1 // SPDX-License-Identifier: GPL-2.0-only <<
2 /* 1 /*
3 * (C) 1997 Linus Torvalds 2 * (C) 1997 Linus Torvalds
4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> 3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 */ 4 */
6 #include <linux/export.h> 5 #include <linux/export.h>
7 #include <linux/fs.h> 6 #include <linux/fs.h>
8 #include <linux/filelock.h> <<
9 #include <linux/mm.h> 7 #include <linux/mm.h>
10 #include <linux/backing-dev.h> 8 #include <linux/backing-dev.h>
11 #include <linux/hash.h> 9 #include <linux/hash.h>
12 #include <linux/swap.h> 10 #include <linux/swap.h>
13 #include <linux/security.h> 11 #include <linux/security.h>
14 #include <linux/cdev.h> 12 #include <linux/cdev.h>
15 #include <linux/memblock.h> !! 13 #include <linux/bootmem.h>
16 #include <linux/fsnotify.h> 14 #include <linux/fsnotify.h>
17 #include <linux/mount.h> 15 #include <linux/mount.h>
18 #include <linux/posix_acl.h> 16 #include <linux/posix_acl.h>
>> 17 #include <linux/prefetch.h>
19 #include <linux/buffer_head.h> /* for inode_ha 18 #include <linux/buffer_head.h> /* for inode_has_buffers */
20 #include <linux/ratelimit.h> 19 #include <linux/ratelimit.h>
21 #include <linux/list_lru.h> 20 #include <linux/list_lru.h>
22 #include <linux/iversion.h> 21 #include <linux/iversion.h>
23 #include <linux/rw_hint.h> <<
24 #include <trace/events/writeback.h> 22 #include <trace/events/writeback.h>
25 #include "internal.h" 23 #include "internal.h"
26 24
27 /* 25 /*
28 * Inode locking rules: 26 * Inode locking rules:
29 * 27 *
30 * inode->i_lock protects: 28 * inode->i_lock protects:
31 * inode->i_state, inode->i_hash, __iget(), !! 29 * inode->i_state, inode->i_hash, __iget()
32 * Inode LRU list locks protect: 30 * Inode LRU list locks protect:
33 * inode->i_sb->s_inode_lru, inode->i_lru 31 * inode->i_sb->s_inode_lru, inode->i_lru
34 * inode->i_sb->s_inode_list_lock protects: 32 * inode->i_sb->s_inode_list_lock protects:
35 * inode->i_sb->s_inodes, inode->i_sb_list 33 * inode->i_sb->s_inodes, inode->i_sb_list
36 * bdi->wb.list_lock protects: 34 * bdi->wb.list_lock protects:
37 * bdi->wb.b_{dirty,io,more_io,dirty_time}, 35 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
38 * inode_hash_lock protects: 36 * inode_hash_lock protects:
39 * inode_hashtable, inode->i_hash 37 * inode_hashtable, inode->i_hash
40 * 38 *
41 * Lock ordering: 39 * Lock ordering:
42 * 40 *
43 * inode->i_sb->s_inode_list_lock 41 * inode->i_sb->s_inode_list_lock
44 * inode->i_lock 42 * inode->i_lock
45 * Inode LRU list locks 43 * Inode LRU list locks
46 * 44 *
47 * bdi->wb.list_lock 45 * bdi->wb.list_lock
48 * inode->i_lock 46 * inode->i_lock
49 * 47 *
50 * inode_hash_lock 48 * inode_hash_lock
51 * inode->i_sb->s_inode_list_lock 49 * inode->i_sb->s_inode_list_lock
52 * inode->i_lock 50 * inode->i_lock
53 * 51 *
54 * iunique_lock 52 * iunique_lock
55 * inode_hash_lock 53 * inode_hash_lock
56 */ 54 */
57 55
58 static unsigned int i_hash_mask __ro_after_ini !! 56 static unsigned int i_hash_mask __read_mostly;
59 static unsigned int i_hash_shift __ro_after_in !! 57 static unsigned int i_hash_shift __read_mostly;
60 static struct hlist_head *inode_hashtable __ro !! 58 static struct hlist_head *inode_hashtable __read_mostly;
61 static __cacheline_aligned_in_smp DEFINE_SPINL 59 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
62 60
63 /* 61 /*
64 * Empty aops. Can be used for the cases where 62 * Empty aops. Can be used for the cases where the user does not
65 * define any of the address_space operations. 63 * define any of the address_space operations.
66 */ 64 */
67 const struct address_space_operations empty_ao 65 const struct address_space_operations empty_aops = {
68 }; 66 };
69 EXPORT_SYMBOL(empty_aops); 67 EXPORT_SYMBOL(empty_aops);
70 68
>> 69 /*
>> 70 * Statistics gathering..
>> 71 */
>> 72 struct inodes_stat_t inodes_stat;
>> 73
71 static DEFINE_PER_CPU(unsigned long, nr_inodes 74 static DEFINE_PER_CPU(unsigned long, nr_inodes);
72 static DEFINE_PER_CPU(unsigned long, nr_unused 75 static DEFINE_PER_CPU(unsigned long, nr_unused);
73 76
74 static struct kmem_cache *inode_cachep __ro_af !! 77 static struct kmem_cache *inode_cachep __read_mostly;
75 78
76 static long get_nr_inodes(void) 79 static long get_nr_inodes(void)
77 { 80 {
78 int i; 81 int i;
79 long sum = 0; 82 long sum = 0;
80 for_each_possible_cpu(i) 83 for_each_possible_cpu(i)
81 sum += per_cpu(nr_inodes, i); 84 sum += per_cpu(nr_inodes, i);
82 return sum < 0 ? 0 : sum; 85 return sum < 0 ? 0 : sum;
83 } 86 }
84 87
85 static inline long get_nr_inodes_unused(void) 88 static inline long get_nr_inodes_unused(void)
86 { 89 {
87 int i; 90 int i;
88 long sum = 0; 91 long sum = 0;
89 for_each_possible_cpu(i) 92 for_each_possible_cpu(i)
90 sum += per_cpu(nr_unused, i); 93 sum += per_cpu(nr_unused, i);
91 return sum < 0 ? 0 : sum; 94 return sum < 0 ? 0 : sum;
92 } 95 }
93 96
94 long get_nr_dirty_inodes(void) 97 long get_nr_dirty_inodes(void)
95 { 98 {
96 /* not actually dirty inodes, but a wi 99 /* not actually dirty inodes, but a wild approximation */
97 long nr_dirty = get_nr_inodes() - get_ 100 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
98 return nr_dirty > 0 ? nr_dirty : 0; 101 return nr_dirty > 0 ? nr_dirty : 0;
99 } 102 }
100 103
101 /* 104 /*
102 * Handle nr_inode sysctl 105 * Handle nr_inode sysctl
103 */ 106 */
104 #ifdef CONFIG_SYSCTL 107 #ifdef CONFIG_SYSCTL
105 /* !! 108 int proc_nr_inodes(struct ctl_table *table, int write,
106 * Statistics gathering.. !! 109 void __user *buffer, size_t *lenp, loff_t *ppos)
107 */ <<
108 static struct inodes_stat_t inodes_stat; <<
109 <<
110 static int proc_nr_inodes(const struct ctl_tab <<
111 size_t *lenp, loff_t <<
112 { 110 {
113 inodes_stat.nr_inodes = get_nr_inodes( 111 inodes_stat.nr_inodes = get_nr_inodes();
114 inodes_stat.nr_unused = get_nr_inodes_ 112 inodes_stat.nr_unused = get_nr_inodes_unused();
115 return proc_doulongvec_minmax(table, w 113 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
116 } 114 }
117 <<
118 static struct ctl_table inodes_sysctls[] = { <<
119 { <<
120 .procname = "inode-nr", <<
121 .data = &inodes_stat <<
122 .maxlen = 2*sizeof(lon <<
123 .mode = 0444, <<
124 .proc_handler = proc_nr_inod <<
125 }, <<
126 { <<
127 .procname = "inode-state <<
128 .data = &inodes_stat <<
129 .maxlen = 7*sizeof(lon <<
130 .mode = 0444, <<
131 .proc_handler = proc_nr_inod <<
132 }, <<
133 }; <<
134 <<
135 static int __init init_fs_inode_sysctls(void) <<
136 { <<
137 register_sysctl_init("fs", inodes_sysc <<
138 return 0; <<
139 } <<
140 early_initcall(init_fs_inode_sysctls); <<
141 #endif 115 #endif
142 116
143 static int no_open(struct inode *inode, struct 117 static int no_open(struct inode *inode, struct file *file)
144 { 118 {
145 return -ENXIO; 119 return -ENXIO;
146 } 120 }
147 121
148 /** 122 /**
149 * inode_init_always_gfp - perform inode struc !! 123 * inode_init_always - perform inode structure initialisation
150 * @sb: superblock inode belongs to 124 * @sb: superblock inode belongs to
151 * @inode: inode to initialise 125 * @inode: inode to initialise
152 * @gfp: allocation flags <<
153 * 126 *
154 * These are initializations that need to be d 127 * These are initializations that need to be done on every inode
155 * allocation as the fields are not initialise 128 * allocation as the fields are not initialised by slab allocation.
156 * If there are additional allocations require <<
157 */ 129 */
158 int inode_init_always_gfp(struct super_block * !! 130 int inode_init_always(struct super_block *sb, struct inode *inode)
159 { 131 {
160 static const struct inode_operations e 132 static const struct inode_operations empty_iops;
161 static const struct file_operations no 133 static const struct file_operations no_open_fops = {.open = no_open};
162 struct address_space *const mapping = 134 struct address_space *const mapping = &inode->i_data;
163 135
164 inode->i_sb = sb; 136 inode->i_sb = sb;
165 inode->i_blkbits = sb->s_blocksize_bit 137 inode->i_blkbits = sb->s_blocksize_bits;
166 inode->i_flags = 0; 138 inode->i_flags = 0;
167 inode->i_state = 0; <<
168 atomic64_set(&inode->i_sequence, 0); <<
169 atomic_set(&inode->i_count, 1); 139 atomic_set(&inode->i_count, 1);
170 inode->i_op = &empty_iops; 140 inode->i_op = &empty_iops;
171 inode->i_fop = &no_open_fops; 141 inode->i_fop = &no_open_fops;
172 inode->i_ino = 0; <<
173 inode->__i_nlink = 1; 142 inode->__i_nlink = 1;
174 inode->i_opflags = 0; 143 inode->i_opflags = 0;
175 if (sb->s_xattr) 144 if (sb->s_xattr)
176 inode->i_opflags |= IOP_XATTR; 145 inode->i_opflags |= IOP_XATTR;
177 i_uid_write(inode, 0); 146 i_uid_write(inode, 0);
178 i_gid_write(inode, 0); 147 i_gid_write(inode, 0);
179 atomic_set(&inode->i_writecount, 0); 148 atomic_set(&inode->i_writecount, 0);
180 inode->i_size = 0; 149 inode->i_size = 0;
181 inode->i_write_hint = WRITE_LIFE_NOT_S 150 inode->i_write_hint = WRITE_LIFE_NOT_SET;
182 inode->i_blocks = 0; 151 inode->i_blocks = 0;
183 inode->i_bytes = 0; 152 inode->i_bytes = 0;
184 inode->i_generation = 0; 153 inode->i_generation = 0;
185 inode->i_pipe = NULL; 154 inode->i_pipe = NULL;
>> 155 inode->i_bdev = NULL;
186 inode->i_cdev = NULL; 156 inode->i_cdev = NULL;
187 inode->i_link = NULL; 157 inode->i_link = NULL;
188 inode->i_dir_seq = 0; 158 inode->i_dir_seq = 0;
189 inode->i_rdev = 0; 159 inode->i_rdev = 0;
190 inode->dirtied_when = 0; 160 inode->dirtied_when = 0;
191 161
192 #ifdef CONFIG_CGROUP_WRITEBACK 162 #ifdef CONFIG_CGROUP_WRITEBACK
193 inode->i_wb_frn_winner = 0; 163 inode->i_wb_frn_winner = 0;
194 inode->i_wb_frn_avg_time = 0; 164 inode->i_wb_frn_avg_time = 0;
195 inode->i_wb_frn_history = 0; 165 inode->i_wb_frn_history = 0;
196 #endif 166 #endif
197 167
>> 168 if (security_inode_alloc(inode))
>> 169 goto out;
198 spin_lock_init(&inode->i_lock); 170 spin_lock_init(&inode->i_lock);
199 lockdep_set_class(&inode->i_lock, &sb- 171 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
200 172
201 init_rwsem(&inode->i_rwsem); 173 init_rwsem(&inode->i_rwsem);
202 lockdep_set_class(&inode->i_rwsem, &sb 174 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
203 175
204 atomic_set(&inode->i_dio_count, 0); 176 atomic_set(&inode->i_dio_count, 0);
205 177
206 mapping->a_ops = &empty_aops; 178 mapping->a_ops = &empty_aops;
207 mapping->host = inode; 179 mapping->host = inode;
208 mapping->flags = 0; 180 mapping->flags = 0;
209 mapping->wb_err = 0; 181 mapping->wb_err = 0;
210 atomic_set(&mapping->i_mmap_writable, 182 atomic_set(&mapping->i_mmap_writable, 0);
211 #ifdef CONFIG_READ_ONLY_THP_FOR_FS <<
212 atomic_set(&mapping->nr_thps, 0); <<
213 #endif <<
214 mapping_set_gfp_mask(mapping, GFP_HIGH 183 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
215 mapping->i_private_data = NULL; !! 184 mapping->private_data = NULL;
216 mapping->writeback_index = 0; 185 mapping->writeback_index = 0;
217 init_rwsem(&mapping->invalidate_lock); <<
218 lockdep_set_class_and_name(&mapping->i <<
219 &sb->s_type <<
220 "mapping.in <<
221 if (sb->s_iflags & SB_I_STABLE_WRITES) <<
222 mapping_set_stable_writes(mapp <<
223 inode->i_private = NULL; 186 inode->i_private = NULL;
224 inode->i_mapping = mapping; 187 inode->i_mapping = mapping;
225 INIT_HLIST_HEAD(&inode->i_dentry); 188 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
226 #ifdef CONFIG_FS_POSIX_ACL 189 #ifdef CONFIG_FS_POSIX_ACL
227 inode->i_acl = inode->i_default_acl = 190 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
228 #endif 191 #endif
229 192
230 #ifdef CONFIG_FSNOTIFY 193 #ifdef CONFIG_FSNOTIFY
231 inode->i_fsnotify_mask = 0; 194 inode->i_fsnotify_mask = 0;
232 #endif 195 #endif
233 inode->i_flctx = NULL; 196 inode->i_flctx = NULL;
234 <<
235 if (unlikely(security_inode_alloc(inod <<
236 return -ENOMEM; <<
237 <<
238 this_cpu_inc(nr_inodes); 197 this_cpu_inc(nr_inodes);
239 198
240 return 0; 199 return 0;
>> 200 out:
>> 201 return -ENOMEM;
241 } 202 }
242 EXPORT_SYMBOL(inode_init_always_gfp); !! 203 EXPORT_SYMBOL(inode_init_always);
243 <<
244 void free_inode_nonrcu(struct inode *inode) <<
245 { <<
246 kmem_cache_free(inode_cachep, inode); <<
247 } <<
248 EXPORT_SYMBOL(free_inode_nonrcu); <<
249 <<
250 static void i_callback(struct rcu_head *head) <<
251 { <<
252 struct inode *inode = container_of(hea <<
253 if (inode->free_inode) <<
254 inode->free_inode(inode); <<
255 else <<
256 free_inode_nonrcu(inode); <<
257 } <<
258 204
259 static struct inode *alloc_inode(struct super_ 205 static struct inode *alloc_inode(struct super_block *sb)
260 { 206 {
261 const struct super_operations *ops = s <<
262 struct inode *inode; 207 struct inode *inode;
263 208
264 if (ops->alloc_inode) !! 209 if (sb->s_op->alloc_inode)
265 inode = ops->alloc_inode(sb); !! 210 inode = sb->s_op->alloc_inode(sb);
266 else 211 else
267 inode = alloc_inode_sb(sb, ino !! 212 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
268 213
269 if (!inode) 214 if (!inode)
270 return NULL; 215 return NULL;
271 216
272 if (unlikely(inode_init_always(sb, ino 217 if (unlikely(inode_init_always(sb, inode))) {
273 if (ops->destroy_inode) { !! 218 if (inode->i_sb->s_op->destroy_inode)
274 ops->destroy_inode(ino !! 219 inode->i_sb->s_op->destroy_inode(inode);
275 if (!ops->free_inode) !! 220 else
276 return NULL; !! 221 kmem_cache_free(inode_cachep, inode);
277 } <<
278 inode->free_inode = ops->free_ <<
279 i_callback(&inode->i_rcu); <<
280 return NULL; 222 return NULL;
281 } 223 }
282 224
283 return inode; 225 return inode;
284 } 226 }
285 227
>> 228 void free_inode_nonrcu(struct inode *inode)
>> 229 {
>> 230 kmem_cache_free(inode_cachep, inode);
>> 231 }
>> 232 EXPORT_SYMBOL(free_inode_nonrcu);
>> 233
286 void __destroy_inode(struct inode *inode) 234 void __destroy_inode(struct inode *inode)
287 { 235 {
288 BUG_ON(inode_has_buffers(inode)); 236 BUG_ON(inode_has_buffers(inode));
289 inode_detach_wb(inode); 237 inode_detach_wb(inode);
290 security_inode_free(inode); 238 security_inode_free(inode);
291 fsnotify_inode_delete(inode); 239 fsnotify_inode_delete(inode);
292 locks_free_lock_context(inode); 240 locks_free_lock_context(inode);
293 if (!inode->i_nlink) { 241 if (!inode->i_nlink) {
294 WARN_ON(atomic_long_read(&inod 242 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
295 atomic_long_dec(&inode->i_sb-> 243 atomic_long_dec(&inode->i_sb->s_remove_count);
296 } 244 }
297 245
298 #ifdef CONFIG_FS_POSIX_ACL 246 #ifdef CONFIG_FS_POSIX_ACL
299 if (inode->i_acl && !is_uncached_acl(i 247 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
300 posix_acl_release(inode->i_acl 248 posix_acl_release(inode->i_acl);
301 if (inode->i_default_acl && !is_uncach 249 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
302 posix_acl_release(inode->i_def 250 posix_acl_release(inode->i_default_acl);
303 #endif 251 #endif
304 this_cpu_dec(nr_inodes); 252 this_cpu_dec(nr_inodes);
305 } 253 }
306 EXPORT_SYMBOL(__destroy_inode); 254 EXPORT_SYMBOL(__destroy_inode);
307 255
308 static void destroy_inode(struct inode *inode) !! 256 static void i_callback(struct rcu_head *head)
309 { 257 {
310 const struct super_operations *ops = i !! 258 struct inode *inode = container_of(head, struct inode, i_rcu);
>> 259 kmem_cache_free(inode_cachep, inode);
>> 260 }
311 261
>> 262 static void destroy_inode(struct inode *inode)
>> 263 {
312 BUG_ON(!list_empty(&inode->i_lru)); 264 BUG_ON(!list_empty(&inode->i_lru));
313 __destroy_inode(inode); 265 __destroy_inode(inode);
314 if (ops->destroy_inode) { !! 266 if (inode->i_sb->s_op->destroy_inode)
315 ops->destroy_inode(inode); !! 267 inode->i_sb->s_op->destroy_inode(inode);
316 if (!ops->free_inode) !! 268 else
317 return; !! 269 call_rcu(&inode->i_rcu, i_callback);
318 } <<
319 inode->free_inode = ops->free_inode; <<
320 call_rcu(&inode->i_rcu, i_callback); <<
321 } 270 }
322 271
323 /** 272 /**
324 * drop_nlink - directly drop an inode's link 273 * drop_nlink - directly drop an inode's link count
325 * @inode: inode 274 * @inode: inode
326 * 275 *
327 * This is a low-level filesystem helper to re 276 * This is a low-level filesystem helper to replace any
328 * direct filesystem manipulation of i_nlink. 277 * direct filesystem manipulation of i_nlink. In cases
329 * where we are attempting to track writes to 278 * where we are attempting to track writes to the
330 * filesystem, a decrement to zero means an im 279 * filesystem, a decrement to zero means an imminent
331 * write when the file is truncated and actual 280 * write when the file is truncated and actually unlinked
332 * on the filesystem. 281 * on the filesystem.
333 */ 282 */
334 void drop_nlink(struct inode *inode) 283 void drop_nlink(struct inode *inode)
335 { 284 {
336 WARN_ON(inode->i_nlink == 0); 285 WARN_ON(inode->i_nlink == 0);
337 inode->__i_nlink--; 286 inode->__i_nlink--;
338 if (!inode->i_nlink) 287 if (!inode->i_nlink)
339 atomic_long_inc(&inode->i_sb-> 288 atomic_long_inc(&inode->i_sb->s_remove_count);
340 } 289 }
341 EXPORT_SYMBOL(drop_nlink); 290 EXPORT_SYMBOL(drop_nlink);
342 291
343 /** 292 /**
344 * clear_nlink - directly zero an inode's link 293 * clear_nlink - directly zero an inode's link count
345 * @inode: inode 294 * @inode: inode
346 * 295 *
347 * This is a low-level filesystem helper to re 296 * This is a low-level filesystem helper to replace any
348 * direct filesystem manipulation of i_nlink. 297 * direct filesystem manipulation of i_nlink. See
349 * drop_nlink() for why we care about i_nlink 298 * drop_nlink() for why we care about i_nlink hitting zero.
350 */ 299 */
351 void clear_nlink(struct inode *inode) 300 void clear_nlink(struct inode *inode)
352 { 301 {
353 if (inode->i_nlink) { 302 if (inode->i_nlink) {
354 inode->__i_nlink = 0; 303 inode->__i_nlink = 0;
355 atomic_long_inc(&inode->i_sb-> 304 atomic_long_inc(&inode->i_sb->s_remove_count);
356 } 305 }
357 } 306 }
358 EXPORT_SYMBOL(clear_nlink); 307 EXPORT_SYMBOL(clear_nlink);
359 308
360 /** 309 /**
361 * set_nlink - directly set an inode's link co 310 * set_nlink - directly set an inode's link count
362 * @inode: inode 311 * @inode: inode
363 * @nlink: new nlink (should be non-zero) 312 * @nlink: new nlink (should be non-zero)
364 * 313 *
365 * This is a low-level filesystem helper to re 314 * This is a low-level filesystem helper to replace any
366 * direct filesystem manipulation of i_nlink. 315 * direct filesystem manipulation of i_nlink.
367 */ 316 */
368 void set_nlink(struct inode *inode, unsigned i 317 void set_nlink(struct inode *inode, unsigned int nlink)
369 { 318 {
370 if (!nlink) { 319 if (!nlink) {
371 clear_nlink(inode); 320 clear_nlink(inode);
372 } else { 321 } else {
373 /* Yes, some filesystems do ch 322 /* Yes, some filesystems do change nlink from zero to one */
374 if (inode->i_nlink == 0) 323 if (inode->i_nlink == 0)
375 atomic_long_dec(&inode 324 atomic_long_dec(&inode->i_sb->s_remove_count);
376 325
377 inode->__i_nlink = nlink; 326 inode->__i_nlink = nlink;
378 } 327 }
379 } 328 }
380 EXPORT_SYMBOL(set_nlink); 329 EXPORT_SYMBOL(set_nlink);
381 330
382 /** 331 /**
383 * inc_nlink - directly increment an inode's l 332 * inc_nlink - directly increment an inode's link count
384 * @inode: inode 333 * @inode: inode
385 * 334 *
386 * This is a low-level filesystem helper to re 335 * This is a low-level filesystem helper to replace any
387 * direct filesystem manipulation of i_nlink. 336 * direct filesystem manipulation of i_nlink. Currently,
388 * it is only here for parity with dec_nlink() 337 * it is only here for parity with dec_nlink().
389 */ 338 */
390 void inc_nlink(struct inode *inode) 339 void inc_nlink(struct inode *inode)
391 { 340 {
392 if (unlikely(inode->i_nlink == 0)) { 341 if (unlikely(inode->i_nlink == 0)) {
393 WARN_ON(!(inode->i_state & I_L 342 WARN_ON(!(inode->i_state & I_LINKABLE));
394 atomic_long_dec(&inode->i_sb-> 343 atomic_long_dec(&inode->i_sb->s_remove_count);
395 } 344 }
396 345
397 inode->__i_nlink++; 346 inode->__i_nlink++;
398 } 347 }
399 EXPORT_SYMBOL(inc_nlink); 348 EXPORT_SYMBOL(inc_nlink);
400 349
401 static void __address_space_init_once(struct a 350 static void __address_space_init_once(struct address_space *mapping)
402 { 351 {
403 xa_init_flags(&mapping->i_pages, XA_FL !! 352 INIT_RADIX_TREE(&mapping->i_pages, GFP_ATOMIC | __GFP_ACCOUNT);
404 init_rwsem(&mapping->i_mmap_rwsem); 353 init_rwsem(&mapping->i_mmap_rwsem);
405 INIT_LIST_HEAD(&mapping->i_private_lis !! 354 INIT_LIST_HEAD(&mapping->private_list);
406 spin_lock_init(&mapping->i_private_loc !! 355 spin_lock_init(&mapping->private_lock);
407 mapping->i_mmap = RB_ROOT_CACHED; 356 mapping->i_mmap = RB_ROOT_CACHED;
408 } 357 }
409 358
410 void address_space_init_once(struct address_sp 359 void address_space_init_once(struct address_space *mapping)
411 { 360 {
412 memset(mapping, 0, sizeof(*mapping)); 361 memset(mapping, 0, sizeof(*mapping));
413 __address_space_init_once(mapping); 362 __address_space_init_once(mapping);
414 } 363 }
415 EXPORT_SYMBOL(address_space_init_once); 364 EXPORT_SYMBOL(address_space_init_once);
416 365
417 /* 366 /*
418 * These are initializations that only need to 367 * These are initializations that only need to be done
419 * once, because the fields are idempotent acr 368 * once, because the fields are idempotent across use
420 * of the inode, so let the slab aware of that 369 * of the inode, so let the slab aware of that.
421 */ 370 */
422 void inode_init_once(struct inode *inode) 371 void inode_init_once(struct inode *inode)
423 { 372 {
424 memset(inode, 0, sizeof(*inode)); 373 memset(inode, 0, sizeof(*inode));
425 INIT_HLIST_NODE(&inode->i_hash); 374 INIT_HLIST_NODE(&inode->i_hash);
426 INIT_LIST_HEAD(&inode->i_devices); 375 INIT_LIST_HEAD(&inode->i_devices);
427 INIT_LIST_HEAD(&inode->i_io_list); 376 INIT_LIST_HEAD(&inode->i_io_list);
428 INIT_LIST_HEAD(&inode->i_wb_list); 377 INIT_LIST_HEAD(&inode->i_wb_list);
429 INIT_LIST_HEAD(&inode->i_lru); 378 INIT_LIST_HEAD(&inode->i_lru);
430 INIT_LIST_HEAD(&inode->i_sb_list); <<
431 __address_space_init_once(&inode->i_da 379 __address_space_init_once(&inode->i_data);
432 i_size_ordered_init(inode); 380 i_size_ordered_init(inode);
433 } 381 }
434 EXPORT_SYMBOL(inode_init_once); 382 EXPORT_SYMBOL(inode_init_once);
435 383
436 static void init_once(void *foo) 384 static void init_once(void *foo)
437 { 385 {
438 struct inode *inode = (struct inode *) 386 struct inode *inode = (struct inode *) foo;
439 387
440 inode_init_once(inode); 388 inode_init_once(inode);
441 } 389 }
442 390
443 /* 391 /*
>> 392 * inode->i_lock must be held
>> 393 */
>> 394 void __iget(struct inode *inode)
>> 395 {
>> 396 atomic_inc(&inode->i_count);
>> 397 }
>> 398
>> 399 /*
444 * get additional reference to inode; caller m 400 * get additional reference to inode; caller must already hold one.
445 */ 401 */
446 void ihold(struct inode *inode) 402 void ihold(struct inode *inode)
447 { 403 {
448 WARN_ON(atomic_inc_return(&inode->i_co 404 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
449 } 405 }
450 EXPORT_SYMBOL(ihold); 406 EXPORT_SYMBOL(ihold);
451 407
452 static void __inode_add_lru(struct inode *inod !! 408 static void inode_lru_list_add(struct inode *inode)
453 { 409 {
454 if (inode->i_state & (I_DIRTY_ALL | I_ !! 410 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
455 return; <<
456 if (atomic_read(&inode->i_count)) <<
457 return; <<
458 if (!(inode->i_sb->s_flags & SB_ACTIVE <<
459 return; <<
460 if (!mapping_shrinkable(&inode->i_data <<
461 return; <<
462 <<
463 if (list_lru_add_obj(&inode->i_sb->s_i <<
464 this_cpu_inc(nr_unused); 411 this_cpu_inc(nr_unused);
465 else if (rotate) !! 412 else
466 inode->i_state |= I_REFERENCED 413 inode->i_state |= I_REFERENCED;
467 } 414 }
468 415
469 struct wait_queue_head *inode_bit_waitqueue(st <<
470 st <<
471 { <<
472 void *bit_address; <<
473 <<
474 bit_address = inode_state_wait_address <<
475 init_wait_var_entry(wqe, bit_address, <<
476 return __var_waitqueue(bit_address); <<
477 } <<
478 EXPORT_SYMBOL(inode_bit_waitqueue); <<
479 <<
480 /* 416 /*
481 * Add inode to LRU if needed (inode is unused 417 * Add inode to LRU if needed (inode is unused and clean).
482 * 418 *
483 * Needs inode->i_lock held. 419 * Needs inode->i_lock held.
484 */ 420 */
485 void inode_add_lru(struct inode *inode) 421 void inode_add_lru(struct inode *inode)
486 { 422 {
487 __inode_add_lru(inode, false); !! 423 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
>> 424 I_FREEING | I_WILL_FREE)) &&
>> 425 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
>> 426 inode_lru_list_add(inode);
488 } 427 }
489 428
490 static void inode_lru_list_del(struct inode *i <<
491 { <<
492 if (list_lru_del_obj(&inode->i_sb->s_i <<
493 this_cpu_dec(nr_unused); <<
494 } <<
495 <<
496 static void inode_pin_lru_isolating(struct ino <<
497 { <<
498 lockdep_assert_held(&inode->i_lock); <<
499 WARN_ON(inode->i_state & (I_LRU_ISOLAT <<
500 inode->i_state |= I_LRU_ISOLATING; <<
501 } <<
502 429
503 static void inode_unpin_lru_isolating(struct i !! 430 static void inode_lru_list_del(struct inode *inode)
504 { <<
505 spin_lock(&inode->i_lock); <<
506 WARN_ON(!(inode->i_state & I_LRU_ISOLA <<
507 inode->i_state &= ~I_LRU_ISOLATING; <<
508 /* Called with inode->i_lock which ens <<
509 inode_wake_up_bit(inode, __I_LRU_ISOLA <<
510 spin_unlock(&inode->i_lock); <<
511 } <<
512 <<
513 static void inode_wait_for_lru_isolating(struc <<
514 { 431 {
515 struct wait_bit_queue_entry wqe; <<
516 struct wait_queue_head *wq_head; <<
517 432
518 lockdep_assert_held(&inode->i_lock); !! 433 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
519 if (!(inode->i_state & I_LRU_ISOLATING !! 434 this_cpu_dec(nr_unused);
520 return; <<
521 <<
522 wq_head = inode_bit_waitqueue(&wqe, in <<
523 for (;;) { <<
524 prepare_to_wait_event(wq_head, <<
525 /* <<
526 * Checking I_LRU_ISOLATING wi <<
527 * memory ordering. <<
528 */ <<
529 if (!(inode->i_state & I_LRU_I <<
530 break; <<
531 spin_unlock(&inode->i_lock); <<
532 schedule(); <<
533 spin_lock(&inode->i_lock); <<
534 } <<
535 finish_wait(wq_head, &wqe.wq_entry); <<
536 WARN_ON(inode->i_state & I_LRU_ISOLATI <<
537 } 435 }
538 436
539 /** 437 /**
540 * inode_sb_list_add - add inode to the superb 438 * inode_sb_list_add - add inode to the superblock list of inodes
541 * @inode: inode to add 439 * @inode: inode to add
542 */ 440 */
543 void inode_sb_list_add(struct inode *inode) 441 void inode_sb_list_add(struct inode *inode)
544 { 442 {
545 spin_lock(&inode->i_sb->s_inode_list_l 443 spin_lock(&inode->i_sb->s_inode_list_lock);
546 list_add(&inode->i_sb_list, &inode->i_ 444 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
547 spin_unlock(&inode->i_sb->s_inode_list 445 spin_unlock(&inode->i_sb->s_inode_list_lock);
548 } 446 }
549 EXPORT_SYMBOL_GPL(inode_sb_list_add); 447 EXPORT_SYMBOL_GPL(inode_sb_list_add);
550 448
551 static inline void inode_sb_list_del(struct in 449 static inline void inode_sb_list_del(struct inode *inode)
552 { 450 {
553 if (!list_empty(&inode->i_sb_list)) { 451 if (!list_empty(&inode->i_sb_list)) {
554 spin_lock(&inode->i_sb->s_inod 452 spin_lock(&inode->i_sb->s_inode_list_lock);
555 list_del_init(&inode->i_sb_lis 453 list_del_init(&inode->i_sb_list);
556 spin_unlock(&inode->i_sb->s_in 454 spin_unlock(&inode->i_sb->s_inode_list_lock);
557 } 455 }
558 } 456 }
559 457
560 static unsigned long hash(struct super_block * 458 static unsigned long hash(struct super_block *sb, unsigned long hashval)
561 { 459 {
562 unsigned long tmp; 460 unsigned long tmp;
563 461
564 tmp = (hashval * (unsigned long)sb) ^ 462 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
565 L1_CACHE_BYTES; 463 L1_CACHE_BYTES;
566 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME 464 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
567 return tmp & i_hash_mask; 465 return tmp & i_hash_mask;
568 } 466 }
569 467
570 /** 468 /**
571 * __insert_inode_hash - hash an inode 469 * __insert_inode_hash - hash an inode
572 * @inode: unhashed inode 470 * @inode: unhashed inode
573 * @hashval: unsigned long value used to 471 * @hashval: unsigned long value used to locate this object in the
574 * inode_hashtable. 472 * inode_hashtable.
575 * 473 *
576 * Add an inode to the inode hash for thi 474 * Add an inode to the inode hash for this superblock.
577 */ 475 */
578 void __insert_inode_hash(struct inode *inode, 476 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
579 { 477 {
580 struct hlist_head *b = inode_hashtable 478 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
581 479
582 spin_lock(&inode_hash_lock); 480 spin_lock(&inode_hash_lock);
583 spin_lock(&inode->i_lock); 481 spin_lock(&inode->i_lock);
584 hlist_add_head_rcu(&inode->i_hash, b); !! 482 hlist_add_head(&inode->i_hash, b);
585 spin_unlock(&inode->i_lock); 483 spin_unlock(&inode->i_lock);
586 spin_unlock(&inode_hash_lock); 484 spin_unlock(&inode_hash_lock);
587 } 485 }
588 EXPORT_SYMBOL(__insert_inode_hash); 486 EXPORT_SYMBOL(__insert_inode_hash);
589 487
590 /** 488 /**
591 * __remove_inode_hash - remove an inode 489 * __remove_inode_hash - remove an inode from the hash
592 * @inode: inode to unhash 490 * @inode: inode to unhash
593 * 491 *
594 * Remove an inode from the superblock. 492 * Remove an inode from the superblock.
595 */ 493 */
596 void __remove_inode_hash(struct inode *inode) 494 void __remove_inode_hash(struct inode *inode)
597 { 495 {
598 spin_lock(&inode_hash_lock); 496 spin_lock(&inode_hash_lock);
599 spin_lock(&inode->i_lock); 497 spin_lock(&inode->i_lock);
600 hlist_del_init_rcu(&inode->i_hash); !! 498 hlist_del_init(&inode->i_hash);
601 spin_unlock(&inode->i_lock); 499 spin_unlock(&inode->i_lock);
602 spin_unlock(&inode_hash_lock); 500 spin_unlock(&inode_hash_lock);
603 } 501 }
604 EXPORT_SYMBOL(__remove_inode_hash); 502 EXPORT_SYMBOL(__remove_inode_hash);
605 503
606 void dump_mapping(const struct address_space * <<
607 { <<
608 struct inode *host; <<
609 const struct address_space_operations <<
610 struct hlist_node *dentry_first; <<
611 struct dentry *dentry_ptr; <<
612 struct dentry dentry; <<
613 char fname[64] = {}; <<
614 unsigned long ino; <<
615 <<
616 /* <<
617 * If mapping is an invalid pointer, w <<
618 * accessing it, so probe everything d <<
619 */ <<
620 if (get_kernel_nofault(host, &mapping- <<
621 get_kernel_nofault(a_ops, &mapping <<
622 pr_warn("invalid mapping:%px\n <<
623 return; <<
624 } <<
625 <<
626 if (!host) { <<
627 pr_warn("aops:%ps\n", a_ops); <<
628 return; <<
629 } <<
630 <<
631 if (get_kernel_nofault(dentry_first, & <<
632 get_kernel_nofault(ino, &host->i_i <<
633 pr_warn("aops:%ps invalid inod <<
634 return; <<
635 } <<
636 <<
637 if (!dentry_first) { <<
638 pr_warn("aops:%ps ino:%lx\n", <<
639 return; <<
640 } <<
641 <<
642 dentry_ptr = container_of(dentry_first <<
643 if (get_kernel_nofault(dentry, dentry_ <<
644 !dentry.d_parent || !dentry.d_name <<
645 pr_warn("aops:%ps ino:%lx inva <<
646 a_ops, ino, de <<
647 return; <<
648 } <<
649 <<
650 if (strncpy_from_kernel_nofault(fname, <<
651 strscpy(fname, "<invalid>"); <<
652 /* <<
653 * Even if strncpy_from_kernel_nofault <<
654 * the fname could be unreliable <<
655 */ <<
656 pr_warn("aops:%ps ino:%lx dentry name( <<
657 a_ops, ino, fname); <<
658 } <<
659 <<
660 void clear_inode(struct inode *inode) 504 void clear_inode(struct inode *inode)
661 { 505 {
662 /* 506 /*
663 * We have to cycle the i_pages lock h 507 * We have to cycle the i_pages lock here because reclaim can be in the
664 * process of removing the last page ( !! 508 * process of removing the last page (in __delete_from_page_cache())
665 * and we must not free the mapping un 509 * and we must not free the mapping under it.
666 */ 510 */
667 xa_lock_irq(&inode->i_data.i_pages); 511 xa_lock_irq(&inode->i_data.i_pages);
668 BUG_ON(inode->i_data.nrpages); 512 BUG_ON(inode->i_data.nrpages);
669 /* !! 513 BUG_ON(inode->i_data.nrexceptional);
670 * Almost always, mapping_empty(&inode <<
671 * two known and long-standing ways in <<
672 * (when deep radix-tree node allocati <<
673 * collapse_file() failed). Until thos <<
674 * or a cleanup function is called her <<
675 * nor even WARN_ON(!mapping_empty). <<
676 */ <<
677 xa_unlock_irq(&inode->i_data.i_pages); 514 xa_unlock_irq(&inode->i_data.i_pages);
678 BUG_ON(!list_empty(&inode->i_data.i_pr !! 515 BUG_ON(!list_empty(&inode->i_data.private_list));
679 BUG_ON(!(inode->i_state & I_FREEING)); 516 BUG_ON(!(inode->i_state & I_FREEING));
680 BUG_ON(inode->i_state & I_CLEAR); 517 BUG_ON(inode->i_state & I_CLEAR);
681 BUG_ON(!list_empty(&inode->i_wb_list)) 518 BUG_ON(!list_empty(&inode->i_wb_list));
682 /* don't need i_lock here, no concurre 519 /* don't need i_lock here, no concurrent mods to i_state */
683 inode->i_state = I_FREEING | I_CLEAR; 520 inode->i_state = I_FREEING | I_CLEAR;
684 } 521 }
685 EXPORT_SYMBOL(clear_inode); 522 EXPORT_SYMBOL(clear_inode);
686 523
687 /* 524 /*
688 * Free the inode passed in, removing it from 525 * Free the inode passed in, removing it from the lists it is still connected
689 * to. We remove any pages still attached to t 526 * to. We remove any pages still attached to the inode and wait for any IO that
690 * is still in progress before finally destroy 527 * is still in progress before finally destroying the inode.
691 * 528 *
692 * An inode must already be marked I_FREEING s 529 * An inode must already be marked I_FREEING so that we avoid the inode being
693 * moved back onto lists if we race with other 530 * moved back onto lists if we race with other code that manipulates the lists
694 * (e.g. writeback_single_inode). The caller i 531 * (e.g. writeback_single_inode). The caller is responsible for setting this.
695 * 532 *
696 * An inode must already be removed from the L 533 * An inode must already be removed from the LRU list before being evicted from
697 * the cache. This should occur atomically wit 534 * the cache. This should occur atomically with setting the I_FREEING state
698 * flag, so no inodes here should ever be on t 535 * flag, so no inodes here should ever be on the LRU when being evicted.
699 */ 536 */
700 static void evict(struct inode *inode) 537 static void evict(struct inode *inode)
701 { 538 {
702 const struct super_operations *op = in 539 const struct super_operations *op = inode->i_sb->s_op;
703 540
704 BUG_ON(!(inode->i_state & I_FREEING)); 541 BUG_ON(!(inode->i_state & I_FREEING));
705 BUG_ON(!list_empty(&inode->i_lru)); 542 BUG_ON(!list_empty(&inode->i_lru));
706 543
707 if (!list_empty(&inode->i_io_list)) 544 if (!list_empty(&inode->i_io_list))
708 inode_io_list_del(inode); 545 inode_io_list_del(inode);
709 546
710 inode_sb_list_del(inode); 547 inode_sb_list_del(inode);
711 548
712 spin_lock(&inode->i_lock); <<
713 inode_wait_for_lru_isolating(inode); <<
714 <<
715 /* 549 /*
716 * Wait for flusher thread to be done 550 * Wait for flusher thread to be done with the inode so that filesystem
717 * does not start destroying it while 551 * does not start destroying it while writeback is still running. Since
718 * the inode has I_FREEING set, flushe 552 * the inode has I_FREEING set, flusher thread won't start new work on
719 * the inode. We just have to wait fo 553 * the inode. We just have to wait for running writeback to finish.
720 */ 554 */
721 inode_wait_for_writeback(inode); 555 inode_wait_for_writeback(inode);
722 spin_unlock(&inode->i_lock); <<
723 556
724 if (op->evict_inode) { 557 if (op->evict_inode) {
725 op->evict_inode(inode); 558 op->evict_inode(inode);
726 } else { 559 } else {
727 truncate_inode_pages_final(&in 560 truncate_inode_pages_final(&inode->i_data);
728 clear_inode(inode); 561 clear_inode(inode);
729 } 562 }
>> 563 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
>> 564 bd_forget(inode);
730 if (S_ISCHR(inode->i_mode) && inode->i 565 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
731 cd_forget(inode); 566 cd_forget(inode);
732 567
733 remove_inode_hash(inode); 568 remove_inode_hash(inode);
734 569
735 /* <<
736 * Wake up waiters in __wait_on_freein <<
737 * <<
738 * Lockless hash lookup may end up fin <<
739 * it above, but only lock it *after* <<
740 * In this case the potential waiter c <<
741 * <<
742 * The inode being unhashed after the <<
743 * used as an indicator whether blocki <<
744 */ <<
745 spin_lock(&inode->i_lock); 570 spin_lock(&inode->i_lock);
746 /* !! 571 wake_up_bit(&inode->i_state, __I_NEW);
747 * Pairs with the barrier in prepare_t <<
748 * ___wait_var_event() either sees the <<
749 * waitqueue_active() check in wake_up <<
750 */ <<
751 smp_mb(); <<
752 inode_wake_up_bit(inode, __I_NEW); <<
753 BUG_ON(inode->i_state != (I_FREEING | 572 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
754 spin_unlock(&inode->i_lock); 573 spin_unlock(&inode->i_lock);
755 574
756 destroy_inode(inode); 575 destroy_inode(inode);
757 } 576 }
758 577
759 /* 578 /*
760 * dispose_list - dispose of the contents of a 579 * dispose_list - dispose of the contents of a local list
761 * @head: the head of the list to free 580 * @head: the head of the list to free
762 * 581 *
763 * Dispose-list gets a local list with local i 582 * Dispose-list gets a local list with local inodes in it, so it doesn't
764 * need to worry about list corruption and SMP 583 * need to worry about list corruption and SMP locks.
765 */ 584 */
766 static void dispose_list(struct list_head *hea 585 static void dispose_list(struct list_head *head)
767 { 586 {
768 while (!list_empty(head)) { 587 while (!list_empty(head)) {
769 struct inode *inode; 588 struct inode *inode;
770 589
771 inode = list_first_entry(head, 590 inode = list_first_entry(head, struct inode, i_lru);
772 list_del_init(&inode->i_lru); 591 list_del_init(&inode->i_lru);
773 592
774 evict(inode); 593 evict(inode);
775 cond_resched(); 594 cond_resched();
776 } 595 }
777 } 596 }
778 597
779 /** 598 /**
780 * evict_inodes - evict all evictable inodes f 599 * evict_inodes - evict all evictable inodes for a superblock
781 * @sb: superblock to operate on 600 * @sb: superblock to operate on
782 * 601 *
783 * Make sure that no inodes with zero refcount 602 * Make sure that no inodes with zero refcount are retained. This is
784 * called by superblock shutdown after having 603 * called by superblock shutdown after having SB_ACTIVE flag removed,
785 * so any inode reaching zero refcount during 604 * so any inode reaching zero refcount during or after that call will
786 * be immediately evicted. 605 * be immediately evicted.
787 */ 606 */
788 void evict_inodes(struct super_block *sb) 607 void evict_inodes(struct super_block *sb)
789 { 608 {
790 struct inode *inode, *next; 609 struct inode *inode, *next;
791 LIST_HEAD(dispose); 610 LIST_HEAD(dispose);
792 611
793 again: 612 again:
794 spin_lock(&sb->s_inode_list_lock); 613 spin_lock(&sb->s_inode_list_lock);
795 list_for_each_entry_safe(inode, next, 614 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
796 if (atomic_read(&inode->i_coun 615 if (atomic_read(&inode->i_count))
797 continue; 616 continue;
798 617
799 spin_lock(&inode->i_lock); 618 spin_lock(&inode->i_lock);
800 if (atomic_read(&inode->i_coun <<
801 spin_unlock(&inode->i_ <<
802 continue; <<
803 } <<
804 if (inode->i_state & (I_NEW | 619 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
805 spin_unlock(&inode->i_ 620 spin_unlock(&inode->i_lock);
806 continue; 621 continue;
807 } 622 }
808 623
809 inode->i_state |= I_FREEING; 624 inode->i_state |= I_FREEING;
810 inode_lru_list_del(inode); 625 inode_lru_list_del(inode);
811 spin_unlock(&inode->i_lock); 626 spin_unlock(&inode->i_lock);
812 list_add(&inode->i_lru, &dispo 627 list_add(&inode->i_lru, &dispose);
813 628
814 /* 629 /*
815 * We can have a ton of inodes 630 * We can have a ton of inodes to evict at unmount time given
816 * enough memory, check to see 631 * enough memory, check to see if we need to go to sleep for a
817 * bit so we don't livelock. 632 * bit so we don't livelock.
818 */ 633 */
819 if (need_resched()) { 634 if (need_resched()) {
820 spin_unlock(&sb->s_ino 635 spin_unlock(&sb->s_inode_list_lock);
821 cond_resched(); 636 cond_resched();
822 dispose_list(&dispose) 637 dispose_list(&dispose);
823 goto again; 638 goto again;
824 } 639 }
825 } 640 }
826 spin_unlock(&sb->s_inode_list_lock); 641 spin_unlock(&sb->s_inode_list_lock);
827 642
828 dispose_list(&dispose); 643 dispose_list(&dispose);
829 } 644 }
830 EXPORT_SYMBOL_GPL(evict_inodes); 645 EXPORT_SYMBOL_GPL(evict_inodes);
831 646
832 /** 647 /**
833 * invalidate_inodes - attempt to free all 648 * invalidate_inodes - attempt to free all inodes on a superblock
834 * @sb: superblock to operate on 649 * @sb: superblock to operate on
>> 650 * @kill_dirty: flag to guide handling of dirty inodes
835 * 651 *
836 * Attempts to free all inodes (including dirt !! 652 * Attempts to free all inodes for a given superblock. If there were any
>> 653 * busy inodes return a non-zero value, else zero.
>> 654 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
>> 655 * them as busy.
837 */ 656 */
838 void invalidate_inodes(struct super_block *sb) !! 657 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
839 { 658 {
>> 659 int busy = 0;
840 struct inode *inode, *next; 660 struct inode *inode, *next;
841 LIST_HEAD(dispose); 661 LIST_HEAD(dispose);
842 662
843 again: <<
844 spin_lock(&sb->s_inode_list_lock); 663 spin_lock(&sb->s_inode_list_lock);
845 list_for_each_entry_safe(inode, next, 664 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
846 spin_lock(&inode->i_lock); 665 spin_lock(&inode->i_lock);
847 if (inode->i_state & (I_NEW | 666 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
848 spin_unlock(&inode->i_ 667 spin_unlock(&inode->i_lock);
849 continue; 668 continue;
850 } 669 }
>> 670 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
>> 671 spin_unlock(&inode->i_lock);
>> 672 busy = 1;
>> 673 continue;
>> 674 }
851 if (atomic_read(&inode->i_coun 675 if (atomic_read(&inode->i_count)) {
852 spin_unlock(&inode->i_ 676 spin_unlock(&inode->i_lock);
>> 677 busy = 1;
853 continue; 678 continue;
854 } 679 }
855 680
856 inode->i_state |= I_FREEING; 681 inode->i_state |= I_FREEING;
857 inode_lru_list_del(inode); 682 inode_lru_list_del(inode);
858 spin_unlock(&inode->i_lock); 683 spin_unlock(&inode->i_lock);
859 list_add(&inode->i_lru, &dispo 684 list_add(&inode->i_lru, &dispose);
860 if (need_resched()) { <<
861 spin_unlock(&sb->s_ino <<
862 cond_resched(); <<
863 dispose_list(&dispose) <<
864 goto again; <<
865 } <<
866 } 685 }
867 spin_unlock(&sb->s_inode_list_lock); 686 spin_unlock(&sb->s_inode_list_lock);
868 687
869 dispose_list(&dispose); 688 dispose_list(&dispose);
>> 689
>> 690 return busy;
870 } 691 }
871 692
872 /* 693 /*
873 * Isolate the inode from the LRU in preparati 694 * Isolate the inode from the LRU in preparation for freeing it.
874 * 695 *
>> 696 * Any inodes which are pinned purely because of attached pagecache have their
>> 697 * pagecache removed. If the inode has metadata buffers attached to
>> 698 * mapping->private_list then try to remove them.
>> 699 *
875 * If the inode has the I_REFERENCED flag set, 700 * If the inode has the I_REFERENCED flag set, then it means that it has been
876 * used recently - the flag is set in iput_fin 701 * used recently - the flag is set in iput_final(). When we encounter such an
877 * inode, clear the flag and move it to the ba 702 * inode, clear the flag and move it to the back of the LRU so it gets another
878 * pass through the LRU before it gets reclaim 703 * pass through the LRU before it gets reclaimed. This is necessary because of
879 * the fact we are doing lazy LRU updates to m 704 * the fact we are doing lazy LRU updates to minimise lock contention so the
880 * LRU does not have strict ordering. Hence we 705 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
881 * with this flag set because they are the ino 706 * with this flag set because they are the inodes that are out of order.
882 */ 707 */
883 static enum lru_status inode_lru_isolate(struc 708 static enum lru_status inode_lru_isolate(struct list_head *item,
884 struct list_lru_one *lru, spin 709 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
885 { 710 {
886 struct list_head *freeable = arg; 711 struct list_head *freeable = arg;
887 struct inode *inode = container_of( 712 struct inode *inode = container_of(item, struct inode, i_lru);
888 713
889 /* 714 /*
890 * We are inverting the lru lock/inode !! 715 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
891 * trylock. If we fail to get the lock !! 716 * If we fail to get the lock, just skip it.
892 */ 717 */
893 if (!spin_trylock(&inode->i_lock)) 718 if (!spin_trylock(&inode->i_lock))
894 return LRU_SKIP; 719 return LRU_SKIP;
895 720
896 /* 721 /*
897 * Inodes can get referenced, redirtie !! 722 * Referenced or dirty inodes are still in use. Give them another pass
898 * they're already on the LRU, and thi !! 723 * through the LRU as we canot reclaim them now.
899 * unreclaimable for a while. Remove t <<
900 * sync, or the last page cache deleti <<
901 */ 724 */
902 if (atomic_read(&inode->i_count) || 725 if (atomic_read(&inode->i_count) ||
903 (inode->i_state & ~I_REFERENCED) | !! 726 (inode->i_state & ~I_REFERENCED)) {
904 !mapping_shrinkable(&inode->i_data <<
905 list_lru_isolate(lru, &inode-> 727 list_lru_isolate(lru, &inode->i_lru);
906 spin_unlock(&inode->i_lock); 728 spin_unlock(&inode->i_lock);
907 this_cpu_dec(nr_unused); 729 this_cpu_dec(nr_unused);
908 return LRU_REMOVED; 730 return LRU_REMOVED;
909 } 731 }
910 732
911 /* Recently referenced inodes get one !! 733 /* recently referenced inodes get one more pass */
912 if (inode->i_state & I_REFERENCED) { 734 if (inode->i_state & I_REFERENCED) {
913 inode->i_state &= ~I_REFERENCE 735 inode->i_state &= ~I_REFERENCED;
914 spin_unlock(&inode->i_lock); 736 spin_unlock(&inode->i_lock);
915 return LRU_ROTATE; 737 return LRU_ROTATE;
916 } 738 }
917 739
918 /* !! 740 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
919 * On highmem systems, mapping_shrinka !! 741 __iget(inode);
920 * page cache in order to free up stru <<
921 * be under pressure before the cache <<
922 */ <<
923 if (inode_has_buffers(inode) || !mappi <<
924 inode_pin_lru_isolating(inode) <<
925 spin_unlock(&inode->i_lock); 742 spin_unlock(&inode->i_lock);
926 spin_unlock(lru_lock); 743 spin_unlock(lru_lock);
927 if (remove_inode_buffers(inode 744 if (remove_inode_buffers(inode)) {
928 unsigned long reap; 745 unsigned long reap;
929 reap = invalidate_mapp 746 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
930 if (current_is_kswapd( 747 if (current_is_kswapd())
931 __count_vm_eve 748 __count_vm_events(KSWAPD_INODESTEAL, reap);
932 else 749 else
933 __count_vm_eve 750 __count_vm_events(PGINODESTEAL, reap);
934 mm_account_reclaimed_p !! 751 if (current->reclaim_state)
>> 752 current->reclaim_state->reclaimed_slab += reap;
935 } 753 }
936 inode_unpin_lru_isolating(inod !! 754 iput(inode);
937 spin_lock(lru_lock); 755 spin_lock(lru_lock);
938 return LRU_RETRY; 756 return LRU_RETRY;
939 } 757 }
940 758
941 WARN_ON(inode->i_state & I_NEW); 759 WARN_ON(inode->i_state & I_NEW);
942 inode->i_state |= I_FREEING; 760 inode->i_state |= I_FREEING;
943 list_lru_isolate_move(lru, &inode->i_l 761 list_lru_isolate_move(lru, &inode->i_lru, freeable);
944 spin_unlock(&inode->i_lock); 762 spin_unlock(&inode->i_lock);
945 763
946 this_cpu_dec(nr_unused); 764 this_cpu_dec(nr_unused);
947 return LRU_REMOVED; 765 return LRU_REMOVED;
948 } 766 }
949 767
950 /* 768 /*
951 * Walk the superblock inode LRU for freeable 769 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
952 * This is called from the superblock shrinker 770 * This is called from the superblock shrinker function with a number of inodes
953 * to trim from the LRU. Inodes to be freed ar 771 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
954 * then are freed outside inode_lock by dispos 772 * then are freed outside inode_lock by dispose_list().
955 */ 773 */
956 long prune_icache_sb(struct super_block *sb, s 774 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
957 { 775 {
958 LIST_HEAD(freeable); 776 LIST_HEAD(freeable);
959 long freed; 777 long freed;
960 778
961 freed = list_lru_shrink_walk(&sb->s_in 779 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
962 inode_lru 780 inode_lru_isolate, &freeable);
963 dispose_list(&freeable); 781 dispose_list(&freeable);
964 return freed; 782 return freed;
965 } 783 }
966 784
967 static void __wait_on_freeing_inode(struct ino !! 785 static void __wait_on_freeing_inode(struct inode *inode);
968 /* 786 /*
969 * Called with the inode lock held. 787 * Called with the inode lock held.
970 */ 788 */
971 static struct inode *find_inode(struct super_b 789 static struct inode *find_inode(struct super_block *sb,
972 struct hlist_h 790 struct hlist_head *head,
973 int (*test)(st 791 int (*test)(struct inode *, void *),
974 void *data, bo !! 792 void *data)
975 { 793 {
976 struct inode *inode = NULL; 794 struct inode *inode = NULL;
977 795
978 if (is_inode_hash_locked) <<
979 lockdep_assert_held(&inode_has <<
980 else <<
981 lockdep_assert_not_held(&inode <<
982 <<
983 rcu_read_lock(); <<
984 repeat: 796 repeat:
985 hlist_for_each_entry_rcu(inode, head, !! 797 hlist_for_each_entry(inode, head, i_hash) {
986 if (inode->i_sb != sb) 798 if (inode->i_sb != sb)
987 continue; 799 continue;
988 if (!test(inode, data)) 800 if (!test(inode, data))
989 continue; 801 continue;
990 spin_lock(&inode->i_lock); 802 spin_lock(&inode->i_lock);
991 if (inode->i_state & (I_FREEIN 803 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
992 __wait_on_freeing_inod !! 804 __wait_on_freeing_inode(inode);
993 goto repeat; 805 goto repeat;
994 } 806 }
995 if (unlikely(inode->i_state & 807 if (unlikely(inode->i_state & I_CREATING)) {
996 spin_unlock(&inode->i_ 808 spin_unlock(&inode->i_lock);
997 rcu_read_unlock(); <<
998 return ERR_PTR(-ESTALE 809 return ERR_PTR(-ESTALE);
999 } 810 }
1000 __iget(inode); 811 __iget(inode);
1001 spin_unlock(&inode->i_lock); 812 spin_unlock(&inode->i_lock);
1002 rcu_read_unlock(); <<
1003 return inode; 813 return inode;
1004 } 814 }
1005 rcu_read_unlock(); <<
1006 return NULL; 815 return NULL;
1007 } 816 }
1008 817
1009 /* 818 /*
1010 * find_inode_fast is the fast path version o 819 * find_inode_fast is the fast path version of find_inode, see the comment at
1011 * iget_locked for details. 820 * iget_locked for details.
1012 */ 821 */
1013 static struct inode *find_inode_fast(struct s 822 static struct inode *find_inode_fast(struct super_block *sb,
1014 struct hlist_ !! 823 struct hlist_head *head, unsigned long ino)
1015 bool is_inode <<
1016 { 824 {
1017 struct inode *inode = NULL; 825 struct inode *inode = NULL;
1018 826
1019 if (is_inode_hash_locked) <<
1020 lockdep_assert_held(&inode_ha <<
1021 else <<
1022 lockdep_assert_not_held(&inod <<
1023 <<
1024 rcu_read_lock(); <<
1025 repeat: 827 repeat:
1026 hlist_for_each_entry_rcu(inode, head, !! 828 hlist_for_each_entry(inode, head, i_hash) {
1027 if (inode->i_ino != ino) 829 if (inode->i_ino != ino)
1028 continue; 830 continue;
1029 if (inode->i_sb != sb) 831 if (inode->i_sb != sb)
1030 continue; 832 continue;
1031 spin_lock(&inode->i_lock); 833 spin_lock(&inode->i_lock);
1032 if (inode->i_state & (I_FREEI 834 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1033 __wait_on_freeing_ino !! 835 __wait_on_freeing_inode(inode);
1034 goto repeat; 836 goto repeat;
1035 } 837 }
1036 if (unlikely(inode->i_state & 838 if (unlikely(inode->i_state & I_CREATING)) {
1037 spin_unlock(&inode->i 839 spin_unlock(&inode->i_lock);
1038 rcu_read_unlock(); <<
1039 return ERR_PTR(-ESTAL 840 return ERR_PTR(-ESTALE);
1040 } 841 }
1041 __iget(inode); 842 __iget(inode);
1042 spin_unlock(&inode->i_lock); 843 spin_unlock(&inode->i_lock);
1043 rcu_read_unlock(); <<
1044 return inode; 844 return inode;
1045 } 845 }
1046 rcu_read_unlock(); <<
1047 return NULL; 846 return NULL;
1048 } 847 }
1049 848
1050 /* 849 /*
1051 * Each cpu owns a range of LAST_INO_BATCH nu 850 * Each cpu owns a range of LAST_INO_BATCH numbers.
1052 * 'shared_last_ino' is dirtied only once out 851 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
1053 * to renew the exhausted range. 852 * to renew the exhausted range.
1054 * 853 *
1055 * This does not significantly increase overf 854 * This does not significantly increase overflow rate because every CPU can
1056 * consume at most LAST_INO_BATCH-1 unused in 855 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
1057 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 409 856 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
1058 * 2^32 range, and is a worst-case. Even a 50 857 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
1059 * overflow rate by 2x, which does not seem t 858 * overflow rate by 2x, which does not seem too significant.
1060 * 859 *
1061 * On a 32bit, non LFS stat() call, glibc wil 860 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1062 * error if st_ino won't fit in target struct 861 * error if st_ino won't fit in target struct field. Use 32bit counter
1063 * here to attempt to avoid that. 862 * here to attempt to avoid that.
1064 */ 863 */
1065 #define LAST_INO_BATCH 1024 864 #define LAST_INO_BATCH 1024
1066 static DEFINE_PER_CPU(unsigned int, last_ino) 865 static DEFINE_PER_CPU(unsigned int, last_ino);
1067 866
1068 unsigned int get_next_ino(void) 867 unsigned int get_next_ino(void)
1069 { 868 {
1070 unsigned int *p = &get_cpu_var(last_i 869 unsigned int *p = &get_cpu_var(last_ino);
1071 unsigned int res = *p; 870 unsigned int res = *p;
1072 871
1073 #ifdef CONFIG_SMP 872 #ifdef CONFIG_SMP
1074 if (unlikely((res & (LAST_INO_BATCH-1 873 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
1075 static atomic_t shared_last_i 874 static atomic_t shared_last_ino;
1076 int next = atomic_add_return( 875 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
1077 876
1078 res = next - LAST_INO_BATCH; 877 res = next - LAST_INO_BATCH;
1079 } 878 }
1080 #endif 879 #endif
1081 880
1082 res++; 881 res++;
1083 /* get_next_ino should not provide a 882 /* get_next_ino should not provide a 0 inode number */
1084 if (unlikely(!res)) 883 if (unlikely(!res))
1085 res++; 884 res++;
1086 *p = res; 885 *p = res;
1087 put_cpu_var(last_ino); 886 put_cpu_var(last_ino);
1088 return res; 887 return res;
1089 } 888 }
1090 EXPORT_SYMBOL(get_next_ino); 889 EXPORT_SYMBOL(get_next_ino);
1091 890
1092 /** 891 /**
1093 * new_inode_pseudo - obtain an i 892 * new_inode_pseudo - obtain an inode
1094 * @sb: superblock 893 * @sb: superblock
1095 * 894 *
1096 * Allocates a new inode for given super 895 * Allocates a new inode for given superblock.
1097 * Inode wont be chained in superblock s 896 * Inode wont be chained in superblock s_inodes list
1098 * This means : 897 * This means :
1099 * - fs can't be unmount 898 * - fs can't be unmount
1100 * - quotas, fsnotify, writeback can't w 899 * - quotas, fsnotify, writeback can't work
1101 */ 900 */
1102 struct inode *new_inode_pseudo(struct super_b 901 struct inode *new_inode_pseudo(struct super_block *sb)
1103 { 902 {
1104 return alloc_inode(sb); !! 903 struct inode *inode = alloc_inode(sb);
>> 904
>> 905 if (inode) {
>> 906 spin_lock(&inode->i_lock);
>> 907 inode->i_state = 0;
>> 908 spin_unlock(&inode->i_lock);
>> 909 INIT_LIST_HEAD(&inode->i_sb_list);
>> 910 }
>> 911 return inode;
1105 } 912 }
1106 913
1107 /** 914 /**
1108 * new_inode - obtain an inode 915 * new_inode - obtain an inode
1109 * @sb: superblock 916 * @sb: superblock
1110 * 917 *
1111 * Allocates a new inode for given super 918 * Allocates a new inode for given superblock. The default gfp_mask
1112 * for allocations related to inode->i_m 919 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1113 * If HIGHMEM pages are unsuitable or it 920 * If HIGHMEM pages are unsuitable or it is known that pages allocated
1114 * for the page cache are not reclaimabl 921 * for the page cache are not reclaimable or migratable,
1115 * mapping_set_gfp_mask() must be called 922 * mapping_set_gfp_mask() must be called with suitable flags on the
1116 * newly created inode's mapping 923 * newly created inode's mapping
1117 * 924 *
1118 */ 925 */
1119 struct inode *new_inode(struct super_block *s 926 struct inode *new_inode(struct super_block *sb)
1120 { 927 {
1121 struct inode *inode; 928 struct inode *inode;
1122 929
>> 930 spin_lock_prefetch(&sb->s_inode_list_lock);
>> 931
1123 inode = new_inode_pseudo(sb); 932 inode = new_inode_pseudo(sb);
1124 if (inode) 933 if (inode)
1125 inode_sb_list_add(inode); 934 inode_sb_list_add(inode);
1126 return inode; 935 return inode;
1127 } 936 }
1128 EXPORT_SYMBOL(new_inode); 937 EXPORT_SYMBOL(new_inode);
1129 938
1130 #ifdef CONFIG_DEBUG_LOCK_ALLOC 939 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1131 void lockdep_annotate_inode_mutex_key(struct 940 void lockdep_annotate_inode_mutex_key(struct inode *inode)
1132 { 941 {
1133 if (S_ISDIR(inode->i_mode)) { 942 if (S_ISDIR(inode->i_mode)) {
1134 struct file_system_type *type 943 struct file_system_type *type = inode->i_sb->s_type;
1135 944
1136 /* Set new key only if filesy 945 /* Set new key only if filesystem hasn't already changed it */
1137 if (lockdep_match_class(&inod 946 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1138 /* 947 /*
1139 * ensure nobody is a 948 * ensure nobody is actually holding i_mutex
1140 */ 949 */
1141 // mutex_destroy(&ino 950 // mutex_destroy(&inode->i_mutex);
1142 init_rwsem(&inode->i_ 951 init_rwsem(&inode->i_rwsem);
1143 lockdep_set_class(&in 952 lockdep_set_class(&inode->i_rwsem,
1144 &ty 953 &type->i_mutex_dir_key);
1145 } 954 }
1146 } 955 }
1147 } 956 }
1148 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_ke 957 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1149 #endif 958 #endif
1150 959
1151 /** 960 /**
1152 * unlock_new_inode - clear the I_NEW state a 961 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1153 * @inode: new inode to unlock 962 * @inode: new inode to unlock
1154 * 963 *
1155 * Called when the inode is fully initialised 964 * Called when the inode is fully initialised to clear the new state of the
1156 * inode and wake up anyone waiting for the i 965 * inode and wake up anyone waiting for the inode to finish initialisation.
1157 */ 966 */
1158 void unlock_new_inode(struct inode *inode) 967 void unlock_new_inode(struct inode *inode)
1159 { 968 {
1160 lockdep_annotate_inode_mutex_key(inod 969 lockdep_annotate_inode_mutex_key(inode);
1161 spin_lock(&inode->i_lock); 970 spin_lock(&inode->i_lock);
1162 WARN_ON(!(inode->i_state & I_NEW)); 971 WARN_ON(!(inode->i_state & I_NEW));
1163 inode->i_state &= ~I_NEW & ~I_CREATIN 972 inode->i_state &= ~I_NEW & ~I_CREATING;
1164 /* <<
1165 * Pairs with the barrier in prepare_ <<
1166 * ___wait_var_event() either sees th <<
1167 * waitqueue_active() check in wake_u <<
1168 */ <<
1169 smp_mb(); 973 smp_mb();
1170 inode_wake_up_bit(inode, __I_NEW); !! 974 wake_up_bit(&inode->i_state, __I_NEW);
1171 spin_unlock(&inode->i_lock); 975 spin_unlock(&inode->i_lock);
1172 } 976 }
1173 EXPORT_SYMBOL(unlock_new_inode); 977 EXPORT_SYMBOL(unlock_new_inode);
1174 978
1175 void discard_new_inode(struct inode *inode) 979 void discard_new_inode(struct inode *inode)
1176 { 980 {
1177 lockdep_annotate_inode_mutex_key(inod 981 lockdep_annotate_inode_mutex_key(inode);
1178 spin_lock(&inode->i_lock); 982 spin_lock(&inode->i_lock);
1179 WARN_ON(!(inode->i_state & I_NEW)); 983 WARN_ON(!(inode->i_state & I_NEW));
1180 inode->i_state &= ~I_NEW; 984 inode->i_state &= ~I_NEW;
1181 /* <<
1182 * Pairs with the barrier in prepare_ <<
1183 * ___wait_var_event() either sees th <<
1184 * waitqueue_active() check in wake_u <<
1185 */ <<
1186 smp_mb(); 985 smp_mb();
1187 inode_wake_up_bit(inode, __I_NEW); !! 986 wake_up_bit(&inode->i_state, __I_NEW);
1188 spin_unlock(&inode->i_lock); 987 spin_unlock(&inode->i_lock);
1189 iput(inode); 988 iput(inode);
1190 } 989 }
1191 EXPORT_SYMBOL(discard_new_inode); 990 EXPORT_SYMBOL(discard_new_inode);
1192 991
1193 /** 992 /**
1194 * lock_two_nondirectories - take two i_mutex 993 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1195 * 994 *
1196 * Lock any non-NULL argument. Passed objects !! 995 * Lock any non-NULL argument that is not a directory.
1197 * Zero, one or two objects may be locked by 996 * Zero, one or two objects may be locked by this function.
1198 * 997 *
1199 * @inode1: first inode to lock 998 * @inode1: first inode to lock
1200 * @inode2: second inode to lock 999 * @inode2: second inode to lock
1201 */ 1000 */
1202 void lock_two_nondirectories(struct inode *in 1001 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1203 { 1002 {
1204 if (inode1) <<
1205 WARN_ON_ONCE(S_ISDIR(inode1-> <<
1206 if (inode2) <<
1207 WARN_ON_ONCE(S_ISDIR(inode2-> <<
1208 if (inode1 > inode2) 1003 if (inode1 > inode2)
1209 swap(inode1, inode2); 1004 swap(inode1, inode2);
1210 if (inode1) !! 1005
>> 1006 if (inode1 && !S_ISDIR(inode1->i_mode))
1211 inode_lock(inode1); 1007 inode_lock(inode1);
1212 if (inode2 && inode2 != inode1) !! 1008 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1213 inode_lock_nested(inode2, I_M 1009 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1214 } 1010 }
1215 EXPORT_SYMBOL(lock_two_nondirectories); 1011 EXPORT_SYMBOL(lock_two_nondirectories);
1216 1012
1217 /** 1013 /**
1218 * unlock_two_nondirectories - release locks 1014 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1219 * @inode1: first inode to unlock 1015 * @inode1: first inode to unlock
1220 * @inode2: second inode to unlock 1016 * @inode2: second inode to unlock
1221 */ 1017 */
1222 void unlock_two_nondirectories(struct inode * 1018 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1223 { 1019 {
1224 if (inode1) { !! 1020 if (inode1 && !S_ISDIR(inode1->i_mode))
1225 WARN_ON_ONCE(S_ISDIR(inode1-> <<
1226 inode_unlock(inode1); 1021 inode_unlock(inode1);
1227 } !! 1022 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1228 if (inode2 && inode2 != inode1) { <<
1229 WARN_ON_ONCE(S_ISDIR(inode2-> <<
1230 inode_unlock(inode2); 1023 inode_unlock(inode2);
1231 } <<
1232 } 1024 }
1233 EXPORT_SYMBOL(unlock_two_nondirectories); 1025 EXPORT_SYMBOL(unlock_two_nondirectories);
1234 1026
1235 /** 1027 /**
1236 * inode_insert5 - obtain an inode from a mou 1028 * inode_insert5 - obtain an inode from a mounted file system
1237 * @inode: pre-allocated inode to use fo 1029 * @inode: pre-allocated inode to use for insert to cache
1238 * @hashval: hash value (usually inode num 1030 * @hashval: hash value (usually inode number) to get
1239 * @test: callback used for comparisons 1031 * @test: callback used for comparisons between inodes
1240 * @set: callback used to initialize a 1032 * @set: callback used to initialize a new struct inode
1241 * @data: opaque data pointer to pass t 1033 * @data: opaque data pointer to pass to @test and @set
1242 * 1034 *
1243 * Search for the inode specified by @hashval 1035 * Search for the inode specified by @hashval and @data in the inode cache,
1244 * and if present it is return it with an inc 1036 * and if present it is return it with an increased reference count. This is
1245 * a variant of iget5_locked() for callers th 1037 * a variant of iget5_locked() for callers that don't want to fail on memory
1246 * allocation of inode. 1038 * allocation of inode.
1247 * 1039 *
1248 * If the inode is not in cache, insert the p 1040 * If the inode is not in cache, insert the pre-allocated inode to cache and
1249 * return it locked, hashed, and with the I_N 1041 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1250 * to fill it in before unlocking it via unlo 1042 * to fill it in before unlocking it via unlock_new_inode().
1251 * 1043 *
1252 * Note both @test and @set are called with t 1044 * Note both @test and @set are called with the inode_hash_lock held, so can't
1253 * sleep. 1045 * sleep.
1254 */ 1046 */
1255 struct inode *inode_insert5(struct inode *ino 1047 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1256 int (*test)(struc 1048 int (*test)(struct inode *, void *),
1257 int (*set)(struct 1049 int (*set)(struct inode *, void *), void *data)
1258 { 1050 {
1259 struct hlist_head *head = inode_hasht 1051 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1260 struct inode *old; 1052 struct inode *old;
>> 1053 bool creating = inode->i_state & I_CREATING;
1261 1054
1262 again: 1055 again:
1263 spin_lock(&inode_hash_lock); 1056 spin_lock(&inode_hash_lock);
1264 old = find_inode(inode->i_sb, head, t !! 1057 old = find_inode(inode->i_sb, head, test, data);
1265 if (unlikely(old)) { 1058 if (unlikely(old)) {
1266 /* 1059 /*
1267 * Uhhuh, somebody else creat 1060 * Uhhuh, somebody else created the same inode under us.
1268 * Use the old inode instead 1061 * Use the old inode instead of the preallocated one.
1269 */ 1062 */
1270 spin_unlock(&inode_hash_lock) 1063 spin_unlock(&inode_hash_lock);
1271 if (IS_ERR(old)) 1064 if (IS_ERR(old))
1272 return NULL; 1065 return NULL;
1273 wait_on_inode(old); 1066 wait_on_inode(old);
1274 if (unlikely(inode_unhashed(o 1067 if (unlikely(inode_unhashed(old))) {
1275 iput(old); 1068 iput(old);
1276 goto again; 1069 goto again;
1277 } 1070 }
1278 return old; 1071 return old;
1279 } 1072 }
1280 1073
1281 if (set && unlikely(set(inode, data)) 1074 if (set && unlikely(set(inode, data))) {
1282 inode = NULL; 1075 inode = NULL;
1283 goto unlock; 1076 goto unlock;
1284 } 1077 }
1285 1078
1286 /* 1079 /*
1287 * Return the locked inode with I_NEW 1080 * Return the locked inode with I_NEW set, the
1288 * caller is responsible for filling 1081 * caller is responsible for filling in the contents
1289 */ 1082 */
1290 spin_lock(&inode->i_lock); 1083 spin_lock(&inode->i_lock);
1291 inode->i_state |= I_NEW; 1084 inode->i_state |= I_NEW;
1292 hlist_add_head_rcu(&inode->i_hash, he !! 1085 hlist_add_head(&inode->i_hash, head);
1293 spin_unlock(&inode->i_lock); 1086 spin_unlock(&inode->i_lock);
1294 !! 1087 if (!creating)
1295 /* <<
1296 * Add inode to the sb list if it's n <<
1297 * point, so it should be safe to tes <<
1298 */ <<
1299 if (list_empty(&inode->i_sb_list)) <<
1300 inode_sb_list_add(inode); 1088 inode_sb_list_add(inode);
1301 unlock: 1089 unlock:
1302 spin_unlock(&inode_hash_lock); 1090 spin_unlock(&inode_hash_lock);
1303 1091
1304 return inode; 1092 return inode;
1305 } 1093 }
1306 EXPORT_SYMBOL(inode_insert5); 1094 EXPORT_SYMBOL(inode_insert5);
1307 1095
1308 /** 1096 /**
1309 * iget5_locked - obtain an inode from a moun 1097 * iget5_locked - obtain an inode from a mounted file system
1310 * @sb: super block of file system 1098 * @sb: super block of file system
1311 * @hashval: hash value (usually inode num 1099 * @hashval: hash value (usually inode number) to get
1312 * @test: callback used for comparisons 1100 * @test: callback used for comparisons between inodes
1313 * @set: callback used to initialize a 1101 * @set: callback used to initialize a new struct inode
1314 * @data: opaque data pointer to pass t 1102 * @data: opaque data pointer to pass to @test and @set
1315 * 1103 *
1316 * Search for the inode specified by @hashval 1104 * Search for the inode specified by @hashval and @data in the inode cache,
1317 * and if present it is return it with an inc 1105 * and if present it is return it with an increased reference count. This is
1318 * a generalized version of iget_locked() for 1106 * a generalized version of iget_locked() for file systems where the inode
1319 * number is not sufficient for unique identi 1107 * number is not sufficient for unique identification of an inode.
1320 * 1108 *
1321 * If the inode is not in cache, allocate a n 1109 * If the inode is not in cache, allocate a new inode and return it locked,
1322 * hashed, and with the I_NEW flag set. The f 1110 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1323 * before unlocking it via unlock_new_inode() 1111 * before unlocking it via unlock_new_inode().
1324 * 1112 *
1325 * Note both @test and @set are called with t 1113 * Note both @test and @set are called with the inode_hash_lock held, so can't
1326 * sleep. 1114 * sleep.
1327 */ 1115 */
1328 struct inode *iget5_locked(struct super_block 1116 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1329 int (*test)(struct inode *, v 1117 int (*test)(struct inode *, void *),
1330 int (*set)(struct inode *, vo 1118 int (*set)(struct inode *, void *), void *data)
1331 { 1119 {
1332 struct inode *inode = ilookup5(sb, ha 1120 struct inode *inode = ilookup5(sb, hashval, test, data);
1333 1121
1334 if (!inode) { 1122 if (!inode) {
1335 struct inode *new = alloc_ino 1123 struct inode *new = alloc_inode(sb);
1336 1124
1337 if (new) { 1125 if (new) {
>> 1126 new->i_state = 0;
1338 inode = inode_insert5 1127 inode = inode_insert5(new, hashval, test, set, data);
1339 if (unlikely(inode != 1128 if (unlikely(inode != new))
1340 destroy_inode 1129 destroy_inode(new);
1341 } 1130 }
1342 } 1131 }
1343 return inode; 1132 return inode;
1344 } 1133 }
1345 EXPORT_SYMBOL(iget5_locked); 1134 EXPORT_SYMBOL(iget5_locked);
1346 1135
1347 /** 1136 /**
1348 * iget5_locked_rcu - obtain an inode from a <<
1349 * @sb: super block of file system <<
1350 * @hashval: hash value (usually inode num <<
1351 * @test: callback used for comparisons <<
1352 * @set: callback used to initialize a <<
1353 * @data: opaque data pointer to pass t <<
1354 * <<
1355 * This is equivalent to iget5_locked, except <<
1356 * tolerate the inode not being stable, inclu <<
1357 */ <<
1358 struct inode *iget5_locked_rcu(struct super_b <<
1359 int (*test)(struct inode *, v <<
1360 int (*set)(struct inode *, vo <<
1361 { <<
1362 struct hlist_head *head = inode_hasht <<
1363 struct inode *inode, *new; <<
1364 <<
1365 again: <<
1366 inode = find_inode(sb, head, test, da <<
1367 if (inode) { <<
1368 if (IS_ERR(inode)) <<
1369 return NULL; <<
1370 wait_on_inode(inode); <<
1371 if (unlikely(inode_unhashed(i <<
1372 iput(inode); <<
1373 goto again; <<
1374 } <<
1375 return inode; <<
1376 } <<
1377 <<
1378 new = alloc_inode(sb); <<
1379 if (new) { <<
1380 inode = inode_insert5(new, ha <<
1381 if (unlikely(inode != new)) <<
1382 destroy_inode(new); <<
1383 } <<
1384 return inode; <<
1385 } <<
1386 EXPORT_SYMBOL_GPL(iget5_locked_rcu); <<
1387 <<
1388 /** <<
1389 * iget_locked - obtain an inode from a mount 1137 * iget_locked - obtain an inode from a mounted file system
1390 * @sb: super block of file system 1138 * @sb: super block of file system
1391 * @ino: inode number to get 1139 * @ino: inode number to get
1392 * 1140 *
1393 * Search for the inode specified by @ino in 1141 * Search for the inode specified by @ino in the inode cache and if present
1394 * return it with an increased reference coun 1142 * return it with an increased reference count. This is for file systems
1395 * where the inode number is sufficient for u 1143 * where the inode number is sufficient for unique identification of an inode.
1396 * 1144 *
1397 * If the inode is not in cache, allocate a n 1145 * If the inode is not in cache, allocate a new inode and return it locked,
1398 * hashed, and with the I_NEW flag set. The 1146 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1399 * before unlocking it via unlock_new_inode() 1147 * before unlocking it via unlock_new_inode().
1400 */ 1148 */
1401 struct inode *iget_locked(struct super_block 1149 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1402 { 1150 {
1403 struct hlist_head *head = inode_hasht 1151 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1404 struct inode *inode; 1152 struct inode *inode;
1405 again: 1153 again:
1406 inode = find_inode_fast(sb, head, ino !! 1154 spin_lock(&inode_hash_lock);
>> 1155 inode = find_inode_fast(sb, head, ino);
>> 1156 spin_unlock(&inode_hash_lock);
1407 if (inode) { 1157 if (inode) {
1408 if (IS_ERR(inode)) 1158 if (IS_ERR(inode))
1409 return NULL; 1159 return NULL;
1410 wait_on_inode(inode); 1160 wait_on_inode(inode);
1411 if (unlikely(inode_unhashed(i 1161 if (unlikely(inode_unhashed(inode))) {
1412 iput(inode); 1162 iput(inode);
1413 goto again; 1163 goto again;
1414 } 1164 }
1415 return inode; 1165 return inode;
1416 } 1166 }
1417 1167
1418 inode = alloc_inode(sb); 1168 inode = alloc_inode(sb);
1419 if (inode) { 1169 if (inode) {
1420 struct inode *old; 1170 struct inode *old;
1421 1171
1422 spin_lock(&inode_hash_lock); 1172 spin_lock(&inode_hash_lock);
1423 /* We released the lock, so.. 1173 /* We released the lock, so.. */
1424 old = find_inode_fast(sb, hea !! 1174 old = find_inode_fast(sb, head, ino);
1425 if (!old) { 1175 if (!old) {
1426 inode->i_ino = ino; 1176 inode->i_ino = ino;
1427 spin_lock(&inode->i_l 1177 spin_lock(&inode->i_lock);
1428 inode->i_state = I_NE 1178 inode->i_state = I_NEW;
1429 hlist_add_head_rcu(&i !! 1179 hlist_add_head(&inode->i_hash, head);
1430 spin_unlock(&inode->i 1180 spin_unlock(&inode->i_lock);
1431 inode_sb_list_add(ino 1181 inode_sb_list_add(inode);
1432 spin_unlock(&inode_ha 1182 spin_unlock(&inode_hash_lock);
1433 1183
1434 /* Return the locked 1184 /* Return the locked inode with I_NEW set, the
1435 * caller is responsi 1185 * caller is responsible for filling in the contents
1436 */ 1186 */
1437 return inode; 1187 return inode;
1438 } 1188 }
1439 1189
1440 /* 1190 /*
1441 * Uhhuh, somebody else creat 1191 * Uhhuh, somebody else created the same inode under
1442 * us. Use the old inode inst 1192 * us. Use the old inode instead of the one we just
1443 * allocated. 1193 * allocated.
1444 */ 1194 */
1445 spin_unlock(&inode_hash_lock) 1195 spin_unlock(&inode_hash_lock);
1446 destroy_inode(inode); 1196 destroy_inode(inode);
1447 if (IS_ERR(old)) 1197 if (IS_ERR(old))
1448 return NULL; 1198 return NULL;
1449 inode = old; 1199 inode = old;
1450 wait_on_inode(inode); 1200 wait_on_inode(inode);
1451 if (unlikely(inode_unhashed(i 1201 if (unlikely(inode_unhashed(inode))) {
1452 iput(inode); 1202 iput(inode);
1453 goto again; 1203 goto again;
1454 } 1204 }
1455 } 1205 }
1456 return inode; 1206 return inode;
1457 } 1207 }
1458 EXPORT_SYMBOL(iget_locked); 1208 EXPORT_SYMBOL(iget_locked);
1459 1209
1460 /* 1210 /*
1461 * search the inode cache for a matching inod 1211 * search the inode cache for a matching inode number.
1462 * If we find one, then the inode number we a 1212 * If we find one, then the inode number we are trying to
1463 * allocate is not unique and so we should no 1213 * allocate is not unique and so we should not use it.
1464 * 1214 *
1465 * Returns 1 if the inode number is unique, 0 1215 * Returns 1 if the inode number is unique, 0 if it is not.
1466 */ 1216 */
1467 static int test_inode_iunique(struct super_bl 1217 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1468 { 1218 {
1469 struct hlist_head *b = inode_hashtabl 1219 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1470 struct inode *inode; 1220 struct inode *inode;
1471 1221
1472 hlist_for_each_entry_rcu(inode, b, i_ !! 1222 spin_lock(&inode_hash_lock);
1473 if (inode->i_ino == ino && in !! 1223 hlist_for_each_entry(inode, b, i_hash) {
>> 1224 if (inode->i_ino == ino && inode->i_sb == sb) {
>> 1225 spin_unlock(&inode_hash_lock);
1474 return 0; 1226 return 0;
>> 1227 }
1475 } 1228 }
>> 1229 spin_unlock(&inode_hash_lock);
>> 1230
1476 return 1; 1231 return 1;
1477 } 1232 }
1478 1233
1479 /** 1234 /**
1480 * iunique - get a unique inode number 1235 * iunique - get a unique inode number
1481 * @sb: superblock 1236 * @sb: superblock
1482 * @max_reserved: highest reserved inode 1237 * @max_reserved: highest reserved inode number
1483 * 1238 *
1484 * Obtain an inode number that is unique 1239 * Obtain an inode number that is unique on the system for a given
1485 * superblock. This is used by file syst 1240 * superblock. This is used by file systems that have no natural
1486 * permanent inode numbering system. An 1241 * permanent inode numbering system. An inode number is returned that
1487 * is higher than the reserved limit but 1242 * is higher than the reserved limit but unique.
1488 * 1243 *
1489 * BUGS: 1244 * BUGS:
1490 * With a large number of inodes live on 1245 * With a large number of inodes live on the file system this function
1491 * currently becomes quite slow. 1246 * currently becomes quite slow.
1492 */ 1247 */
1493 ino_t iunique(struct super_block *sb, ino_t m 1248 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1494 { 1249 {
1495 /* 1250 /*
1496 * On a 32bit, non LFS stat() call, g 1251 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1497 * error if st_ino won't fit in targe 1252 * error if st_ino won't fit in target struct field. Use 32bit counter
1498 * here to attempt to avoid that. 1253 * here to attempt to avoid that.
1499 */ 1254 */
1500 static DEFINE_SPINLOCK(iunique_lock); 1255 static DEFINE_SPINLOCK(iunique_lock);
1501 static unsigned int counter; 1256 static unsigned int counter;
1502 ino_t res; 1257 ino_t res;
1503 1258
1504 rcu_read_lock(); <<
1505 spin_lock(&iunique_lock); 1259 spin_lock(&iunique_lock);
1506 do { 1260 do {
1507 if (counter <= max_reserved) 1261 if (counter <= max_reserved)
1508 counter = max_reserve 1262 counter = max_reserved + 1;
1509 res = counter++; 1263 res = counter++;
1510 } while (!test_inode_iunique(sb, res) 1264 } while (!test_inode_iunique(sb, res));
1511 spin_unlock(&iunique_lock); 1265 spin_unlock(&iunique_lock);
1512 rcu_read_unlock(); <<
1513 1266
1514 return res; 1267 return res;
1515 } 1268 }
1516 EXPORT_SYMBOL(iunique); 1269 EXPORT_SYMBOL(iunique);
1517 1270
1518 struct inode *igrab(struct inode *inode) 1271 struct inode *igrab(struct inode *inode)
1519 { 1272 {
1520 spin_lock(&inode->i_lock); 1273 spin_lock(&inode->i_lock);
1521 if (!(inode->i_state & (I_FREEING|I_W 1274 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1522 __iget(inode); 1275 __iget(inode);
1523 spin_unlock(&inode->i_lock); 1276 spin_unlock(&inode->i_lock);
1524 } else { 1277 } else {
1525 spin_unlock(&inode->i_lock); 1278 spin_unlock(&inode->i_lock);
1526 /* 1279 /*
1527 * Handle the case where s_op 1280 * Handle the case where s_op->clear_inode is not been
1528 * called yet, and somebody i 1281 * called yet, and somebody is calling igrab
1529 * while the inode is getting 1282 * while the inode is getting freed.
1530 */ 1283 */
1531 inode = NULL; 1284 inode = NULL;
1532 } 1285 }
1533 return inode; 1286 return inode;
1534 } 1287 }
1535 EXPORT_SYMBOL(igrab); 1288 EXPORT_SYMBOL(igrab);
1536 1289
1537 /** 1290 /**
1538 * ilookup5_nowait - search for an inode in t 1291 * ilookup5_nowait - search for an inode in the inode cache
1539 * @sb: super block of file system to 1292 * @sb: super block of file system to search
1540 * @hashval: hash value (usually inode num 1293 * @hashval: hash value (usually inode number) to search for
1541 * @test: callback used for comparisons 1294 * @test: callback used for comparisons between inodes
1542 * @data: opaque data pointer to pass t 1295 * @data: opaque data pointer to pass to @test
1543 * 1296 *
1544 * Search for the inode specified by @hashval 1297 * Search for the inode specified by @hashval and @data in the inode cache.
1545 * If the inode is in the cache, the inode is 1298 * If the inode is in the cache, the inode is returned with an incremented
1546 * reference count. 1299 * reference count.
1547 * 1300 *
1548 * Note: I_NEW is not waited upon so you have 1301 * Note: I_NEW is not waited upon so you have to be very careful what you do
1549 * with the returned inode. You probably sho 1302 * with the returned inode. You probably should be using ilookup5() instead.
1550 * 1303 *
1551 * Note2: @test is called with the inode_hash 1304 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1552 */ 1305 */
1553 struct inode *ilookup5_nowait(struct super_bl 1306 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1554 int (*test)(struct inode *, v 1307 int (*test)(struct inode *, void *), void *data)
1555 { 1308 {
1556 struct hlist_head *head = inode_hasht 1309 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1557 struct inode *inode; 1310 struct inode *inode;
1558 1311
1559 spin_lock(&inode_hash_lock); 1312 spin_lock(&inode_hash_lock);
1560 inode = find_inode(sb, head, test, da !! 1313 inode = find_inode(sb, head, test, data);
1561 spin_unlock(&inode_hash_lock); 1314 spin_unlock(&inode_hash_lock);
1562 1315
1563 return IS_ERR(inode) ? NULL : inode; 1316 return IS_ERR(inode) ? NULL : inode;
1564 } 1317 }
1565 EXPORT_SYMBOL(ilookup5_nowait); 1318 EXPORT_SYMBOL(ilookup5_nowait);
1566 1319
1567 /** 1320 /**
1568 * ilookup5 - search for an inode in the inod 1321 * ilookup5 - search for an inode in the inode cache
1569 * @sb: super block of file system to 1322 * @sb: super block of file system to search
1570 * @hashval: hash value (usually inode num 1323 * @hashval: hash value (usually inode number) to search for
1571 * @test: callback used for comparisons 1324 * @test: callback used for comparisons between inodes
1572 * @data: opaque data pointer to pass t 1325 * @data: opaque data pointer to pass to @test
1573 * 1326 *
1574 * Search for the inode specified by @hashval 1327 * Search for the inode specified by @hashval and @data in the inode cache,
1575 * and if the inode is in the cache, return t 1328 * and if the inode is in the cache, return the inode with an incremented
1576 * reference count. Waits on I_NEW before re 1329 * reference count. Waits on I_NEW before returning the inode.
1577 * returned with an incremented reference cou 1330 * returned with an incremented reference count.
1578 * 1331 *
1579 * This is a generalized version of ilookup() 1332 * This is a generalized version of ilookup() for file systems where the
1580 * inode number is not sufficient for unique 1333 * inode number is not sufficient for unique identification of an inode.
1581 * 1334 *
1582 * Note: @test is called with the inode_hash_ 1335 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1583 */ 1336 */
1584 struct inode *ilookup5(struct super_block *sb 1337 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1585 int (*test)(struct inode *, v 1338 int (*test)(struct inode *, void *), void *data)
1586 { 1339 {
1587 struct inode *inode; 1340 struct inode *inode;
1588 again: 1341 again:
1589 inode = ilookup5_nowait(sb, hashval, 1342 inode = ilookup5_nowait(sb, hashval, test, data);
1590 if (inode) { 1343 if (inode) {
1591 wait_on_inode(inode); 1344 wait_on_inode(inode);
1592 if (unlikely(inode_unhashed(i 1345 if (unlikely(inode_unhashed(inode))) {
1593 iput(inode); 1346 iput(inode);
1594 goto again; 1347 goto again;
1595 } 1348 }
1596 } 1349 }
1597 return inode; 1350 return inode;
1598 } 1351 }
1599 EXPORT_SYMBOL(ilookup5); 1352 EXPORT_SYMBOL(ilookup5);
1600 1353
1601 /** 1354 /**
1602 * ilookup - search for an inode in the inode 1355 * ilookup - search for an inode in the inode cache
1603 * @sb: super block of file system to 1356 * @sb: super block of file system to search
1604 * @ino: inode number to search for 1357 * @ino: inode number to search for
1605 * 1358 *
1606 * Search for the inode @ino in the inode cac 1359 * Search for the inode @ino in the inode cache, and if the inode is in the
1607 * cache, the inode is returned with an incre 1360 * cache, the inode is returned with an incremented reference count.
1608 */ 1361 */
1609 struct inode *ilookup(struct super_block *sb, 1362 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1610 { 1363 {
1611 struct hlist_head *head = inode_hasht 1364 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1612 struct inode *inode; 1365 struct inode *inode;
1613 again: 1366 again:
1614 inode = find_inode_fast(sb, head, ino !! 1367 spin_lock(&inode_hash_lock);
>> 1368 inode = find_inode_fast(sb, head, ino);
>> 1369 spin_unlock(&inode_hash_lock);
1615 1370
1616 if (inode) { 1371 if (inode) {
1617 if (IS_ERR(inode)) 1372 if (IS_ERR(inode))
1618 return NULL; 1373 return NULL;
1619 wait_on_inode(inode); 1374 wait_on_inode(inode);
1620 if (unlikely(inode_unhashed(i 1375 if (unlikely(inode_unhashed(inode))) {
1621 iput(inode); 1376 iput(inode);
1622 goto again; 1377 goto again;
1623 } 1378 }
1624 } 1379 }
1625 return inode; 1380 return inode;
1626 } 1381 }
1627 EXPORT_SYMBOL(ilookup); 1382 EXPORT_SYMBOL(ilookup);
1628 1383
1629 /** 1384 /**
1630 * find_inode_nowait - find an inode in the i 1385 * find_inode_nowait - find an inode in the inode cache
1631 * @sb: super block of file system to 1386 * @sb: super block of file system to search
1632 * @hashval: hash value (usually inode num 1387 * @hashval: hash value (usually inode number) to search for
1633 * @match: callback used for comparisons 1388 * @match: callback used for comparisons between inodes
1634 * @data: opaque data pointer to pass t 1389 * @data: opaque data pointer to pass to @match
1635 * 1390 *
1636 * Search for the inode specified by @hashval 1391 * Search for the inode specified by @hashval and @data in the inode
1637 * cache, where the helper function @match wi 1392 * cache, where the helper function @match will return 0 if the inode
1638 * does not match, 1 if the inode does match, 1393 * does not match, 1 if the inode does match, and -1 if the search
1639 * should be stopped. The @match function mu 1394 * should be stopped. The @match function must be responsible for
1640 * taking the i_lock spin_lock and checking i 1395 * taking the i_lock spin_lock and checking i_state for an inode being
1641 * freed or being initialized, and incrementi 1396 * freed or being initialized, and incrementing the reference count
1642 * before returning 1. It also must not slee 1397 * before returning 1. It also must not sleep, since it is called with
1643 * the inode_hash_lock spinlock held. 1398 * the inode_hash_lock spinlock held.
1644 * 1399 *
1645 * This is a even more generalized version of 1400 * This is a even more generalized version of ilookup5() when the
1646 * function must never block --- find_inode() 1401 * function must never block --- find_inode() can block in
1647 * __wait_on_freeing_inode() --- or when the 1402 * __wait_on_freeing_inode() --- or when the caller can not increment
1648 * the reference count because the resulting 1403 * the reference count because the resulting iput() might cause an
1649 * inode eviction. The tradeoff is that the 1404 * inode eviction. The tradeoff is that the @match funtion must be
1650 * very carefully implemented. 1405 * very carefully implemented.
1651 */ 1406 */
1652 struct inode *find_inode_nowait(struct super_ 1407 struct inode *find_inode_nowait(struct super_block *sb,
1653 unsigned long 1408 unsigned long hashval,
1654 int (*match)( 1409 int (*match)(struct inode *, unsigned long,
1655 1410 void *),
1656 void *data) 1411 void *data)
1657 { 1412 {
1658 struct hlist_head *head = inode_hasht 1413 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1659 struct inode *inode, *ret_inode = NUL 1414 struct inode *inode, *ret_inode = NULL;
1660 int mval; 1415 int mval;
1661 1416
1662 spin_lock(&inode_hash_lock); 1417 spin_lock(&inode_hash_lock);
1663 hlist_for_each_entry(inode, head, i_h 1418 hlist_for_each_entry(inode, head, i_hash) {
1664 if (inode->i_sb != sb) 1419 if (inode->i_sb != sb)
1665 continue; 1420 continue;
1666 mval = match(inode, hashval, 1421 mval = match(inode, hashval, data);
1667 if (mval == 0) 1422 if (mval == 0)
1668 continue; 1423 continue;
1669 if (mval == 1) 1424 if (mval == 1)
1670 ret_inode = inode; 1425 ret_inode = inode;
1671 goto out; 1426 goto out;
1672 } 1427 }
1673 out: 1428 out:
1674 spin_unlock(&inode_hash_lock); 1429 spin_unlock(&inode_hash_lock);
1675 return ret_inode; 1430 return ret_inode;
1676 } 1431 }
1677 EXPORT_SYMBOL(find_inode_nowait); 1432 EXPORT_SYMBOL(find_inode_nowait);
1678 1433
1679 /** <<
1680 * find_inode_rcu - find an inode in the inod <<
1681 * @sb: Super block of file system to <<
1682 * @hashval: Key to hash <<
1683 * @test: Function to test match on an <<
1684 * @data: Data for test function <<
1685 * <<
1686 * Search for the inode specified by @hashval <<
1687 * where the helper function @test will retur <<
1688 * and 1 if it does. The @test function must <<
1689 * i_lock spin_lock and checking i_state for <<
1690 * initialized. <<
1691 * <<
1692 * If successful, this will return the inode <<
1693 * returned 1 and NULL otherwise. <<
1694 * <<
1695 * The @test function is not permitted to tak <<
1696 * It is also not permitted to sleep. <<
1697 * <<
1698 * The caller must hold the RCU read lock. <<
1699 */ <<
1700 struct inode *find_inode_rcu(struct super_blo <<
1701 int (*test)(stru <<
1702 { <<
1703 struct hlist_head *head = inode_hasht <<
1704 struct inode *inode; <<
1705 <<
1706 RCU_LOCKDEP_WARN(!rcu_read_lock_held( <<
1707 "suspicious find_ino <<
1708 <<
1709 hlist_for_each_entry_rcu(inode, head, <<
1710 if (inode->i_sb == sb && <<
1711 !(READ_ONCE(inode->i_stat <<
1712 test(inode, data)) <<
1713 return inode; <<
1714 } <<
1715 return NULL; <<
1716 } <<
1717 EXPORT_SYMBOL(find_inode_rcu); <<
1718 <<
1719 /** <<
1720 * find_inode_by_ino_rcu - Find an inode in t <<
1721 * @sb: Super block of file system to <<
1722 * @ino: The inode number to match <<
1723 * <<
1724 * Search for the inode specified by @hashval <<
1725 * where the helper function @test will retur <<
1726 * and 1 if it does. The @test function must <<
1727 * i_lock spin_lock and checking i_state for <<
1728 * initialized. <<
1729 * <<
1730 * If successful, this will return the inode <<
1731 * returned 1 and NULL otherwise. <<
1732 * <<
1733 * The @test function is not permitted to tak <<
1734 * It is also not permitted to sleep. <<
1735 * <<
1736 * The caller must hold the RCU read lock. <<
1737 */ <<
1738 struct inode *find_inode_by_ino_rcu(struct su <<
1739 unsigned <<
1740 { <<
1741 struct hlist_head *head = inode_hasht <<
1742 struct inode *inode; <<
1743 <<
1744 RCU_LOCKDEP_WARN(!rcu_read_lock_held( <<
1745 "suspicious find_ino <<
1746 <<
1747 hlist_for_each_entry_rcu(inode, head, <<
1748 if (inode->i_ino == ino && <<
1749 inode->i_sb == sb && <<
1750 !(READ_ONCE(inode->i_stat <<
1751 return inode; <<
1752 } <<
1753 return NULL; <<
1754 } <<
1755 EXPORT_SYMBOL(find_inode_by_ino_rcu); <<
1756 <<
1757 int insert_inode_locked(struct inode *inode) 1434 int insert_inode_locked(struct inode *inode)
1758 { 1435 {
1759 struct super_block *sb = inode->i_sb; 1436 struct super_block *sb = inode->i_sb;
1760 ino_t ino = inode->i_ino; 1437 ino_t ino = inode->i_ino;
1761 struct hlist_head *head = inode_hasht 1438 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1762 1439
1763 while (1) { 1440 while (1) {
1764 struct inode *old = NULL; 1441 struct inode *old = NULL;
1765 spin_lock(&inode_hash_lock); 1442 spin_lock(&inode_hash_lock);
1766 hlist_for_each_entry(old, hea 1443 hlist_for_each_entry(old, head, i_hash) {
1767 if (old->i_ino != ino 1444 if (old->i_ino != ino)
1768 continue; 1445 continue;
1769 if (old->i_sb != sb) 1446 if (old->i_sb != sb)
1770 continue; 1447 continue;
1771 spin_lock(&old->i_loc 1448 spin_lock(&old->i_lock);
1772 if (old->i_state & (I 1449 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1773 spin_unlock(& 1450 spin_unlock(&old->i_lock);
1774 continue; 1451 continue;
1775 } 1452 }
1776 break; 1453 break;
1777 } 1454 }
1778 if (likely(!old)) { 1455 if (likely(!old)) {
1779 spin_lock(&inode->i_l 1456 spin_lock(&inode->i_lock);
1780 inode->i_state |= I_N 1457 inode->i_state |= I_NEW | I_CREATING;
1781 hlist_add_head_rcu(&i !! 1458 hlist_add_head(&inode->i_hash, head);
1782 spin_unlock(&inode->i 1459 spin_unlock(&inode->i_lock);
1783 spin_unlock(&inode_ha 1460 spin_unlock(&inode_hash_lock);
1784 return 0; 1461 return 0;
1785 } 1462 }
1786 if (unlikely(old->i_state & I 1463 if (unlikely(old->i_state & I_CREATING)) {
1787 spin_unlock(&old->i_l 1464 spin_unlock(&old->i_lock);
1788 spin_unlock(&inode_ha 1465 spin_unlock(&inode_hash_lock);
1789 return -EBUSY; 1466 return -EBUSY;
1790 } 1467 }
1791 __iget(old); 1468 __iget(old);
1792 spin_unlock(&old->i_lock); 1469 spin_unlock(&old->i_lock);
1793 spin_unlock(&inode_hash_lock) 1470 spin_unlock(&inode_hash_lock);
1794 wait_on_inode(old); 1471 wait_on_inode(old);
1795 if (unlikely(!inode_unhashed( 1472 if (unlikely(!inode_unhashed(old))) {
1796 iput(old); 1473 iput(old);
1797 return -EBUSY; 1474 return -EBUSY;
1798 } 1475 }
1799 iput(old); 1476 iput(old);
1800 } 1477 }
1801 } 1478 }
1802 EXPORT_SYMBOL(insert_inode_locked); 1479 EXPORT_SYMBOL(insert_inode_locked);
1803 1480
1804 int insert_inode_locked4(struct inode *inode, 1481 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1805 int (*test)(struct inode *, v 1482 int (*test)(struct inode *, void *), void *data)
1806 { 1483 {
1807 struct inode *old; 1484 struct inode *old;
1808 1485
1809 inode->i_state |= I_CREATING; 1486 inode->i_state |= I_CREATING;
1810 old = inode_insert5(inode, hashval, t 1487 old = inode_insert5(inode, hashval, test, NULL, data);
1811 1488
1812 if (old != inode) { 1489 if (old != inode) {
1813 iput(old); 1490 iput(old);
1814 return -EBUSY; 1491 return -EBUSY;
1815 } 1492 }
1816 return 0; 1493 return 0;
1817 } 1494 }
1818 EXPORT_SYMBOL(insert_inode_locked4); 1495 EXPORT_SYMBOL(insert_inode_locked4);
1819 1496
1820 1497
1821 int generic_delete_inode(struct inode *inode) 1498 int generic_delete_inode(struct inode *inode)
1822 { 1499 {
1823 return 1; 1500 return 1;
1824 } 1501 }
1825 EXPORT_SYMBOL(generic_delete_inode); 1502 EXPORT_SYMBOL(generic_delete_inode);
1826 1503
1827 /* 1504 /*
1828 * Called when we're dropping the last refere 1505 * Called when we're dropping the last reference
1829 * to an inode. 1506 * to an inode.
1830 * 1507 *
1831 * Call the FS "drop_inode()" function, defau 1508 * Call the FS "drop_inode()" function, defaulting to
1832 * the legacy UNIX filesystem behaviour. If 1509 * the legacy UNIX filesystem behaviour. If it tells
1833 * us to evict inode, do so. Otherwise, reta 1510 * us to evict inode, do so. Otherwise, retain inode
1834 * in cache if fs is alive, sync and evict if 1511 * in cache if fs is alive, sync and evict if fs is
1835 * shutting down. 1512 * shutting down.
1836 */ 1513 */
1837 static void iput_final(struct inode *inode) 1514 static void iput_final(struct inode *inode)
1838 { 1515 {
1839 struct super_block *sb = inode->i_sb; 1516 struct super_block *sb = inode->i_sb;
1840 const struct super_operations *op = i 1517 const struct super_operations *op = inode->i_sb->s_op;
1841 unsigned long state; <<
1842 int drop; 1518 int drop;
1843 1519
1844 WARN_ON(inode->i_state & I_NEW); 1520 WARN_ON(inode->i_state & I_NEW);
1845 1521
1846 if (op->drop_inode) 1522 if (op->drop_inode)
1847 drop = op->drop_inode(inode); 1523 drop = op->drop_inode(inode);
1848 else 1524 else
1849 drop = generic_drop_inode(ino 1525 drop = generic_drop_inode(inode);
1850 1526
1851 if (!drop && !! 1527 if (!drop && (sb->s_flags & SB_ACTIVE)) {
1852 !(inode->i_state & I_DONTCACHE) & !! 1528 inode_add_lru(inode);
1853 (sb->s_flags & SB_ACTIVE)) { <<
1854 __inode_add_lru(inode, true); <<
1855 spin_unlock(&inode->i_lock); 1529 spin_unlock(&inode->i_lock);
1856 return; 1530 return;
1857 } 1531 }
1858 1532
1859 state = inode->i_state; <<
1860 if (!drop) { 1533 if (!drop) {
1861 WRITE_ONCE(inode->i_state, st !! 1534 inode->i_state |= I_WILL_FREE;
1862 spin_unlock(&inode->i_lock); 1535 spin_unlock(&inode->i_lock);
1863 <<
1864 write_inode_now(inode, 1); 1536 write_inode_now(inode, 1);
1865 <<
1866 spin_lock(&inode->i_lock); 1537 spin_lock(&inode->i_lock);
1867 state = inode->i_state; !! 1538 WARN_ON(inode->i_state & I_NEW);
1868 WARN_ON(state & I_NEW); !! 1539 inode->i_state &= ~I_WILL_FREE;
1869 state &= ~I_WILL_FREE; <<
1870 } 1540 }
1871 1541
1872 WRITE_ONCE(inode->i_state, state | I_ !! 1542 inode->i_state |= I_FREEING;
1873 if (!list_empty(&inode->i_lru)) 1543 if (!list_empty(&inode->i_lru))
1874 inode_lru_list_del(inode); 1544 inode_lru_list_del(inode);
1875 spin_unlock(&inode->i_lock); 1545 spin_unlock(&inode->i_lock);
1876 1546
1877 evict(inode); 1547 evict(inode);
1878 } 1548 }
1879 1549
1880 /** 1550 /**
1881 * iput - put an inode 1551 * iput - put an inode
1882 * @inode: inode to put 1552 * @inode: inode to put
1883 * 1553 *
1884 * Puts an inode, dropping its usage cou 1554 * Puts an inode, dropping its usage count. If the inode use count hits
1885 * zero, the inode is then freed and may 1555 * zero, the inode is then freed and may also be destroyed.
1886 * 1556 *
1887 * Consequently, iput() can sleep. 1557 * Consequently, iput() can sleep.
1888 */ 1558 */
1889 void iput(struct inode *inode) 1559 void iput(struct inode *inode)
1890 { 1560 {
1891 if (!inode) 1561 if (!inode)
1892 return; 1562 return;
1893 BUG_ON(inode->i_state & I_CLEAR); 1563 BUG_ON(inode->i_state & I_CLEAR);
1894 retry: 1564 retry:
1895 if (atomic_dec_and_lock(&inode->i_cou 1565 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1896 if (inode->i_nlink && (inode- 1566 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1897 atomic_inc(&inode->i_ 1567 atomic_inc(&inode->i_count);
1898 spin_unlock(&inode->i 1568 spin_unlock(&inode->i_lock);
1899 trace_writeback_lazyt 1569 trace_writeback_lazytime_iput(inode);
1900 mark_inode_dirty_sync 1570 mark_inode_dirty_sync(inode);
1901 goto retry; 1571 goto retry;
1902 } 1572 }
1903 iput_final(inode); 1573 iput_final(inode);
1904 } 1574 }
1905 } 1575 }
1906 EXPORT_SYMBOL(iput); 1576 EXPORT_SYMBOL(iput);
1907 1577
1908 #ifdef CONFIG_BLOCK <<
1909 /** 1578 /**
1910 * bmap - find a block number in a fi 1579 * bmap - find a block number in a file
1911 * @inode: inode owning the block numbe !! 1580 * @inode: inode of file
1912 * @block: pointer containing the block !! 1581 * @block: block to find
1913 * 1582 *
1914 * Replaces the value in ``*block`` with !! 1583 * Returns the block number on the device holding the inode that
1915 * corresponding to the requested block !! 1584 * is the disk block number for the block of the file requested.
1916 * That is, asked for block 4 of inode 1 !! 1585 * That is, asked for block 4 of inode 1 the function will return the
1917 * 4 in ``*block``, with disk block rela !! 1586 * disk block relative to the disk start that holds that block of the
1918 * block of the file. !! 1587 * file.
1919 * !! 1588 */
1920 * Returns -EINVAL in case of error, 0 o !! 1589 sector_t bmap(struct inode *inode, sector_t block)
1921 * hole, returns 0 and ``*block`` is als !! 1590 {
>> 1591 sector_t res = 0;
>> 1592 if (inode->i_mapping->a_ops->bmap)
>> 1593 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
>> 1594 return res;
>> 1595 }
>> 1596 EXPORT_SYMBOL(bmap);
>> 1597
>> 1598 /*
>> 1599 * Update times in overlayed inode from underlying real inode
1922 */ 1600 */
1923 int bmap(struct inode *inode, sector_t *block !! 1601 static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode,
>> 1602 bool rcu)
1924 { 1603 {
1925 if (!inode->i_mapping->a_ops->bmap) !! 1604 struct dentry *upperdentry;
1926 return -EINVAL; <<
1927 1605
1928 *block = inode->i_mapping->a_ops->bma !! 1606 /*
1929 return 0; !! 1607 * Nothing to do if in rcu or if non-overlayfs
>> 1608 */
>> 1609 if (rcu || likely(!(dentry->d_flags & DCACHE_OP_REAL)))
>> 1610 return;
>> 1611
>> 1612 upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER);
>> 1613
>> 1614 /*
>> 1615 * If file is on lower then we can't update atime, so no worries about
>> 1616 * stale mtime/ctime.
>> 1617 */
>> 1618 if (upperdentry) {
>> 1619 struct inode *realinode = d_inode(upperdentry);
>> 1620
>> 1621 if ((!timespec64_equal(&inode->i_mtime, &realinode->i_mtime) ||
>> 1622 !timespec64_equal(&inode->i_ctime, &realinode->i_ctime))) {
>> 1623 inode->i_mtime = realinode->i_mtime;
>> 1624 inode->i_ctime = realinode->i_ctime;
>> 1625 }
>> 1626 }
1930 } 1627 }
1931 EXPORT_SYMBOL(bmap); <<
1932 #endif <<
1933 1628
1934 /* 1629 /*
1935 * With relative atime, only update atime if 1630 * With relative atime, only update atime if the previous atime is
1936 * earlier than or equal to either the ctime !! 1631 * earlier than either the ctime or mtime or if at least a day has
1937 * or if at least a day has passed since the !! 1632 * passed since the last atime update.
1938 */ 1633 */
1939 static bool relatime_need_update(struct vfsmo !! 1634 static int relatime_need_update(const struct path *path, struct inode *inode,
1940 struct timespec6 !! 1635 struct timespec now, bool rcu)
1941 { 1636 {
1942 struct timespec64 atime, mtime, ctime <<
1943 1637
1944 if (!(mnt->mnt_flags & MNT_RELATIME)) !! 1638 if (!(path->mnt->mnt_flags & MNT_RELATIME))
1945 return true; !! 1639 return 1;
>> 1640
>> 1641 update_ovl_inode_times(path->dentry, inode, rcu);
1946 /* 1642 /*
1947 * Is mtime younger than or equal to !! 1643 * Is mtime younger than atime? If yes, update atime:
1948 */ 1644 */
1949 atime = inode_get_atime(inode); !! 1645 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1950 mtime = inode_get_mtime(inode); !! 1646 return 1;
1951 if (timespec64_compare(&mtime, &atime <<
1952 return true; <<
1953 /* 1647 /*
1954 * Is ctime younger than or equal to !! 1648 * Is ctime younger than atime? If yes, update atime:
1955 */ 1649 */
1956 ctime = inode_get_ctime(inode); !! 1650 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1957 if (timespec64_compare(&ctime, &atime !! 1651 return 1;
1958 return true; <<
1959 1652
1960 /* 1653 /*
1961 * Is the previous atime value older 1654 * Is the previous atime value older than a day? If yes,
1962 * update atime: 1655 * update atime:
1963 */ 1656 */
1964 if ((long)(now.tv_sec - atime.tv_sec) !! 1657 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1965 return true; !! 1658 return 1;
1966 /* 1659 /*
1967 * Good, we can skip the atime update 1660 * Good, we can skip the atime update:
1968 */ 1661 */
1969 return false; !! 1662 return 0;
1970 } <<
1971 <<
1972 /** <<
1973 * inode_update_timestamps - update the times <<
1974 * @inode: inode to be updated <<
1975 * @flags: S_* flags that needed to be update <<
1976 * <<
1977 * The update_time function is called when an <<
1978 * updated for a read or write operation. Thi <<
1979 * actual timestamps. It's up to the caller t <<
1980 * dirty appropriately. <<
1981 * <<
1982 * In the case where any of S_MTIME, S_CTIME, <<
1983 * attempt to update all three of them. S_ATI <<
1984 * independently of the rest. <<
1985 * <<
1986 * Returns a set of S_* flags indicating whic <<
1987 */ <<
1988 int inode_update_timestamps(struct inode *ino <<
1989 { <<
1990 int updated = 0; <<
1991 struct timespec64 now; <<
1992 <<
1993 if (flags & (S_MTIME|S_CTIME|S_VERSIO <<
1994 struct timespec64 ctime = ino <<
1995 struct timespec64 mtime = ino <<
1996 <<
1997 now = inode_set_ctime_current <<
1998 if (!timespec64_equal(&now, & <<
1999 updated |= S_CTIME; <<
2000 if (!timespec64_equal(&now, & <<
2001 inode_set_mtime_to_ts <<
2002 updated |= S_MTIME; <<
2003 } <<
2004 if (IS_I_VERSION(inode) && in <<
2005 updated |= S_VERSION; <<
2006 } else { <<
2007 now = current_time(inode); <<
2008 } <<
2009 <<
2010 if (flags & S_ATIME) { <<
2011 struct timespec64 atime = ino <<
2012 <<
2013 if (!timespec64_equal(&now, & <<
2014 inode_set_atime_to_ts <<
2015 updated |= S_ATIME; <<
2016 } <<
2017 } <<
2018 return updated; <<
2019 } 1663 }
2020 EXPORT_SYMBOL(inode_update_timestamps); <<
2021 1664
2022 /** !! 1665 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
2023 * generic_update_time - update the timestamp <<
2024 * @inode: inode to be updated <<
2025 * @flags: S_* flags that needed to be update <<
2026 * <<
2027 * The update_time function is called when an <<
2028 * updated for a read or write operation. In <<
2029 * or S_VERSION need to be updated we attempt <<
2030 * updates can be handled done independently <<
2031 * <<
2032 * Returns a S_* mask indicating which fields <<
2033 */ <<
2034 int generic_update_time(struct inode *inode, <<
2035 { 1666 {
2036 int updated = inode_update_timestamps !! 1667 int iflags = I_DIRTY_TIME;
2037 int dirty_flags = 0; !! 1668 bool dirty = false;
2038 1669
2039 if (updated & (S_ATIME|S_MTIME|S_CTIM !! 1670 if (flags & S_ATIME)
2040 dirty_flags = inode->i_sb->s_ !! 1671 inode->i_atime = *time;
2041 if (updated & S_VERSION) !! 1672 if (flags & S_VERSION)
2042 dirty_flags |= I_DIRTY_SYNC; !! 1673 dirty = inode_maybe_inc_iversion(inode, false);
2043 __mark_inode_dirty(inode, dirty_flags !! 1674 if (flags & S_CTIME)
2044 return updated; !! 1675 inode->i_ctime = *time;
>> 1676 if (flags & S_MTIME)
>> 1677 inode->i_mtime = *time;
>> 1678 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
>> 1679 !(inode->i_sb->s_flags & SB_LAZYTIME))
>> 1680 dirty = true;
>> 1681
>> 1682 if (dirty)
>> 1683 iflags |= I_DIRTY_SYNC;
>> 1684 __mark_inode_dirty(inode, iflags);
>> 1685 return 0;
2045 } 1686 }
2046 EXPORT_SYMBOL(generic_update_time); 1687 EXPORT_SYMBOL(generic_update_time);
2047 1688
2048 /* 1689 /*
2049 * This does the actual work of updating an i 1690 * This does the actual work of updating an inodes time or version. Must have
2050 * had called mnt_want_write() before calling 1691 * had called mnt_want_write() before calling this.
2051 */ 1692 */
2052 int inode_update_time(struct inode *inode, in !! 1693 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
2053 { 1694 {
2054 if (inode->i_op->update_time) !! 1695 int (*update_time)(struct inode *, struct timespec64 *, int);
2055 return inode->i_op->update_ti !! 1696
2056 generic_update_time(inode, flags); !! 1697 update_time = inode->i_op->update_time ? inode->i_op->update_time :
2057 return 0; !! 1698 generic_update_time;
>> 1699
>> 1700 return update_time(inode, time, flags);
2058 } 1701 }
2059 EXPORT_SYMBOL(inode_update_time); <<
2060 1702
2061 /** 1703 /**
2062 * atime_needs_update - updat !! 1704 * touch_atime - update the access time
2063 * @path: the &struct path to update 1705 * @path: the &struct path to update
2064 * @inode: inode to update 1706 * @inode: inode to update
2065 * 1707 *
2066 * Update the accessed time on an inode 1708 * Update the accessed time on an inode and mark it for writeback.
2067 * This function automatically handles r 1709 * This function automatically handles read only file systems and media,
2068 * as well as the "noatime" flag and ino 1710 * as well as the "noatime" flag and inode specific "noatime" markers.
2069 */ 1711 */
2070 bool atime_needs_update(const struct path *pa !! 1712 bool __atime_needs_update(const struct path *path, struct inode *inode,
>> 1713 bool rcu)
2071 { 1714 {
2072 struct vfsmount *mnt = path->mnt; 1715 struct vfsmount *mnt = path->mnt;
2073 struct timespec64 now, atime; !! 1716 struct timespec64 now;
2074 1717
2075 if (inode->i_flags & S_NOATIME) 1718 if (inode->i_flags & S_NOATIME)
2076 return false; 1719 return false;
2077 1720
2078 /* Atime updates will likely cause i_ 1721 /* Atime updates will likely cause i_uid and i_gid to be written
2079 * back improprely if their true valu 1722 * back improprely if their true value is unknown to the vfs.
2080 */ 1723 */
2081 if (HAS_UNMAPPED_ID(mnt_idmap(mnt), i !! 1724 if (HAS_UNMAPPED_ID(inode))
2082 return false; 1725 return false;
2083 1726
2084 if (IS_NOATIME(inode)) 1727 if (IS_NOATIME(inode))
2085 return false; 1728 return false;
2086 if ((inode->i_sb->s_flags & SB_NODIRA 1729 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
2087 return false; 1730 return false;
2088 1731
2089 if (mnt->mnt_flags & MNT_NOATIME) 1732 if (mnt->mnt_flags & MNT_NOATIME)
2090 return false; 1733 return false;
2091 if ((mnt->mnt_flags & MNT_NODIRATIME) 1734 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
2092 return false; 1735 return false;
2093 1736
2094 now = current_time(inode); 1737 now = current_time(inode);
2095 1738
2096 if (!relatime_need_update(mnt, inode, !! 1739 if (!relatime_need_update(path, inode, timespec64_to_timespec(now), rcu))
2097 return false; 1740 return false;
2098 1741
2099 atime = inode_get_atime(inode); !! 1742 if (timespec64_equal(&inode->i_atime, &now))
2100 if (timespec64_equal(&atime, &now)) <<
2101 return false; 1743 return false;
2102 1744
2103 return true; 1745 return true;
2104 } 1746 }
2105 1747
2106 void touch_atime(const struct path *path) 1748 void touch_atime(const struct path *path)
2107 { 1749 {
2108 struct vfsmount *mnt = path->mnt; 1750 struct vfsmount *mnt = path->mnt;
2109 struct inode *inode = d_inode(path->d 1751 struct inode *inode = d_inode(path->dentry);
>> 1752 struct timespec64 now;
2110 1753
2111 if (!atime_needs_update(path, inode)) !! 1754 if (!__atime_needs_update(path, inode, false))
2112 return; 1755 return;
2113 1756
2114 if (!sb_start_write_trylock(inode->i_ 1757 if (!sb_start_write_trylock(inode->i_sb))
2115 return; 1758 return;
2116 1759
2117 if (mnt_get_write_access(mnt) != 0) !! 1760 if (__mnt_want_write(mnt) != 0)
2118 goto skip_update; 1761 goto skip_update;
2119 /* 1762 /*
2120 * File systems can error out when up 1763 * File systems can error out when updating inodes if they need to
2121 * allocate new space to modify an in 1764 * allocate new space to modify an inode (such is the case for
2122 * Btrfs), but since we touch atime w 1765 * Btrfs), but since we touch atime while walking down the path we
2123 * really don't care if we failed to 1766 * really don't care if we failed to update the atime of the file,
2124 * so just ignore the return value. 1767 * so just ignore the return value.
2125 * We may also fail on filesystems th 1768 * We may also fail on filesystems that have the ability to make parts
2126 * of the fs read only, e.g. subvolum 1769 * of the fs read only, e.g. subvolumes in Btrfs.
2127 */ 1770 */
2128 inode_update_time(inode, S_ATIME); !! 1771 now = current_time(inode);
2129 mnt_put_write_access(mnt); !! 1772 update_time(inode, &now, S_ATIME);
>> 1773 __mnt_drop_write(mnt);
2130 skip_update: 1774 skip_update:
2131 sb_end_write(inode->i_sb); 1775 sb_end_write(inode->i_sb);
2132 } 1776 }
2133 EXPORT_SYMBOL(touch_atime); 1777 EXPORT_SYMBOL(touch_atime);
2134 1778
2135 /* 1779 /*
>> 1780 * The logic we want is
>> 1781 *
>> 1782 * if suid or (sgid and xgrp)
>> 1783 * remove privs
>> 1784 */
>> 1785 int should_remove_suid(struct dentry *dentry)
>> 1786 {
>> 1787 umode_t mode = d_inode(dentry)->i_mode;
>> 1788 int kill = 0;
>> 1789
>> 1790 /* suid always must be killed */
>> 1791 if (unlikely(mode & S_ISUID))
>> 1792 kill = ATTR_KILL_SUID;
>> 1793
>> 1794 /*
>> 1795 * sgid without any exec bits is just a mandatory locking mark; leave
>> 1796 * it alone. If some exec bits are set, it's a real sgid; kill it.
>> 1797 */
>> 1798 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
>> 1799 kill |= ATTR_KILL_SGID;
>> 1800
>> 1801 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
>> 1802 return kill;
>> 1803
>> 1804 return 0;
>> 1805 }
>> 1806 EXPORT_SYMBOL(should_remove_suid);
>> 1807
>> 1808 /*
2136 * Return mask of changes for notify_change() 1809 * Return mask of changes for notify_change() that need to be done as a
2137 * response to write or truncate. Return 0 if 1810 * response to write or truncate. Return 0 if nothing has to be changed.
2138 * Negative value on error (change should be 1811 * Negative value on error (change should be denied).
2139 */ 1812 */
2140 int dentry_needs_remove_privs(struct mnt_idma !! 1813 int dentry_needs_remove_privs(struct dentry *dentry)
2141 struct dentry * <<
2142 { 1814 {
2143 struct inode *inode = d_inode(dentry) 1815 struct inode *inode = d_inode(dentry);
2144 int mask = 0; 1816 int mask = 0;
2145 int ret; 1817 int ret;
2146 1818
2147 if (IS_NOSEC(inode)) 1819 if (IS_NOSEC(inode))
2148 return 0; 1820 return 0;
2149 1821
2150 mask = setattr_should_drop_suidgid(id !! 1822 mask = should_remove_suid(dentry);
2151 ret = security_inode_need_killpriv(de 1823 ret = security_inode_need_killpriv(dentry);
2152 if (ret < 0) 1824 if (ret < 0)
2153 return ret; 1825 return ret;
2154 if (ret) 1826 if (ret)
2155 mask |= ATTR_KILL_PRIV; 1827 mask |= ATTR_KILL_PRIV;
2156 return mask; 1828 return mask;
2157 } 1829 }
2158 1830
2159 static int __remove_privs(struct mnt_idmap *i !! 1831 static int __remove_privs(struct dentry *dentry, int kill)
2160 struct dentry *dent <<
2161 { 1832 {
2162 struct iattr newattrs; 1833 struct iattr newattrs;
2163 1834
2164 newattrs.ia_valid = ATTR_FORCE | kill 1835 newattrs.ia_valid = ATTR_FORCE | kill;
2165 /* 1836 /*
2166 * Note we call this on write, so not 1837 * Note we call this on write, so notify_change will not
2167 * encounter any conflicting delegati 1838 * encounter any conflicting delegations:
2168 */ 1839 */
2169 return notify_change(idmap, dentry, & !! 1840 return notify_change(dentry, &newattrs, NULL);
2170 } 1841 }
2171 1842
2172 int file_remove_privs_flags(struct file *file !! 1843 /*
>> 1844 * Remove special file priviledges (suid, capabilities) when file is written
>> 1845 * to or truncated.
>> 1846 */
>> 1847 int file_remove_privs(struct file *file)
2173 { 1848 {
2174 struct dentry *dentry = file_dentry(f 1849 struct dentry *dentry = file_dentry(file);
2175 struct inode *inode = file_inode(file 1850 struct inode *inode = file_inode(file);
2176 int error = 0; <<
2177 int kill; 1851 int kill;
>> 1852 int error = 0;
2178 1853
2179 if (IS_NOSEC(inode) || !S_ISREG(inode !! 1854 /* Fast path for nothing security related */
>> 1855 if (IS_NOSEC(inode))
2180 return 0; 1856 return 0;
2181 1857
2182 kill = dentry_needs_remove_privs(file !! 1858 kill = dentry_needs_remove_privs(dentry);
2183 if (kill < 0) 1859 if (kill < 0)
2184 return kill; 1860 return kill;
2185 !! 1861 if (kill)
2186 if (kill) { !! 1862 error = __remove_privs(dentry, kill);
2187 if (flags & IOCB_NOWAIT) <<
2188 return -EAGAIN; <<
2189 <<
2190 error = __remove_privs(file_m <<
2191 } <<
2192 <<
2193 if (!error) 1863 if (!error)
2194 inode_has_no_xattr(inode); 1864 inode_has_no_xattr(inode);
>> 1865
2195 return error; 1866 return error;
2196 } 1867 }
2197 EXPORT_SYMBOL_GPL(file_remove_privs_flags); !! 1868 EXPORT_SYMBOL(file_remove_privs);
2198 1869
2199 /** 1870 /**
2200 * file_remove_privs - remove special file pr !! 1871 * file_update_time - update mtime and ctime time
2201 * @file: file to remove privileges from !! 1872 * @file: file accessed
2202 * 1873 *
2203 * When file is modified by a write or trunca !! 1874 * Update the mtime and ctime members of an inode and mark the inode
2204 * file privileges are removed. !! 1875 * for writeback. Note that this function is meant exclusively for
2205 * !! 1876 * usage in the file write path of filesystems, and filesystems may
2206 * Return: 0 on success, negative errno on fa !! 1877 * choose to explicitly ignore update via this function with the
>> 1878 * S_NOCMTIME inode flag, e.g. for network filesystem where these
>> 1879 * timestamps are handled by the server. This can return an error for
>> 1880 * file systems who need to allocate space in order to update an inode.
2207 */ 1881 */
2208 int file_remove_privs(struct file *file) <<
2209 { <<
2210 return file_remove_privs_flags(file, <<
2211 } <<
2212 EXPORT_SYMBOL(file_remove_privs); <<
2213 1882
2214 static int inode_needs_update_time(struct ino !! 1883 int file_update_time(struct file *file)
2215 { 1884 {
>> 1885 struct inode *inode = file_inode(file);
>> 1886 struct timespec64 now;
2216 int sync_it = 0; 1887 int sync_it = 0;
2217 struct timespec64 now = current_time( !! 1888 int ret;
2218 struct timespec64 ts; <<
2219 1889
2220 /* First try to exhaust all avenues t 1890 /* First try to exhaust all avenues to not sync */
2221 if (IS_NOCMTIME(inode)) 1891 if (IS_NOCMTIME(inode))
2222 return 0; 1892 return 0;
2223 1893
2224 ts = inode_get_mtime(inode); !! 1894 now = current_time(inode);
2225 if (!timespec64_equal(&ts, &now)) !! 1895 if (!timespec64_equal(&inode->i_mtime, &now))
2226 sync_it = S_MTIME; 1896 sync_it = S_MTIME;
2227 1897
2228 ts = inode_get_ctime(inode); !! 1898 if (!timespec64_equal(&inode->i_ctime, &now))
2229 if (!timespec64_equal(&ts, &now)) <<
2230 sync_it |= S_CTIME; 1899 sync_it |= S_CTIME;
2231 1900
2232 if (IS_I_VERSION(inode) && inode_iver 1901 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2233 sync_it |= S_VERSION; 1902 sync_it |= S_VERSION;
2234 1903
2235 return sync_it; !! 1904 if (!sync_it)
2236 } !! 1905 return 0;
2237 1906
2238 static int __file_update_time(struct file *fi !! 1907 /* Finally allowed to write? Takes lock. */
2239 { !! 1908 if (__mnt_want_write_file(file))
2240 int ret = 0; !! 1909 return 0;
2241 struct inode *inode = file_inode(file <<
2242 1910
2243 /* try to update time settings */ !! 1911 ret = update_time(inode, &now, sync_it);
2244 if (!mnt_get_write_access_file(file)) !! 1912 __mnt_drop_write_file(file);
2245 ret = inode_update_time(inode <<
2246 mnt_put_write_access_file(fil <<
2247 } <<
2248 1913
2249 return ret; 1914 return ret;
2250 } 1915 }
2251 <<
2252 /** <<
2253 * file_update_time - update mtime and ctime <<
2254 * @file: file accessed <<
2255 * <<
2256 * Update the mtime and ctime members of an i <<
2257 * writeback. Note that this function is mean <<
2258 * the file write path of filesystems, and fi <<
2259 * explicitly ignore updates via this functio <<
2260 * flag, e.g. for network filesystem where th <<
2261 * by the server. This can return an error fo <<
2262 * allocate space in order to update an inode <<
2263 * <<
2264 * Return: 0 on success, negative errno on fa <<
2265 */ <<
2266 int file_update_time(struct file *file) <<
2267 { <<
2268 int ret; <<
2269 struct inode *inode = file_inode(file <<
2270 <<
2271 ret = inode_needs_update_time(inode); <<
2272 if (ret <= 0) <<
2273 return ret; <<
2274 <<
2275 return __file_update_time(file, ret); <<
2276 } <<
2277 EXPORT_SYMBOL(file_update_time); 1916 EXPORT_SYMBOL(file_update_time);
2278 1917
2279 /** <<
2280 * file_modified_flags - handle mandated vfs <<
2281 * @file: file that was modified <<
2282 * @flags: kiocb flags <<
2283 * <<
2284 * When file has been modified ensure that sp <<
2285 * file privileges are removed and time setti <<
2286 * <<
2287 * If IOCB_NOWAIT is set, special file privil <<
2288 * time settings will not be updated. It will <<
2289 * <<
2290 * Context: Caller must hold the file's inode <<
2291 * <<
2292 * Return: 0 on success, negative errno on fa <<
2293 */ <<
2294 static int file_modified_flags(struct file *f <<
2295 { <<
2296 int ret; <<
2297 struct inode *inode = file_inode(file <<
2298 <<
2299 /* <<
2300 * Clear the security bits if the pro <<
2301 * This keeps people from modifying s <<
2302 */ <<
2303 ret = file_remove_privs_flags(file, f <<
2304 if (ret) <<
2305 return ret; <<
2306 <<
2307 if (unlikely(file->f_mode & FMODE_NOC <<
2308 return 0; <<
2309 <<
2310 ret = inode_needs_update_time(inode); <<
2311 if (ret <= 0) <<
2312 return ret; <<
2313 if (flags & IOCB_NOWAIT) <<
2314 return -EAGAIN; <<
2315 <<
2316 return __file_update_time(file, ret); <<
2317 } <<
2318 <<
2319 /** <<
2320 * file_modified - handle mandated vfs change <<
2321 * @file: file that was modified <<
2322 * <<
2323 * When file has been modified ensure that sp <<
2324 * file privileges are removed and time setti <<
2325 * <<
2326 * Context: Caller must hold the file's inode <<
2327 * <<
2328 * Return: 0 on success, negative errno on fa <<
2329 */ <<
2330 int file_modified(struct file *file) <<
2331 { <<
2332 return file_modified_flags(file, 0); <<
2333 } <<
2334 EXPORT_SYMBOL(file_modified); <<
2335 <<
2336 /** <<
2337 * kiocb_modified - handle mandated vfs chang <<
2338 * @iocb: iocb that was modified <<
2339 * <<
2340 * When file has been modified ensure that sp <<
2341 * file privileges are removed and time setti <<
2342 * <<
2343 * Context: Caller must hold the file's inode <<
2344 * <<
2345 * Return: 0 on success, negative errno on fa <<
2346 */ <<
2347 int kiocb_modified(struct kiocb *iocb) <<
2348 { <<
2349 return file_modified_flags(iocb->ki_f <<
2350 } <<
2351 EXPORT_SYMBOL_GPL(kiocb_modified); <<
2352 <<
2353 int inode_needs_sync(struct inode *inode) 1918 int inode_needs_sync(struct inode *inode)
2354 { 1919 {
2355 if (IS_SYNC(inode)) 1920 if (IS_SYNC(inode))
2356 return 1; 1921 return 1;
2357 if (S_ISDIR(inode->i_mode) && IS_DIRS 1922 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2358 return 1; 1923 return 1;
2359 return 0; 1924 return 0;
2360 } 1925 }
2361 EXPORT_SYMBOL(inode_needs_sync); 1926 EXPORT_SYMBOL(inode_needs_sync);
2362 1927
2363 /* 1928 /*
2364 * If we try to find an inode in the inode ha 1929 * If we try to find an inode in the inode hash while it is being
2365 * deleted, we have to wait until the filesys 1930 * deleted, we have to wait until the filesystem completes its
2366 * deletion before reporting that it isn't fo 1931 * deletion before reporting that it isn't found. This function waits
2367 * until the deletion _might_ have completed. 1932 * until the deletion _might_ have completed. Callers are responsible
2368 * to recheck inode state. 1933 * to recheck inode state.
2369 * 1934 *
2370 * It doesn't matter if I_NEW is not set init 1935 * It doesn't matter if I_NEW is not set initially, a call to
2371 * wake_up_bit(&inode->i_state, __I_NEW) afte 1936 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2372 * will DTRT. 1937 * will DTRT.
2373 */ 1938 */
2374 static void __wait_on_freeing_inode(struct in !! 1939 static void __wait_on_freeing_inode(struct inode *inode)
2375 { 1940 {
2376 struct wait_bit_queue_entry wqe; !! 1941 wait_queue_head_t *wq;
2377 struct wait_queue_head *wq_head; !! 1942 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2378 !! 1943 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2379 /* !! 1944 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2380 * Handle racing against evict(), see <<
2381 */ <<
2382 if (unlikely(inode_unhashed(inode))) <<
2383 WARN_ON(is_inode_hash_locked) <<
2384 spin_unlock(&inode->i_lock); <<
2385 return; <<
2386 } <<
2387 <<
2388 wq_head = inode_bit_waitqueue(&wqe, i <<
2389 prepare_to_wait_event(wq_head, &wqe.w <<
2390 spin_unlock(&inode->i_lock); 1945 spin_unlock(&inode->i_lock);
2391 rcu_read_unlock(); !! 1946 spin_unlock(&inode_hash_lock);
2392 if (is_inode_hash_locked) <<
2393 spin_unlock(&inode_hash_lock) <<
2394 schedule(); 1947 schedule();
2395 finish_wait(wq_head, &wqe.wq_entry); !! 1948 finish_wait(wq, &wait.wq_entry);
2396 if (is_inode_hash_locked) !! 1949 spin_lock(&inode_hash_lock);
2397 spin_lock(&inode_hash_lock); <<
2398 rcu_read_lock(); <<
2399 } 1950 }
2400 1951
2401 static __initdata unsigned long ihash_entries 1952 static __initdata unsigned long ihash_entries;
2402 static int __init set_ihash_entries(char *str 1953 static int __init set_ihash_entries(char *str)
2403 { 1954 {
2404 if (!str) 1955 if (!str)
2405 return 0; 1956 return 0;
2406 ihash_entries = simple_strtoul(str, & 1957 ihash_entries = simple_strtoul(str, &str, 0);
2407 return 1; 1958 return 1;
2408 } 1959 }
2409 __setup("ihash_entries=", set_ihash_entries); 1960 __setup("ihash_entries=", set_ihash_entries);
2410 1961
2411 /* 1962 /*
2412 * Initialize the waitqueues and inode hash t 1963 * Initialize the waitqueues and inode hash table.
2413 */ 1964 */
2414 void __init inode_init_early(void) 1965 void __init inode_init_early(void)
2415 { 1966 {
2416 /* If hashes are distributed across N 1967 /* If hashes are distributed across NUMA nodes, defer
2417 * hash allocation until vmalloc spac 1968 * hash allocation until vmalloc space is available.
2418 */ 1969 */
2419 if (hashdist) 1970 if (hashdist)
2420 return; 1971 return;
2421 1972
2422 inode_hashtable = 1973 inode_hashtable =
2423 alloc_large_system_hash("Inod 1974 alloc_large_system_hash("Inode-cache",
2424 sizeo 1975 sizeof(struct hlist_head),
2425 ihash 1976 ihash_entries,
2426 14, 1977 14,
2427 HASH_ 1978 HASH_EARLY | HASH_ZERO,
2428 &i_ha 1979 &i_hash_shift,
2429 &i_ha 1980 &i_hash_mask,
2430 0, 1981 0,
2431 0); 1982 0);
2432 } 1983 }
2433 1984
2434 void __init inode_init(void) 1985 void __init inode_init(void)
2435 { 1986 {
2436 /* inode slab cache */ 1987 /* inode slab cache */
2437 inode_cachep = kmem_cache_create("ino 1988 inode_cachep = kmem_cache_create("inode_cache",
2438 size 1989 sizeof(struct inode),
2439 0, 1990 0,
2440 (SLA 1991 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2441 SLAB !! 1992 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2442 init 1993 init_once);
2443 1994
2444 /* Hash may have been set up in inode 1995 /* Hash may have been set up in inode_init_early */
2445 if (!hashdist) 1996 if (!hashdist)
2446 return; 1997 return;
2447 1998
2448 inode_hashtable = 1999 inode_hashtable =
2449 alloc_large_system_hash("Inod 2000 alloc_large_system_hash("Inode-cache",
2450 sizeo 2001 sizeof(struct hlist_head),
2451 ihash 2002 ihash_entries,
2452 14, 2003 14,
2453 HASH_ 2004 HASH_ZERO,
2454 &i_ha 2005 &i_hash_shift,
2455 &i_ha 2006 &i_hash_mask,
2456 0, 2007 0,
2457 0); 2008 0);
2458 } 2009 }
2459 2010
2460 void init_special_inode(struct inode *inode, 2011 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2461 { 2012 {
2462 inode->i_mode = mode; 2013 inode->i_mode = mode;
2463 if (S_ISCHR(mode)) { 2014 if (S_ISCHR(mode)) {
2464 inode->i_fop = &def_chr_fops; 2015 inode->i_fop = &def_chr_fops;
2465 inode->i_rdev = rdev; 2016 inode->i_rdev = rdev;
2466 } else if (S_ISBLK(mode)) { 2017 } else if (S_ISBLK(mode)) {
2467 if (IS_ENABLED(CONFIG_BLOCK)) !! 2018 inode->i_fop = &def_blk_fops;
2468 inode->i_fop = &def_b <<
2469 inode->i_rdev = rdev; 2019 inode->i_rdev = rdev;
2470 } else if (S_ISFIFO(mode)) 2020 } else if (S_ISFIFO(mode))
2471 inode->i_fop = &pipefifo_fops 2021 inode->i_fop = &pipefifo_fops;
2472 else if (S_ISSOCK(mode)) 2022 else if (S_ISSOCK(mode))
2473 ; /* leave it no_open_f 2023 ; /* leave it no_open_fops */
2474 else 2024 else
2475 printk(KERN_DEBUG "init_speci 2025 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2476 " inode %s: 2026 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2477 inode->i_in 2027 inode->i_ino);
2478 } 2028 }
2479 EXPORT_SYMBOL(init_special_inode); 2029 EXPORT_SYMBOL(init_special_inode);
2480 2030
2481 /** 2031 /**
2482 * inode_init_owner - Init uid,gid,mode for n 2032 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2483 * @idmap: idmap of the mount the inode was c <<
2484 * @inode: New inode 2033 * @inode: New inode
2485 * @dir: Directory inode 2034 * @dir: Directory inode
2486 * @mode: mode of the new inode 2035 * @mode: mode of the new inode
2487 * <<
2488 * If the inode has been created through an i <<
2489 * the vfsmount must be passed through @idmap <<
2490 * care to map the inode according to @idmap <<
2491 * and initializing i_uid and i_gid. On non-i <<
2492 * checking is to be performed on the raw ino <<
2493 */ 2036 */
2494 void inode_init_owner(struct mnt_idmap *idmap !! 2037 void inode_init_owner(struct inode *inode, const struct inode *dir,
2495 const struct inode *dir !! 2038 umode_t mode)
2496 { 2039 {
2497 inode_fsuid_set(inode, idmap); !! 2040 inode->i_uid = current_fsuid();
2498 if (dir && dir->i_mode & S_ISGID) { 2041 if (dir && dir->i_mode & S_ISGID) {
2499 inode->i_gid = dir->i_gid; 2042 inode->i_gid = dir->i_gid;
2500 2043
2501 /* Directories are special, a 2044 /* Directories are special, and always inherit S_ISGID */
2502 if (S_ISDIR(mode)) 2045 if (S_ISDIR(mode))
2503 mode |= S_ISGID; 2046 mode |= S_ISGID;
>> 2047 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
>> 2048 !in_group_p(inode->i_gid) &&
>> 2049 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
>> 2050 mode &= ~S_ISGID;
2504 } else 2051 } else
2505 inode_fsgid_set(inode, idmap) !! 2052 inode->i_gid = current_fsgid();
2506 inode->i_mode = mode; 2053 inode->i_mode = mode;
2507 } 2054 }
2508 EXPORT_SYMBOL(inode_init_owner); 2055 EXPORT_SYMBOL(inode_init_owner);
2509 2056
2510 /** 2057 /**
2511 * inode_owner_or_capable - check current tas 2058 * inode_owner_or_capable - check current task permissions to inode
2512 * @idmap: idmap of the mount the inode was f <<
2513 * @inode: inode being checked 2059 * @inode: inode being checked
2514 * 2060 *
2515 * Return true if current either has CAP_FOWN 2061 * Return true if current either has CAP_FOWNER in a namespace with the
2516 * inode owner uid mapped, or owns the file. 2062 * inode owner uid mapped, or owns the file.
2517 * <<
2518 * If the inode has been found through an idm <<
2519 * the vfsmount must be passed through @idmap <<
2520 * care to map the inode according to @idmap <<
2521 * On non-idmapped mounts or if permission ch <<
2522 * raw inode simply pass @nop_mnt_idmap. <<
2523 */ 2063 */
2524 bool inode_owner_or_capable(struct mnt_idmap !! 2064 bool inode_owner_or_capable(const struct inode *inode)
2525 const struct inod <<
2526 { 2065 {
2527 vfsuid_t vfsuid; <<
2528 struct user_namespace *ns; 2066 struct user_namespace *ns;
2529 2067
2530 vfsuid = i_uid_into_vfsuid(idmap, ino !! 2068 if (uid_eq(current_fsuid(), inode->i_uid))
2531 if (vfsuid_eq_kuid(vfsuid, current_fs <<
2532 return true; 2069 return true;
2533 2070
2534 ns = current_user_ns(); 2071 ns = current_user_ns();
2535 if (vfsuid_has_mapping(ns, vfsuid) && !! 2072 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2536 return true; 2073 return true;
2537 return false; 2074 return false;
2538 } 2075 }
2539 EXPORT_SYMBOL(inode_owner_or_capable); 2076 EXPORT_SYMBOL(inode_owner_or_capable);
2540 2077
2541 /* 2078 /*
2542 * Direct i/o helper functions 2079 * Direct i/o helper functions
2543 */ 2080 */
2544 bool inode_dio_finished(const struct inode *i !! 2081 static void __inode_dio_wait(struct inode *inode)
2545 { 2082 {
2546 return atomic_read(&inode->i_dio_coun !! 2083 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
>> 2084 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
>> 2085
>> 2086 do {
>> 2087 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
>> 2088 if (atomic_read(&inode->i_dio_count))
>> 2089 schedule();
>> 2090 } while (atomic_read(&inode->i_dio_count));
>> 2091 finish_wait(wq, &q.wq_entry);
2547 } 2092 }
2548 EXPORT_SYMBOL(inode_dio_finished); <<
2549 2093
2550 /** 2094 /**
2551 * inode_dio_wait - wait for outstanding DIO 2095 * inode_dio_wait - wait for outstanding DIO requests to finish
2552 * @inode: inode to wait for 2096 * @inode: inode to wait for
2553 * 2097 *
2554 * Waits for all pending direct I/O requests 2098 * Waits for all pending direct I/O requests to finish so that we can
2555 * proceed with a truncate or equivalent oper 2099 * proceed with a truncate or equivalent operation.
2556 * 2100 *
2557 * Must be called under a lock that serialize 2101 * Must be called under a lock that serializes taking new references
2558 * to i_dio_count, usually by inode->i_mutex. 2102 * to i_dio_count, usually by inode->i_mutex.
2559 */ 2103 */
2560 void inode_dio_wait(struct inode *inode) 2104 void inode_dio_wait(struct inode *inode)
2561 { 2105 {
2562 wait_var_event(&inode->i_dio_count, i !! 2106 if (atomic_read(&inode->i_dio_count))
>> 2107 __inode_dio_wait(inode);
2563 } 2108 }
2564 EXPORT_SYMBOL(inode_dio_wait); 2109 EXPORT_SYMBOL(inode_dio_wait);
2565 2110
2566 void inode_dio_wait_interruptible(struct inod <<
2567 { <<
2568 wait_var_event_interruptible(&inode-> <<
2569 inode_di <<
2570 } <<
2571 EXPORT_SYMBOL(inode_dio_wait_interruptible); <<
2572 <<
2573 /* 2111 /*
2574 * inode_set_flags - atomically set some inod 2112 * inode_set_flags - atomically set some inode flags
2575 * 2113 *
2576 * Note: the caller should be holding i_mutex 2114 * Note: the caller should be holding i_mutex, or else be sure that
2577 * they have exclusive access to the inode st 2115 * they have exclusive access to the inode structure (i.e., while the
2578 * inode is being instantiated). The reason 2116 * inode is being instantiated). The reason for the cmpxchg() loop
2579 * --- which wouldn't be necessary if all cod 2117 * --- which wouldn't be necessary if all code paths which modify
2580 * i_flags actually followed this rule, is th 2118 * i_flags actually followed this rule, is that there is at least one
2581 * code path which doesn't today so we use cm 2119 * code path which doesn't today so we use cmpxchg() out of an abundance
2582 * of caution. 2120 * of caution.
2583 * 2121 *
2584 * In the long run, i_mutex is overkill, and 2122 * In the long run, i_mutex is overkill, and we should probably look
2585 * at using the i_lock spinlock to protect i_ 2123 * at using the i_lock spinlock to protect i_flags, and then make sure
2586 * it is so documented in include/linux/fs.h 2124 * it is so documented in include/linux/fs.h and that all code follows
2587 * the locking convention!! 2125 * the locking convention!!
2588 */ 2126 */
2589 void inode_set_flags(struct inode *inode, uns 2127 void inode_set_flags(struct inode *inode, unsigned int flags,
2590 unsigned int mask) 2128 unsigned int mask)
2591 { 2129 {
>> 2130 unsigned int old_flags, new_flags;
>> 2131
2592 WARN_ON_ONCE(flags & ~mask); 2132 WARN_ON_ONCE(flags & ~mask);
2593 set_mask_bits(&inode->i_flags, mask, !! 2133 do {
>> 2134 old_flags = READ_ONCE(inode->i_flags);
>> 2135 new_flags = (old_flags & ~mask) | flags;
>> 2136 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
>> 2137 new_flags) != old_flags));
2594 } 2138 }
2595 EXPORT_SYMBOL(inode_set_flags); 2139 EXPORT_SYMBOL(inode_set_flags);
2596 2140
2597 void inode_nohighmem(struct inode *inode) 2141 void inode_nohighmem(struct inode *inode)
2598 { 2142 {
2599 mapping_set_gfp_mask(inode->i_mapping 2143 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2600 } 2144 }
2601 EXPORT_SYMBOL(inode_nohighmem); 2145 EXPORT_SYMBOL(inode_nohighmem);
2602 2146
2603 /** 2147 /**
2604 * timestamp_truncate - Truncate timespec to !! 2148 * timespec64_trunc - Truncate timespec64 to a granularity
2605 * @t: Timespec !! 2149 * @t: Timespec64
2606 * @inode: inode being updated !! 2150 * @gran: Granularity in ns.
2607 * 2151 *
2608 * Truncate a timespec to the granularity sup !! 2152 * Truncate a timespec64 to a granularity. Always rounds down. gran must
2609 * containing the inode. Always rounds down. <<
2610 * not be 0 nor greater than a second (NSEC_P 2153 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2611 */ 2154 */
2612 struct timespec64 timestamp_truncate(struct t !! 2155 struct timespec64 timespec64_trunc(struct timespec64 t, unsigned gran)
2613 { 2156 {
2614 struct super_block *sb = inode->i_sb; <<
2615 unsigned int gran = sb->s_time_gran; <<
2616 <<
2617 t.tv_sec = clamp(t.tv_sec, sb->s_time <<
2618 if (unlikely(t.tv_sec == sb->s_time_m <<
2619 t.tv_nsec = 0; <<
2620 <<
2621 /* Avoid division in the common cases 2157 /* Avoid division in the common cases 1 ns and 1 s. */
2622 if (gran == 1) !! 2158 if (gran == 1) {
2623 ; /* nothing */ !! 2159 /* nothing */
2624 else if (gran == NSEC_PER_SEC) !! 2160 } else if (gran == NSEC_PER_SEC) {
2625 t.tv_nsec = 0; 2161 t.tv_nsec = 0;
2626 else if (gran > 1 && gran < NSEC_PER_ !! 2162 } else if (gran > 1 && gran < NSEC_PER_SEC) {
2627 t.tv_nsec -= t.tv_nsec % gran 2163 t.tv_nsec -= t.tv_nsec % gran;
2628 else !! 2164 } else {
2629 WARN(1, "invalid file time gr !! 2165 WARN(1, "illegal file time granularity: %u", gran);
>> 2166 }
2630 return t; 2167 return t;
2631 } 2168 }
2632 EXPORT_SYMBOL(timestamp_truncate); !! 2169 EXPORT_SYMBOL(timespec64_trunc);
2633 2170
2634 /** 2171 /**
2635 * current_time - Return FS time 2172 * current_time - Return FS time
2636 * @inode: inode. 2173 * @inode: inode.
2637 * 2174 *
2638 * Return the current time truncated to the t 2175 * Return the current time truncated to the time granularity supported by
2639 * the fs. 2176 * the fs.
2640 * 2177 *
2641 * Note that inode and inode->sb cannot be NU 2178 * Note that inode and inode->sb cannot be NULL.
2642 * Otherwise, the function warns and returns 2179 * Otherwise, the function warns and returns time without truncation.
2643 */ 2180 */
2644 struct timespec64 current_time(struct inode * 2181 struct timespec64 current_time(struct inode *inode)
2645 { 2182 {
2646 struct timespec64 now; !! 2183 struct timespec64 now = current_kernel_time64();
2647 <<
2648 ktime_get_coarse_real_ts64(&now); <<
2649 return timestamp_truncate(now, inode) <<
2650 } <<
2651 EXPORT_SYMBOL(current_time); <<
2652 <<
2653 /** <<
2654 * inode_set_ctime_current - set the ctime to <<
2655 * @inode: inode <<
2656 * <<
2657 * Set the inode->i_ctime to the current valu <<
2658 * the current value that was assigned to i_c <<
2659 */ <<
2660 struct timespec64 inode_set_ctime_current(str <<
2661 { <<
2662 struct timespec64 now = current_time( <<
2663 2184
2664 inode_set_ctime_to_ts(inode, now); !! 2185 if (unlikely(!inode->i_sb)) {
2665 return now; !! 2186 WARN(1, "current_time() called with uninitialized super_block in the inode");
2666 } !! 2187 return now;
2667 EXPORT_SYMBOL(inode_set_ctime_current); !! 2188 }
2668 <<
2669 /** <<
2670 * in_group_or_capable - check whether caller <<
2671 * @idmap: idmap of the mount @inode was <<
2672 * @inode: inode to check <<
2673 * @vfsgid: the new/current vfsgid of @in <<
2674 * <<
2675 * Check wether @vfsgid is in the caller's gr <<
2676 * privileged with CAP_FSETID over @inode. Th <<
2677 * whether the setgid bit can be kept or must <<
2678 * <<
2679 * Return: true if the caller is sufficiently <<
2680 */ <<
2681 bool in_group_or_capable(struct mnt_idmap *id <<
2682 const struct inode * <<
2683 { <<
2684 if (vfsgid_in_group_p(vfsgid)) <<
2685 return true; <<
2686 if (capable_wrt_inode_uidgid(idmap, i <<
2687 return true; <<
2688 return false; <<
2689 } <<
2690 EXPORT_SYMBOL(in_group_or_capable); <<
2691 2189
2692 /** !! 2190 return timespec64_trunc(now, inode->i_sb->s_time_gran);
2693 * mode_strip_sgid - handle the sgid bit for <<
2694 * @idmap: idmap of the mount the inode was c <<
2695 * @dir: parent directory inode <<
2696 * @mode: mode of the file to be created in @ <<
2697 * <<
2698 * If the @mode of the new file has both the <<
2699 * raised and @dir has the S_ISGID bit raised <<
2700 * either in the group of the parent director <<
2701 * in their user namespace and are privileged <<
2702 * In all other cases, strip the S_ISGID bit <<
2703 * <<
2704 * Return: the new mode to use for the file <<
2705 */ <<
2706 umode_t mode_strip_sgid(struct mnt_idmap *idm <<
2707 const struct inode *d <<
2708 { <<
2709 if ((mode & (S_ISGID | S_IXGRP)) != ( <<
2710 return mode; <<
2711 if (S_ISDIR(mode) || !dir || !(dir->i <<
2712 return mode; <<
2713 if (in_group_or_capable(idmap, dir, i <<
2714 return mode; <<
2715 return mode & ~S_ISGID; <<
2716 } 2191 }
2717 EXPORT_SYMBOL(mode_strip_sgid); !! 2192 EXPORT_SYMBOL(current_time);
2718 2193
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.