orangefs: do not invalidate attributes on inode create
[linux-block.git] / fs / orangefs / orangefs-kernel.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
f7ab093f
MM
2/*
3 * (C) 2001 Clemson University and The University of Chicago
4 *
5 * See COPYING in top-level directory.
6 */
7
8/*
8bb8aefd 9 * The ORANGEFS Linux kernel support allows ORANGEFS volumes to be mounted and
f7ab093f
MM
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/*
8bb8aefd 16 * Declarations and macros for the ORANGEFS Linux kernel support.
f7ab093f
MM
17 */
18
8bb8aefd
YL
19#ifndef __ORANGEFSKERNEL_H
20#define __ORANGEFSKERNEL_H
f7ab093f
MM
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>
174cd4b1 45#include <linux/sched/signal.h>
f7ab093f
MM
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>
fc2e2e9c 54#include <linux/hashtable.h>
f7ab093f
MM
55
56#include <asm/unaligned.h>
57
575e9461 58#include "orangefs-dev-proto.h"
f7ab093f 59
8bb8aefd 60#define ORANGEFS_DEFAULT_OP_TIMEOUT_SECS 20
f7ab093f 61
8bb8aefd 62#define ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS 30
f7ab093f 63
8bb8aefd 64#define ORANGEFS_DEFAULT_SLOT_TIMEOUT_SECS 900 /* 15 minutes */
f7ab093f 65
8bb8aefd 66#define ORANGEFS_REQDEVICE_NAME "pvfs2-req"
f7ab093f 67
8bb8aefd 68#define ORANGEFS_DEVREQ_MAGIC 0x20030529
8bb8aefd 69#define ORANGEFS_PURGE_RETRY_COUNT 0x00000005
f7ab093f 70
cf0c2771 71#define MAX_DEV_REQ_UPSIZE (2 * sizeof(__s32) + \
8bb8aefd 72sizeof(__u64) + sizeof(struct orangefs_upcall_s))
cf0c2771 73#define MAX_DEV_REQ_DOWNSIZE (2 * sizeof(__s32) + \
8bb8aefd 74sizeof(__u64) + sizeof(struct orangefs_downcall_s))
f7ab093f 75
f7ab093f 76/*
8bb8aefd 77 * valid orangefs kernel operation states
f7ab093f
MM
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
ed42fe05 85 * given up - submitter has given up waiting for it
f7ab093f 86 */
8bb8aefd 87enum orangefs_vfs_op_states {
f7ab093f
MM
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,
ed42fe05 93 OP_VFS_STATE_GIVEN_UP = 16,
f7ab093f
MM
94};
95
f7ab093f 96/*
8bb8aefd 97 * orangefs kernel memory related flags
f7ab093f
MM
98 */
99
538e3048 100#if (defined CONFIG_DEBUG_SLAB)
8bb8aefd 101#define ORANGEFS_CACHE_CREATE_FLAGS SLAB_RED_ZONE
f7ab093f 102#else
8bb8aefd 103#define ORANGEFS_CACHE_CREATE_FLAGS 0
538e3048 104#endif
f7ab093f 105
8bb8aefd
YL
106extern int orangefs_init_acl(struct inode *inode, struct inode *dir);
107extern const struct xattr_handler *orangefs_xattr_handlers[];
f7ab093f 108
8bb8aefd
YL
109extern struct posix_acl *orangefs_get_acl(struct inode *inode, int type);
110extern int orangefs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
f7ab093f 111
f7ab093f 112/*
8bb8aefd 113 * orangefs data structures
f7ab093f 114 */
8bb8aefd
YL
115struct orangefs_kernel_op_s {
116 enum orangefs_vfs_op_states op_state;
f7ab093f
MM
117 __u64 tag;
118
119 /*
d37c0f30
MM
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.
f7ab093f 126 */
78699e29
AV
127 union {
128 int uses_shared_memory;
129 int slot_to_free;
130 };
f7ab093f 131
8bb8aefd
YL
132 struct orangefs_upcall_s upcall;
133 struct orangefs_downcall_s downcall;
f7ab093f 134
d2d87a3b 135 struct completion waitq;
f7ab093f
MM
136 spinlock_t lock;
137
f7ab093f
MM
138 int attempts;
139
140 struct list_head list;
141};
142
60831949
AV
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)
2d4cae0d 145#define set_op_state_given_up(op) ((op)->op_state = OP_VFS_STATE_GIVEN_UP)
60831949
AV
146static inline void set_op_state_serviced(struct orangefs_kernel_op_s *op)
147{
148 op->op_state = OP_VFS_STATE_SERVICED;
d2d87a3b 149 complete(&op->waitq);
60831949 150}
60831949
AV
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)
ed42fe05 156#define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP)
78699e29 157#define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL)
ed42fe05 158
c1223ca4 159void op_release(struct orangefs_kernel_op_s *op);
60831949 160
78699e29
AV
161extern void orangefs_bufmap_put(int);
162static 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
168static 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))) {
05a50a5b 172 list_del_init(&op->list);
78699e29
AV
173 spin_unlock(&op->lock);
174 put_cancel(op);
175 } else {
176 op->op_state |= OP_VFS_STATE_PURGED;
d2d87a3b 177 complete(&op->waitq);
78699e29
AV
178 spin_unlock(&op->lock);
179 }
180}
181
8bb8aefd
YL
182/* per inode private orangefs info */
183struct orangefs_inode_s {
184 struct orangefs_object_kref refn;
185 char link_target[ORANGEFS_NAME_MAX];
f7ab093f
MM
186 /*
187 * Reading/Writing Extended attributes need to acquire the appropriate
8bb8aefd 188 * reader/writer semaphore on the orangefs_inode_s structure.
f7ab093f
MM
189 */
190 struct rw_semaphore xattr_sem;
191
192 struct inode vfs_inode;
193 sector_t last_failed_block_index_read;
194
71680c18 195 unsigned long getattr_time;
68a24a6c 196 u32 getattr_mask;
fc2e2e9c
MB
197
198 DECLARE_HASHTABLE(xattr_cache, 4);
f7ab093f
MM
199};
200
8bb8aefd
YL
201/* per superblock private orangefs info */
202struct orangefs_sb_info_s {
203 struct orangefs_khandle root_khandle;
f7ab093f
MM
204 __s32 fs_id;
205 int id;
206 int flags;
8bb8aefd
YL
207#define ORANGEFS_OPT_INTR 0x01
208#define ORANGEFS_OPT_LOCAL_LOCK 0x02
209 char devname[ORANGEFS_MAX_SERVER_ADDR_LEN];
f7ab093f
MM
210 struct super_block *sb;
211 int mount_pending;
1ec1688c 212 int no_list;
f7ab093f
MM
213 struct list_head list;
214};
215
8bb8aefd 216struct orangefs_stats {
f7ab093f
MM
217 unsigned long cache_hits;
218 unsigned long cache_misses;
219 unsigned long reads;
220 unsigned long writes;
221};
222
fc2e2e9c
MB
223struct orangefs_cached_xattr {
224 struct hlist_node node;
225 char key[ORANGEFS_MAX_XATTR_NAMELEN];
226 char val[ORANGEFS_MAX_XATTR_VALUELEN];
227 ssize_t length;
228 unsigned long timeout;
229};
230
889d5f1b 231extern struct orangefs_stats orangefs_stats;
f7ab093f
MM
232
233/*
54804949
MM
234 * NOTE: See Documentation/filesystems/porting for information
235 * on implementing FOO_I and properly accessing fs private data
236 */
8bb8aefd 237static inline struct orangefs_inode_s *ORANGEFS_I(struct inode *inode)
f7ab093f 238{
8bb8aefd 239 return container_of(inode, struct orangefs_inode_s, vfs_inode);
f7ab093f
MM
240}
241
8bb8aefd 242static inline struct orangefs_sb_info_s *ORANGEFS_SB(struct super_block *sb)
f7ab093f 243{
8bb8aefd 244 return (struct orangefs_sb_info_s *) sb->s_fs_info;
f7ab093f
MM
245}
246
247/* ino_t descends from "unsigned long", 8 bytes, 64 bits. */
8bb8aefd 248static inline ino_t orangefs_khandle_to_ino(struct orangefs_khandle *khandle)
f7ab093f
MM
249{
250 union {
251 unsigned char u[8];
252 __u64 ino;
253 } ihandle;
254
255 ihandle.u[0] = khandle->u[0] ^ khandle->u[4];
256 ihandle.u[1] = khandle->u[1] ^ khandle->u[5];
257 ihandle.u[2] = khandle->u[2] ^ khandle->u[6];
258 ihandle.u[3] = khandle->u[3] ^ khandle->u[7];
259 ihandle.u[4] = khandle->u[12] ^ khandle->u[8];
260 ihandle.u[5] = khandle->u[13] ^ khandle->u[9];
261 ihandle.u[6] = khandle->u[14] ^ khandle->u[10];
262 ihandle.u[7] = khandle->u[15] ^ khandle->u[11];
263
264 return ihandle.ino;
265}
266
8bb8aefd 267static inline struct orangefs_khandle *get_khandle_from_ino(struct inode *inode)
f7ab093f 268{
8bb8aefd 269 return &(ORANGEFS_I(inode)->refn.khandle);
f7ab093f
MM
270}
271
f7ab093f
MM
272static inline int is_root_handle(struct inode *inode)
273{
274 gossip_debug(GOSSIP_DCACHE_DEBUG,
275 "%s: root handle: %pU, this handle: %pU:\n",
276 __func__,
8bb8aefd 277 &ORANGEFS_SB(inode->i_sb)->root_khandle,
f7ab093f
MM
278 get_khandle_from_ino(inode));
279
8bb8aefd 280 if (ORANGEFS_khandle_cmp(&(ORANGEFS_SB(inode->i_sb)->root_khandle),
f7ab093f
MM
281 get_khandle_from_ino(inode)))
282 return 0;
283 else
284 return 1;
285}
286
8bb8aefd 287static inline int match_handle(struct orangefs_khandle resp_handle,
f7ab093f
MM
288 struct inode *inode)
289{
290 gossip_debug(GOSSIP_DCACHE_DEBUG,
291 "%s: one handle: %pU, another handle:%pU:\n",
292 __func__,
293 &resp_handle,
294 get_khandle_from_ino(inode));
295
8bb8aefd 296 if (ORANGEFS_khandle_cmp(&resp_handle, get_khandle_from_ino(inode)))
f7ab093f
MM
297 return 0;
298 else
299 return 1;
300}
301
302/*
8bb8aefd 303 * defined in orangefs-cache.c
f7ab093f
MM
304 */
305int op_cache_initialize(void);
306int op_cache_finalize(void);
8bb8aefd 307struct orangefs_kernel_op_s *op_alloc(__s32 type);
78699e29 308void orangefs_new_tag(struct orangefs_kernel_op_s *op);
8bb8aefd 309char *get_opname_string(struct orangefs_kernel_op_s *new_op);
f7ab093f 310
8bb8aefd
YL
311int orangefs_inode_cache_initialize(void);
312int orangefs_inode_cache_finalize(void);
f7ab093f 313
f7ab093f 314/*
8bb8aefd 315 * defined in orangefs-mod.c
f7ab093f
MM
316 */
317void purge_inprogress_ops(void);
318
319/*
320 * defined in waitqueue.c
321 */
f7ab093f
MM
322void purge_waiting_ops(void);
323
324/*
325 * defined in super.c
326 */
482664dd
MB
327extern uint64_t orangefs_features;
328
8bb8aefd 329struct dentry *orangefs_mount(struct file_system_type *fst,
f7ab093f
MM
330 int flags,
331 const char *devname,
332 void *data);
333
8bb8aefd 334void orangefs_kill_sb(struct super_block *sb);
45996492 335int orangefs_remount(struct orangefs_sb_info_s *);
f7ab093f
MM
336
337int fsid_key_table_initialize(void);
338void fsid_key_table_finalize(void);
339
340/*
341 * defined in inode.c
342 */
8bb8aefd 343struct inode *orangefs_new_inode(struct super_block *sb,
f7ab093f
MM
344 struct inode *dir,
345 int mode,
346 dev_t dev,
8bb8aefd 347 struct orangefs_object_kref *ref);
f7ab093f 348
8bb8aefd 349int orangefs_setattr(struct dentry *dentry, struct iattr *iattr);
f7ab093f 350
a528d35e
DH
351int orangefs_getattr(const struct path *path, struct kstat *stat,
352 u32 request_mask, unsigned int flags);
f7ab093f 353
933287da
MB
354int orangefs_permission(struct inode *inode, int mask);
355
95582b00 356int orangefs_update_time(struct inode *, struct timespec64 *, int);
a55f2d86 357
f7ab093f
MM
358/*
359 * defined in xattr.c
360 */
8bb8aefd 361ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size);
f7ab093f
MM
362
363/*
364 * defined in namei.c
365 */
8bb8aefd
YL
366struct inode *orangefs_iget(struct super_block *sb,
367 struct orangefs_object_kref *ref);
f7ab093f 368
8bb8aefd
YL
369ssize_t orangefs_inode_read(struct inode *inode,
370 struct iov_iter *iter,
371 loff_t *offset,
372 loff_t readahead_size);
f7ab093f
MM
373
374/*
8bb8aefd 375 * defined in devorangefs-req.c
f7ab093f 376 */
482664dd 377extern uint32_t orangefs_userspace_version;
f2ee3b75 378
8bb8aefd
YL
379int orangefs_dev_init(void);
380void orangefs_dev_cleanup(void);
f7ab093f 381int is_daemon_in_service(void);
78699e29 382bool __is_daemon_in_service(void);
f7ab093f
MM
383
384/*
8bb8aefd 385 * defined in orangefs-utils.c
f7ab093f 386 */
8bb8aefd 387__s32 fsid_of_op(struct orangefs_kernel_op_s *op);
f7ab093f 388
8bb8aefd 389ssize_t orangefs_inode_getxattr(struct inode *inode,
f7ab093f
MM
390 const char *name,
391 void *buffer,
392 size_t size);
393
8bb8aefd 394int orangefs_inode_setxattr(struct inode *inode,
f7ab093f
MM
395 const char *name,
396 const void *value,
397 size_t size,
398 int flags);
399
68a24a6c
MB
400int orangefs_inode_getattr(struct inode *inode, int new, int bypass,
401 u32 request_mask);
f7ab093f 402
5859d77e
MB
403int orangefs_inode_check_changed(struct inode *inode);
404
8bb8aefd 405int orangefs_inode_setattr(struct inode *inode, struct iattr *iattr);
f7ab093f 406
78699e29 407bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op);
f7ab093f 408
8bb8aefd 409int orangefs_normalize_to_errno(__s32 error_code);
f7ab093f 410
1d503617 411extern struct mutex orangefs_request_mutex;
f7ab093f
MM
412extern int op_timeout_secs;
413extern int slot_timeout_secs;
1d503617
MB
414extern int orangefs_dcache_timeout_msecs;
415extern int orangefs_getattr_timeout_msecs;
8bb8aefd
YL
416extern struct list_head orangefs_superblocks;
417extern spinlock_t orangefs_superblocks_lock;
418extern struct list_head orangefs_request_list;
419extern spinlock_t orangefs_request_list_lock;
420extern wait_queue_head_t orangefs_request_list_waitq;
1d503617
MB
421extern struct list_head *orangefs_htable_ops_in_progress;
422extern spinlock_t orangefs_htable_ops_in_progress_lock;
f7ab093f
MM
423extern int hash_table_size;
424
8bb8aefd 425extern const struct file_operations orangefs_file_operations;
6f3fc107
AV
426extern const struct inode_operations orangefs_symlink_inode_operations;
427extern const struct inode_operations orangefs_dir_inode_operations;
8bb8aefd
YL
428extern const struct file_operations orangefs_dir_operations;
429extern const struct dentry_operations orangefs_dentry_operations;
f7ab093f 430
f7ab093f
MM
431/*
432 * misc convenience macros
433 */
f7ab093f 434
8bb8aefd
YL
435#define ORANGEFS_OP_INTERRUPTIBLE 1 /* service_operation() is interruptible */
436#define ORANGEFS_OP_PRIORITY 2 /* service_operation() is high priority */
437#define ORANGEFS_OP_CANCELLATION 4 /* this is a cancellation */
adcf34a2 438#define ORANGEFS_OP_NO_MUTEX 8 /* don't acquire request_mutex */
8bb8aefd 439#define ORANGEFS_OP_ASYNC 16 /* Queue it, but don't wait */
f7ab093f 440
8bb8aefd 441int service_operation(struct orangefs_kernel_op_s *op,
f7ab093f
MM
442 const char *op_name,
443 int flags);
444
f7ab093f 445#define get_interruptible_flag(inode) \
8bb8aefd
YL
446 ((ORANGEFS_SB(inode->i_sb)->flags & ORANGEFS_OPT_INTR) ? \
447 ORANGEFS_OP_INTERRUPTIBLE : 0)
f7ab093f 448
f7ab093f
MM
449#define fill_default_sys_attrs(sys_attr, type, mode) \
450do { \
78fee0b6
JH
451 sys_attr.owner = from_kuid(&init_user_ns, current_fsuid()); \
452 sys_attr.group = from_kgid(&init_user_ns, current_fsgid()); \
8bb8aefd 453 sys_attr.perms = ORANGEFS_util_translate_mode(mode); \
69a23de2
MB
454 sys_attr.mtime = 0; \
455 sys_attr.atime = 0; \
456 sys_attr.ctime = 0; \
8bb8aefd 457 sys_attr.mask = ORANGEFS_ATTR_SYS_ALL_SETABLE; \
f7ab093f
MM
458} while (0)
459
804b1737
MS
460static inline void orangefs_set_timeout(struct dentry *dentry)
461{
462 unsigned long time = jiffies + orangefs_dcache_timeout_msecs*HZ/1000;
463
464 dentry->d_fsdata = (void *) time;
465}
466
8bb8aefd 467#endif /* __ORANGEFSKERNEL_H */