1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * (C) 2001 Clemson University and The University of Chicago 4 * 5 * See COPYING in top-level directory. 6 */ 7 8/* 9 * The ORANGEFS Linux kernel support allows ORANGEFS volumes to be mounted and 10 * accessed through the Linux VFS (i.e. using standard I/O system calls). 11 * This support is only needed on clients that wish to mount the file system. 12 * 13 */ 14 15/* 16 * Declarations and macros for the ORANGEFS Linux kernel support. 17 */ 18 19#ifndef __ORANGEFSKERNEL_H 20#define __ORANGEFSKERNEL_H 21 22#include <linux/kernel.h> 23#include <linux/moduleparam.h> 24#include <linux/statfs.h> 25#include <linux/backing-dev.h> 26#include <linux/device.h> 27#include <linux/mpage.h> 28#include <linux/namei.h> 29#include <linux/errno.h> 30#include <linux/init.h> 31#include <linux/module.h> 32#include <linux/slab.h> 33#include <linux/types.h> 34#include <linux/fs.h> 35#include <linux/vmalloc.h> 36 37#include <linux/aio.h> 38#include <linux/posix_acl.h> 39#include <linux/posix_acl_xattr.h> 40#include <linux/compat.h> 41#include <linux/mount.h> 42#include <linux/uaccess.h> 43#include <linux/atomic.h> 44#include <linux/uio.h> 45#include <linux/sched/signal.h> 46#include <linux/mm.h> 47#include <linux/wait.h> 48#include <linux/dcache.h> 49#include <linux/pagemap.h> 50#include <linux/poll.h> 51#include <linux/rwsem.h> 52#include <linux/xattr.h> 53#include <linux/exportfs.h> 54#include <linux/hashtable.h> 55 56#include <asm/unaligned.h> 57 58#include "orangefs-dev-proto.h" 59 60#define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 20 61 62#define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS 30 63 64#define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS 900 /* 15 minutes */ 65 66#define ORANGEFS_REQDEVICE_NAME "pvfs2-req" 67 68#define ORANGEFS_DEVREQ_MAGIC 0x20030529 69#define ORANGEFS_PURGE_RETRY_COUNT 0x00000005 70 71#define MAX_DEV_REQ_UPSIZE (2 * sizeof(__s32) + \ 72sizeof(__u64) + sizeof(struct orangefs_upcall_s)) 73#define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \ 74sizeof(__u64) + sizeof(struct orangefs_downcall_s)) 75 76/* 77 * valid orangefs kernel operation states 78 * 79 * unknown - op was just initialized 80 * waiting - op is on request_list (upward bound) 81 * inprogr - op is in progress (waiting for downcall) 82 * serviced - op has matching downcall; ok 83 * purged - op has to start a timer since client-core 84 * exited uncleanly before servicing op 85 * given up - submitter has given up waiting for it 86 */ 87enum orangefs_vfs_op_states { 88 OP_VFS_STATE_UNKNOWN = 0, 89 OP_VFS_STATE_WAITING = 1, 90 OP_VFS_STATE_INPROGR = 2, 91 OP_VFS_STATE_SERVICED = 4, 92 OP_VFS_STATE_PURGED = 8, 93 OP_VFS_STATE_GIVEN_UP = 16, 94}; 95 96extern const struct xattr_handler * const orangefs_xattr_handlers[]; 97 98extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type, bool rcu); 99extern int orangefs_set_acl(struct mnt_idmap *idmap, 100 struct dentry *dentry, struct posix_acl *acl, 101 int type); 102int __orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type); 103 104/* 105 * orangefs data structures 106 */ 107struct orangefs_kernel_op_s { 108 enum orangefs_vfs_op_states op_state; 109 __u64 tag; 110 111 /* 112 * Set uses_shared_memory to non zero if this operation uses 113 * shared memory. If true, then a retry on the op must also 114 * get a new shared memory buffer and re-populate it. 115 * Cancels don't care - it only matters for service_operation() 116 * retry logics and cancels don't go through it anymore. It 117 * safely stays non-zero when we use it as slot_to_free. 118 */ 119 union { 120 int uses_shared_memory; 121 int slot_to_free; 122 }; 123 124 struct orangefs_upcall_s upcall; 125 struct orangefs_downcall_s downcall; 126 127 struct completion waitq; 128 spinlock_t lock; 129 130 int attempts; 131 132 struct list_head list; 133}; 134 135#define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING) 136#define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR) 137#define set_op_state_given_up(op) ((op)->op_state = OP_VFS_STATE_GIVEN_UP) 138static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op) 139{ 140 op->op_state = OP_VFS_STATE_SERVICED; 141 complete(&op->waitq); 142} 143 144#define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING) 145#define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR) 146#define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED) 147#define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED) 148#define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP) 149#define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL) 150 151void op_release(struct orangefs_kernel_op_s *op); 152 153extern void orangefs_bufmap_put(int); 154static inline void put_cancel(struct orangefs_kernel_op_s *op) 155{ 156 orangefs_bufmap_put(op->slot_to_free); 157 op_release(op); 158} 159 160static inline void set_op_state_purged(struct orangefs_kernel_op_s *op) 161{ 162 spin_lock(&op->lock); 163 if (unlikely(op_is_cancel(op))) { 164 list_del_init(&op->list); 165 spin_unlock(&op->lock); 166 put_cancel(op); 167 } else { 168 op->op_state |= OP_VFS_STATE_PURGED; 169 complete(&op->waitq); 170 spin_unlock(&op->lock); 171 } 172} 173 174/* per inode private orangefs info */ 175struct orangefs_inode_s { 176 struct orangefs_object_kref refn; 177 char link_target[ORANGEFS_NAME_MAX]; 178 /* 179 * Reading/Writing Extended attributes need to acquire the appropriate 180 * reader/writer semaphore on the orangefs_inode_s structure. 181 */ 182 struct rw_semaphore xattr_sem; 183 184 struct inode vfs_inode; 185 sector_t last_failed_block_index_read; 186 187 unsigned long getattr_time; 188 unsigned long mapping_time; 189 int attr_valid; 190 kuid_t attr_uid; 191 kgid_t attr_gid; 192 unsigned long bitlock; 193 194 DECLARE_HASHTABLE(xattr_cache, 4); 195}; 196 197/* per superblock private orangefs info */ 198struct orangefs_sb_info_s { 199 struct orangefs_khandle root_khandle; 200 __s32 fs_id; 201 int id; 202 int flags; 203#define ORANGEFS_OPT_INTR 0x01 204#define ORANGEFS_OPT_LOCAL_LOCK 0x02 205 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN]; 206 struct super_block *sb; 207 int mount_pending; 208 int no_list; 209 struct list_head list; 210}; 211 212struct orangefs_stats { 213 unsigned long cache_hits; 214 unsigned long cache_misses; 215 unsigned long reads; 216 unsigned long writes; 217}; 218 219struct orangefs_cached_xattr { 220 struct hlist_node node; 221 char key[ORANGEFS_MAX_XATTR_NAMELEN]; 222 char val[ORANGEFS_MAX_XATTR_VALUELEN]; 223 ssize_t length; 224 unsigned long timeout; 225}; 226 227struct orangefs_write_range { 228 loff_t pos; 229 size_t len; 230 kuid_t uid; 231 kgid_t gid; 232}; 233 234extern struct orangefs_stats orangefs_stats; 235 236/* 237 * NOTE: See Documentation/filesystems/porting.rst for information 238 * on implementing FOO_I and properly accessing fs private data 239 */ 240static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode) 241{ 242 return container_of(inode, struct orangefs_inode_s, vfs_inode); 243} 244 245static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb) 246{ 247 return (struct orangefs_sb_info_s *) sb->s_fs_info; 248} 249 250/* ino_t descends from "unsigned long", 8 bytes, 64 bits. */ 251static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle) 252{ 253 union { 254 unsigned char u[8]; 255 __u64 ino; 256 } ihandle; 257 258 ihandle.u[0] = khandle->u[0] ^ khandle->u[4]; 259 ihandle.u[1] = khandle->u[1] ^ khandle->u[5]; 260 ihandle.u[2] = khandle->u[2] ^ khandle->u[6]; 261 ihandle.u[3] = khandle->u[3] ^ khandle->u[7]; 262 ihandle.u[4] = khandle->u[12] ^ khandle->u[8]; 263 ihandle.u[5] = khandle->u[13] ^ khandle->u[9]; 264 ihandle.u[6] = khandle->u[14] ^ khandle->u[10]; 265 ihandle.u[7] = khandle->u[15] ^ khandle->u[11]; 266 267 return ihandle.ino; 268} 269 270static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode) 271{ 272 return &(ORANGEFS_I(inode)->refn.khandle); 273} 274 275static inline int is_root_handle(struct inode *inode) 276{ 277 gossip_debug(GOSSIP_DCACHE_DEBUG, 278 "%s: root handle: %pU, this handle: %pU:\n", 279 __func__, 280 &ORANGEFS_SB(inode->i_sb)->root_khandle, 281 get_khandle_from_ino(inode)); 282 283 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle), 284 get_khandle_from_ino(inode))) 285 return 0; 286 else 287 return 1; 288} 289 290static inline int match_handle(struct orangefs_khandle resp_handle, 291 struct inode *inode) 292{ 293 gossip_debug(GOSSIP_DCACHE_DEBUG, 294 "%s: one handle: %pU, another handle:%pU:\n", 295 __func__, 296 &resp_handle, 297 get_khandle_from_ino(inode)); 298 299 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode))) 300 return 0; 301 else 302 return 1; 303} 304 305/* 306 * defined in orangefs-cache.c 307 */ 308int op_cache_initialize(void); 309int op_cache_finalize(void); 310struct orangefs_kernel_op_s *op_alloc(__s32 type); 311void orangefs_new_tag(struct orangefs_kernel_op_s *op); 312char *get_opname_string(struct orangefs_kernel_op_s *new_op); 313 314int orangefs_inode_cache_initialize(void); 315int orangefs_inode_cache_finalize(void); 316 317/* 318 * defined in orangefs-mod.c 319 */ 320void purge_inprogress_ops(void); 321 322/* 323 * defined in waitqueue.c 324 */ 325void purge_waiting_ops(void); 326 327/* 328 * defined in super.c 329 */ 330extern uint64_t orangefs_features; 331 332struct dentry *orangefs_mount(struct file_system_type *fst, 333 int flags, 334 const char *devname, 335 void *data); 336 337void orangefs_kill_sb(struct super_block *sb); 338int orangefs_remount(struct orangefs_sb_info_s *); 339 340int fsid_key_table_initialize(void); 341void fsid_key_table_finalize(void); 342 343/* 344 * defined in inode.c 345 */ 346vm_fault_t orangefs_page_mkwrite(struct vm_fault *); 347struct inode *orangefs_new_inode(struct super_block *sb, 348 struct inode *dir, 349 umode_t mode, 350 dev_t dev, 351 struct orangefs_object_kref *ref); 352 353int __orangefs_setattr(struct inode *, struct iattr *); 354int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr); 355int orangefs_setattr(struct mnt_idmap *, struct dentry *, struct iattr *); 356 357int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path, 358 struct kstat *stat, u32 request_mask, unsigned int flags); 359 360int orangefs_permission(struct mnt_idmap *idmap, 361 struct inode *inode, int mask); 362 363int orangefs_update_time(struct inode *, int); 364 365/* 366 * defined in xattr.c 367 */ 368ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size); 369 370/* 371 * defined in namei.c 372 */ 373struct inode *orangefs_iget(struct super_block *sb, 374 struct orangefs_object_kref *ref); 375 376/* 377 * defined in devorangefs-req.c 378 */ 379extern uint32_t orangefs_userspace_version; 380 381int orangefs_dev_init(void); 382void orangefs_dev_cleanup(void); 383int is_daemon_in_service(void); 384bool __is_daemon_in_service(void); 385 386/* 387 * defined in file.c 388 */ 389int orangefs_revalidate_mapping(struct inode *); 390ssize_t wait_for_direct_io(enum ORANGEFS_io_type, struct inode *, loff_t *, 391 struct iov_iter *, size_t, loff_t, struct orangefs_write_range *, int *, 392 struct file *); 393ssize_t do_readv_writev(enum ORANGEFS_io_type, struct file *, loff_t *, 394 struct iov_iter *); 395 396/* 397 * defined in orangefs-utils.c 398 */ 399__s32 fsid_of_op(struct orangefs_kernel_op_s *op); 400 401ssize_t orangefs_inode_getxattr(struct inode *inode, 402 const char *name, 403 void *buffer, 404 size_t size); 405 406int orangefs_inode_setxattr(struct inode *inode, 407 const char *name, 408 const void *value, 409 size_t size, 410 int flags); 411 412#define ORANGEFS_GETATTR_NEW 1 413#define ORANGEFS_GETATTR_SIZE 2 414 415int orangefs_inode_getattr(struct inode *, int); 416 417int orangefs_inode_check_changed(struct inode *inode); 418 419int orangefs_inode_setattr(struct inode *inode); 420 421bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op); 422 423int orangefs_normalize_to_errno(__s32 error_code); 424 425extern struct mutex orangefs_request_mutex; 426extern int op_timeout_secs; 427extern int slot_timeout_secs; 428extern int orangefs_cache_timeout_msecs; 429extern int orangefs_dcache_timeout_msecs; 430extern int orangefs_getattr_timeout_msecs; 431extern struct list_head orangefs_superblocks; 432extern spinlock_t orangefs_superblocks_lock; 433extern struct list_head orangefs_request_list; 434extern spinlock_t orangefs_request_list_lock; 435extern wait_queue_head_t orangefs_request_list_waitq; 436extern struct list_head *orangefs_htable_ops_in_progress; 437extern spinlock_t orangefs_htable_ops_in_progress_lock; 438extern int hash_table_size; 439 440extern const struct file_operations orangefs_file_operations; 441extern const struct inode_operations orangefs_symlink_inode_operations; 442extern const struct inode_operations orangefs_dir_inode_operations; 443extern const struct file_operations orangefs_dir_operations; 444extern const struct dentry_operations orangefs_dentry_operations; 445 446/* 447 * misc convenience macros 448 */ 449 450#define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */ 451#define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */ 452#define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */ 453#define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */ 454#define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */ 455#define ORANGEFS_OP_WRITEBACK 32 456 457int service_operation(struct orangefs_kernel_op_s *op, 458 const char *op_name, 459 int flags); 460 461#define get_interruptible_flag(inode) \ 462 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \ 463 ORANGEFS_OP_INTERRUPTIBLE : 0) 464 465#define fill_default_sys_attrs(sys_attr, type, mode) \ 466do { \ 467 sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \ 468 sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \ 469 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \ 470 sys_attr.mtime = 0; \ 471 sys_attr.atime = 0; \ 472 sys_attr.ctime = 0; \ 473 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \ 474} while (0) 475 476static inline void orangefs_set_timeout(struct dentry *dentry) 477{ 478 unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000; 479 480 dentry->d_fsdata = (void *) time; 481} 482 483#endif /* __ORANGEFSKERNEL_H */ 484