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