Commit | Line | Data |
---|---|---|
fb18a575 LH |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * quota.c - CephFS quota | |
4 | * | |
5 | * Copyright (C) 2017-2018 SUSE | |
fb18a575 LH |
6 | */ |
7 | ||
9122eed5 LH |
8 | #include <linux/statfs.h> |
9 | ||
fb18a575 LH |
10 | #include "super.h" |
11 | #include "mds_client.h" | |
12 | ||
d557c48d | 13 | void ceph_adjust_quota_realms_count(struct inode *inode, bool inc) |
cafe21a4 | 14 | { |
d557c48d LH |
15 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
16 | if (inc) | |
17 | atomic64_inc(&mdsc->quotarealms_count); | |
18 | else | |
19 | atomic64_dec(&mdsc->quotarealms_count); | |
20 | } | |
21 | ||
22 | static inline bool ceph_has_realms_with_quotas(struct inode *inode) | |
23 | { | |
24 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; | |
25 | return atomic64_read(&mdsc->quotarealms_count) > 0; | |
cafe21a4 LH |
26 | } |
27 | ||
fb18a575 LH |
28 | void ceph_handle_quota(struct ceph_mds_client *mdsc, |
29 | struct ceph_mds_session *session, | |
30 | struct ceph_msg *msg) | |
31 | { | |
32 | struct super_block *sb = mdsc->fsc->sb; | |
33 | struct ceph_mds_quota *h = msg->front.iov_base; | |
34 | struct ceph_vino vino; | |
35 | struct inode *inode; | |
36 | struct ceph_inode_info *ci; | |
37 | ||
0fcf6c02 | 38 | if (msg->front.iov_len < sizeof(*h)) { |
fb18a575 LH |
39 | pr_err("%s corrupt message mds%d len %d\n", __func__, |
40 | session->s_mds, (int)msg->front.iov_len); | |
41 | ceph_msg_dump(msg); | |
42 | return; | |
43 | } | |
44 | ||
45 | /* increment msg sequence number */ | |
46 | mutex_lock(&session->s_mutex); | |
47 | session->s_seq++; | |
48 | mutex_unlock(&session->s_mutex); | |
49 | ||
50 | /* lookup inode */ | |
51 | vino.ino = le64_to_cpu(h->ino); | |
52 | vino.snap = CEPH_NOSNAP; | |
53 | inode = ceph_find_inode(sb, vino); | |
54 | if (!inode) { | |
55 | pr_warn("Failed to find inode %llu\n", vino.ino); | |
56 | return; | |
57 | } | |
58 | ci = ceph_inode(inode); | |
59 | ||
60 | spin_lock(&ci->i_ceph_lock); | |
61 | ci->i_rbytes = le64_to_cpu(h->rbytes); | |
62 | ci->i_rfiles = le64_to_cpu(h->rfiles); | |
63 | ci->i_rsubdirs = le64_to_cpu(h->rsubdirs); | |
d557c48d LH |
64 | __ceph_update_quota(ci, le64_to_cpu(h->max_bytes), |
65 | le64_to_cpu(h->max_files)); | |
fb18a575 LH |
66 | spin_unlock(&ci->i_ceph_lock); |
67 | ||
68 | iput(inode); | |
69 | } | |
b7a29217 | 70 | |
cafe21a4 LH |
71 | /* |
72 | * This function walks through the snaprealm for an inode and returns the | |
73 | * ceph_snap_realm for the first snaprealm that has quotas set (either max_files | |
74 | * or max_bytes). If the root is reached, return the root ceph_snap_realm | |
75 | * instead. | |
76 | * | |
77 | * Note that the caller is responsible for calling ceph_put_snap_realm() on the | |
78 | * returned realm. | |
79 | */ | |
80 | static struct ceph_snap_realm *get_quota_realm(struct ceph_mds_client *mdsc, | |
81 | struct inode *inode) | |
82 | { | |
83 | struct ceph_inode_info *ci = NULL; | |
84 | struct ceph_snap_realm *realm, *next; | |
cafe21a4 | 85 | struct inode *in; |
0eb6bbe4 | 86 | bool has_quota; |
cafe21a4 | 87 | |
25963669 YZ |
88 | if (ceph_snap(inode) != CEPH_NOSNAP) |
89 | return NULL; | |
90 | ||
cafe21a4 | 91 | realm = ceph_inode(inode)->i_snap_realm; |
25963669 YZ |
92 | if (realm) |
93 | ceph_get_snap_realm(mdsc, realm); | |
94 | else | |
95 | pr_err_ratelimited("get_quota_realm: ino (%llx.%llx) " | |
96 | "null i_snap_realm\n", ceph_vinop(inode)); | |
cafe21a4 | 97 | while (realm) { |
e3161f17 LH |
98 | spin_lock(&realm->inodes_with_caps_lock); |
99 | in = realm->inode ? igrab(realm->inode) : NULL; | |
100 | spin_unlock(&realm->inodes_with_caps_lock); | |
101 | if (!in) | |
cafe21a4 | 102 | break; |
e3161f17 | 103 | |
cafe21a4 | 104 | ci = ceph_inode(in); |
d557c48d | 105 | has_quota = __ceph_has_any_quota(ci); |
cafe21a4 | 106 | iput(in); |
0eb6bbe4 | 107 | |
cafe21a4 | 108 | next = realm->parent; |
0eb6bbe4 YZ |
109 | if (has_quota || !next) |
110 | return realm; | |
111 | ||
cafe21a4 LH |
112 | ceph_get_snap_realm(mdsc, next); |
113 | ceph_put_snap_realm(mdsc, realm); | |
114 | realm = next; | |
115 | } | |
116 | if (realm) | |
117 | ceph_put_snap_realm(mdsc, realm); | |
118 | ||
119 | return NULL; | |
120 | } | |
121 | ||
122 | bool ceph_quota_is_same_realm(struct inode *old, struct inode *new) | |
123 | { | |
124 | struct ceph_mds_client *mdsc = ceph_inode_to_client(old)->mdsc; | |
125 | struct ceph_snap_realm *old_realm, *new_realm; | |
126 | bool is_same; | |
127 | ||
128 | down_read(&mdsc->snap_rwsem); | |
129 | old_realm = get_quota_realm(mdsc, old); | |
130 | new_realm = get_quota_realm(mdsc, new); | |
131 | is_same = (old_realm == new_realm); | |
132 | up_read(&mdsc->snap_rwsem); | |
133 | ||
134 | if (old_realm) | |
135 | ceph_put_snap_realm(mdsc, old_realm); | |
136 | if (new_realm) | |
137 | ceph_put_snap_realm(mdsc, new_realm); | |
138 | ||
139 | return is_same; | |
140 | } | |
141 | ||
b7a29217 | 142 | enum quota_check_op { |
2b83845f | 143 | QUOTA_CHECK_MAX_FILES_OP, /* check quota max_files limit */ |
1ab302a0 LH |
144 | QUOTA_CHECK_MAX_BYTES_OP, /* check quota max_files limit */ |
145 | QUOTA_CHECK_MAX_BYTES_APPROACHING_OP /* check if quota max_files | |
146 | limit is approaching */ | |
b7a29217 LH |
147 | }; |
148 | ||
149 | /* | |
150 | * check_quota_exceeded() will walk up the snaprealm hierarchy and, for each | |
151 | * realm, it will execute quota check operation defined by the 'op' parameter. | |
152 | * The snaprealm walk is interrupted if the quota check detects that the quota | |
153 | * is exceeded or if the root inode is reached. | |
154 | */ | |
155 | static bool check_quota_exceeded(struct inode *inode, enum quota_check_op op, | |
156 | loff_t delta) | |
157 | { | |
158 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; | |
159 | struct ceph_inode_info *ci; | |
160 | struct ceph_snap_realm *realm, *next; | |
b7a29217 LH |
161 | struct inode *in; |
162 | u64 max, rvalue; | |
b7a29217 LH |
163 | bool exceeded = false; |
164 | ||
25963669 YZ |
165 | if (ceph_snap(inode) != CEPH_NOSNAP) |
166 | return false; | |
167 | ||
b7a29217 LH |
168 | down_read(&mdsc->snap_rwsem); |
169 | realm = ceph_inode(inode)->i_snap_realm; | |
25963669 YZ |
170 | if (realm) |
171 | ceph_get_snap_realm(mdsc, realm); | |
172 | else | |
173 | pr_err_ratelimited("check_quota_exceeded: ino (%llx.%llx) " | |
174 | "null i_snap_realm\n", ceph_vinop(inode)); | |
b7a29217 | 175 | while (realm) { |
e3161f17 LH |
176 | spin_lock(&realm->inodes_with_caps_lock); |
177 | in = realm->inode ? igrab(realm->inode) : NULL; | |
178 | spin_unlock(&realm->inodes_with_caps_lock); | |
179 | if (!in) | |
b7a29217 | 180 | break; |
e3161f17 | 181 | |
b7a29217 LH |
182 | ci = ceph_inode(in); |
183 | spin_lock(&ci->i_ceph_lock); | |
184 | if (op == QUOTA_CHECK_MAX_FILES_OP) { | |
185 | max = ci->i_max_files; | |
186 | rvalue = ci->i_rfiles + ci->i_rsubdirs; | |
2b83845f LH |
187 | } else { |
188 | max = ci->i_max_bytes; | |
189 | rvalue = ci->i_rbytes; | |
b7a29217 | 190 | } |
b7a29217 LH |
191 | spin_unlock(&ci->i_ceph_lock); |
192 | switch (op) { | |
193 | case QUOTA_CHECK_MAX_FILES_OP: | |
194 | exceeded = (max && (rvalue >= max)); | |
195 | break; | |
2b83845f LH |
196 | case QUOTA_CHECK_MAX_BYTES_OP: |
197 | exceeded = (max && (rvalue + delta > max)); | |
198 | break; | |
1ab302a0 LH |
199 | case QUOTA_CHECK_MAX_BYTES_APPROACHING_OP: |
200 | if (max) { | |
201 | if (rvalue >= max) | |
202 | exceeded = true; | |
203 | else { | |
204 | /* | |
205 | * when we're writing more that 1/16th | |
206 | * of the available space | |
207 | */ | |
208 | exceeded = | |
209 | (((max - rvalue) >> 4) < delta); | |
210 | } | |
211 | } | |
212 | break; | |
b7a29217 LH |
213 | default: |
214 | /* Shouldn't happen */ | |
215 | pr_warn("Invalid quota check op (%d)\n", op); | |
216 | exceeded = true; /* Just break the loop */ | |
217 | } | |
218 | iput(in); | |
219 | ||
b7a29217 | 220 | next = realm->parent; |
0eb6bbe4 YZ |
221 | if (exceeded || !next) |
222 | break; | |
b7a29217 LH |
223 | ceph_get_snap_realm(mdsc, next); |
224 | ceph_put_snap_realm(mdsc, realm); | |
225 | realm = next; | |
226 | } | |
71f2cc64 LH |
227 | if (realm) |
228 | ceph_put_snap_realm(mdsc, realm); | |
b7a29217 LH |
229 | up_read(&mdsc->snap_rwsem); |
230 | ||
231 | return exceeded; | |
232 | } | |
233 | ||
234 | /* | |
235 | * ceph_quota_is_max_files_exceeded - check if we can create a new file | |
236 | * @inode: directory where a new file is being created | |
237 | * | |
238 | * This functions returns true is max_files quota allows a new file to be | |
239 | * created. It is necessary to walk through the snaprealm hierarchy (until the | |
240 | * FS root) to check all realms with quotas set. | |
241 | */ | |
242 | bool ceph_quota_is_max_files_exceeded(struct inode *inode) | |
243 | { | |
d557c48d LH |
244 | if (!ceph_has_realms_with_quotas(inode)) |
245 | return false; | |
246 | ||
b7a29217 LH |
247 | WARN_ON(!S_ISDIR(inode->i_mode)); |
248 | ||
249 | return check_quota_exceeded(inode, QUOTA_CHECK_MAX_FILES_OP, 0); | |
250 | } | |
2b83845f LH |
251 | |
252 | /* | |
253 | * ceph_quota_is_max_bytes_exceeded - check if we can write to a file | |
254 | * @inode: inode being written | |
255 | * @newsize: new size if write succeeds | |
256 | * | |
257 | * This functions returns true is max_bytes quota allows a file size to reach | |
258 | * @newsize; it returns false otherwise. | |
259 | */ | |
260 | bool ceph_quota_is_max_bytes_exceeded(struct inode *inode, loff_t newsize) | |
261 | { | |
262 | loff_t size = i_size_read(inode); | |
263 | ||
d557c48d LH |
264 | if (!ceph_has_realms_with_quotas(inode)) |
265 | return false; | |
266 | ||
2b83845f LH |
267 | /* return immediately if we're decreasing file size */ |
268 | if (newsize <= size) | |
269 | return false; | |
270 | ||
271 | return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_OP, (newsize - size)); | |
272 | } | |
1ab302a0 LH |
273 | |
274 | /* | |
275 | * ceph_quota_is_max_bytes_approaching - check if we're reaching max_bytes | |
276 | * @inode: inode being written | |
277 | * @newsize: new size if write succeeds | |
278 | * | |
279 | * This function returns true if the new file size @newsize will be consuming | |
280 | * more than 1/16th of the available quota space; it returns false otherwise. | |
281 | */ | |
282 | bool ceph_quota_is_max_bytes_approaching(struct inode *inode, loff_t newsize) | |
283 | { | |
284 | loff_t size = ceph_inode(inode)->i_reported_size; | |
285 | ||
d557c48d LH |
286 | if (!ceph_has_realms_with_quotas(inode)) |
287 | return false; | |
288 | ||
1ab302a0 LH |
289 | /* return immediately if we're decreasing file size */ |
290 | if (newsize <= size) | |
291 | return false; | |
292 | ||
293 | return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_APPROACHING_OP, | |
294 | (newsize - size)); | |
295 | } | |
9122eed5 LH |
296 | |
297 | /* | |
298 | * ceph_quota_update_statfs - if root has quota update statfs with quota status | |
299 | * @fsc: filesystem client instance | |
300 | * @buf: statfs to update | |
301 | * | |
302 | * If the mounted filesystem root has max_bytes quota set, update the filesystem | |
303 | * statistics with the quota status. | |
304 | * | |
305 | * This function returns true if the stats have been updated, false otherwise. | |
306 | */ | |
307 | bool ceph_quota_update_statfs(struct ceph_fs_client *fsc, struct kstatfs *buf) | |
308 | { | |
309 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
310 | struct ceph_inode_info *ci; | |
311 | struct ceph_snap_realm *realm; | |
312 | struct inode *in; | |
313 | u64 total = 0, used, free; | |
314 | bool is_updated = false; | |
315 | ||
316 | down_read(&mdsc->snap_rwsem); | |
317 | realm = get_quota_realm(mdsc, d_inode(fsc->sb->s_root)); | |
318 | up_read(&mdsc->snap_rwsem); | |
319 | if (!realm) | |
320 | return false; | |
321 | ||
322 | spin_lock(&realm->inodes_with_caps_lock); | |
323 | in = realm->inode ? igrab(realm->inode) : NULL; | |
324 | spin_unlock(&realm->inodes_with_caps_lock); | |
325 | if (in) { | |
326 | ci = ceph_inode(in); | |
327 | spin_lock(&ci->i_ceph_lock); | |
328 | if (ci->i_max_bytes) { | |
329 | total = ci->i_max_bytes >> CEPH_BLOCK_SHIFT; | |
330 | used = ci->i_rbytes >> CEPH_BLOCK_SHIFT; | |
331 | /* It is possible for a quota to be exceeded. | |
332 | * Report 'zero' in that case | |
333 | */ | |
334 | free = total > used ? total - used : 0; | |
335 | } | |
336 | spin_unlock(&ci->i_ceph_lock); | |
337 | if (total) { | |
338 | buf->f_blocks = total; | |
339 | buf->f_bfree = free; | |
340 | buf->f_bavail = free; | |
341 | is_updated = true; | |
342 | } | |
343 | iput(in); | |
344 | } | |
345 | ceph_put_snap_realm(mdsc, realm); | |
346 | ||
347 | return is_updated; | |
348 | } | |
349 |