include/linux/kernfs.h

   1 /* SPDX-License-Identifier: GPL-2.0-only */
   2 /*
   3  * kernfs.h - pseudo filesystem decoupled from vfs locking
   4  */
   5
   6 #ifndef __LINUX_KERNFS_H
   7 #define __LINUX_KERNFS_H
   8
   9 #include <linux/kernel.h>
  10 #include <linux/err.h>
  11 #include <linux/list.h>
  12 #include <linux/mutex.h>
  13 #include <linux/idr.h>
  14 #include <linux/lockdep.h>
  15 #include <linux/rbtree.h>
  16 #include <linux/atomic.h>
  17 #include <linux/uidgid.h>
  18 #include <linux/wait.h>
  19
  20 struct file;
  21 struct dentry;
  22 struct iattr;
  23 struct seq_file;
  24 struct vm_area_struct;
  25 struct super_block;
  26 struct file_system_type;
  27 struct poll_table_struct;
  28 struct fs_context;
  29
  30 struct kernfs_fs_context;
  31 struct kernfs_open_node;
  32 struct kernfs_iattrs;
  33
  34 enum kernfs_node_type {
  35         KERNFS_DIR              = 0x0001,
  36         KERNFS_FILE             = 0x0002,
  37         KERNFS_LINK             = 0x0004,
  38 };
  39
  40 #define KERNFS_TYPE_MASK        0x000f
  41 #define KERNFS_FLAG_MASK        ~KERNFS_TYPE_MASK
  42
  43 enum kernfs_node_flag {
  44         KERNFS_ACTIVATED        = 0x0010,
  45         KERNFS_NS               = 0x0020,
  46         KERNFS_HAS_SEQ_SHOW     = 0x0040,
  47         KERNFS_HAS_MMAP         = 0x0080,
  48         KERNFS_LOCKDEP          = 0x0100,
  49         KERNFS_SUICIDAL         = 0x0400,
  50         KERNFS_SUICIDED         = 0x0800,
  51         KERNFS_EMPTY_DIR        = 0x1000,
  52         KERNFS_HAS_RELEASE      = 0x2000,
  53 };
  54
  55 /* @flags for kernfs_create_root() */
  56 enum kernfs_root_flag {
  57         /*
  58          * kernfs_nodes are created in the deactivated state and invisible.
  59          * They require explicit kernfs_activate() to become visible.  This
  60          * can be used to make related nodes become visible atomically
  61          * after all nodes are created successfully.
  62          */
  63         KERNFS_ROOT_CREATE_DEACTIVATED          = 0x0001,
  64
  65         /*
  66          * For regular files, if the opener has CAP_DAC_OVERRIDE, open(2)
  67          * succeeds regardless of the RW permissions.  sysfs had an extra
  68          * layer of enforcement where open(2) fails with -EACCES regardless
  69          * of CAP_DAC_OVERRIDE if the permission doesn't have the
  70          * respective read or write access at all (none of S_IRUGO or
  71          * S_IWUGO) or the respective operation isn't implemented.  The
  72          * following flag enables that behavior.
  73          */
  74         KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK       = 0x0002,
  75
  76         /*
  77          * The filesystem supports exportfs operation, so userspace can use
  78          * fhandle to access nodes of the fs.
  79          */
  80         KERNFS_ROOT_SUPPORT_EXPORTOP            = 0x0004,
  81 };
  82
  83 /* type-specific structures for kernfs_node union members */
  84 struct kernfs_elem_dir {
  85         unsigned long           subdirs;
  86         /* children rbtree starts here and goes through kn->rb */
  87         struct rb_root          children;
  88
  89         /*
  90          * The kernfs hierarchy this directory belongs to.  This fits
  91          * better directly in kernfs_node but is here to save space.
  92          */
  93         struct kernfs_root      *root;
  94 };
  95
  96 struct kernfs_elem_symlink {
  97         struct kernfs_node      *target_kn;
  98 };
  99
 100 struct kernfs_elem_attr {
 101         const struct kernfs_ops *ops;
 102         struct kernfs_open_node *open;
 103         loff_t                  size;
 104         struct kernfs_node      *notify_next;   /* for kernfs_notify() */
 105 };
 106
 107 /* represent a kernfs node */
 108 union kernfs_node_id {
 109         struct {
 110                 /*
 111                  * blktrace will export this struct as a simplified 'struct
 112                  * fid' (which is a big data struction), so userspace can use
 113                  * it to find kernfs node. The layout must match the first two
 114                  * fields of 'struct fid' exactly.
 115                  */
 116                 u32             ino;
 117                 u32             generation;
 118         };
 119         u64                     id;
 120 };
 121
 122 /*
 123  * kernfs_node - the building block of kernfs hierarchy.  Each and every
 124  * kernfs node is represented by single kernfs_node.  Most fields are
 125  * private to kernfs and shouldn't be accessed directly by kernfs users.
 126  *
 127  * As long as s_count reference is held, the kernfs_node itself is
 128  * accessible.  Dereferencing elem or any other outer entity requires
 129  * active reference.
 130  */
 131 struct kernfs_node {
 132         atomic_t                count;
 133         atomic_t                active;
 134 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 135         struct lockdep_map      dep_map;
 136 #endif
 137         /*
 138          * Use kernfs_get_parent() and kernfs_name/path() instead of
 139          * accessing the following two fields directly.  If the node is
 140          * never moved to a different parent, it is safe to access the
 141          * parent directly.
 142          */
 143         struct kernfs_node      *parent;
 144         const char              *name;
 145
 146         struct rb_node          rb;
 147
 148         const void              *ns;    /* namespace tag */
 149         unsigned int            hash;   /* ns + name hash */
 150         union {
 151                 struct kernfs_elem_dir          dir;
 152                 struct kernfs_elem_symlink      symlink;
 153                 struct kernfs_elem_attr         attr;
 154         };
 155
 156         void                    *priv;
 157
 158         union kernfs_node_id    id;
 159         unsigned short          flags;
 160         umode_t                 mode;
 161         struct kernfs_iattrs    *iattr;
 162 };
 163
 164 /*
 165  * kernfs_syscall_ops may be specified on kernfs_create_root() to support
 166  * syscalls.  These optional callbacks are invoked on the matching syscalls
 167  * and can perform any kernfs operations which don't necessarily have to be
 168  * the exact operation requested.  An active reference is held for each
 169  * kernfs_node parameter.
 170  */
 171 struct kernfs_syscall_ops {
 172         int (*show_options)(struct seq_file *sf, struct kernfs_root *root);
 173
 174         int (*mkdir)(struct kernfs_node *parent, const char *name,
 175                      umode_t mode);
 176         int (*rmdir)(struct kernfs_node *kn);
 177         int (*rename)(struct kernfs_node *kn, struct kernfs_node *new_parent,
 178                       const char *new_name);
 179         int (*show_path)(struct seq_file *sf, struct kernfs_node *kn,
 180                          struct kernfs_root *root);
 181 };
 182
 183 struct kernfs_root {
 184         /* published fields */
 185         struct kernfs_node      *kn;
 186         unsigned int            flags;  /* KERNFS_ROOT_* flags */
 187
 188         /* private fields, do not use outside kernfs proper */
 189         struct idr              ino_idr;
 190         u32                     last_ino;
 191         u32                     next_generation;
 192         struct kernfs_syscall_ops *syscall_ops;
 193
 194         /* list of kernfs_super_info of this root, protected by kernfs_mutex */
 195         struct list_head        supers;
 196
 197         wait_queue_head_t       deactivate_waitq;
 198 };
 199
 200 struct kernfs_open_file {
 201         /* published fields */
 202         struct kernfs_node      *kn;
 203         struct file             *file;
 204         struct seq_file         *seq_file;
 205         void                    *priv;
 206
 207         /* private fields, do not use outside kernfs proper */
 208         struct mutex            mutex;
 209         struct mutex            prealloc_mutex;
 210         int                     event;
 211         struct list_head        list;
 212         char                    *prealloc_buf;
 213
 214         size_t                  atomic_write_len;
 215         bool                    mmapped:1;
 216         bool                    released:1;
 217         const struct vm_operations_struct *vm_ops;
 218 };
 219
 220 struct kernfs_ops {
 221         /*
 222          * Optional open/release methods.  Both are called with
 223          * @of->seq_file populated.
 224          */
 225         int (*open)(struct kernfs_open_file *of);
 226         void (*release)(struct kernfs_open_file *of);
 227
 228         /*
 229          * Read is handled by either seq_file or raw_read().
 230          *
 231          * If seq_show() is present, seq_file path is active.  Other seq
 232          * operations are optional and if not implemented, the behavior is
 233          * equivalent to single_open().  @sf->private points to the
 234          * associated kernfs_open_file.
 235          *
 236          * read() is bounced through kernel buffer and a read larger than
 237          * PAGE_SIZE results in partial operation of PAGE_SIZE.
 238          */
 239         int (*seq_show)(struct seq_file *sf, void *v);
 240
 241         void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
 242         void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
 243         void (*seq_stop)(struct seq_file *sf, void *v);
 244
 245         ssize_t (*read)(struct kernfs_open_file *of, char *buf, size_t bytes,
 246                         loff_t off);
 247
 248         /*
 249          * write() is bounced through kernel buffer.  If atomic_write_len
 250          * is not set, a write larger than PAGE_SIZE results in partial
 251          * operations of PAGE_SIZE chunks.  If atomic_write_len is set,
 252          * writes upto the specified size are executed atomically but
 253          * larger ones are rejected with -E2BIG.
 254          */
 255         size_t atomic_write_len;
 256         /*
 257          * "prealloc" causes a buffer to be allocated at open for
 258          * all read/write requests.  As ->seq_show uses seq_read()
 259          * which does its own allocation, it is incompatible with
 260          * ->prealloc.  Provide ->read and ->write with ->prealloc.
 261          */
 262         bool prealloc;
 263         ssize_t (*write)(struct kernfs_open_file *of, char *buf, size_t bytes,
 264                          loff_t off);
 265
 266         __poll_t (*poll)(struct kernfs_open_file *of,
 267                          struct poll_table_struct *pt);
 268
 269         int (*mmap)(struct kernfs_open_file *of, struct vm_area_struct *vma);
 270
 271 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 272         struct lock_class_key   lockdep_key;
 273 #endif
 274 };
 275
 276 /*
 277  * The kernfs superblock creation/mount parameter context.
 278  */
 279 struct kernfs_fs_context {
 280         struct kernfs_root      *root;          /* Root of the hierarchy being mounted */
 281         void                    *ns_tag;        /* Namespace tag of the mount (or NULL) */
 282         unsigned long           magic;          /* File system specific magic number */
 283
 284         /* The following are set/used by kernfs_mount() */
 285         bool                    new_sb_created; /* Set to T if we allocated a new sb */
 286 };
 287
 288 #ifdef CONFIG_KERNFS
 289
 290 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
 291 {
 292         return kn->flags & KERNFS_TYPE_MASK;
 293 }
 294
 295 /**
 296  * kernfs_enable_ns - enable namespace under a directory
 297  * @kn: directory of interest, should be empty
 298  *
 299  * This is to be called right after @kn is created to enable namespace
 300  * under it.  All children of @kn must have non-NULL namespace tags and
 301  * only the ones which match the super_block's tag will be visible.
 302  */
 303 static inline void kernfs_enable_ns(struct kernfs_node *kn)
 304 {
 305         WARN_ON_ONCE(kernfs_type(kn) != KERNFS_DIR);
 306         WARN_ON_ONCE(!RB_EMPTY_ROOT(&kn->dir.children));
 307         kn->flags |= KERNFS_NS;
 308 }
 309
 310 /**
 311  * kernfs_ns_enabled - test whether namespace is enabled
 312  * @kn: the node to test
 313  *
 314  * Test whether namespace filtering is enabled for the children of @ns.
 315  */
 316 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
 317 {
 318         return kn->flags & KERNFS_NS;
 319 }
 320
 321 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen);
 322 int kernfs_path_from_node(struct kernfs_node *root_kn, struct kernfs_node *kn,
 323                           char *buf, size_t buflen);
 324 void pr_cont_kernfs_name(struct kernfs_node *kn);
 325 void pr_cont_kernfs_path(struct kernfs_node *kn);
 326 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn);
 327 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
 328                                            const char *name, const void *ns);
 329 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
 330                                            const char *path, const void *ns);
 331 void kernfs_get(struct kernfs_node *kn);
 332 void kernfs_put(struct kernfs_node *kn);
 333
 334 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry);
 335 struct kernfs_root *kernfs_root_from_sb(struct super_block *sb);
 336 struct inode *kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn);
 337
 338 struct dentry *kernfs_node_dentry(struct kernfs_node *kn,
 339                                   struct super_block *sb);
 340 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
 341                                        unsigned int flags, void *priv);
 342 void kernfs_destroy_root(struct kernfs_root *root);
 343
 344 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
 345                                          const char *name, umode_t mode,
 346                                          kuid_t uid, kgid_t gid,
 347                                          void *priv, const void *ns);
 348 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
 349                                             const char *name);
 350 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
 351                                          const char *name, umode_t mode,
 352                                          kuid_t uid, kgid_t gid,
 353                                          loff_t size,
 354                                          const struct kernfs_ops *ops,
 355                                          void *priv, const void *ns,
 356                                          struct lock_class_key *key);
 357 struct kernfs_node *kernfs_create_link(struct kernfs_node *parent,
 358                                        const char *name,
 359                                        struct kernfs_node *target);
 360 void kernfs_activate(struct kernfs_node *kn);
 361 void kernfs_remove(struct kernfs_node *kn);
 362 void kernfs_break_active_protection(struct kernfs_node *kn);
 363 void kernfs_unbreak_active_protection(struct kernfs_node *kn);
 364 bool kernfs_remove_self(struct kernfs_node *kn);
 365 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
 366                              const void *ns);
 367 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
 368                      const char *new_name, const void *new_ns);
 369 int kernfs_setattr(struct kernfs_node *kn, const struct iattr *iattr);
 370 __poll_t kernfs_generic_poll(struct kernfs_open_file *of,
 371                              struct poll_table_struct *pt);
 372 void kernfs_notify(struct kernfs_node *kn);
 373
 374 int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
 375                      void *value, size_t size);
 376 int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
 377                      const void *value, size_t size, int flags);
 378
 379 const void *kernfs_super_ns(struct super_block *sb);
 380 int kernfs_get_tree(struct fs_context *fc);
 381 void kernfs_free_fs_context(struct fs_context *fc);
 382 void kernfs_kill_sb(struct super_block *sb);
 383
 384 void kernfs_init(void);
 385
 386 struct kernfs_node *kernfs_get_node_by_id(struct kernfs_root *root,
 387         const union kernfs_node_id *id);
 388 #else   /* CONFIG_KERNFS */
 389
 390 static inline enum kernfs_node_type kernfs_type(struct kernfs_node *kn)
 391 { return 0; }   /* whatever */
 392
 393 static inline void kernfs_enable_ns(struct kernfs_node *kn) { }
 394
 395 static inline bool kernfs_ns_enabled(struct kernfs_node *kn)
 396 { return false; }
 397
 398 static inline int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
 399 { return -ENOSYS; }
 400
 401 static inline int kernfs_path_from_node(struct kernfs_node *root_kn,
 402                                         struct kernfs_node *kn,
 403                                         char *buf, size_t buflen)
 404 { return -ENOSYS; }
 405
 406 static inline void pr_cont_kernfs_name(struct kernfs_node *kn) { }
 407 static inline void pr_cont_kernfs_path(struct kernfs_node *kn) { }
 408
 409 static inline struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
 410 { return NULL; }
 411
 412 static inline struct kernfs_node *
 413 kernfs_find_and_get_ns(struct kernfs_node *parent, const char *name,
 414                        const void *ns)
 415 { return NULL; }
 416 static inline struct kernfs_node *
 417 kernfs_walk_and_get_ns(struct kernfs_node *parent, const char *path,
 418                        const void *ns)
 419 { return NULL; }
 420
 421 static inline void kernfs_get(struct kernfs_node *kn) { }
 422 static inline void kernfs_put(struct kernfs_node *kn) { }
 423
 424 static inline struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
 425 { return NULL; }
 426
 427 static inline struct kernfs_root *kernfs_root_from_sb(struct super_block *sb)
 428 { return NULL; }
 429
 430 static inline struct inode *
 431 kernfs_get_inode(struct super_block *sb, struct kernfs_node *kn)
 432 { return NULL; }
 433
 434 static inline struct kernfs_root *
 435 kernfs_create_root(struct kernfs_syscall_ops *scops, unsigned int flags,
 436                    void *priv)
 437 { return ERR_PTR(-ENOSYS); }
 438
 439 static inline void kernfs_destroy_root(struct kernfs_root *root) { }
 440
 441 static inline struct kernfs_node *
 442 kernfs_create_dir_ns(struct kernfs_node *parent, const char *name,
 443                      umode_t mode, kuid_t uid, kgid_t gid,
 444                      void *priv, const void *ns)
 445 { return ERR_PTR(-ENOSYS); }
 446
 447 static inline struct kernfs_node *
 448 __kernfs_create_file(struct kernfs_node *parent, const char *name,
 449                      umode_t mode, kuid_t uid, kgid_t gid,
 450                      loff_t size, const struct kernfs_ops *ops,
 451                      void *priv, const void *ns, struct lock_class_key *key)
 452 { return ERR_PTR(-ENOSYS); }
 453
 454 static inline struct kernfs_node *
 455 kernfs_create_link(struct kernfs_node *parent, const char *name,
 456                    struct kernfs_node *target)
 457 { return ERR_PTR(-ENOSYS); }
 458
 459 static inline void kernfs_activate(struct kernfs_node *kn) { }
 460
 461 static inline void kernfs_remove(struct kernfs_node *kn) { }
 462
 463 static inline bool kernfs_remove_self(struct kernfs_node *kn)
 464 { return false; }
 465
 466 static inline int kernfs_remove_by_name_ns(struct kernfs_node *kn,
 467                                            const char *name, const void *ns)
 468 { return -ENOSYS; }
 469
 470 static inline int kernfs_rename_ns(struct kernfs_node *kn,
 471                                    struct kernfs_node *new_parent,
 472                                    const char *new_name, const void *new_ns)
 473 { return -ENOSYS; }
 474
 475 static inline int kernfs_setattr(struct kernfs_node *kn,
 476                                  const struct iattr *iattr)
 477 { return -ENOSYS; }
 478
 479 static inline void kernfs_notify(struct kernfs_node *kn) { }
 480
 481 static inline int kernfs_xattr_get(struct kernfs_node *kn, const char *name,
 482                                    void *value, size_t size)
 483 { return -ENOSYS; }
 484
 485 static inline int kernfs_xattr_set(struct kernfs_node *kn, const char *name,
 486                                    const void *value, size_t size, int flags)
 487 { return -ENOSYS; }
 488
 489 static inline const void *kernfs_super_ns(struct super_block *sb)
 490 { return NULL; }
 491
 492 static inline int kernfs_get_tree(struct fs_context *fc)
 493 { return -ENOSYS; }
 494
 495 static inline void kernfs_free_fs_context(struct fs_context *fc) { }
 496
 497 static inline void kernfs_kill_sb(struct super_block *sb) { }
 498
 499 static inline void kernfs_init(void) { }
 500
 501 #endif  /* CONFIG_KERNFS */
 502
 503 /**
 504  * kernfs_path - build full path of a given node
 505  * @kn: kernfs_node of interest
 506  * @buf: buffer to copy @kn's name into
 507  * @buflen: size of @buf
 508  *
 509  * If @kn is NULL result will be "(null)".
 510  *
 511  * Returns the length of the full path.  If the full length is equal to or
 512  * greater than @buflen, @buf contains the truncated path with the trailing
 513  * '\0'.  On error, -errno is returned.
 514  */
 515 static inline int kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
 516 {
 517         return kernfs_path_from_node(kn, NULL, buf, buflen);
 518 }
 519
 520 static inline struct kernfs_node *
 521 kernfs_find_and_get(struct kernfs_node *kn, const char *name)
 522 {
 523         return kernfs_find_and_get_ns(kn, name, NULL);
 524 }
 525
 526 static inline struct kernfs_node *
 527 kernfs_walk_and_get(struct kernfs_node *kn, const char *path)
 528 {
 529         return kernfs_walk_and_get_ns(kn, path, NULL);
 530 }
 531
 532 static inline struct kernfs_node *
 533 kernfs_create_dir(struct kernfs_node *parent, const char *name, umode_t mode,
 534                   void *priv)
 535 {
 536         return kernfs_create_dir_ns(parent, name, mode,
 537                                     GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
 538                                     priv, NULL);
 539 }
 540
 541 static inline struct kernfs_node *
 542 kernfs_create_file_ns(struct kernfs_node *parent, const char *name,
 543                       umode_t mode, kuid_t uid, kgid_t gid,
 544                       loff_t size, const struct kernfs_ops *ops,
 545                       void *priv, const void *ns)
 546 {
 547         struct lock_class_key *key = NULL;
 548
 549 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 550         key = (struct lock_class_key *)&ops->lockdep_key;
 551 #endif
 552         return __kernfs_create_file(parent, name, mode, uid, gid,
 553                                     size, ops, priv, ns, key);
 554 }
 555
 556 static inline struct kernfs_node *
 557 kernfs_create_file(struct kernfs_node *parent, const char *name, umode_t mode,
 558                    loff_t size, const struct kernfs_ops *ops, void *priv)
 559 {
 560         return kernfs_create_file_ns(parent, name, mode,
 561                                      GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
 562                                      size, ops, priv, NULL);
 563 }
 564
 565 static inline int kernfs_remove_by_name(struct kernfs_node *parent,
 566                                         const char *name)
 567 {
 568         return kernfs_remove_by_name_ns(parent, name, NULL);
 569 }
 570
 571 static inline int kernfs_rename(struct kernfs_node *kn,
 572                                 struct kernfs_node *new_parent,
 573                                 const char *new_name)
 574 {
 575         return kernfs_rename_ns(kn, new_parent, new_name, NULL);
 576 }
 577
 578 #endif  /* __LINUX_KERNFS_H */