1 /* 2 * (C) 2001 Clemson University and The University of Chicago 3 * 4 * See COPYING in top-level directory. 5 */ 6 7 /* 8 * The ORANGEFS Linux kernel support allows ORANGEFS volumes to be mounted and 9 * accessed through the Linux VFS (i.e. using standard I/O system calls). 10 * This support is only needed on clients that wish to mount the file system. 11 * 12 */ 13 14 /* 15 * Declarations and macros for the ORANGEFS Linux kernel support. 16 */ 17 18 #ifndef __ORANGEFSKERNEL_H 19 #define __ORANGEFSKERNEL_H 20 21 #include <linux/kernel.h> 22 #include <linux/moduleparam.h> 23 #include <linux/statfs.h> 24 #include <linux/backing-dev.h> 25 #include <linux/device.h> 26 #include <linux/mpage.h> 27 #include <linux/namei.h> 28 #include <linux/errno.h> 29 #include <linux/init.h> 30 #include <linux/module.h> 31 #include <linux/slab.h> 32 #include <linux/types.h> 33 #include <linux/fs.h> 34 #include <linux/vmalloc.h> 35 36 #include <linux/aio.h> 37 #include <linux/posix_acl.h> 38 #include <linux/posix_acl_xattr.h> 39 #include <linux/compat.h> 40 #include <linux/mount.h> 41 #include <linux/uaccess.h> 42 #include <linux/atomic.h> 43 #include <linux/uio.h> 44 #include <linux/sched.h> 45 #include <linux/mm.h> 46 #include <linux/wait.h> 47 #include <linux/dcache.h> 48 #include <linux/pagemap.h> 49 #include <linux/poll.h> 50 #include <linux/rwsem.h> 51 #include <linux/xattr.h> 52 #include <linux/exportfs.h> 53 54 #include <asm/unaligned.h> 55 56 #include "orangefs-dev-proto.h" 57 58 #ifdef ORANGEFS_KERNEL_DEBUG 59 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 10 60 #else 61 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 20 62 #endif 63 64 #define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS 30 65 66 #define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS 900 /* 15 minutes */ 67 68 #define ORANGEFS_REQDEVICE_NAME "pvfs2-req" 69 70 #define ORANGEFS_DEVREQ_MAGIC 0x20030529 71 #define ORANGEFS_LINK_MAX 0x000000FF 72 #define ORANGEFS_PURGE_RETRY_COUNT 0x00000005 73 #define ORANGEFS_MAX_NUM_OPTIONS 0x00000004 74 #define ORANGEFS_MAX_MOUNT_OPT_LEN 0x00000080 75 #define ORANGEFS_MAX_FSKEY_LEN 64 76 77 #define MAX_DEV_REQ_UPSIZE (2 * sizeof(__s32) + \ 78 sizeof(__u64) + sizeof(struct orangefs_upcall_s)) 79 #define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \ 80 sizeof(__u64) + sizeof(struct orangefs_downcall_s)) 81 82 /* 83 * valid orangefs kernel operation states 84 * 85 * unknown - op was just initialized 86 * waiting - op is on request_list (upward bound) 87 * inprogr - op is in progress (waiting for downcall) 88 * serviced - op has matching downcall; ok 89 * purged - op has to start a timer since client-core 90 * exited uncleanly before servicing op 91 * given up - submitter has given up waiting for it 92 */ 93 enum orangefs_vfs_op_states { 94 OP_VFS_STATE_UNKNOWN = 0, 95 OP_VFS_STATE_WAITING = 1, 96 OP_VFS_STATE_INPROGR = 2, 97 OP_VFS_STATE_SERVICED = 4, 98 OP_VFS_STATE_PURGED = 8, 99 OP_VFS_STATE_GIVEN_UP = 16, 100 }; 101 102 /* 103 * An array of client_debug_mask will be built to hold debug keyword/mask 104 * values fetched from userspace. 105 */ 106 struct client_debug_mask { 107 char *keyword; 108 __u64 mask1; 109 __u64 mask2; 110 }; 111 112 /* 113 * orangefs kernel memory related flags 114 */ 115 116 #if ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) 117 #define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE 118 #else 119 #define ORANGEFS_CACHE_CREATE_FLAGS 0 120 #endif /* ((defined ORANGEFS_KERNEL_DEBUG) && (defined CONFIG_DEBUG_SLAB)) */ 121 122 /* orangefs xattr and acl related defines */ 123 #define ORANGEFS_XATTR_INDEX_POSIX_ACL_ACCESS 1 124 #define ORANGEFS_XATTR_INDEX_POSIX_ACL_DEFAULT 2 125 #define ORANGEFS_XATTR_INDEX_TRUSTED 3 126 #define ORANGEFS_XATTR_INDEX_DEFAULT 4 127 128 #define ORANGEFS_XATTR_NAME_ACL_ACCESS XATTR_NAME_POSIX_ACL_ACCESS 129 #define ORANGEFS_XATTR_NAME_ACL_DEFAULT XATTR_NAME_POSIX_ACL_DEFAULT 130 #define ORANGEFS_XATTR_NAME_TRUSTED_PREFIX "trusted." 131 #define ORANGEFS_XATTR_NAME_DEFAULT_PREFIX "" 132 133 /* these functions are defined in orangefs-utils.c */ 134 int orangefs_prepare_cdm_array(char *debug_array_string); 135 int orangefs_prepare_debugfs_help_string(int); 136 137 /* defined in orangefs-debugfs.c */ 138 int orangefs_client_debug_init(void); 139 140 void debug_string_to_mask(char *, void *, int); 141 void do_c_mask(int, char *, struct client_debug_mask **); 142 void do_k_mask(int, char *, __u64 **); 143 144 void debug_mask_to_string(void *, int); 145 void do_k_string(void *, int); 146 void do_c_string(void *, int); 147 int check_amalgam_keyword(void *, int); 148 int keyword_is_amalgam(char *); 149 150 /*these variables are defined in orangefs-mod.c */ 151 extern char kernel_debug_string[ORANGEFS_MAX_DEBUG_STRING_LEN]; 152 extern char client_debug_string[ORANGEFS_MAX_DEBUG_STRING_LEN]; 153 extern char client_debug_array_string[ORANGEFS_MAX_DEBUG_STRING_LEN]; 154 extern unsigned int kernel_mask_set_mod_init; 155 156 extern int orangefs_init_acl(struct inode *inode, struct inode *dir); 157 extern const struct xattr_handler *orangefs_xattr_handlers[]; 158 159 extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type); 160 extern int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type); 161 162 /* 163 * Redefine xtvec structure so that we could move helper functions out of 164 * the define 165 */ 166 struct xtvec { 167 __kernel_off_t xtv_off; /* must be off_t */ 168 __kernel_size_t xtv_len; /* must be size_t */ 169 }; 170 171 /* 172 * orangefs data structures 173 */ 174 struct orangefs_kernel_op_s { 175 enum orangefs_vfs_op_states op_state; 176 __u64 tag; 177 178 /* 179 * Set uses_shared_memory to non zero if this operation uses 180 * shared memory. If true, then a retry on the op must also 181 * get a new shared memory buffer and re-populate it. 182 * Cancels don't care - it only matters for service_operation() 183 * retry logics and cancels don't go through it anymore. It 184 * safely stays non-zero when we use it as slot_to_free. 185 */ 186 union { 187 int uses_shared_memory; 188 int slot_to_free; 189 }; 190 191 struct orangefs_upcall_s upcall; 192 struct orangefs_downcall_s downcall; 193 194 struct completion waitq; 195 spinlock_t lock; 196 197 int attempts; 198 199 struct list_head list; 200 }; 201 202 #define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING) 203 #define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR) 204 #define set_op_state_given_up(op) ((op)->op_state = OP_VFS_STATE_GIVEN_UP) 205 static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op) 206 { 207 op->op_state = OP_VFS_STATE_SERVICED; 208 complete(&op->waitq); 209 } 210 211 #define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING) 212 #define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR) 213 #define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED) 214 #define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED) 215 #define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP) 216 #define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL) 217 218 void op_release(struct orangefs_kernel_op_s *op); 219 220 extern void orangefs_bufmap_put(int); 221 static inline void put_cancel(struct orangefs_kernel_op_s *op) 222 { 223 orangefs_bufmap_put(op->slot_to_free); 224 op_release(op); 225 } 226 227 static inline void set_op_state_purged(struct orangefs_kernel_op_s *op) 228 { 229 spin_lock(&op->lock); 230 if (unlikely(op_is_cancel(op))) { 231 list_del_init(&op->list); 232 spin_unlock(&op->lock); 233 put_cancel(op); 234 } else { 235 op->op_state |= OP_VFS_STATE_PURGED; 236 complete(&op->waitq); 237 spin_unlock(&op->lock); 238 } 239 } 240 241 /* per inode private orangefs info */ 242 struct orangefs_inode_s { 243 struct orangefs_object_kref refn; 244 char link_target[ORANGEFS_NAME_MAX]; 245 __s64 blksize; 246 /* 247 * Reading/Writing Extended attributes need to acquire the appropriate 248 * reader/writer semaphore on the orangefs_inode_s structure. 249 */ 250 struct rw_semaphore xattr_sem; 251 252 struct inode vfs_inode; 253 sector_t last_failed_block_index_read; 254 255 /* 256 * State of in-memory attributes not yet flushed to disk associated 257 * with this object 258 */ 259 unsigned long pinode_flags; 260 }; 261 262 #define P_ATIME_FLAG 0 263 #define P_MTIME_FLAG 1 264 #define P_CTIME_FLAG 2 265 #define P_MODE_FLAG 3 266 267 #define ClearAtimeFlag(pinode) clear_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 268 #define SetAtimeFlag(pinode) set_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 269 #define AtimeFlag(pinode) test_bit(P_ATIME_FLAG, &(pinode)->pinode_flags) 270 271 #define ClearMtimeFlag(pinode) clear_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 272 #define SetMtimeFlag(pinode) set_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 273 #define MtimeFlag(pinode) test_bit(P_MTIME_FLAG, &(pinode)->pinode_flags) 274 275 #define ClearCtimeFlag(pinode) clear_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 276 #define SetCtimeFlag(pinode) set_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 277 #define CtimeFlag(pinode) test_bit(P_CTIME_FLAG, &(pinode)->pinode_flags) 278 279 #define ClearModeFlag(pinode) clear_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 280 #define SetModeFlag(pinode) set_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 281 #define ModeFlag(pinode) test_bit(P_MODE_FLAG, &(pinode)->pinode_flags) 282 283 /* per superblock private orangefs info */ 284 struct orangefs_sb_info_s { 285 struct orangefs_khandle root_khandle; 286 __s32 fs_id; 287 int id; 288 int flags; 289 #define ORANGEFS_OPT_INTR 0x01 290 #define ORANGEFS_OPT_LOCAL_LOCK 0x02 291 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN]; 292 struct super_block *sb; 293 int mount_pending; 294 struct list_head list; 295 }; 296 297 /* 298 * structure that holds the state of any async I/O operation issued 299 * through the VFS. Needed especially to handle cancellation requests 300 * or even completion notification so that the VFS client-side daemon 301 * can free up its vfs_request slots. 302 */ 303 struct orangefs_kiocb_s { 304 /* the pointer to the task that initiated the AIO */ 305 struct task_struct *tsk; 306 307 /* pointer to the kiocb that kicked this operation */ 308 struct kiocb *kiocb; 309 310 /* buffer index that was used for the I/O */ 311 struct orangefs_bufmap *bufmap; 312 int buffer_index; 313 314 /* orangefs kernel operation type */ 315 struct orangefs_kernel_op_s *op; 316 317 /* The user space buffers from/to which I/O is being staged */ 318 struct iovec *iov; 319 320 /* number of elements in the iovector */ 321 unsigned long nr_segs; 322 323 /* set to indicate the type of the operation */ 324 int rw; 325 326 /* file offset */ 327 loff_t offset; 328 329 /* and the count in bytes */ 330 size_t bytes_to_be_copied; 331 332 ssize_t bytes_copied; 333 int needs_cleanup; 334 }; 335 336 struct orangefs_stats { 337 unsigned long cache_hits; 338 unsigned long cache_misses; 339 unsigned long reads; 340 unsigned long writes; 341 }; 342 343 extern struct orangefs_stats g_orangefs_stats; 344 345 /* 346 * NOTE: See Documentation/filesystems/porting for information 347 * on implementing FOO_I and properly accessing fs private data 348 */ 349 static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode) 350 { 351 return container_of(inode, struct orangefs_inode_s, vfs_inode); 352 } 353 354 static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb) 355 { 356 return (struct orangefs_sb_info_s *) sb->s_fs_info; 357 } 358 359 /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */ 360 static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle) 361 { 362 union { 363 unsigned char u[8]; 364 __u64 ino; 365 } ihandle; 366 367 ihandle.u[0] = khandle->u[0] ^ khandle->u[4]; 368 ihandle.u[1] = khandle->u[1] ^ khandle->u[5]; 369 ihandle.u[2] = khandle->u[2] ^ khandle->u[6]; 370 ihandle.u[3] = khandle->u[3] ^ khandle->u[7]; 371 ihandle.u[4] = khandle->u[12] ^ khandle->u[8]; 372 ihandle.u[5] = khandle->u[13] ^ khandle->u[9]; 373 ihandle.u[6] = khandle->u[14] ^ khandle->u[10]; 374 ihandle.u[7] = khandle->u[15] ^ khandle->u[11]; 375 376 return ihandle.ino; 377 } 378 379 static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode) 380 { 381 return &(ORANGEFS_I(inode)->refn.khandle); 382 } 383 384 static inline __s32 get_fsid_from_ino(struct inode *inode) 385 { 386 return ORANGEFS_I(inode)->refn.fs_id; 387 } 388 389 static inline ino_t get_ino_from_khandle(struct inode *inode) 390 { 391 struct orangefs_khandle *khandle; 392 ino_t ino; 393 394 khandle = get_khandle_from_ino(inode); 395 ino = orangefs_khandle_to_ino(khandle); 396 return ino; 397 } 398 399 static inline ino_t get_parent_ino_from_dentry(struct dentry *dentry) 400 { 401 return get_ino_from_khandle(dentry->d_parent->d_inode); 402 } 403 404 static inline int is_root_handle(struct inode *inode) 405 { 406 gossip_debug(GOSSIP_DCACHE_DEBUG, 407 "%s: root handle: %pU, this handle: %pU:\n", 408 __func__, 409 &ORANGEFS_SB(inode->i_sb)->root_khandle, 410 get_khandle_from_ino(inode)); 411 412 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle), 413 get_khandle_from_ino(inode))) 414 return 0; 415 else 416 return 1; 417 } 418 419 static inline int match_handle(struct orangefs_khandle resp_handle, 420 struct inode *inode) 421 { 422 gossip_debug(GOSSIP_DCACHE_DEBUG, 423 "%s: one handle: %pU, another handle:%pU:\n", 424 __func__, 425 &resp_handle, 426 get_khandle_from_ino(inode)); 427 428 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode))) 429 return 0; 430 else 431 return 1; 432 } 433 434 /* 435 * defined in orangefs-cache.c 436 */ 437 int op_cache_initialize(void); 438 int op_cache_finalize(void); 439 struct orangefs_kernel_op_s *op_alloc(__s32 type); 440 void orangefs_new_tag(struct orangefs_kernel_op_s *op); 441 char *get_opname_string(struct orangefs_kernel_op_s *new_op); 442 443 int orangefs_inode_cache_initialize(void); 444 int orangefs_inode_cache_finalize(void); 445 446 /* 447 * defined in orangefs-mod.c 448 */ 449 void purge_inprogress_ops(void); 450 451 /* 452 * defined in waitqueue.c 453 */ 454 void purge_waiting_ops(void); 455 456 /* 457 * defined in super.c 458 */ 459 struct dentry *orangefs_mount(struct file_system_type *fst, 460 int flags, 461 const char *devname, 462 void *data); 463 464 void orangefs_kill_sb(struct super_block *sb); 465 int orangefs_remount(struct orangefs_sb_info_s *); 466 467 int fsid_key_table_initialize(void); 468 void fsid_key_table_finalize(void); 469 470 /* 471 * defined in inode.c 472 */ 473 __u32 convert_to_orangefs_mask(unsigned long lite_mask); 474 struct inode *orangefs_new_inode(struct super_block *sb, 475 struct inode *dir, 476 int mode, 477 dev_t dev, 478 struct orangefs_object_kref *ref); 479 480 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr); 481 482 int orangefs_getattr(struct vfsmount *mnt, 483 struct dentry *dentry, 484 struct kstat *kstat); 485 486 int orangefs_permission(struct inode *inode, int mask); 487 488 /* 489 * defined in xattr.c 490 */ 491 int orangefs_setxattr(struct dentry *dentry, 492 const char *name, 493 const void *value, 494 size_t size, 495 int flags); 496 497 ssize_t orangefs_getxattr(struct dentry *dentry, 498 const char *name, 499 void *buffer, 500 size_t size); 501 502 ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size); 503 504 /* 505 * defined in namei.c 506 */ 507 struct inode *orangefs_iget(struct super_block *sb, 508 struct orangefs_object_kref *ref); 509 510 ssize_t orangefs_inode_read(struct inode *inode, 511 struct iov_iter *iter, 512 loff_t *offset, 513 loff_t readahead_size); 514 515 /* 516 * defined in devorangefs-req.c 517 */ 518 int orangefs_dev_init(void); 519 void orangefs_dev_cleanup(void); 520 int is_daemon_in_service(void); 521 bool __is_daemon_in_service(void); 522 523 /* 524 * defined in orangefs-utils.c 525 */ 526 __s32 fsid_of_op(struct orangefs_kernel_op_s *op); 527 528 int orangefs_flush_inode(struct inode *inode); 529 530 ssize_t orangefs_inode_getxattr(struct inode *inode, 531 const char *prefix, 532 const char *name, 533 void *buffer, 534 size_t size); 535 536 int orangefs_inode_setxattr(struct inode *inode, 537 const char *prefix, 538 const char *name, 539 const void *value, 540 size_t size, 541 int flags); 542 543 int orangefs_inode_getattr(struct inode *inode, int new, int size); 544 545 int orangefs_inode_check_changed(struct inode *inode); 546 547 int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr); 548 549 void orangefs_make_bad_inode(struct inode *inode); 550 551 int orangefs_unmount_sb(struct super_block *sb); 552 553 bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op); 554 555 int orangefs_normalize_to_errno(__s32 error_code); 556 557 extern struct mutex devreq_mutex; 558 extern struct mutex request_mutex; 559 extern int debug; 560 extern int op_timeout_secs; 561 extern int slot_timeout_secs; 562 extern struct list_head orangefs_superblocks; 563 extern spinlock_t orangefs_superblocks_lock; 564 extern struct list_head orangefs_request_list; 565 extern spinlock_t orangefs_request_list_lock; 566 extern wait_queue_head_t orangefs_request_list_waitq; 567 extern struct list_head *htable_ops_in_progress; 568 extern spinlock_t htable_ops_in_progress_lock; 569 extern int hash_table_size; 570 571 extern const struct address_space_operations orangefs_address_operations; 572 extern struct backing_dev_info orangefs_backing_dev_info; 573 extern struct inode_operations orangefs_file_inode_operations; 574 extern const struct file_operations orangefs_file_operations; 575 extern struct inode_operations orangefs_symlink_inode_operations; 576 extern struct inode_operations orangefs_dir_inode_operations; 577 extern const struct file_operations orangefs_dir_operations; 578 extern const struct dentry_operations orangefs_dentry_operations; 579 extern const struct file_operations orangefs_devreq_file_operations; 580 581 extern wait_queue_head_t orangefs_bufmap_init_waitq; 582 583 /* 584 * misc convenience macros 585 */ 586 587 #define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */ 588 #define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */ 589 #define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */ 590 #define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */ 591 #define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */ 592 593 int service_operation(struct orangefs_kernel_op_s *op, 594 const char *op_name, 595 int flags); 596 597 #define get_interruptible_flag(inode) \ 598 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \ 599 ORANGEFS_OP_INTERRUPTIBLE : 0) 600 601 #define fill_default_sys_attrs(sys_attr, type, mode) \ 602 do { \ 603 sys_attr.owner = from_kuid(current_user_ns(), current_fsuid()); \ 604 sys_attr.group = from_kgid(current_user_ns(), current_fsgid()); \ 605 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \ 606 sys_attr.mtime = 0; \ 607 sys_attr.atime = 0; \ 608 sys_attr.ctime = 0; \ 609 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \ 610 } while (0) 611 612 static inline void orangefs_i_size_write(struct inode *inode, loff_t i_size) 613 { 614 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP) 615 inode_lock(inode); 616 #endif 617 i_size_write(inode, i_size); 618 #if BITS_PER_LONG == 32 && defined(CONFIG_SMP) 619 inode_unlock(inode); 620 #endif 621 } 622 623 #endif /* __ORANGEFSKERNEL_H */ 624