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);
77 /* avoid calling iput_final() in dispatch thread */
78 ceph_async_iput(inode);
81 static struct ceph_quotarealm_inode *
82 find_quotarealm_inode(struct ceph_mds_client *mdsc, u64 ino)
84 struct ceph_quotarealm_inode *qri = NULL;
85 struct rb_node **node, *parent = NULL;
87 mutex_lock(&mdsc->quotarealms_inodes_mutex);
88 node = &(mdsc->quotarealms_inodes.rb_node);
91 qri = container_of(*node, struct ceph_quotarealm_inode, node);
94 node = &((*node)->rb_left);
95 else if (ino > qri->ino)
96 node = &((*node)->rb_right);
100 if (!qri || (qri->ino != ino)) {
101 /* Not found, create a new one and insert it */
102 qri = kmalloc(sizeof(*qri), GFP_KERNEL);
107 mutex_init(&qri->mutex);
108 rb_link_node(&qri->node, parent, node);
109 rb_insert_color(&qri->node, &mdsc->quotarealms_inodes);
111 pr_warn("Failed to alloc quotarealms_inode\n");
113 mutex_unlock(&mdsc->quotarealms_inodes_mutex);
119 * This function will try to lookup a realm inode which isn't visible in the
120 * filesystem mountpoint. A list of these kind of inodes (not visible) is
121 * maintained in the mdsc and freed only when the filesystem is umounted.
123 * Note that these inodes are kept in this list even if the lookup fails, which
124 * allows to prevent useless lookup requests.
126 static struct inode *lookup_quotarealm_inode(struct ceph_mds_client *mdsc,
127 struct super_block *sb,
128 struct ceph_snap_realm *realm)
130 struct ceph_quotarealm_inode *qri;
133 qri = find_quotarealm_inode(mdsc, realm->ino);
137 mutex_lock(&qri->mutex);
138 if (qri->inode && ceph_is_any_caps(qri->inode)) {
139 /* A request has already returned the inode */
140 mutex_unlock(&qri->mutex);
143 /* Check if this inode lookup has failed recently */
145 time_before_eq(jiffies, qri->timeout)) {
146 mutex_unlock(&qri->mutex);
151 int ret = __ceph_do_getattr(qri->inode, NULL,
152 CEPH_STAT_CAP_INODE, true);
158 in = ceph_lookup_inode(sb, realm->ino);
162 dout("Can't lookup inode %llx (err: %ld)\n",
163 realm->ino, PTR_ERR(in));
164 qri->timeout = jiffies + msecs_to_jiffies(60 * 1000); /* XXX */
169 mutex_unlock(&qri->mutex);
174 void ceph_cleanup_quotarealms_inodes(struct ceph_mds_client *mdsc)
176 struct ceph_quotarealm_inode *qri;
177 struct rb_node *node;
180 * It should now be safe to clean quotarealms_inode tree without holding
181 * mdsc->quotarealms_inodes_mutex...
183 mutex_lock(&mdsc->quotarealms_inodes_mutex);
184 while (!RB_EMPTY_ROOT(&mdsc->quotarealms_inodes)) {
185 node = rb_first(&mdsc->quotarealms_inodes);
186 qri = rb_entry(node, struct ceph_quotarealm_inode, node);
187 rb_erase(node, &mdsc->quotarealms_inodes);
191 mutex_unlock(&mdsc->quotarealms_inodes_mutex);
195 * This function walks through the snaprealm for an inode and returns the
196 * ceph_snap_realm for the first snaprealm that has quotas set (either max_files
197 * or max_bytes). If the root is reached, return the root ceph_snap_realm
200 * Note that the caller is responsible for calling ceph_put_snap_realm() on the
203 * Callers of this function need to hold mdsc->snap_rwsem. However, if there's
204 * a need to do an inode lookup, this rwsem will be temporarily dropped. Hence
205 * the 'retry' argument: if rwsem needs to be dropped and 'retry' is 'false'
206 * this function will return -EAGAIN; otherwise, the snaprealms walk-through
209 static struct ceph_snap_realm *get_quota_realm(struct ceph_mds_client *mdsc,
210 struct inode *inode, bool retry)
212 struct ceph_inode_info *ci = NULL;
213 struct ceph_snap_realm *realm, *next;
217 if (ceph_snap(inode) != CEPH_NOSNAP)
221 realm = ceph_inode(inode)->i_snap_realm;
223 ceph_get_snap_realm(mdsc, realm);
225 pr_err_ratelimited("get_quota_realm: ino (%llx.%llx) "
226 "null i_snap_realm\n", ceph_vinop(inode));
230 spin_lock(&realm->inodes_with_caps_lock);
231 has_inode = realm->inode;
232 in = has_inode ? igrab(realm->inode) : NULL;
233 spin_unlock(&realm->inodes_with_caps_lock);
234 if (has_inode && !in)
237 up_read(&mdsc->snap_rwsem);
238 in = lookup_quotarealm_inode(mdsc, inode->i_sb, realm);
239 down_read(&mdsc->snap_rwsem);
240 if (IS_ERR_OR_NULL(in))
242 ceph_put_snap_realm(mdsc, realm);
244 return ERR_PTR(-EAGAIN);
249 has_quota = __ceph_has_any_quota(ci);
250 /* avoid calling iput_final() while holding mdsc->snap_rwsem */
253 next = realm->parent;
254 if (has_quota || !next)
257 ceph_get_snap_realm(mdsc, next);
258 ceph_put_snap_realm(mdsc, realm);
262 ceph_put_snap_realm(mdsc, realm);
267 bool ceph_quota_is_same_realm(struct inode *old, struct inode *new)
269 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old->i_sb);
270 struct ceph_snap_realm *old_realm, *new_realm;
275 * We need to lookup 2 quota realms atomically, i.e. with snap_rwsem.
276 * However, get_quota_realm may drop it temporarily. By setting the
277 * 'retry' parameter to 'false', we'll get -EAGAIN if the rwsem was
278 * dropped and we can then restart the whole operation.
280 down_read(&mdsc->snap_rwsem);
281 old_realm = get_quota_realm(mdsc, old, true);
282 new_realm = get_quota_realm(mdsc, new, false);
283 if (PTR_ERR(new_realm) == -EAGAIN) {
284 up_read(&mdsc->snap_rwsem);
286 ceph_put_snap_realm(mdsc, old_realm);
289 is_same = (old_realm == new_realm);
290 up_read(&mdsc->snap_rwsem);
293 ceph_put_snap_realm(mdsc, old_realm);
295 ceph_put_snap_realm(mdsc, new_realm);
300 enum quota_check_op {
301 QUOTA_CHECK_MAX_FILES_OP, /* check quota max_files limit */
302 QUOTA_CHECK_MAX_BYTES_OP, /* check quota max_files limit */
303 QUOTA_CHECK_MAX_BYTES_APPROACHING_OP /* check if quota max_files
304 limit is approaching */
308 * check_quota_exceeded() will walk up the snaprealm hierarchy and, for each
309 * realm, it will execute quota check operation defined by the 'op' parameter.
310 * The snaprealm walk is interrupted if the quota check detects that the quota
311 * is exceeded or if the root inode is reached.
313 static bool check_quota_exceeded(struct inode *inode, enum quota_check_op op,
316 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
317 struct ceph_inode_info *ci;
318 struct ceph_snap_realm *realm, *next;
321 bool exceeded = false;
323 if (ceph_snap(inode) != CEPH_NOSNAP)
326 down_read(&mdsc->snap_rwsem);
328 realm = ceph_inode(inode)->i_snap_realm;
330 ceph_get_snap_realm(mdsc, realm);
332 pr_err_ratelimited("check_quota_exceeded: ino (%llx.%llx) "
333 "null i_snap_realm\n", ceph_vinop(inode));
337 spin_lock(&realm->inodes_with_caps_lock);
338 has_inode = realm->inode;
339 in = has_inode ? igrab(realm->inode) : NULL;
340 spin_unlock(&realm->inodes_with_caps_lock);
341 if (has_inode && !in)
344 up_read(&mdsc->snap_rwsem);
345 in = lookup_quotarealm_inode(mdsc, inode->i_sb, realm);
346 down_read(&mdsc->snap_rwsem);
347 if (IS_ERR_OR_NULL(in))
349 ceph_put_snap_realm(mdsc, realm);
353 spin_lock(&ci->i_ceph_lock);
354 if (op == QUOTA_CHECK_MAX_FILES_OP) {
355 max = ci->i_max_files;
356 rvalue = ci->i_rfiles + ci->i_rsubdirs;
358 max = ci->i_max_bytes;
359 rvalue = ci->i_rbytes;
361 spin_unlock(&ci->i_ceph_lock);
363 case QUOTA_CHECK_MAX_FILES_OP:
364 case QUOTA_CHECK_MAX_BYTES_OP:
365 exceeded = (max && (rvalue + delta > max));
367 case QUOTA_CHECK_MAX_BYTES_APPROACHING_OP:
373 * when we're writing more that 1/16th
374 * of the available space
377 (((max - rvalue) >> 4) < delta);
382 /* Shouldn't happen */
383 pr_warn("Invalid quota check op (%d)\n", op);
384 exceeded = true; /* Just break the loop */
386 /* avoid calling iput_final() while holding mdsc->snap_rwsem */
389 next = realm->parent;
390 if (exceeded || !next)
392 ceph_get_snap_realm(mdsc, next);
393 ceph_put_snap_realm(mdsc, realm);
397 ceph_put_snap_realm(mdsc, realm);
398 up_read(&mdsc->snap_rwsem);
404 * ceph_quota_is_max_files_exceeded - check if we can create a new file
405 * @inode: directory where a new file is being created
407 * This functions returns true is max_files quota allows a new file to be
408 * created. It is necessary to walk through the snaprealm hierarchy (until the
409 * FS root) to check all realms with quotas set.
411 bool ceph_quota_is_max_files_exceeded(struct inode *inode)
413 if (!ceph_has_realms_with_quotas(inode))
416 WARN_ON(!S_ISDIR(inode->i_mode));
418 return check_quota_exceeded(inode, QUOTA_CHECK_MAX_FILES_OP, 1);
422 * ceph_quota_is_max_bytes_exceeded - check if we can write to a file
423 * @inode: inode being written
424 * @newsize: new size if write succeeds
426 * This functions returns true is max_bytes quota allows a file size to reach
427 * @newsize; it returns false otherwise.
429 bool ceph_quota_is_max_bytes_exceeded(struct inode *inode, loff_t newsize)
431 loff_t size = i_size_read(inode);
433 if (!ceph_has_realms_with_quotas(inode))
436 /* return immediately if we're decreasing file size */
440 return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_OP, (newsize - size));
444 * ceph_quota_is_max_bytes_approaching - check if we're reaching max_bytes
445 * @inode: inode being written
446 * @newsize: new size if write succeeds
448 * This function returns true if the new file size @newsize will be consuming
449 * more than 1/16th of the available quota space; it returns false otherwise.
451 bool ceph_quota_is_max_bytes_approaching(struct inode *inode, loff_t newsize)
453 loff_t size = ceph_inode(inode)->i_reported_size;
455 if (!ceph_has_realms_with_quotas(inode))
458 /* return immediately if we're decreasing file size */
462 return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_APPROACHING_OP,
467 * ceph_quota_update_statfs - if root has quota update statfs with quota status
468 * @fsc: filesystem client instance
469 * @buf: statfs to update
471 * If the mounted filesystem root has max_bytes quota set, update the filesystem
472 * statistics with the quota status.
474 * This function returns true if the stats have been updated, false otherwise.
476 bool ceph_quota_update_statfs(struct ceph_fs_client *fsc, struct kstatfs *buf)
478 struct ceph_mds_client *mdsc = fsc->mdsc;
479 struct ceph_inode_info *ci;
480 struct ceph_snap_realm *realm;
482 u64 total = 0, used, free;
483 bool is_updated = false;
485 down_read(&mdsc->snap_rwsem);
486 realm = get_quota_realm(mdsc, d_inode(fsc->sb->s_root), true);
487 up_read(&mdsc->snap_rwsem);
491 spin_lock(&realm->inodes_with_caps_lock);
492 in = realm->inode ? igrab(realm->inode) : NULL;
493 spin_unlock(&realm->inodes_with_caps_lock);
496 spin_lock(&ci->i_ceph_lock);
497 if (ci->i_max_bytes) {
498 total = ci->i_max_bytes >> CEPH_BLOCK_SHIFT;
499 used = ci->i_rbytes >> CEPH_BLOCK_SHIFT;
500 /* It is possible for a quota to be exceeded.
501 * Report 'zero' in that case
503 free = total > used ? total - used : 0;
505 spin_unlock(&ci->i_ceph_lock);
507 buf->f_blocks = total;
509 buf->f_bavail = free;
514 ceph_put_snap_realm(mdsc, realm);