projects/linux·blob master · refs

300 lines · 11.0 KiB

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