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) + \
72 sizeof(__u64) + sizeof(struct orangefs_upcall_s))
73 #define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \
74 sizeof(__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 */
87 enum 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
96 /*
97 * orangefs kernel memory related flags
98 */
99
100 #if (defined CONFIG_DEBUG_SLAB)
101 #define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE
102 #else
103 #define ORANGEFS_CACHE_CREATE_FLAGS 0
104 #endif
105
106 extern int orangefs_init_acl(struct inode *inode, struct inode *dir);
107 extern const struct xattr_handler *orangefs_xattr_handlers[];
108
109 extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type);
110 extern int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
111
112 /*
113 * orangefs data structures
114 */
115 struct orangefs_kernel_op_s {
116 enum orangefs_vfs_op_states op_state;
117 __u64 tag;
118
119 /*
120 * Set uses_shared_memory to non zero if this operation uses
121 * shared memory. If true, then a retry on the op must also
122 * get a new shared memory buffer and re-populate it.
123 * Cancels don't care - it only matters for service_operation()
124 * retry logics and cancels don't go through it anymore. It
125 * safely stays non-zero when we use it as slot_to_free.
126 */
127 union {
128 int uses_shared_memory;
129 int slot_to_free;
130 };
131
132 struct orangefs_upcall_s upcall;
133 struct orangefs_downcall_s downcall;
134
135 struct completion waitq;
136 spinlock_t lock;
137
138 int attempts;
139
140 struct list_head list;
141 };
142
143 #define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING)
144 #define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR)
145 #define set_op_state_given_up(op) ((op)->op_state = OP_VFS_STATE_GIVEN_UP)
set_op_state_serviced(struct orangefs_kernel_op_s * op)146 static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op)
147 {
148 op->op_state = OP_VFS_STATE_SERVICED;
149 complete(&op->waitq);
150 }
151
152 #define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING)
153 #define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR)
154 #define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED)
155 #define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED)
156 #define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP)
157 #define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL)
158
159 void op_release(struct orangefs_kernel_op_s *op);
160
161 extern void orangefs_bufmap_put(int);
put_cancel(struct orangefs_kernel_op_s * op)162 static inline void put_cancel(struct orangefs_kernel_op_s *op)
163 {
164 orangefs_bufmap_put(op->slot_to_free);
165 op_release(op);
166 }
167
set_op_state_purged(struct orangefs_kernel_op_s * op)168 static inline void set_op_state_purged(struct orangefs_kernel_op_s *op)
169 {
170 spin_lock(&op->lock);
171 if (unlikely(op_is_cancel(op))) {
172 list_del_init(&op->list);
173 spin_unlock(&op->lock);
174 put_cancel(op);
175 } else {
176 op->op_state |= OP_VFS_STATE_PURGED;
177 complete(&op->waitq);
178 spin_unlock(&op->lock);
179 }
180 }
181
182 /* per inode private orangefs info */
183 struct orangefs_inode_s {
184 struct orangefs_object_kref refn;
185 char link_target[ORANGEFS_NAME_MAX];
186 /*
187 * Reading/Writing Extended attributes need to acquire the appropriate
188 * reader/writer semaphore on the orangefs_inode_s structure.
189 */
190 struct rw_semaphore xattr_sem;
191
192 struct inode vfs_inode;
193 sector_t last_failed_block_index_read;
194
195 unsigned long getattr_time;
196 unsigned long mapping_time;
197 int attr_valid;
198 kuid_t attr_uid;
199 kgid_t attr_gid;
200 unsigned long bitlock;
201
202 DECLARE_HASHTABLE(xattr_cache, 4);
203 };
204
205 /* per superblock private orangefs info */
206 struct orangefs_sb_info_s {
207 struct orangefs_khandle root_khandle;
208 __s32 fs_id;
209 int id;
210 int flags;
211 #define ORANGEFS_OPT_INTR 0x01
212 #define ORANGEFS_OPT_LOCAL_LOCK 0x02
213 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN];
214 struct super_block *sb;
215 int mount_pending;
216 int no_list;
217 struct list_head list;
218 };
219
220 struct orangefs_stats {
221 unsigned long cache_hits;
222 unsigned long cache_misses;
223 unsigned long reads;
224 unsigned long writes;
225 };
226
227 struct orangefs_cached_xattr {
228 struct hlist_node node;
229 char key[ORANGEFS_MAX_XATTR_NAMELEN];
230 char val[ORANGEFS_MAX_XATTR_VALUELEN];
231 ssize_t length;
232 unsigned long timeout;
233 };
234
235 struct orangefs_write_range {
236 loff_t pos;
237 size_t len;
238 kuid_t uid;
239 kgid_t gid;
240 };
241
242 extern struct orangefs_stats orangefs_stats;
243
244 /*
245 * NOTE: See Documentation/filesystems/porting.rst for information
246 * on implementing FOO_I and properly accessing fs private data
247 */
ORANGEFS_I(struct inode * inode)248 static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode)
249 {
250 return container_of(inode, struct orangefs_inode_s, vfs_inode);
251 }
252
ORANGEFS_SB(struct super_block * sb)253 static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb)
254 {
255 return (struct orangefs_sb_info_s *) sb->s_fs_info;
256 }
257
258 /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */
orangefs_khandle_to_ino(struct orangefs_khandle * khandle)259 static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle)
260 {
261 union {
262 unsigned char u[8];
263 __u64 ino;
264 } ihandle;
265
266 ihandle.u[0] = khandle->u[0] ^ khandle->u[4];
267 ihandle.u[1] = khandle->u[1] ^ khandle->u[5];
268 ihandle.u[2] = khandle->u[2] ^ khandle->u[6];
269 ihandle.u[3] = khandle->u[3] ^ khandle->u[7];
270 ihandle.u[4] = khandle->u[12] ^ khandle->u[8];
271 ihandle.u[5] = khandle->u[13] ^ khandle->u[9];
272 ihandle.u[6] = khandle->u[14] ^ khandle->u[10];
273 ihandle.u[7] = khandle->u[15] ^ khandle->u[11];
274
275 return ihandle.ino;
276 }
277
get_khandle_from_ino(struct inode * inode)278 static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode)
279 {
280 return &(ORANGEFS_I(inode)->refn.khandle);
281 }
282
is_root_handle(struct inode * inode)283 static inline int is_root_handle(struct inode *inode)
284 {
285 gossip_debug(GOSSIP_DCACHE_DEBUG,
286 "%s: root handle: %pU, this handle: %pU:\n",
287 __func__,
288 &ORANGEFS_SB(inode->i_sb)->root_khandle,
289 get_khandle_from_ino(inode));
290
291 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle),
292 get_khandle_from_ino(inode)))
293 return 0;
294 else
295 return 1;
296 }
297
match_handle(struct orangefs_khandle resp_handle,struct inode * inode)298 static inline int match_handle(struct orangefs_khandle resp_handle,
299 struct inode *inode)
300 {
301 gossip_debug(GOSSIP_DCACHE_DEBUG,
302 "%s: one handle: %pU, another handle:%pU:\n",
303 __func__,
304 &resp_handle,
305 get_khandle_from_ino(inode));
306
307 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode)))
308 return 0;
309 else
310 return 1;
311 }
312
313 /*
314 * defined in orangefs-cache.c
315 */
316 int op_cache_initialize(void);
317 int op_cache_finalize(void);
318 struct orangefs_kernel_op_s *op_alloc(__s32 type);
319 void orangefs_new_tag(struct orangefs_kernel_op_s *op);
320 char *get_opname_string(struct orangefs_kernel_op_s *new_op);
321
322 int orangefs_inode_cache_initialize(void);
323 int orangefs_inode_cache_finalize(void);
324
325 /*
326 * defined in orangefs-mod.c
327 */
328 void purge_inprogress_ops(void);
329
330 /*
331 * defined in waitqueue.c
332 */
333 void purge_waiting_ops(void);
334
335 /*
336 * defined in super.c
337 */
338 extern uint64_t orangefs_features;
339
340 struct dentry *orangefs_mount(struct file_system_type *fst,
341 int flags,
342 const char *devname,
343 void *data);
344
345 void orangefs_kill_sb(struct super_block *sb);
346 int orangefs_remount(struct orangefs_sb_info_s *);
347
348 int fsid_key_table_initialize(void);
349 void fsid_key_table_finalize(void);
350
351 /*
352 * defined in inode.c
353 */
354 vm_fault_t orangefs_page_mkwrite(struct vm_fault *);
355 struct inode *orangefs_new_inode(struct super_block *sb,
356 struct inode *dir,
357 int mode,
358 dev_t dev,
359 struct orangefs_object_kref *ref);
360
361 int __orangefs_setattr(struct inode *, struct iattr *);
362 int orangefs_setattr(struct dentry *, struct iattr *);
363
364 int orangefs_getattr(const struct path *path, struct kstat *stat,
365 u32 request_mask, unsigned int flags);
366
367 int orangefs_permission(struct inode *inode, int mask);
368
369 int orangefs_update_time(struct inode *, struct timespec64 *, int);
370
371 /*
372 * defined in xattr.c
373 */
374 ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size);
375
376 /*
377 * defined in namei.c
378 */
379 struct inode *orangefs_iget(struct super_block *sb,
380 struct orangefs_object_kref *ref);
381
382 /*
383 * defined in devorangefs-req.c
384 */
385 extern uint32_t orangefs_userspace_version;
386
387 int orangefs_dev_init(void);
388 void orangefs_dev_cleanup(void);
389 int is_daemon_in_service(void);
390 bool __is_daemon_in_service(void);
391
392 /*
393 * defined in file.c
394 */
395 int orangefs_revalidate_mapping(struct inode *);
396 ssize_t wait_for_direct_io(enum ORANGEFS_io_type, struct inode *, loff_t *,
397 struct iov_iter *, size_t, loff_t, struct orangefs_write_range *, int *,
398 struct file *);
399 ssize_t do_readv_writev(enum ORANGEFS_io_type, struct file *, loff_t *,
400 struct iov_iter *);
401
402 /*
403 * defined in orangefs-utils.c
404 */
405 __s32 fsid_of_op(struct orangefs_kernel_op_s *op);
406
407 ssize_t orangefs_inode_getxattr(struct inode *inode,
408 const char *name,
409 void *buffer,
410 size_t size);
411
412 int orangefs_inode_setxattr(struct inode *inode,
413 const char *name,
414 const void *value,
415 size_t size,
416 int flags);
417
418 #define ORANGEFS_GETATTR_NEW 1
419 #define ORANGEFS_GETATTR_SIZE 2
420
421 int orangefs_inode_getattr(struct inode *, int);
422
423 int orangefs_inode_check_changed(struct inode *inode);
424
425 int orangefs_inode_setattr(struct inode *inode);
426
427 bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op);
428
429 int orangefs_normalize_to_errno(__s32 error_code);
430
431 extern struct mutex orangefs_request_mutex;
432 extern int op_timeout_secs;
433 extern int slot_timeout_secs;
434 extern int orangefs_cache_timeout_msecs;
435 extern int orangefs_dcache_timeout_msecs;
436 extern int orangefs_getattr_timeout_msecs;
437 extern struct list_head orangefs_superblocks;
438 extern spinlock_t orangefs_superblocks_lock;
439 extern struct list_head orangefs_request_list;
440 extern spinlock_t orangefs_request_list_lock;
441 extern wait_queue_head_t orangefs_request_list_waitq;
442 extern struct list_head *orangefs_htable_ops_in_progress;
443 extern spinlock_t orangefs_htable_ops_in_progress_lock;
444 extern int hash_table_size;
445
446 extern const struct file_operations orangefs_file_operations;
447 extern const struct inode_operations orangefs_symlink_inode_operations;
448 extern const struct inode_operations orangefs_dir_inode_operations;
449 extern const struct file_operations orangefs_dir_operations;
450 extern const struct dentry_operations orangefs_dentry_operations;
451
452 /*
453 * misc convenience macros
454 */
455
456 #define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */
457 #define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */
458 #define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */
459 #define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */
460 #define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */
461 #define ORANGEFS_OP_WRITEBACK 32
462
463 int service_operation(struct orangefs_kernel_op_s *op,
464 const char *op_name,
465 int flags);
466
467 #define get_interruptible_flag(inode) \
468 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \
469 ORANGEFS_OP_INTERRUPTIBLE : 0)
470
471 #define fill_default_sys_attrs(sys_attr, type, mode) \
472 do { \
473 sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \
474 sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \
475 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \
476 sys_attr.mtime = 0; \
477 sys_attr.atime = 0; \
478 sys_attr.ctime = 0; \
479 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \
480 } while (0)
481
orangefs_set_timeout(struct dentry * dentry)482 static inline void orangefs_set_timeout(struct dentry *dentry)
483 {
484 unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000;
485
486 dentry->d_fsdata = (void *) time;
487 }
488
489 #endif /* __ORANGEFSKERNEL_H */
490