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
55 #include <asm/unaligned.h>
56
57 #include "orangefs-dev-proto.h"
58
59 #define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 20
60
61 #define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS 30
62
63 #define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS 900 /* 15 minutes */
64
65 #define ORANGEFS_REQDEVICE_NAME "pvfs2-req"
66
67 #define ORANGEFS_DEVREQ_MAGIC 0x20030529
68 #define ORANGEFS_PURGE_RETRY_COUNT 0x00000005
69
70 #define MAX_DEV_REQ_UPSIZE (2 * sizeof(__s32) + \
71 sizeof(__u64) + sizeof(struct orangefs_upcall_s))
72 #define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \
73 sizeof(__u64) + sizeof(struct orangefs_downcall_s))
74
75 /*
76 * valid orangefs kernel operation states
77 *
78 * unknown - op was just initialized
79 * waiting - op is on request_list (upward bound)
80 * inprogr - op is in progress (waiting for downcall)
81 * serviced - op has matching downcall; ok
82 * purged - op has to start a timer since client-core
83 * exited uncleanly before servicing op
84 * given up - submitter has given up waiting for it
85 */
86 enum orangefs_vfs_op_states {
87 OP_VFS_STATE_UNKNOWN = 0,
88 OP_VFS_STATE_WAITING = 1,
89 OP_VFS_STATE_INPROGR = 2,
90 OP_VFS_STATE_SERVICED = 4,
91 OP_VFS_STATE_PURGED = 8,
92 OP_VFS_STATE_GIVEN_UP = 16,
93 };
94
95 /*
96 * orangefs kernel memory related flags
97 */
98
99 #if (defined CONFIG_DEBUG_SLAB)
100 #define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE
101 #else
102 #define ORANGEFS_CACHE_CREATE_FLAGS 0
103 #endif
104
105 extern int orangefs_init_acl(struct inode *inode, struct inode *dir);
106 extern const struct xattr_handler *orangefs_xattr_handlers[];
107
108 extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type);
109 extern int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
110
111 /*
112 * orangefs data structures
113 */
114 struct orangefs_kernel_op_s {
115 enum orangefs_vfs_op_states op_state;
116 __u64 tag;
117
118 /*
119 * Set uses_shared_memory to non zero if this operation uses
120 * shared memory. If true, then a retry on the op must also
121 * get a new shared memory buffer and re-populate it.
122 * Cancels don't care - it only matters for service_operation()
123 * retry logics and cancels don't go through it anymore. It
124 * safely stays non-zero when we use it as slot_to_free.
125 */
126 union {
127 int uses_shared_memory;
128 int slot_to_free;
129 };
130
131 struct orangefs_upcall_s upcall;
132 struct orangefs_downcall_s downcall;
133
134 struct completion waitq;
135 spinlock_t lock;
136
137 int attempts;
138
139 struct list_head list;
140 };
141
142 #define set_op_state_waiting(op) ((op)->op_state = OP_VFS_STATE_WAITING)
143 #define set_op_state_inprogress(op) ((op)->op_state = OP_VFS_STATE_INPROGR)
144 #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)145 static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op)
146 {
147 op->op_state = OP_VFS_STATE_SERVICED;
148 complete(&op->waitq);
149 }
150
151 #define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING)
152 #define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR)
153 #define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED)
154 #define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED)
155 #define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP)
156 #define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL)
157
158 void op_release(struct orangefs_kernel_op_s *op);
159
160 extern void orangefs_bufmap_put(int);
put_cancel(struct orangefs_kernel_op_s * op)161 static inline void put_cancel(struct orangefs_kernel_op_s *op)
162 {
163 orangefs_bufmap_put(op->slot_to_free);
164 op_release(op);
165 }
166
set_op_state_purged(struct orangefs_kernel_op_s * op)167 static inline void set_op_state_purged(struct orangefs_kernel_op_s *op)
168 {
169 spin_lock(&op->lock);
170 if (unlikely(op_is_cancel(op))) {
171 list_del_init(&op->list);
172 spin_unlock(&op->lock);
173 put_cancel(op);
174 } else {
175 op->op_state |= OP_VFS_STATE_PURGED;
176 complete(&op->waitq);
177 spin_unlock(&op->lock);
178 }
179 }
180
181 /* per inode private orangefs info */
182 struct orangefs_inode_s {
183 struct orangefs_object_kref refn;
184 char link_target[ORANGEFS_NAME_MAX];
185 /*
186 * Reading/Writing Extended attributes need to acquire the appropriate
187 * reader/writer semaphore on the orangefs_inode_s structure.
188 */
189 struct rw_semaphore xattr_sem;
190
191 struct inode vfs_inode;
192 sector_t last_failed_block_index_read;
193
194 unsigned long getattr_time;
195 u32 getattr_mask;
196 };
197
198 /* per superblock private orangefs info */
199 struct orangefs_sb_info_s {
200 struct orangefs_khandle root_khandle;
201 __s32 fs_id;
202 int id;
203 int flags;
204 #define ORANGEFS_OPT_INTR 0x01
205 #define ORANGEFS_OPT_LOCAL_LOCK 0x02
206 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN];
207 struct super_block *sb;
208 int mount_pending;
209 int no_list;
210 struct list_head list;
211 };
212
213 struct orangefs_stats {
214 unsigned long cache_hits;
215 unsigned long cache_misses;
216 unsigned long reads;
217 unsigned long writes;
218 };
219
220 extern struct orangefs_stats orangefs_stats;
221
222 /*
223 * NOTE: See Documentation/filesystems/porting for information
224 * on implementing FOO_I and properly accessing fs private data
225 */
ORANGEFS_I(struct inode * inode)226 static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode)
227 {
228 return container_of(inode, struct orangefs_inode_s, vfs_inode);
229 }
230
ORANGEFS_SB(struct super_block * sb)231 static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb)
232 {
233 return (struct orangefs_sb_info_s *) sb->s_fs_info;
234 }
235
236 /* ino_t descends from "unsigned long", 8 bytes, 64 bits. */
orangefs_khandle_to_ino(struct orangefs_khandle * khandle)237 static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle)
238 {
239 union {
240 unsigned char u[8];
241 __u64 ino;
242 } ihandle;
243
244 ihandle.u[0] = khandle->u[0] ^ khandle->u[4];
245 ihandle.u[1] = khandle->u[1] ^ khandle->u[5];
246 ihandle.u[2] = khandle->u[2] ^ khandle->u[6];
247 ihandle.u[3] = khandle->u[3] ^ khandle->u[7];
248 ihandle.u[4] = khandle->u[12] ^ khandle->u[8];
249 ihandle.u[5] = khandle->u[13] ^ khandle->u[9];
250 ihandle.u[6] = khandle->u[14] ^ khandle->u[10];
251 ihandle.u[7] = khandle->u[15] ^ khandle->u[11];
252
253 return ihandle.ino;
254 }
255
get_khandle_from_ino(struct inode * inode)256 static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode)
257 {
258 return &(ORANGEFS_I(inode)->refn.khandle);
259 }
260
is_root_handle(struct inode * inode)261 static inline int is_root_handle(struct inode *inode)
262 {
263 gossip_debug(GOSSIP_DCACHE_DEBUG,
264 "%s: root handle: %pU, this handle: %pU:\n",
265 __func__,
266 &ORANGEFS_SB(inode->i_sb)->root_khandle,
267 get_khandle_from_ino(inode));
268
269 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle),
270 get_khandle_from_ino(inode)))
271 return 0;
272 else
273 return 1;
274 }
275
match_handle(struct orangefs_khandle resp_handle,struct inode * inode)276 static inline int match_handle(struct orangefs_khandle resp_handle,
277 struct inode *inode)
278 {
279 gossip_debug(GOSSIP_DCACHE_DEBUG,
280 "%s: one handle: %pU, another handle:%pU:\n",
281 __func__,
282 &resp_handle,
283 get_khandle_from_ino(inode));
284
285 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode)))
286 return 0;
287 else
288 return 1;
289 }
290
291 /*
292 * defined in orangefs-cache.c
293 */
294 int op_cache_initialize(void);
295 int op_cache_finalize(void);
296 struct orangefs_kernel_op_s *op_alloc(__s32 type);
297 void orangefs_new_tag(struct orangefs_kernel_op_s *op);
298 char *get_opname_string(struct orangefs_kernel_op_s *new_op);
299
300 int orangefs_inode_cache_initialize(void);
301 int orangefs_inode_cache_finalize(void);
302
303 /*
304 * defined in orangefs-mod.c
305 */
306 void purge_inprogress_ops(void);
307
308 /*
309 * defined in waitqueue.c
310 */
311 void purge_waiting_ops(void);
312
313 /*
314 * defined in super.c
315 */
316 extern uint64_t orangefs_features;
317
318 struct dentry *orangefs_mount(struct file_system_type *fst,
319 int flags,
320 const char *devname,
321 void *data);
322
323 void orangefs_kill_sb(struct super_block *sb);
324 int orangefs_remount(struct orangefs_sb_info_s *);
325
326 int fsid_key_table_initialize(void);
327 void fsid_key_table_finalize(void);
328
329 /*
330 * defined in inode.c
331 */
332 struct inode *orangefs_new_inode(struct super_block *sb,
333 struct inode *dir,
334 int mode,
335 dev_t dev,
336 struct orangefs_object_kref *ref);
337
338 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr);
339
340 int orangefs_getattr(const struct path *path, struct kstat *stat,
341 u32 request_mask, unsigned int flags);
342
343 int orangefs_permission(struct inode *inode, int mask);
344
345 int orangefs_update_time(struct inode *, struct timespec64 *, int);
346
347 /*
348 * defined in xattr.c
349 */
350 ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size);
351
352 /*
353 * defined in namei.c
354 */
355 struct inode *orangefs_iget(struct super_block *sb,
356 struct orangefs_object_kref *ref);
357
358 ssize_t orangefs_inode_read(struct inode *inode,
359 struct iov_iter *iter,
360 loff_t *offset,
361 loff_t readahead_size);
362
363 /*
364 * defined in devorangefs-req.c
365 */
366 extern uint32_t orangefs_userspace_version;
367
368 int orangefs_dev_init(void);
369 void orangefs_dev_cleanup(void);
370 int is_daemon_in_service(void);
371 bool __is_daemon_in_service(void);
372
373 /*
374 * defined in orangefs-utils.c
375 */
376 __s32 fsid_of_op(struct orangefs_kernel_op_s *op);
377
378 ssize_t orangefs_inode_getxattr(struct inode *inode,
379 const char *name,
380 void *buffer,
381 size_t size);
382
383 int orangefs_inode_setxattr(struct inode *inode,
384 const char *name,
385 const void *value,
386 size_t size,
387 int flags);
388
389 int orangefs_inode_getattr(struct inode *inode, int new, int bypass,
390 u32 request_mask);
391
392 int orangefs_inode_check_changed(struct inode *inode);
393
394 int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr);
395
396 bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op);
397
398 int orangefs_normalize_to_errno(__s32 error_code);
399
400 extern struct mutex orangefs_request_mutex;
401 extern int op_timeout_secs;
402 extern int slot_timeout_secs;
403 extern int orangefs_dcache_timeout_msecs;
404 extern int orangefs_getattr_timeout_msecs;
405 extern struct list_head orangefs_superblocks;
406 extern spinlock_t orangefs_superblocks_lock;
407 extern struct list_head orangefs_request_list;
408 extern spinlock_t orangefs_request_list_lock;
409 extern wait_queue_head_t orangefs_request_list_waitq;
410 extern struct list_head *orangefs_htable_ops_in_progress;
411 extern spinlock_t orangefs_htable_ops_in_progress_lock;
412 extern int hash_table_size;
413
414 extern const struct file_operations orangefs_file_operations;
415 extern const struct inode_operations orangefs_symlink_inode_operations;
416 extern const struct inode_operations orangefs_dir_inode_operations;
417 extern const struct file_operations orangefs_dir_operations;
418 extern const struct dentry_operations orangefs_dentry_operations;
419
420 /*
421 * misc convenience macros
422 */
423
424 #define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */
425 #define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */
426 #define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */
427 #define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */
428 #define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */
429
430 int service_operation(struct orangefs_kernel_op_s *op,
431 const char *op_name,
432 int flags);
433
434 #define get_interruptible_flag(inode) \
435 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \
436 ORANGEFS_OP_INTERRUPTIBLE : 0)
437
438 #define fill_default_sys_attrs(sys_attr, type, mode) \
439 do { \
440 sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \
441 sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \
442 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \
443 sys_attr.mtime = 0; \
444 sys_attr.atime = 0; \
445 sys_attr.ctime = 0; \
446 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \
447 } while (0)
448
orangefs_set_timeout(struct dentry * dentry)449 static inline void orangefs_set_timeout(struct dentry *dentry)
450 {
451 unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000;
452
453 dentry->d_fsdata = (void *) time;
454 }
455
456 #endif /* __ORANGEFSKERNEL_H */
457