it is incompatible with
 259          * ->prealloc.  Provide ->read and ->write with ->prealloc.
 260          */
 261         bool prealloc;
 262         ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
 263                          loff_t off);
 264
 265         int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
 266
 267 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 268         struct lock_class_key   lockdep_key;
 269 #endif
 270 };
 271
 272 #ifdef CONFIG_KERNFS
 273
 274 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
 275 {
 276         return kn->flags & KERNFS_TYPE_MASK;
 277 }
 278
 279 /**
 280  * kernfs_enable_ns - enable namespace under a directory
 281  * @kn: directory of interest, should be empty
 282  *
 283  * This is to be called right after @kn is created to enable namespace
 284  * under it.  All children of @kn must have non-NULL namespace tags and
 285  * only the ones which match the super_block's tag will be visible.
 286  */
 287 static inline void kernfs_enable_ns(struct kernfs_node *kn)
 288 {
 289         WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
 290         WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
 291         kn->flags |= KERNFS_NS;
 292 }
 293
 294 /**
 295  * kernfs_ns_enabled - test whether namespace is enabled
 296  * @kn: the node to test
 297  *
 298  * Test whether namespace filtering is enabled for the children of @ns.
 299  */
 300 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
 301 {
 302         return kn->flags & KERNFS_NS;
 303 }
 304
 305 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
 306 int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
 307                           char *buf, size_t buflen);
 308 void pr_cont_kernfs_name(struct kernfs_node *kn);
 309 void pr_cont_kernfs_path(struct kernfs_node *kn);
 310 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
 311 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 312                                            const char *name, const void *ns);
 313 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 314                                            const char *path, const void *ns);
 315 void kernfs_get(struct kernfs_node *kn);
 316 void kernfs_put(struct kernfs_node *kn);
 317
 318 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
 319 struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
 320 struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
 321
 322 struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
 323                                   struct super_block *sb);
 324 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 325                                        unsigned int flags, void *priv);
 326 void kernfs_destroy_root(struct kernfs_root *root);
 327
 328 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
 329                                          const char *name, umode_t mode,
 330                                          kuid_t uid, kgid_t gid,
 331                                          void *priv, const void *ns);
 332 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
 333                                             const char *name);
 334 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
 335                                          const char *name, umode_t mode,
 336                                          kuid_t uid, kgid_t gid,
 337                                          loff_t size,
 338                                          const struct kernfs_ops *ops,
 339                                          void *priv, const void *ns,
 340                                          struct lock_class_key *key);
 341 struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
 342                                        const char *name,
 343                                        struct kernfs_node *target);
 344 void kernfs_activate(struct kernfs_node *kn);
 345 void kernfs_remove(struct kernfs_node *kn);
 346 void kernfs_break_active_protection(struct kernfs_node *kn);
 347 void kernfs_unbreak_active_protection(struct kernfs_node *kn);
 348 bool kernfs_remove_self(struct kernfs_node *kn);
 349 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
 350                              const void *ns);
 351 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
 352                      const char *new_name, const void *new_ns);
 353 int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
 354 void kernfs_notify(struct kernfs_node *kn);
 355
 356 const void *kernfs_super_ns(struct super_block *sb);
 357 struct dentry *kernfs_mount_ns(struct file_system_type *fs_type, int flags,
 358                                struct kernfs_root *root, unsigned long magic,
 359                                bool *new_sb_created, const void *ns);
 360 void kernfs_kill_sb(struct super_block *sb);
 361 struct super_block *kernfs_pin_sb(struct kernfs_root *root, const void *ns);
 362
 363 void kernfs_init(void);
 364
 365 struct kernfs_node *kernfs_get_node_by_id(struct kernfs_root *root,
 366         const union kernfs_node_id *id);
 367 #else   /* CONFIG_KERNFS */
 368
 369 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
 370 { return 0; }   /* whatever */
 371
 372 static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
 373
 374 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
 375 { return false; }
 376
 377 static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
 378 { return -ENOSYS; }
 379
 380 static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
 381                                         struct kernfs_node *kn,
 382                                         char *buf, size_t buflen)
 383 { return -ENOSYS; }
 384
 385 static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
 386 static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
 387
 388 static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
 389 { return NULL; }
 390
 391 static inline struct kernfs_node *
 392 kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
 393                        const void *ns)
 394 { return NULL; }
 395 static inline struct kernfs_node *
 396 kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
 397                        const void *ns)
 398 { return NULL; }
 399
 400 static inline void kernfs_get(struct kernfs_node *kn) { }
 401 static inline void kernfs_put(struct kernfs_node *kn) { }
 402
 403 static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 404 { return NULL; }
 405
 406 static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
 407 { return NULL; }
 408
 409 static inline struct inode *
 410 kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
 411 { return NULL; }
 412
 413 static inline struct kernfs_root *
 414 kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
 415                    void *priv)
 416 { return ERR_PTR(-ENOSYS); }
 417
 418 static inline void kernfs_destroy_root(struct kernfs_root *root) { }
 419
 420 static inline struct kernfs_node *
 421 kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
 422                      umode_t mode, kuid_t uid, kgid_t gid,
 423                      void *priv, const void *ns)
 424 { return ERR_PTR(-ENOSYS); }
 425
 426 static inline struct kernfs_node *
 427 __kernfs_create_file(struct kernfs_node *parent, const char *name,
 428                      umode_t mode, kuid_t uid, kgid_t gid,
 429                      loff_t size, const struct kernfs_ops *ops,
 430                      void *priv, const void *ns, struct lock_class_key *key)
 431 { return ERR_PTR(-ENOSYS); }
 432
 433 static inline struct kernfs_node *
 434 kernfs_create_link(struct kernfs_node *parent, const char *name,
 435                    struct kernfs_node *target)
 436 { return ERR_PTR(-ENOSYS); }
 437
 438 static inline void kernfs_activate(struct kernfs_node *kn) { }
 439
 440 static inline void kernfs_remove(struct kernfs_node *kn) { }
 441
 442 static inline bool kernfs_remove_self(struct kernfs_node *kn)
 443 { return false; }
 444
 445 static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
 446                                            const char *name, const void *ns)
 447 { return -ENOSYS; }
 448
 449 static inline int kernfs_rename_ns(struct kernfs_node *kn,
 450                                    struct kernfs_node *new_parent,
 451                                    const char *new_name, const void *new_ns)
 452 { return -ENOSYS; }
 453
 454 static inline int kernfs_setattr(struct kernfs_node *kn,
 455                                  const struct iattr *iattr)
 456 { return -ENOSYS; }
 457
 458 static inline void kernfs_notify(struct kernfs_node *kn) { }
 459
 460 static inline const void *kernfs_super_ns(struct super_block *sb)
 461 { return NULL; }
 462
 463 static inline struct dentry *
 464 kernfs_mount_ns(struct file_system_type *fs_type, int flags,
 465                 struct kernfs_root *root, unsigned long magic,
 466                 bool *new_sb_created, const void *ns)
 467 { return ERR_PTR(-ENOSYS); }
 468
 469 static inline void kernfs_kill_sb(struct super_block *sb) { }
 470
 471 static inline void kernfs_init(void) { }
 472
 473 #endif  /* CONFIG_KERNFS */
 474
 475 /**
 476  * kernfs_path - build full path of a given node
 477  * @kn: kernfs_node of interest
 478  * @buf: buffer to copy @kn's name into
 479  * @buflen: size of @buf
 480  *
 481  * Builds and returns the full path of @kn in @buf of @buflen bytes.  The
 482  * path is built from the end of @buf so the returned pointer usually
 483  * doesn't match @buf.  If @buf isn't long enough, @buf is nul terminated
 484  * and %NULL is returned.
 485  */
 486 static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
 487 {
 488         return kernfs_path_from_node(kn, NULL, buf, buflen);
 489 }
 490
 491 static inline struct kernfs_node *
 492 kernfs_find_and_get(struct kernfs_node *kn, const char *name)
 493 {
 494         return kernfs_find_and_get_ns(kn, name, NULL);
 495 }
 496
 497 static inline struct kernfs_node *
 498 kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
 499 {
 500         return kernfs_walk_and_get_ns(kn, path, NULL);
 501 }
 502
 503 static inline struct kernfs_node *
 504 kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
 505                   void *priv)
 506 {
 507         return kernfs_create_dir_ns(parent, name, mode,
 508                                     GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
 509                                     priv, NULL);
 510 }
 511
 512 static inline struct kernfs_node *
 513 kernfs_create_file_ns(struct kernfs_node *parent, const char *name,
 514                       umode_t mode, kuid_t uid, kgid_t gid,
 515                       loff_t size, const struct kernfs_ops *ops,
 516                       void *priv, const void *ns)
 517 {
 518         struct lock_class_key *key = NULL;
 519
 520 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 521         key = (struct lock_class_key *)&ops->lockdep_key;
 522 #endif
 523         return __kernfs_create_file(parent, name, mode, uid, gid,
 524                                     size, ops, priv, ns, key);
 525 }
 526
 527 static inline struct kernfs_node *
 528 kernfs_create_file(struct kernfs_node *parent, const char *name, umode_t mode,
 529                    loff_t size, const struct kernfs_ops *ops, void *priv)
 530 {
 531         return kernfs_create_file_ns(parent, name, mode,
 532                                      GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
 533                                      size, ops, priv, NULL);
 534 }
 535
 536 static inline int kernfs_remove_by_name(struct kernfs_node *parent,
 537                                         const char *name)
 538 {
 539         return kernfs_remove_by_name_ns(parent, name, NULL);
 540 }
 541
 542 static inline int kernfs_rename(struct kernfs_node *kn,
 543                                 struct kernfs_node *new_parent,
 544                                 const char *new_name)
 545 {
 546         return kernfs_rename_ns(kn, new_parent, new_name, NULL);
 547 }
 548
 549 static inline struct dentry *
 550 kernfs_mount(struct file_system_type *fs_type, int flags,
 551                 struct kernfs_root *root, unsigned long magic,
 552                 bool *new_sb_created)
 553 {
 554         return kernfs_mount_ns(fs_type, flags, root,
 555                                 magic, new_sb_created, NULL);
 556 }
 557
 558 #endif  /* __LINUX_KERNFS_H */