1 // SPDX-License-Identifier: GPL-2.0
3 * quota.c - CephFS quota
5 * Copyright (C) 2017-2018 SUSE
8 #include <linux/statfs.h>
11 #include "mds_client.h"
13 void ceph_adjust_quota_realms_count(struct inode *inode, bool inc)
15 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
17 atomic64_inc(&mdsc->quotarealms_count);
19 atomic64_dec(&mdsc->quotarealms_count);
22 static inline bool ceph_has_realms_with_quotas(struct inode *inode)
24 struct super_block *sb = inode->i_sb;
25 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(sb);
26 struct inode *root = d_inode(sb->s_root);
28 if (atomic64_read(&mdsc->quotarealms_count) > 0)
30 /* if root is the real CephFS root, we don't have quota realms */
31 if (root && ceph_ino(root) == CEPH_INO_ROOT)
33 /* otherwise, we can't know for sure */
37 void ceph_handle_quota(struct ceph_mds_client *mdsc,
38 struct ceph_mds_session *session,
41 struct super_block *sb = mdsc->fsc->sb;
42 struct ceph_mds_quota *h = msg->front.iov_base;
43 struct ceph_vino vino;
45 struct ceph_inode_info *ci;
47 if (msg->front.iov_len < sizeof(*h)) {
48 pr_err("%s corrupt message mds%d len %d\n", __func__,
49 session->s_mds, (int)msg->front.iov_len);
54 /* increment msg sequence number */
55 mutex_lock(&session->s_mutex);
56 inc_session_sequence(session);
57 mutex_unlock(&session->s_mutex);
60 vino.ino = le64_to_cpu(h->ino);
61 vino.snap = CEPH_NOSNAP;
62 inode = ceph_find_inode(sb, vino);
64 pr_warn("Failed to find inode %llu\n", vino.ino);
67 ci = ceph_inode(inode);
69 spin_lock(&ci->i_ceph_lock);
70 ci->i_rbytes = le64_to_cpu(h->rbytes);
71 ci->i_rfiles = le64_to_cpu(h->rfiles);
72 ci->i_rsubdirs = le64_to_cpu(h->rsubdirs);
73 __ceph_update_quota(ci, le64_to_cpu(h->max_bytes),
74 le64_to_cpu(h->max_files));
75 spin_unlock(&ci->i_ceph_lock);
80 static struct ceph_quotarealm_inode *
81 find_quotarealm_inode(struct ceph_mds_client *mdsc, u64 ino)
83 struct ceph_quotarealm_inode *qri = NULL;
84 struct rb_node **node, *parent = NULL;
86 mutex_lock(&mdsc->quotarealms_inodes_mutex);
87 node = &(mdsc->quotarealms_inodes.rb_node);
90 qri = container_of(*node, struct ceph_quotarealm_inode, node);
93 node = &((*node)->rb_left);
94 else if (ino > qri->ino)
95 node = &((*node)->rb_right);
99 if (!qri || (qri->ino != ino)) {
100 /* Not found, create a new one and insert it */
101 qri = kmalloc(sizeof(*qri), GFP_KERNEL);
106 mutex_init(&qri->mutex);
107 rb_link_node(&qri->node, parent, node);
108 rb_insert_color(&qri->node, &mdsc->quotarealms_inodes);
110 pr_warn("Failed to alloc quotarealms_inode\n");
112 mutex_unlock(&mdsc->quotarealms_inodes_mutex);
118 * This function will try to lookup a realm inode which isn't visible in the
119 * filesystem mountpoint. A list of these kind of inodes (not visible) is
120 * maintained in the mdsc and freed only when the filesystem is umounted.
122 * Note that these inodes are kept in this list even if the lookup fails, which
123 * allows to prevent useless lookup requests.
125 static struct inode *lookup_quotarealm_inode(struct ceph_mds_client *mdsc,
126 struct super_block *sb,
127 struct ceph_snap_realm *realm)
129 struct ceph_quotarealm_inode *qri;
132 qri = find_quotarealm_inode(mdsc, realm->ino);
136 mutex_lock(&qri->mutex);
137 if (qri->inode && ceph_is_any_caps(qri->inode)) {
138 /* A request has already returned the inode */
139 mutex_unlock(&qri->mutex);
142 /* Check if this inode lookup has failed recently */
144 time_before_eq(jiffies, qri->timeout)) {
145 mutex_unlock(&qri->mutex);
150 int ret = __ceph_do_getattr(qri->inode, NULL,
151 CEPH_STAT_CAP_INODE, true);
157 in = ceph_lookup_inode(sb, realm->ino);
161 dout("Can't lookup inode %llx (err: %ld)\n",
162 realm->ino, PTR_ERR(in));
163 qri->timeout = jiffies + msecs_to_jiffies(60 * 1000); /* XXX */
168 mutex_unlock(&qri->mutex);
173 void ceph_cleanup_quotarealms_inodes(struct ceph_mds_client *mdsc)
175 struct ceph_quotarealm_inode *qri;
176 struct rb_node *node;
179 * It should now be safe to clean quotarealms_inode tree without holding
180 * mdsc->quotarealms_inodes_mutex...
182 mutex_lock(&mdsc->quotarealms_inodes_mutex);
183 while (!RB_EMPTY_ROOT(&mdsc->quotarealms_inodes)) {
184 node = rb_first(&mdsc->quotarealms_inodes);
185 qri = rb_entry(node, struct ceph_quotarealm_inode, node);
186 rb_erase(node, &mdsc->quotarealms_inodes);
190 mutex_unlock(&mdsc->quotarealms_inodes_mutex);
194 * This function walks through the snaprealm for an inode and returns the
195 * ceph_snap_realm for the first snaprealm that has quotas set (either max_files
196 * or max_bytes). If the root is reached, return the root ceph_snap_realm
199 * Note that the caller is responsible for calling ceph_put_snap_realm() on the
202 * Callers of this function need to hold mdsc->snap_rwsem. However, if there's
203 * a need to do an inode lookup, this rwsem will be temporarily dropped. Hence
204 * the 'retry' argument: if rwsem needs to be dropped and 'retry' is 'false'
205 * this function will return -EAGAIN; otherwise, the snaprealms walk-through
208 static struct ceph_snap_realm *get_quota_realm(struct ceph_mds_client *mdsc,
209 struct inode *inode, bool retry)
211 struct ceph_inode_info *ci = NULL;
212 struct ceph_snap_realm *realm, *next;
216 if (ceph_snap(inode) != CEPH_NOSNAP)
220 realm = ceph_inode(inode)->i_snap_realm;
222 ceph_get_snap_realm(mdsc, realm);
224 pr_err_ratelimited("get_quota_realm: ino (%llx.%llx) "
225 "null i_snap_realm\n", ceph_vinop(inode));
229 spin_lock(&realm->inodes_with_caps_lock);
230 has_inode = realm->inode;
231 in = has_inode ? igrab(realm->inode) : NULL;
232 spin_unlock(&realm->inodes_with_caps_lock);
233 if (has_inode && !in)
236 up_read(&mdsc->snap_rwsem);
237 in = lookup_quotarealm_inode(mdsc, inode->i_sb, realm);
238 down_read(&mdsc->snap_rwsem);
239 if (IS_ERR_OR_NULL(in))
241 ceph_put_snap_realm(mdsc, realm);
243 return ERR_PTR(-EAGAIN);
248 has_quota = __ceph_has_any_quota(ci);
251 next = realm->parent;
252 if (has_quota || !next)
255 ceph_get_snap_realm(mdsc, next);
256 ceph_put_snap_realm(mdsc, realm);
260 ceph_put_snap_realm(mdsc, realm);
265 bool ceph_quota_is_same_realm(struct inode *old, struct inode *new)
267 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old->i_sb);
268 struct ceph_snap_realm *old_realm, *new_realm;
273 * We need to lookup 2 quota realms atomically, i.e. with snap_rwsem.
274 * However, get_quota_realm may drop it temporarily. By setting the
275 * 'retry' parameter to 'false', we'll get -EAGAIN if the rwsem was
276 * dropped and we can then restart the whole operation.
278 down_read(&mdsc->snap_rwsem);
279 old_realm = get_quota_realm(mdsc, old, true);
280 new_realm = get_quota_realm(mdsc, new, false);
281 if (PTR_ERR(new_realm) == -EAGAIN) {
282 up_read(&mdsc->snap_rwsem);
284 ceph_put_snap_realm(mdsc, old_realm);
287 is_same = (old_realm == new_realm);
288 up_read(&mdsc->snap_rwsem);
291 ceph_put_snap_realm(mdsc, old_realm);
293 ceph_put_snap_realm(mdsc, new_realm);
298 enum quota_check_op {
299 QUOTA_CHECK_MAX_FILES_OP, /* check quota max_files limit */
300 QUOTA_CHECK_MAX_BYTES_OP, /* check quota max_files limit */
301 QUOTA_CHECK_MAX_BYTES_APPROACHING_OP /* check if quota max_files
302 limit is approaching */
306 * check_quota_exceeded() will walk up the snaprealm hierarchy and, for each
307 * realm, it will execute quota check operation defined by the 'op' parameter.
308 * The snaprealm walk is interrupted if the quota check detects that the quota
309 * is exceeded or if the root inode is reached.
311 static bool check_quota_exceeded(struct inode *inode, enum quota_check_op op,
314 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
315 struct ceph_inode_info *ci;
316 struct ceph_snap_realm *realm, *next;
319 bool exceeded = false;
321 if (ceph_snap(inode) != CEPH_NOSNAP)
324 down_read(&mdsc->snap_rwsem);
326 realm = ceph_inode(inode)->i_snap_realm;
328 ceph_get_snap_realm(mdsc, realm);
330 pr_err_ratelimited("check_quota_exceeded: ino (%llx.%llx) "
331 "null i_snap_realm\n", ceph_vinop(inode));
335 spin_lock(&realm->inodes_with_caps_lock);
336 has_inode = realm->inode;
337 in = has_inode ? igrab(realm->inode) : NULL;
338 spin_unlock(&realm->inodes_with_caps_lock);
339 if (has_inode && !in)
342 up_read(&mdsc->snap_rwsem);
343 in = lookup_quotarealm_inode(mdsc, inode->i_sb, realm);
344 down_read(&mdsc->snap_rwsem);
345 if (IS_ERR_OR_NULL(in))
347 ceph_put_snap_realm(mdsc, realm);
351 spin_lock(&ci->i_ceph_lock);
352 if (op == QUOTA_CHECK_MAX_FILES_OP) {
353 max = ci->i_max_files;
354 rvalue = ci->i_rfiles + ci->i_rsubdirs;
356 max = ci->i_max_bytes;
357 rvalue = ci->i_rbytes;
359 spin_unlock(&ci->i_ceph_lock);
361 case QUOTA_CHECK_MAX_FILES_OP:
362 case QUOTA_CHECK_MAX_BYTES_OP:
363 exceeded = (max && (rvalue + delta > max));
365 case QUOTA_CHECK_MAX_BYTES_APPROACHING_OP:
371 * when we're writing more that 1/16th
372 * of the available space
375 (((max - rvalue) >> 4) < delta);
380 /* Shouldn't happen */
381 pr_warn("Invalid quota check op (%d)\n", op);
382 exceeded = true; /* Just break the loop */
386 next = realm->parent;
387 if (exceeded || !next)
389 ceph_get_snap_realm(mdsc, next);
390 ceph_put_snap_realm(mdsc, realm);
394 ceph_put_snap_realm(mdsc, realm);
395 up_read(&mdsc->snap_rwsem);
401 * ceph_quota_is_max_files_exceeded - check if we can create a new file
402 * @inode: directory where a new file is being created
404 * This functions returns true is max_files quota allows a new file to be
405 * created. It is necessary to walk through the snaprealm hierarchy (until the
406 * FS root) to check all realms with quotas set.
408 bool ceph_quota_is_max_files_exceeded(struct inode *inode)
410 if (!ceph_has_realms_with_quotas(inode))
413 WARN_ON(!S_ISDIR(inode->i_mode));
415 return check_quota_exceeded(inode, QUOTA_CHECK_MAX_FILES_OP, 1);
419 * ceph_quota_is_max_bytes_exceeded - check if we can write to a file
420 * @inode: inode being written
421 * @newsize: new size if write succeeds
423 * This functions returns true is max_bytes quota allows a file size to reach
424 * @newsize; it returns false otherwise.
426 bool ceph_quota_is_max_bytes_exceeded(struct inode *inode, loff_t newsize)
428 loff_t size = i_size_read(inode);
430 if (!ceph_has_realms_with_quotas(inode))
433 /* return immediately if we're decreasing file size */
437 return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_OP, (newsize - size));
441 * ceph_quota_is_max_bytes_approaching - check if we're reaching max_bytes
442 * @inode: inode being written
443 * @newsize: new size if write succeeds
445 * This function returns true if the new file size @newsize will be consuming
446 * more than 1/16th of the available quota space; it returns false otherwise.
448 bool ceph_quota_is_max_bytes_approaching(struct inode *inode, loff_t newsize)
450 loff_t size = ceph_inode(inode)->i_reported_size;
452 if (!ceph_has_realms_with_quotas(inode))
455 /* return immediately if we're decreasing file size */
459 return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_APPROACHING_OP,
464 * ceph_quota_update_statfs - if root has quota update statfs with quota status
465 * @fsc: filesystem client instance
466 * @buf: statfs to update
468 * If the mounted filesystem root has max_bytes quota set, update the filesystem
469 * statistics with the quota status.
471 * This function returns true if the stats have been updated, false otherwise.
473 bool ceph_quota_update_statfs(struct ceph_fs_client *fsc, struct kstatfs *buf)
475 struct ceph_mds_client *mdsc = fsc->mdsc;
476 struct ceph_inode_info *ci;
477 struct ceph_snap_realm *realm;
479 u64 total = 0, used, free;
480 bool is_updated = false;
482 down_read(&mdsc->snap_rwsem);
483 realm = get_quota_realm(mdsc, d_inode(fsc->sb->s_root), true);
484 up_read(&mdsc->snap_rwsem);
488 spin_lock(&realm->inodes_with_caps_lock);
489 in = realm->inode ? igrab(realm->inode) : NULL;
490 spin_unlock(&realm->inodes_with_caps_lock);
493 spin_lock(&ci->i_ceph_lock);
494 if (ci->i_max_bytes) {
495 total = ci->i_max_bytes >> CEPH_BLOCK_SHIFT;
496 used = ci->i_rbytes >> CEPH_BLOCK_SHIFT;
497 /* It is possible for a quota to be exceeded.
498 * Report 'zero' in that case
500 free = total > used ? total - used : 0;
502 spin_unlock(&ci->i_ceph_lock);
504 buf->f_blocks = total;
506 buf->f_bavail = free;
511 ceph_put_snap_realm(mdsc, realm);