1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * quota.c - CephFS quota
4  *
5  * Copyright (C) 2017-2018 SUSE
6  */
7 
8 #include <linux/statfs.h>
9 
10 #include "super.h"
11 #include "mds_client.h"
12 
ceph_adjust_quota_realms_count(struct inode * inode,bool inc)13 void ceph_adjust_quota_realms_count(struct inode *inode, bool inc)
14 {
15 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
16 	if (inc)
17 		atomic64_inc(&mdsc->quotarealms_count);
18 	else
19 		atomic64_dec(&mdsc->quotarealms_count);
20 }
21 
ceph_has_realms_with_quotas(struct inode * inode)22 static inline bool ceph_has_realms_with_quotas(struct inode *inode)
23 {
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);
27 
28 	if (atomic64_read(&mdsc->quotarealms_count) > 0)
29 		return true;
30 	/* if root is the real CephFS root, we don't have quota realms */
31 	if (root && ceph_ino(root) == CEPH_INO_ROOT)
32 		return false;
33 	/* otherwise, we can't know for sure */
34 	return true;
35 }
36 
ceph_handle_quota(struct ceph_mds_client * mdsc,struct ceph_mds_session * session,struct ceph_msg * msg)37 void ceph_handle_quota(struct ceph_mds_client *mdsc,
38 		       struct ceph_mds_session *session,
39 		       struct ceph_msg *msg)
40 {
41 	struct super_block *sb = mdsc->fsc->sb;
42 	struct ceph_mds_quota *h = msg->front.iov_base;
43 	struct ceph_vino vino;
44 	struct inode *inode;
45 	struct ceph_inode_info *ci;
46 
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);
50 		ceph_msg_dump(msg);
51 		return;
52 	}
53 
54 	/* increment msg sequence number */
55 	mutex_lock(&session->s_mutex);
56 	inc_session_sequence(session);
57 	mutex_unlock(&session->s_mutex);
58 
59 	/* lookup inode */
60 	vino.ino = le64_to_cpu(h->ino);
61 	vino.snap = CEPH_NOSNAP;
62 	inode = ceph_find_inode(sb, vino);
63 	if (!inode) {
64 		pr_warn("Failed to find inode %llu\n", vino.ino);
65 		return;
66 	}
67 	ci = ceph_inode(inode);
68 
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);
76 
77 	iput(inode);
78 }
79 
80 static struct ceph_quotarealm_inode *
find_quotarealm_inode(struct ceph_mds_client * mdsc,u64 ino)81 find_quotarealm_inode(struct ceph_mds_client *mdsc, u64 ino)
82 {
83 	struct ceph_quotarealm_inode *qri = NULL;
84 	struct rb_node **node, *parent = NULL;
85 
86 	mutex_lock(&mdsc->quotarealms_inodes_mutex);
87 	node = &(mdsc->quotarealms_inodes.rb_node);
88 	while (*node) {
89 		parent = *node;
90 		qri = container_of(*node, struct ceph_quotarealm_inode, node);
91 
92 		if (ino < qri->ino)
93 			node = &((*node)->rb_left);
94 		else if (ino > qri->ino)
95 			node = &((*node)->rb_right);
96 		else
97 			break;
98 	}
99 	if (!qri || (qri->ino != ino)) {
100 		/* Not found, create a new one and insert it */
101 		qri = kmalloc(sizeof(*qri), GFP_KERNEL);
102 		if (qri) {
103 			qri->ino = ino;
104 			qri->inode = NULL;
105 			qri->timeout = 0;
106 			mutex_init(&qri->mutex);
107 			rb_link_node(&qri->node, parent, node);
108 			rb_insert_color(&qri->node, &mdsc->quotarealms_inodes);
109 		} else
110 			pr_warn("Failed to alloc quotarealms_inode\n");
111 	}
112 	mutex_unlock(&mdsc->quotarealms_inodes_mutex);
113 
114 	return qri;
115 }
116 
117 /*
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.
121  *
122  * Note that these inodes are kept in this list even if the lookup fails, which
123  * allows to prevent useless lookup requests.
124  */
lookup_quotarealm_inode(struct ceph_mds_client * mdsc,struct super_block * sb,struct ceph_snap_realm * realm)125 static struct inode *lookup_quotarealm_inode(struct ceph_mds_client *mdsc,
126 					     struct super_block *sb,
127 					     struct ceph_snap_realm *realm)
128 {
129 	struct ceph_quotarealm_inode *qri;
130 	struct inode *in;
131 
132 	qri = find_quotarealm_inode(mdsc, realm->ino);
133 	if (!qri)
134 		return NULL;
135 
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);
140 		return qri->inode;
141 	}
142 	/* Check if this inode lookup has failed recently */
143 	if (qri->timeout &&
144 	    time_before_eq(jiffies, qri->timeout)) {
145 		mutex_unlock(&qri->mutex);
146 		return NULL;
147 	}
148 	if (qri->inode) {
149 		/* get caps */
150 		int ret = __ceph_do_getattr(qri->inode, NULL,
151 					    CEPH_STAT_CAP_INODE, true);
152 		if (ret >= 0)
153 			in = qri->inode;
154 		else
155 			in = ERR_PTR(ret);
156 	}  else {
157 		in = ceph_lookup_inode(sb, realm->ino);
158 	}
159 
160 	if (IS_ERR(in)) {
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 */
164 	} else {
165 		qri->timeout = 0;
166 		qri->inode = in;
167 	}
168 	mutex_unlock(&qri->mutex);
169 
170 	return in;
171 }
172 
ceph_cleanup_quotarealms_inodes(struct ceph_mds_client * mdsc)173 void ceph_cleanup_quotarealms_inodes(struct ceph_mds_client *mdsc)
174 {
175 	struct ceph_quotarealm_inode *qri;
176 	struct rb_node *node;
177 
178 	/*
179 	 * It should now be safe to clean quotarealms_inode tree without holding
180 	 * mdsc->quotarealms_inodes_mutex...
181 	 */
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);
187 		iput(qri->inode);
188 		kfree(qri);
189 	}
190 	mutex_unlock(&mdsc->quotarealms_inodes_mutex);
191 }
192 
193 /*
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
197  * instead.
198  *
199  * Note that the caller is responsible for calling ceph_put_snap_realm() on the
200  * returned realm.
201  *
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
206  * will be restarted.
207  */
get_quota_realm(struct ceph_mds_client * mdsc,struct inode * inode,bool retry)208 static struct ceph_snap_realm *get_quota_realm(struct ceph_mds_client *mdsc,
209 					       struct inode *inode, bool retry)
210 {
211 	struct ceph_inode_info *ci = NULL;
212 	struct ceph_snap_realm *realm, *next;
213 	struct inode *in;
214 	bool has_quota;
215 
216 	if (ceph_snap(inode) != CEPH_NOSNAP)
217 		return NULL;
218 
219 restart:
220 	realm = ceph_inode(inode)->i_snap_realm;
221 	if (realm)
222 		ceph_get_snap_realm(mdsc, realm);
223 	else
224 		pr_err_ratelimited("get_quota_realm: ino (%llx.%llx) "
225 				   "null i_snap_realm\n", ceph_vinop(inode));
226 	while (realm) {
227 		bool has_inode;
228 
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)
234 			break;
235 		if (!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))
240 				break;
241 			ceph_put_snap_realm(mdsc, realm);
242 			if (!retry)
243 				return ERR_PTR(-EAGAIN);
244 			goto restart;
245 		}
246 
247 		ci = ceph_inode(in);
248 		has_quota = __ceph_has_any_quota(ci);
249 		iput(in);
250 
251 		next = realm->parent;
252 		if (has_quota || !next)
253 		       return realm;
254 
255 		ceph_get_snap_realm(mdsc, next);
256 		ceph_put_snap_realm(mdsc, realm);
257 		realm = next;
258 	}
259 	if (realm)
260 		ceph_put_snap_realm(mdsc, realm);
261 
262 	return NULL;
263 }
264 
ceph_quota_is_same_realm(struct inode * old,struct inode * new)265 bool ceph_quota_is_same_realm(struct inode *old, struct inode *new)
266 {
267 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old->i_sb);
268 	struct ceph_snap_realm *old_realm, *new_realm;
269 	bool is_same;
270 
271 restart:
272 	/*
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.
277 	 */
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);
283 		if (old_realm)
284 			ceph_put_snap_realm(mdsc, old_realm);
285 		goto restart;
286 	}
287 	is_same = (old_realm == new_realm);
288 	up_read(&mdsc->snap_rwsem);
289 
290 	if (old_realm)
291 		ceph_put_snap_realm(mdsc, old_realm);
292 	if (new_realm)
293 		ceph_put_snap_realm(mdsc, new_realm);
294 
295 	return is_same;
296 }
297 
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 */
303 };
304 
305 /*
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.
310  */
check_quota_exceeded(struct inode * inode,enum quota_check_op op,loff_t delta)311 static bool check_quota_exceeded(struct inode *inode, enum quota_check_op op,
312 				 loff_t delta)
313 {
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;
317 	struct inode *in;
318 	u64 max, rvalue;
319 	bool exceeded = false;
320 
321 	if (ceph_snap(inode) != CEPH_NOSNAP)
322 		return false;
323 
324 	down_read(&mdsc->snap_rwsem);
325 restart:
326 	realm = ceph_inode(inode)->i_snap_realm;
327 	if (realm)
328 		ceph_get_snap_realm(mdsc, realm);
329 	else
330 		pr_err_ratelimited("check_quota_exceeded: ino (%llx.%llx) "
331 				   "null i_snap_realm\n", ceph_vinop(inode));
332 	while (realm) {
333 		bool has_inode;
334 
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)
340 			break;
341 		if (!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))
346 				break;
347 			ceph_put_snap_realm(mdsc, realm);
348 			goto restart;
349 		}
350 		ci = ceph_inode(in);
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;
355 		} else {
356 			max = ci->i_max_bytes;
357 			rvalue = ci->i_rbytes;
358 		}
359 		spin_unlock(&ci->i_ceph_lock);
360 		switch (op) {
361 		case QUOTA_CHECK_MAX_FILES_OP:
362 		case QUOTA_CHECK_MAX_BYTES_OP:
363 			exceeded = (max && (rvalue + delta > max));
364 			break;
365 		case QUOTA_CHECK_MAX_BYTES_APPROACHING_OP:
366 			if (max) {
367 				if (rvalue >= max)
368 					exceeded = true;
369 				else {
370 					/*
371 					 * when we're writing more that 1/16th
372 					 * of the available space
373 					 */
374 					exceeded =
375 						(((max - rvalue) >> 4) < delta);
376 				}
377 			}
378 			break;
379 		default:
380 			/* Shouldn't happen */
381 			pr_warn("Invalid quota check op (%d)\n", op);
382 			exceeded = true; /* Just break the loop */
383 		}
384 		iput(in);
385 
386 		next = realm->parent;
387 		if (exceeded || !next)
388 			break;
389 		ceph_get_snap_realm(mdsc, next);
390 		ceph_put_snap_realm(mdsc, realm);
391 		realm = next;
392 	}
393 	if (realm)
394 		ceph_put_snap_realm(mdsc, realm);
395 	up_read(&mdsc->snap_rwsem);
396 
397 	return exceeded;
398 }
399 
400 /*
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
403  *
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.
407  */
ceph_quota_is_max_files_exceeded(struct inode * inode)408 bool ceph_quota_is_max_files_exceeded(struct inode *inode)
409 {
410 	if (!ceph_has_realms_with_quotas(inode))
411 		return false;
412 
413 	WARN_ON(!S_ISDIR(inode->i_mode));
414 
415 	return check_quota_exceeded(inode, QUOTA_CHECK_MAX_FILES_OP, 1);
416 }
417 
418 /*
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
422  *
423  * This functions returns true is max_bytes quota allows a file size to reach
424  * @newsize; it returns false otherwise.
425  */
ceph_quota_is_max_bytes_exceeded(struct inode * inode,loff_t newsize)426 bool ceph_quota_is_max_bytes_exceeded(struct inode *inode, loff_t newsize)
427 {
428 	loff_t size = i_size_read(inode);
429 
430 	if (!ceph_has_realms_with_quotas(inode))
431 		return false;
432 
433 	/* return immediately if we're decreasing file size */
434 	if (newsize <= size)
435 		return false;
436 
437 	return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_OP, (newsize - size));
438 }
439 
440 /*
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
444  *
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.
447  */
ceph_quota_is_max_bytes_approaching(struct inode * inode,loff_t newsize)448 bool ceph_quota_is_max_bytes_approaching(struct inode *inode, loff_t newsize)
449 {
450 	loff_t size = ceph_inode(inode)->i_reported_size;
451 
452 	if (!ceph_has_realms_with_quotas(inode))
453 		return false;
454 
455 	/* return immediately if we're decreasing file size */
456 	if (newsize <= size)
457 		return false;
458 
459 	return check_quota_exceeded(inode, QUOTA_CHECK_MAX_BYTES_APPROACHING_OP,
460 				    (newsize - size));
461 }
462 
463 /*
464  * ceph_quota_update_statfs - if root has quota update statfs with quota status
465  * @fsc:	filesystem client instance
466  * @buf:	statfs to update
467  *
468  * If the mounted filesystem root has max_bytes quota set, update the filesystem
469  * statistics with the quota status.
470  *
471  * This function returns true if the stats have been updated, false otherwise.
472  */
ceph_quota_update_statfs(struct ceph_fs_client * fsc,struct kstatfs * buf)473 bool ceph_quota_update_statfs(struct ceph_fs_client *fsc, struct kstatfs *buf)
474 {
475 	struct ceph_mds_client *mdsc = fsc->mdsc;
476 	struct ceph_inode_info *ci;
477 	struct ceph_snap_realm *realm;
478 	struct inode *in;
479 	u64 total = 0, used, free;
480 	bool is_updated = false;
481 
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);
485 	if (!realm)
486 		return false;
487 
488 	spin_lock(&realm->inodes_with_caps_lock);
489 	in = realm->inode ? igrab(realm->inode) : NULL;
490 	spin_unlock(&realm->inodes_with_caps_lock);
491 	if (in) {
492 		ci = ceph_inode(in);
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
499 			 */
500 			free = total > used ? total - used : 0;
501 		}
502 		spin_unlock(&ci->i_ceph_lock);
503 		if (total) {
504 			buf->f_blocks = total;
505 			buf->f_bfree = free;
506 			buf->f_bavail = free;
507 			is_updated = true;
508 		}
509 		iput(in);
510 	}
511 	ceph_put_snap_realm(mdsc, realm);
512 
513 	return is_updated;
514 }
515 
516