1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_IVERSION_H 3 #define _LINUX_IVERSION_H 4 5 #include <linux/fs.h> 6 7 /* 8 * The inode->i_version field: 9 * --------------------------- 10 * The change attribute (i_version) is mandated by NFSv4 and is mostly for 11 * knfsd, but is also used for other purposes (e.g. IMA). The i_version must 12 * appear different to observers if there was a change to the inode's data or 13 * metadata since it was last queried. 14 * 15 * Observers see the i_version as a 64-bit number that never decreases. If it 16 * remains the same since it was last checked, then nothing has changed in the 17 * inode. If it's different then something has changed. Observers cannot infer 18 * anything about the nature or magnitude of the changes from the value, only 19 * that the inode has changed in some fashion. 20 * 21 * Not all filesystems properly implement the i_version counter. Subsystems that 22 * want to use i_version field on an inode should first check whether the 23 * filesystem sets the SB_I_VERSION flag (usually via the IS_I_VERSION macro). 24 * 25 * Those that set SB_I_VERSION will automatically have their i_version counter 26 * incremented on writes to normal files. If the SB_I_VERSION is not set, then 27 * the VFS will not touch it on writes, and the filesystem can use it how it 28 * wishes. Note that the filesystem is always responsible for updating the 29 * i_version on namespace changes in directories (mkdir, rmdir, unlink, etc.). 30 * We consider these sorts of filesystems to have a kernel-managed i_version. 31 * 32 * It may be impractical for filesystems to keep i_version updates atomic with 33 * respect to the changes that cause them. They should, however, guarantee 34 * that i_version updates are never visible before the changes that caused 35 * them. Also, i_version updates should never be delayed longer than it takes 36 * the original change to reach disk. 37 * 38 * This implementation uses the low bit in the i_version field as a flag to 39 * track when the value has been queried. If it has not been queried since it 40 * was last incremented, we can skip the increment in most cases. 41 * 42 * In the event that we're updating the ctime, we will usually go ahead and 43 * bump the i_version anyway. Since that has to go to stable storage in some 44 * fashion, we might as well increment it as well. 45 * 46 * With this implementation, the value should always appear to observers to 47 * increase over time if the file has changed. It's recommended to use 48 * inode_eq_iversion() helper to compare values. 49 * 50 * Note that some filesystems (e.g. NFS and AFS) just use the field to store 51 * a server-provided value (for the most part). For that reason, those 52 * filesystems do not set SB_I_VERSION. These filesystems are considered to 53 * have a self-managed i_version. 54 * 55 * Persistently storing the i_version 56 * ---------------------------------- 57 * Queries of the i_version field are not gated on them hitting the backing 58 * store. It's always possible that the host could crash after allowing 59 * a query of the value but before it has made it to disk. 60 * 61 * To mitigate this problem, filesystems should always use 62 * inode_set_iversion_queried when loading an existing inode from disk. This 63 * ensures that the next attempted inode increment will result in the value 64 * changing. 65 * 66 * Storing the value to disk therefore does not count as a query, so those 67 * filesystems should use inode_peek_iversion to grab the value to be stored. 68 * There is no need to flag the value as having been queried in that case. 69 */ 70 71 /* 72 * We borrow the lowest bit in the i_version to use as a flag to tell whether 73 * it has been queried since we last incremented it. If it has, then we must 74 * increment it on the next change. After that, we can clear the flag and 75 * avoid incrementing it again until it has again been queried. 76 */ 77 #define I_VERSION_QUERIED_SHIFT (1) 78 #define I_VERSION_QUERIED (1ULL << (I_VERSION_QUERIED_SHIFT - 1)) 79 #define I_VERSION_INCREMENT (1ULL << I_VERSION_QUERIED_SHIFT) 80 81 /** 82 * inode_set_iversion_raw - set i_version to the specified raw value 83 * @inode: inode to set 84 * @val: new i_version value to set 85 * 86 * Set @inode's i_version field to @val. This function is for use by 87 * filesystems that self-manage the i_version. 88 * 89 * For example, the NFS client stores its NFSv4 change attribute in this way, 90 * and the AFS client stores the data_version from the server here. 91 */ 92 static inline void 93 inode_set_iversion_raw(struct inode *inode, u64 val) 94 { 95 atomic64_set(&inode->i_version, val); 96 } 97 98 /** 99 * inode_peek_iversion_raw - grab a "raw" iversion value 100 * @inode: inode from which i_version should be read 101 * 102 * Grab a "raw" inode->i_version value and return it. The i_version is not 103 * flagged or converted in any way. This is mostly used to access a self-managed 104 * i_version. 105 * 106 * With those filesystems, we want to treat the i_version as an entirely 107 * opaque value. 108 */ 109 static inline u64 110 inode_peek_iversion_raw(const struct inode *inode) 111 { 112 return atomic64_read(&inode->i_version); 113 } 114 115 /** 116 * inode_set_max_iversion_raw - update i_version new value is larger 117 * @inode: inode to set 118 * @val: new i_version to set 119 * 120 * Some self-managed filesystems (e.g Ceph) will only update the i_version 121 * value if the new value is larger than the one we already have. 122 */ 123 static inline void 124 inode_set_max_iversion_raw(struct inode *inode, u64 val) 125 { 126 u64 cur = inode_peek_iversion_raw(inode); 127 128 do { 129 if (cur > val) 130 break; 131 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, val)); 132 } 133 134 /** 135 * inode_set_iversion - set i_version to a particular value 136 * @inode: inode to set 137 * @val: new i_version value to set 138 * 139 * Set @inode's i_version field to @val. This function is for filesystems with 140 * a kernel-managed i_version, for initializing a newly-created inode from 141 * scratch. 142 * 143 * In this case, we do not set the QUERIED flag since we know that this value 144 * has never been queried. 145 */ 146 static inline void 147 inode_set_iversion(struct inode *inode, u64 val) 148 { 149 inode_set_iversion_raw(inode, val << I_VERSION_QUERIED_SHIFT); 150 } 151 152 /** 153 * inode_set_iversion_queried - set i_version to a particular value as quereied 154 * @inode: inode to set 155 * @val: new i_version value to set 156 * 157 * Set @inode's i_version field to @val, and flag it for increment on the next 158 * change. 159 * 160 * Filesystems that persistently store the i_version on disk should use this 161 * when loading an existing inode from disk. 162 * 163 * When loading in an i_version value from a backing store, we can't be certain 164 * that it wasn't previously viewed before being stored. Thus, we must assume 165 * that it was, to ensure that we don't end up handing out the same value for 166 * different versions of the same inode. 167 */ 168 static inline void 169 inode_set_iversion_queried(struct inode *inode, u64 val) 170 { 171 inode_set_iversion_raw(inode, (val << I_VERSION_QUERIED_SHIFT) | 172 I_VERSION_QUERIED); 173 } 174 175 bool inode_maybe_inc_iversion(struct inode *inode, bool force); 176 177 /** 178 * inode_inc_iversion - forcibly increment i_version 179 * @inode: inode that needs to be updated 180 * 181 * Forcbily increment the i_version field. This always results in a change to 182 * the observable value. 183 */ 184 static inline void 185 inode_inc_iversion(struct inode *inode) 186 { 187 inode_maybe_inc_iversion(inode, true); 188 } 189 190 /** 191 * inode_iversion_need_inc - is the i_version in need of being incremented? 192 * @inode: inode to check 193 * 194 * Returns whether the inode->i_version counter needs incrementing on the next 195 * change. Just fetch the value and check the QUERIED flag. 196 */ 197 static inline bool 198 inode_iversion_need_inc(struct inode *inode) 199 { 200 return inode_peek_iversion_raw(inode) & I_VERSION_QUERIED; 201 } 202 203 /** 204 * inode_inc_iversion_raw - forcibly increment raw i_version 205 * @inode: inode that needs to be updated 206 * 207 * Forcbily increment the raw i_version field. This always results in a change 208 * to the raw value. 209 * 210 * NFS will use the i_version field to store the value from the server. It 211 * mostly treats it as opaque, but in the case where it holds a write 212 * delegation, it must increment the value itself. This function does that. 213 */ 214 static inline void 215 inode_inc_iversion_raw(struct inode *inode) 216 { 217 atomic64_inc(&inode->i_version); 218 } 219 220 /** 221 * inode_peek_iversion - read i_version without flagging it to be incremented 222 * @inode: inode from which i_version should be read 223 * 224 * Read the inode i_version counter for an inode without registering it as a 225 * query. 226 * 227 * This is typically used by local filesystems that need to store an i_version 228 * on disk. In that situation, it's not necessary to flag it as having been 229 * viewed, as the result won't be used to gauge changes from that point. 230 */ 231 static inline u64 232 inode_peek_iversion(const struct inode *inode) 233 { 234 return inode_peek_iversion_raw(inode) >> I_VERSION_QUERIED_SHIFT; 235 } 236 237 /** 238 * inode_query_iversion - read i_version for later use 239 * @inode: inode from which i_version should be read 240 * 241 * Read the inode i_version counter. This should be used by callers that wish 242 * to store the returned i_version for later comparison. This will guarantee 243 * that a later query of the i_version will result in a different value if 244 * anything has changed. 245 * 246 * In this implementation, we fetch the current value, set the QUERIED flag and 247 * then try to swap it into place with a cmpxchg, if it wasn't already set. If 248 * that fails, we try again with the newly fetched value from the cmpxchg. 249 */ 250 static inline u64 251 inode_query_iversion(struct inode *inode) 252 { 253 u64 cur, new; 254 255 cur = inode_peek_iversion_raw(inode); 256 do { 257 /* If flag is already set, then no need to swap */ 258 if (cur & I_VERSION_QUERIED) { 259 /* 260 * This barrier (and the implicit barrier in the 261 * cmpxchg below) pairs with the barrier in 262 * inode_maybe_inc_iversion(). 263 */ 264 smp_mb(); 265 break; 266 } 267 268 new = cur | I_VERSION_QUERIED; 269 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new)); 270 return cur >> I_VERSION_QUERIED_SHIFT; 271 } 272 273 /* 274 * For filesystems without any sort of change attribute, the best we can 275 * do is fake one up from the ctime: 276 */ 277 static inline u64 time_to_chattr(struct timespec64 *t) 278 { 279 u64 chattr = t->tv_sec; 280 281 chattr <<= 32; 282 chattr += t->tv_nsec; 283 return chattr; 284 } 285 286 /** 287 * inode_eq_iversion_raw - check whether the raw i_version counter has changed 288 * @inode: inode to check 289 * @old: old value to check against its i_version 290 * 291 * Compare the current raw i_version counter with a previous one. Returns true 292 * if they are the same or false if they are different. 293 */ 294 static inline bool 295 inode_eq_iversion_raw(const struct inode *inode, u64 old) 296 { 297 return inode_peek_iversion_raw(inode) == old; 298 } 299 300 /** 301 * inode_eq_iversion - check whether the i_version counter has changed 302 * @inode: inode to check 303 * @old: old value to check against its i_version 304 * 305 * Compare an i_version counter with a previous one. Returns true if they are 306 * the same, and false if they are different. 307 * 308 * Note that we don't need to set the QUERIED flag in this case, as the value 309 * in the inode is not being recorded for later use. 310 */ 311 static inline bool 312 inode_eq_iversion(const struct inode *inode, u64 old) 313 { 314 return inode_peek_iversion(inode) == old; 315 } 316 #endif 317