xref: /openbmc/linux/fs/ceph/inode.c (revision e5bd61e8)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/module.h>
5 #include <linux/fs.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/xattr.h>
13 #include <linux/posix_acl.h>
14 #include <linux/random.h>
15 #include <linux/sort.h>
16 #include <linux/iversion.h>
17 
18 #include "super.h"
19 #include "mds_client.h"
20 #include "cache.h"
21 #include <linux/ceph/decode.h>
22 
23 /*
24  * Ceph inode operations
25  *
26  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
27  * setattr, etc.), xattr helpers, and helpers for assimilating
28  * metadata returned by the MDS into our cache.
29  *
30  * Also define helpers for doing asynchronous writeback, invalidation,
31  * and truncation for the benefit of those who can't afford to block
32  * (typically because they are in the message handler path).
33  */
34 
35 static const struct inode_operations ceph_symlink_iops;
36 
37 static void ceph_inode_work(struct work_struct *work);
38 
39 /*
40  * find or create an inode, given the ceph ino number
41  */
42 static int ceph_set_ino_cb(struct inode *inode, void *data)
43 {
44 	struct ceph_inode_info *ci = ceph_inode(inode);
45 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
46 
47 	ci->i_vino = *(struct ceph_vino *)data;
48 	inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
49 	inode_set_iversion_raw(inode, 0);
50 	percpu_counter_inc(&mdsc->metric.total_inodes);
51 
52 	return 0;
53 }
54 
55 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
56 {
57 	struct inode *inode;
58 
59 	if (ceph_vino_is_reserved(vino))
60 		return ERR_PTR(-EREMOTEIO);
61 
62 	inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
63 			     ceph_set_ino_cb, &vino);
64 	if (!inode)
65 		return ERR_PTR(-ENOMEM);
66 
67 	dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
68 	     ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
69 	return inode;
70 }
71 
72 /*
73  * get/constuct snapdir inode for a given directory
74  */
75 struct inode *ceph_get_snapdir(struct inode *parent)
76 {
77 	struct ceph_vino vino = {
78 		.ino = ceph_ino(parent),
79 		.snap = CEPH_SNAPDIR,
80 	};
81 	struct inode *inode = ceph_get_inode(parent->i_sb, vino);
82 	struct ceph_inode_info *ci = ceph_inode(inode);
83 
84 	if (IS_ERR(inode))
85 		return inode;
86 
87 	if (!S_ISDIR(parent->i_mode)) {
88 		pr_warn_once("bad snapdir parent type (mode=0%o)\n",
89 			     parent->i_mode);
90 		return ERR_PTR(-ENOTDIR);
91 	}
92 
93 	if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
94 		pr_warn_once("bad snapdir inode type (mode=0%o)\n",
95 			     inode->i_mode);
96 		return ERR_PTR(-ENOTDIR);
97 	}
98 
99 	inode->i_mode = parent->i_mode;
100 	inode->i_uid = parent->i_uid;
101 	inode->i_gid = parent->i_gid;
102 	inode->i_mtime = parent->i_mtime;
103 	inode->i_ctime = parent->i_ctime;
104 	inode->i_atime = parent->i_atime;
105 	ci->i_rbytes = 0;
106 	ci->i_btime = ceph_inode(parent)->i_btime;
107 
108 	if (inode->i_state & I_NEW) {
109 		inode->i_op = &ceph_snapdir_iops;
110 		inode->i_fop = &ceph_snapdir_fops;
111 		ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
112 		unlock_new_inode(inode);
113 	}
114 
115 	return inode;
116 }
117 
118 const struct inode_operations ceph_file_iops = {
119 	.permission = ceph_permission,
120 	.setattr = ceph_setattr,
121 	.getattr = ceph_getattr,
122 	.listxattr = ceph_listxattr,
123 	.get_acl = ceph_get_acl,
124 	.set_acl = ceph_set_acl,
125 };
126 
127 
128 /*
129  * We use a 'frag tree' to keep track of the MDS's directory fragments
130  * for a given inode (usually there is just a single fragment).  We
131  * need to know when a child frag is delegated to a new MDS, or when
132  * it is flagged as replicated, so we can direct our requests
133  * accordingly.
134  */
135 
136 /*
137  * find/create a frag in the tree
138  */
139 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
140 						    u32 f)
141 {
142 	struct rb_node **p;
143 	struct rb_node *parent = NULL;
144 	struct ceph_inode_frag *frag;
145 	int c;
146 
147 	p = &ci->i_fragtree.rb_node;
148 	while (*p) {
149 		parent = *p;
150 		frag = rb_entry(parent, struct ceph_inode_frag, node);
151 		c = ceph_frag_compare(f, frag->frag);
152 		if (c < 0)
153 			p = &(*p)->rb_left;
154 		else if (c > 0)
155 			p = &(*p)->rb_right;
156 		else
157 			return frag;
158 	}
159 
160 	frag = kmalloc(sizeof(*frag), GFP_NOFS);
161 	if (!frag)
162 		return ERR_PTR(-ENOMEM);
163 
164 	frag->frag = f;
165 	frag->split_by = 0;
166 	frag->mds = -1;
167 	frag->ndist = 0;
168 
169 	rb_link_node(&frag->node, parent, p);
170 	rb_insert_color(&frag->node, &ci->i_fragtree);
171 
172 	dout("get_or_create_frag added %llx.%llx frag %x\n",
173 	     ceph_vinop(&ci->vfs_inode), f);
174 	return frag;
175 }
176 
177 /*
178  * find a specific frag @f
179  */
180 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
181 {
182 	struct rb_node *n = ci->i_fragtree.rb_node;
183 
184 	while (n) {
185 		struct ceph_inode_frag *frag =
186 			rb_entry(n, struct ceph_inode_frag, node);
187 		int c = ceph_frag_compare(f, frag->frag);
188 		if (c < 0)
189 			n = n->rb_left;
190 		else if (c > 0)
191 			n = n->rb_right;
192 		else
193 			return frag;
194 	}
195 	return NULL;
196 }
197 
198 /*
199  * Choose frag containing the given value @v.  If @pfrag is
200  * specified, copy the frag delegation info to the caller if
201  * it is present.
202  */
203 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
204 			      struct ceph_inode_frag *pfrag, int *found)
205 {
206 	u32 t = ceph_frag_make(0, 0);
207 	struct ceph_inode_frag *frag;
208 	unsigned nway, i;
209 	u32 n;
210 
211 	if (found)
212 		*found = 0;
213 
214 	while (1) {
215 		WARN_ON(!ceph_frag_contains_value(t, v));
216 		frag = __ceph_find_frag(ci, t);
217 		if (!frag)
218 			break; /* t is a leaf */
219 		if (frag->split_by == 0) {
220 			if (pfrag)
221 				memcpy(pfrag, frag, sizeof(*pfrag));
222 			if (found)
223 				*found = 1;
224 			break;
225 		}
226 
227 		/* choose child */
228 		nway = 1 << frag->split_by;
229 		dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
230 		     frag->split_by, nway);
231 		for (i = 0; i < nway; i++) {
232 			n = ceph_frag_make_child(t, frag->split_by, i);
233 			if (ceph_frag_contains_value(n, v)) {
234 				t = n;
235 				break;
236 			}
237 		}
238 		BUG_ON(i == nway);
239 	}
240 	dout("choose_frag(%x) = %x\n", v, t);
241 
242 	return t;
243 }
244 
245 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
246 		     struct ceph_inode_frag *pfrag, int *found)
247 {
248 	u32 ret;
249 	mutex_lock(&ci->i_fragtree_mutex);
250 	ret = __ceph_choose_frag(ci, v, pfrag, found);
251 	mutex_unlock(&ci->i_fragtree_mutex);
252 	return ret;
253 }
254 
255 /*
256  * Process dirfrag (delegation) info from the mds.  Include leaf
257  * fragment in tree ONLY if ndist > 0.  Otherwise, only
258  * branches/splits are included in i_fragtree)
259  */
260 static int ceph_fill_dirfrag(struct inode *inode,
261 			     struct ceph_mds_reply_dirfrag *dirinfo)
262 {
263 	struct ceph_inode_info *ci = ceph_inode(inode);
264 	struct ceph_inode_frag *frag;
265 	u32 id = le32_to_cpu(dirinfo->frag);
266 	int mds = le32_to_cpu(dirinfo->auth);
267 	int ndist = le32_to_cpu(dirinfo->ndist);
268 	int diri_auth = -1;
269 	int i;
270 	int err = 0;
271 
272 	spin_lock(&ci->i_ceph_lock);
273 	if (ci->i_auth_cap)
274 		diri_auth = ci->i_auth_cap->mds;
275 	spin_unlock(&ci->i_ceph_lock);
276 
277 	if (mds == -1) /* CDIR_AUTH_PARENT */
278 		mds = diri_auth;
279 
280 	mutex_lock(&ci->i_fragtree_mutex);
281 	if (ndist == 0 && mds == diri_auth) {
282 		/* no delegation info needed. */
283 		frag = __ceph_find_frag(ci, id);
284 		if (!frag)
285 			goto out;
286 		if (frag->split_by == 0) {
287 			/* tree leaf, remove */
288 			dout("fill_dirfrag removed %llx.%llx frag %x"
289 			     " (no ref)\n", ceph_vinop(inode), id);
290 			rb_erase(&frag->node, &ci->i_fragtree);
291 			kfree(frag);
292 		} else {
293 			/* tree branch, keep and clear */
294 			dout("fill_dirfrag cleared %llx.%llx frag %x"
295 			     " referral\n", ceph_vinop(inode), id);
296 			frag->mds = -1;
297 			frag->ndist = 0;
298 		}
299 		goto out;
300 	}
301 
302 
303 	/* find/add this frag to store mds delegation info */
304 	frag = __get_or_create_frag(ci, id);
305 	if (IS_ERR(frag)) {
306 		/* this is not the end of the world; we can continue
307 		   with bad/inaccurate delegation info */
308 		pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
309 		       ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
310 		err = -ENOMEM;
311 		goto out;
312 	}
313 
314 	frag->mds = mds;
315 	frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
316 	for (i = 0; i < frag->ndist; i++)
317 		frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
318 	dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
319 	     ceph_vinop(inode), frag->frag, frag->ndist);
320 
321 out:
322 	mutex_unlock(&ci->i_fragtree_mutex);
323 	return err;
324 }
325 
326 static int frag_tree_split_cmp(const void *l, const void *r)
327 {
328 	struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
329 	struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
330 	return ceph_frag_compare(le32_to_cpu(ls->frag),
331 				 le32_to_cpu(rs->frag));
332 }
333 
334 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
335 {
336 	if (!frag)
337 		return f == ceph_frag_make(0, 0);
338 	if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
339 		return false;
340 	return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
341 }
342 
343 static int ceph_fill_fragtree(struct inode *inode,
344 			      struct ceph_frag_tree_head *fragtree,
345 			      struct ceph_mds_reply_dirfrag *dirinfo)
346 {
347 	struct ceph_inode_info *ci = ceph_inode(inode);
348 	struct ceph_inode_frag *frag, *prev_frag = NULL;
349 	struct rb_node *rb_node;
350 	unsigned i, split_by, nsplits;
351 	u32 id;
352 	bool update = false;
353 
354 	mutex_lock(&ci->i_fragtree_mutex);
355 	nsplits = le32_to_cpu(fragtree->nsplits);
356 	if (nsplits != ci->i_fragtree_nsplits) {
357 		update = true;
358 	} else if (nsplits) {
359 		i = prandom_u32() % nsplits;
360 		id = le32_to_cpu(fragtree->splits[i].frag);
361 		if (!__ceph_find_frag(ci, id))
362 			update = true;
363 	} else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
364 		rb_node = rb_first(&ci->i_fragtree);
365 		frag = rb_entry(rb_node, struct ceph_inode_frag, node);
366 		if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
367 			update = true;
368 	}
369 	if (!update && dirinfo) {
370 		id = le32_to_cpu(dirinfo->frag);
371 		if (id != __ceph_choose_frag(ci, id, NULL, NULL))
372 			update = true;
373 	}
374 	if (!update)
375 		goto out_unlock;
376 
377 	if (nsplits > 1) {
378 		sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
379 		     frag_tree_split_cmp, NULL);
380 	}
381 
382 	dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
383 	rb_node = rb_first(&ci->i_fragtree);
384 	for (i = 0; i < nsplits; i++) {
385 		id = le32_to_cpu(fragtree->splits[i].frag);
386 		split_by = le32_to_cpu(fragtree->splits[i].by);
387 		if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
388 			pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
389 			       "frag %x split by %d\n", ceph_vinop(inode),
390 			       i, nsplits, id, split_by);
391 			continue;
392 		}
393 		frag = NULL;
394 		while (rb_node) {
395 			frag = rb_entry(rb_node, struct ceph_inode_frag, node);
396 			if (ceph_frag_compare(frag->frag, id) >= 0) {
397 				if (frag->frag != id)
398 					frag = NULL;
399 				else
400 					rb_node = rb_next(rb_node);
401 				break;
402 			}
403 			rb_node = rb_next(rb_node);
404 			/* delete stale split/leaf node */
405 			if (frag->split_by > 0 ||
406 			    !is_frag_child(frag->frag, prev_frag)) {
407 				rb_erase(&frag->node, &ci->i_fragtree);
408 				if (frag->split_by > 0)
409 					ci->i_fragtree_nsplits--;
410 				kfree(frag);
411 			}
412 			frag = NULL;
413 		}
414 		if (!frag) {
415 			frag = __get_or_create_frag(ci, id);
416 			if (IS_ERR(frag))
417 				continue;
418 		}
419 		if (frag->split_by == 0)
420 			ci->i_fragtree_nsplits++;
421 		frag->split_by = split_by;
422 		dout(" frag %x split by %d\n", frag->frag, frag->split_by);
423 		prev_frag = frag;
424 	}
425 	while (rb_node) {
426 		frag = rb_entry(rb_node, struct ceph_inode_frag, node);
427 		rb_node = rb_next(rb_node);
428 		/* delete stale split/leaf node */
429 		if (frag->split_by > 0 ||
430 		    !is_frag_child(frag->frag, prev_frag)) {
431 			rb_erase(&frag->node, &ci->i_fragtree);
432 			if (frag->split_by > 0)
433 				ci->i_fragtree_nsplits--;
434 			kfree(frag);
435 		}
436 	}
437 out_unlock:
438 	mutex_unlock(&ci->i_fragtree_mutex);
439 	return 0;
440 }
441 
442 /*
443  * initialize a newly allocated inode.
444  */
445 struct inode *ceph_alloc_inode(struct super_block *sb)
446 {
447 	struct ceph_inode_info *ci;
448 	int i;
449 
450 	ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
451 	if (!ci)
452 		return NULL;
453 
454 	dout("alloc_inode %p\n", &ci->vfs_inode);
455 
456 	spin_lock_init(&ci->i_ceph_lock);
457 
458 	ci->i_version = 0;
459 	ci->i_inline_version = 0;
460 	ci->i_time_warp_seq = 0;
461 	ci->i_ceph_flags = 0;
462 	atomic64_set(&ci->i_ordered_count, 1);
463 	atomic64_set(&ci->i_release_count, 1);
464 	atomic64_set(&ci->i_complete_seq[0], 0);
465 	atomic64_set(&ci->i_complete_seq[1], 0);
466 	ci->i_symlink = NULL;
467 
468 	ci->i_max_bytes = 0;
469 	ci->i_max_files = 0;
470 
471 	memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
472 	memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
473 	RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
474 
475 	ci->i_fragtree = RB_ROOT;
476 	mutex_init(&ci->i_fragtree_mutex);
477 
478 	ci->i_xattrs.blob = NULL;
479 	ci->i_xattrs.prealloc_blob = NULL;
480 	ci->i_xattrs.dirty = false;
481 	ci->i_xattrs.index = RB_ROOT;
482 	ci->i_xattrs.count = 0;
483 	ci->i_xattrs.names_size = 0;
484 	ci->i_xattrs.vals_size = 0;
485 	ci->i_xattrs.version = 0;
486 	ci->i_xattrs.index_version = 0;
487 
488 	ci->i_caps = RB_ROOT;
489 	ci->i_auth_cap = NULL;
490 	ci->i_dirty_caps = 0;
491 	ci->i_flushing_caps = 0;
492 	INIT_LIST_HEAD(&ci->i_dirty_item);
493 	INIT_LIST_HEAD(&ci->i_flushing_item);
494 	ci->i_prealloc_cap_flush = NULL;
495 	INIT_LIST_HEAD(&ci->i_cap_flush_list);
496 	init_waitqueue_head(&ci->i_cap_wq);
497 	ci->i_hold_caps_max = 0;
498 	INIT_LIST_HEAD(&ci->i_cap_delay_list);
499 	INIT_LIST_HEAD(&ci->i_cap_snaps);
500 	ci->i_head_snapc = NULL;
501 	ci->i_snap_caps = 0;
502 
503 	ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
504 	for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
505 		ci->i_nr_by_mode[i] = 0;
506 
507 	mutex_init(&ci->i_truncate_mutex);
508 	ci->i_truncate_seq = 0;
509 	ci->i_truncate_size = 0;
510 	ci->i_truncate_pending = 0;
511 
512 	ci->i_max_size = 0;
513 	ci->i_reported_size = 0;
514 	ci->i_wanted_max_size = 0;
515 	ci->i_requested_max_size = 0;
516 
517 	ci->i_pin_ref = 0;
518 	ci->i_rd_ref = 0;
519 	ci->i_rdcache_ref = 0;
520 	ci->i_wr_ref = 0;
521 	ci->i_wb_ref = 0;
522 	ci->i_fx_ref = 0;
523 	ci->i_wrbuffer_ref = 0;
524 	ci->i_wrbuffer_ref_head = 0;
525 	atomic_set(&ci->i_filelock_ref, 0);
526 	atomic_set(&ci->i_shared_gen, 1);
527 	ci->i_rdcache_gen = 0;
528 	ci->i_rdcache_revoking = 0;
529 
530 	INIT_LIST_HEAD(&ci->i_unsafe_dirops);
531 	INIT_LIST_HEAD(&ci->i_unsafe_iops);
532 	spin_lock_init(&ci->i_unsafe_lock);
533 
534 	ci->i_snap_realm = NULL;
535 	INIT_LIST_HEAD(&ci->i_snap_realm_item);
536 	INIT_LIST_HEAD(&ci->i_snap_flush_item);
537 
538 	INIT_WORK(&ci->i_work, ceph_inode_work);
539 	ci->i_work_mask = 0;
540 	memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
541 
542 	ceph_fscache_inode_init(ci);
543 
544 	ci->i_meta_err = 0;
545 
546 	return &ci->vfs_inode;
547 }
548 
549 void ceph_free_inode(struct inode *inode)
550 {
551 	struct ceph_inode_info *ci = ceph_inode(inode);
552 
553 	kfree(ci->i_symlink);
554 	kmem_cache_free(ceph_inode_cachep, ci);
555 }
556 
557 void ceph_evict_inode(struct inode *inode)
558 {
559 	struct ceph_inode_info *ci = ceph_inode(inode);
560 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
561 	struct ceph_inode_frag *frag;
562 	struct rb_node *n;
563 
564 	dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
565 
566 	percpu_counter_dec(&mdsc->metric.total_inodes);
567 
568 	truncate_inode_pages_final(&inode->i_data);
569 	clear_inode(inode);
570 
571 	ceph_fscache_unregister_inode_cookie(ci);
572 
573 	__ceph_remove_caps(ci);
574 
575 	if (__ceph_has_any_quota(ci))
576 		ceph_adjust_quota_realms_count(inode, false);
577 
578 	/*
579 	 * we may still have a snap_realm reference if there are stray
580 	 * caps in i_snap_caps.
581 	 */
582 	if (ci->i_snap_realm) {
583 		if (ceph_snap(inode) == CEPH_NOSNAP) {
584 			struct ceph_snap_realm *realm = ci->i_snap_realm;
585 			dout(" dropping residual ref to snap realm %p\n",
586 			     realm);
587 			spin_lock(&realm->inodes_with_caps_lock);
588 			list_del_init(&ci->i_snap_realm_item);
589 			ci->i_snap_realm = NULL;
590 			if (realm->ino == ci->i_vino.ino)
591 				realm->inode = NULL;
592 			spin_unlock(&realm->inodes_with_caps_lock);
593 			ceph_put_snap_realm(mdsc, realm);
594 		} else {
595 			ceph_put_snapid_map(mdsc, ci->i_snapid_map);
596 			ci->i_snap_realm = NULL;
597 		}
598 	}
599 
600 	while ((n = rb_first(&ci->i_fragtree)) != NULL) {
601 		frag = rb_entry(n, struct ceph_inode_frag, node);
602 		rb_erase(n, &ci->i_fragtree);
603 		kfree(frag);
604 	}
605 	ci->i_fragtree_nsplits = 0;
606 
607 	__ceph_destroy_xattrs(ci);
608 	if (ci->i_xattrs.blob)
609 		ceph_buffer_put(ci->i_xattrs.blob);
610 	if (ci->i_xattrs.prealloc_blob)
611 		ceph_buffer_put(ci->i_xattrs.prealloc_blob);
612 
613 	ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
614 	ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
615 }
616 
617 static inline blkcnt_t calc_inode_blocks(u64 size)
618 {
619 	return (size + (1<<9) - 1) >> 9;
620 }
621 
622 /*
623  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
624  * careful because either the client or MDS may have more up to date
625  * info, depending on which capabilities are held, and whether
626  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
627  * and size are monotonically increasing, except when utimes() or
628  * truncate() increments the corresponding _seq values.)
629  */
630 int ceph_fill_file_size(struct inode *inode, int issued,
631 			u32 truncate_seq, u64 truncate_size, u64 size)
632 {
633 	struct ceph_inode_info *ci = ceph_inode(inode);
634 	int queue_trunc = 0;
635 	loff_t isize = i_size_read(inode);
636 
637 	if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
638 	    (truncate_seq == ci->i_truncate_seq && size > isize)) {
639 		dout("size %lld -> %llu\n", isize, size);
640 		if (size > 0 && S_ISDIR(inode->i_mode)) {
641 			pr_err("fill_file_size non-zero size for directory\n");
642 			size = 0;
643 		}
644 		i_size_write(inode, size);
645 		inode->i_blocks = calc_inode_blocks(size);
646 		ci->i_reported_size = size;
647 		if (truncate_seq != ci->i_truncate_seq) {
648 			dout("truncate_seq %u -> %u\n",
649 			     ci->i_truncate_seq, truncate_seq);
650 			ci->i_truncate_seq = truncate_seq;
651 
652 			/* the MDS should have revoked these caps */
653 			WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
654 					       CEPH_CAP_FILE_RD |
655 					       CEPH_CAP_FILE_WR |
656 					       CEPH_CAP_FILE_LAZYIO));
657 			/*
658 			 * If we hold relevant caps, or in the case where we're
659 			 * not the only client referencing this file and we
660 			 * don't hold those caps, then we need to check whether
661 			 * the file is either opened or mmaped
662 			 */
663 			if ((issued & (CEPH_CAP_FILE_CACHE|
664 				       CEPH_CAP_FILE_BUFFER)) ||
665 			    mapping_mapped(inode->i_mapping) ||
666 			    __ceph_is_file_opened(ci)) {
667 				ci->i_truncate_pending++;
668 				queue_trunc = 1;
669 			}
670 		}
671 	}
672 	if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
673 	    ci->i_truncate_size != truncate_size) {
674 		dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
675 		     truncate_size);
676 		ci->i_truncate_size = truncate_size;
677 	}
678 
679 	if (queue_trunc)
680 		ceph_fscache_invalidate(inode);
681 
682 	return queue_trunc;
683 }
684 
685 void ceph_fill_file_time(struct inode *inode, int issued,
686 			 u64 time_warp_seq, struct timespec64 *ctime,
687 			 struct timespec64 *mtime, struct timespec64 *atime)
688 {
689 	struct ceph_inode_info *ci = ceph_inode(inode);
690 	int warn = 0;
691 
692 	if (issued & (CEPH_CAP_FILE_EXCL|
693 		      CEPH_CAP_FILE_WR|
694 		      CEPH_CAP_FILE_BUFFER|
695 		      CEPH_CAP_AUTH_EXCL|
696 		      CEPH_CAP_XATTR_EXCL)) {
697 		if (ci->i_version == 0 ||
698 		    timespec64_compare(ctime, &inode->i_ctime) > 0) {
699 			dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
700 			     inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
701 			     ctime->tv_sec, ctime->tv_nsec);
702 			inode->i_ctime = *ctime;
703 		}
704 		if (ci->i_version == 0 ||
705 		    ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
706 			/* the MDS did a utimes() */
707 			dout("mtime %lld.%09ld -> %lld.%09ld "
708 			     "tw %d -> %d\n",
709 			     inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
710 			     mtime->tv_sec, mtime->tv_nsec,
711 			     ci->i_time_warp_seq, (int)time_warp_seq);
712 
713 			inode->i_mtime = *mtime;
714 			inode->i_atime = *atime;
715 			ci->i_time_warp_seq = time_warp_seq;
716 		} else if (time_warp_seq == ci->i_time_warp_seq) {
717 			/* nobody did utimes(); take the max */
718 			if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
719 				dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
720 				     inode->i_mtime.tv_sec,
721 				     inode->i_mtime.tv_nsec,
722 				     mtime->tv_sec, mtime->tv_nsec);
723 				inode->i_mtime = *mtime;
724 			}
725 			if (timespec64_compare(atime, &inode->i_atime) > 0) {
726 				dout("atime %lld.%09ld -> %lld.%09ld inc\n",
727 				     inode->i_atime.tv_sec,
728 				     inode->i_atime.tv_nsec,
729 				     atime->tv_sec, atime->tv_nsec);
730 				inode->i_atime = *atime;
731 			}
732 		} else if (issued & CEPH_CAP_FILE_EXCL) {
733 			/* we did a utimes(); ignore mds values */
734 		} else {
735 			warn = 1;
736 		}
737 	} else {
738 		/* we have no write|excl caps; whatever the MDS says is true */
739 		if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
740 			inode->i_ctime = *ctime;
741 			inode->i_mtime = *mtime;
742 			inode->i_atime = *atime;
743 			ci->i_time_warp_seq = time_warp_seq;
744 		} else {
745 			warn = 1;
746 		}
747 	}
748 	if (warn) /* time_warp_seq shouldn't go backwards */
749 		dout("%p mds time_warp_seq %llu < %u\n",
750 		     inode, time_warp_seq, ci->i_time_warp_seq);
751 }
752 
753 /*
754  * Populate an inode based on info from mds.  May be called on new or
755  * existing inodes.
756  */
757 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
758 		    struct ceph_mds_reply_info_in *iinfo,
759 		    struct ceph_mds_reply_dirfrag *dirinfo,
760 		    struct ceph_mds_session *session, int cap_fmode,
761 		    struct ceph_cap_reservation *caps_reservation)
762 {
763 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
764 	struct ceph_mds_reply_inode *info = iinfo->in;
765 	struct ceph_inode_info *ci = ceph_inode(inode);
766 	int issued, new_issued, info_caps;
767 	struct timespec64 mtime, atime, ctime;
768 	struct ceph_buffer *xattr_blob = NULL;
769 	struct ceph_buffer *old_blob = NULL;
770 	struct ceph_string *pool_ns = NULL;
771 	struct ceph_cap *new_cap = NULL;
772 	int err = 0;
773 	bool wake = false;
774 	bool queue_trunc = false;
775 	bool new_version = false;
776 	bool fill_inline = false;
777 	umode_t mode = le32_to_cpu(info->mode);
778 	dev_t rdev = le32_to_cpu(info->rdev);
779 
780 	dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
781 	     inode, ceph_vinop(inode), le64_to_cpu(info->version),
782 	     ci->i_version);
783 
784 	/* Once I_NEW is cleared, we can't change type or dev numbers */
785 	if (inode->i_state & I_NEW) {
786 		inode->i_mode = mode;
787 	} else {
788 		if (inode_wrong_type(inode, mode)) {
789 			pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
790 				     ceph_vinop(inode), inode->i_mode, mode);
791 			return -ESTALE;
792 		}
793 
794 		if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
795 			pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
796 				     ceph_vinop(inode), MAJOR(inode->i_rdev),
797 				     MINOR(inode->i_rdev), MAJOR(rdev),
798 				     MINOR(rdev));
799 			return -ESTALE;
800 		}
801 	}
802 
803 	info_caps = le32_to_cpu(info->cap.caps);
804 
805 	/* prealloc new cap struct */
806 	if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
807 		new_cap = ceph_get_cap(mdsc, caps_reservation);
808 		if (!new_cap)
809 			return -ENOMEM;
810 	}
811 
812 	/*
813 	 * prealloc xattr data, if it looks like we'll need it.  only
814 	 * if len > 4 (meaning there are actually xattrs; the first 4
815 	 * bytes are the xattr count).
816 	 */
817 	if (iinfo->xattr_len > 4) {
818 		xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
819 		if (!xattr_blob)
820 			pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
821 			       iinfo->xattr_len);
822 	}
823 
824 	if (iinfo->pool_ns_len > 0)
825 		pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
826 						     iinfo->pool_ns_len);
827 
828 	if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
829 		ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
830 
831 	spin_lock(&ci->i_ceph_lock);
832 
833 	/*
834 	 * provided version will be odd if inode value is projected,
835 	 * even if stable.  skip the update if we have newer stable
836 	 * info (ours>=theirs, e.g. due to racing mds replies), unless
837 	 * we are getting projected (unstable) info (in which case the
838 	 * version is odd, and we want ours>theirs).
839 	 *   us   them
840 	 *   2    2     skip
841 	 *   3    2     skip
842 	 *   3    3     update
843 	 */
844 	if (ci->i_version == 0 ||
845 	    ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
846 	     le64_to_cpu(info->version) > (ci->i_version & ~1)))
847 		new_version = true;
848 
849 	/* Update change_attribute */
850 	inode_set_max_iversion_raw(inode, iinfo->change_attr);
851 
852 	__ceph_caps_issued(ci, &issued);
853 	issued |= __ceph_caps_dirty(ci);
854 	new_issued = ~issued & info_caps;
855 
856 	/* directories have fl_stripe_unit set to zero */
857 	if (le32_to_cpu(info->layout.fl_stripe_unit))
858 		inode->i_blkbits =
859 			fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
860 	else
861 		inode->i_blkbits = CEPH_BLOCK_SHIFT;
862 
863 	__ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
864 
865 	if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
866 	    (issued & CEPH_CAP_AUTH_EXCL) == 0) {
867 		inode->i_mode = mode;
868 		inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
869 		inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
870 		dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
871 		     from_kuid(&init_user_ns, inode->i_uid),
872 		     from_kgid(&init_user_ns, inode->i_gid));
873 		ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
874 		ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
875 	}
876 
877 	if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
878 	    (issued & CEPH_CAP_LINK_EXCL) == 0)
879 		set_nlink(inode, le32_to_cpu(info->nlink));
880 
881 	if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
882 		/* be careful with mtime, atime, size */
883 		ceph_decode_timespec64(&atime, &info->atime);
884 		ceph_decode_timespec64(&mtime, &info->mtime);
885 		ceph_decode_timespec64(&ctime, &info->ctime);
886 		ceph_fill_file_time(inode, issued,
887 				le32_to_cpu(info->time_warp_seq),
888 				&ctime, &mtime, &atime);
889 	}
890 
891 	if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
892 		ci->i_files = le64_to_cpu(info->files);
893 		ci->i_subdirs = le64_to_cpu(info->subdirs);
894 	}
895 
896 	if (new_version ||
897 	    (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
898 		s64 old_pool = ci->i_layout.pool_id;
899 		struct ceph_string *old_ns;
900 
901 		ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
902 		old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
903 					lockdep_is_held(&ci->i_ceph_lock));
904 		rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
905 
906 		if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
907 			ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
908 
909 		pool_ns = old_ns;
910 
911 		queue_trunc = ceph_fill_file_size(inode, issued,
912 					le32_to_cpu(info->truncate_seq),
913 					le64_to_cpu(info->truncate_size),
914 					le64_to_cpu(info->size));
915 		/* only update max_size on auth cap */
916 		if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
917 		    ci->i_max_size != le64_to_cpu(info->max_size)) {
918 			dout("max_size %lld -> %llu\n", ci->i_max_size,
919 					le64_to_cpu(info->max_size));
920 			ci->i_max_size = le64_to_cpu(info->max_size);
921 		}
922 	}
923 
924 	/* layout and rstat are not tracked by capability, update them if
925 	 * the inode info is from auth mds */
926 	if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
927 		if (S_ISDIR(inode->i_mode)) {
928 			ci->i_dir_layout = iinfo->dir_layout;
929 			ci->i_rbytes = le64_to_cpu(info->rbytes);
930 			ci->i_rfiles = le64_to_cpu(info->rfiles);
931 			ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
932 			ci->i_dir_pin = iinfo->dir_pin;
933 			ci->i_rsnaps = iinfo->rsnaps;
934 			ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
935 		}
936 	}
937 
938 	/* xattrs */
939 	/* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
940 	if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
941 	    le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
942 		if (ci->i_xattrs.blob)
943 			old_blob = ci->i_xattrs.blob;
944 		ci->i_xattrs.blob = xattr_blob;
945 		if (xattr_blob)
946 			memcpy(ci->i_xattrs.blob->vec.iov_base,
947 			       iinfo->xattr_data, iinfo->xattr_len);
948 		ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
949 		ceph_forget_all_cached_acls(inode);
950 		ceph_security_invalidate_secctx(inode);
951 		xattr_blob = NULL;
952 	}
953 
954 	/* finally update i_version */
955 	if (le64_to_cpu(info->version) > ci->i_version)
956 		ci->i_version = le64_to_cpu(info->version);
957 
958 	inode->i_mapping->a_ops = &ceph_aops;
959 
960 	switch (inode->i_mode & S_IFMT) {
961 	case S_IFIFO:
962 	case S_IFBLK:
963 	case S_IFCHR:
964 	case S_IFSOCK:
965 		inode->i_blkbits = PAGE_SHIFT;
966 		init_special_inode(inode, inode->i_mode, rdev);
967 		inode->i_op = &ceph_file_iops;
968 		break;
969 	case S_IFREG:
970 		inode->i_op = &ceph_file_iops;
971 		inode->i_fop = &ceph_file_fops;
972 		break;
973 	case S_IFLNK:
974 		inode->i_op = &ceph_symlink_iops;
975 		if (!ci->i_symlink) {
976 			u32 symlen = iinfo->symlink_len;
977 			char *sym;
978 
979 			spin_unlock(&ci->i_ceph_lock);
980 
981 			if (symlen != i_size_read(inode)) {
982 				pr_err("%s %llx.%llx BAD symlink "
983 					"size %lld\n", __func__,
984 					ceph_vinop(inode),
985 					i_size_read(inode));
986 				i_size_write(inode, symlen);
987 				inode->i_blocks = calc_inode_blocks(symlen);
988 			}
989 
990 			err = -ENOMEM;
991 			sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
992 			if (!sym)
993 				goto out;
994 
995 			spin_lock(&ci->i_ceph_lock);
996 			if (!ci->i_symlink)
997 				ci->i_symlink = sym;
998 			else
999 				kfree(sym); /* lost a race */
1000 		}
1001 		inode->i_link = ci->i_symlink;
1002 		break;
1003 	case S_IFDIR:
1004 		inode->i_op = &ceph_dir_iops;
1005 		inode->i_fop = &ceph_dir_fops;
1006 		break;
1007 	default:
1008 		pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1009 		       ceph_vinop(inode), inode->i_mode);
1010 	}
1011 
1012 	/* were we issued a capability? */
1013 	if (info_caps) {
1014 		if (ceph_snap(inode) == CEPH_NOSNAP) {
1015 			ceph_add_cap(inode, session,
1016 				     le64_to_cpu(info->cap.cap_id),
1017 				     info_caps,
1018 				     le32_to_cpu(info->cap.wanted),
1019 				     le32_to_cpu(info->cap.seq),
1020 				     le32_to_cpu(info->cap.mseq),
1021 				     le64_to_cpu(info->cap.realm),
1022 				     info->cap.flags, &new_cap);
1023 
1024 			/* set dir completion flag? */
1025 			if (S_ISDIR(inode->i_mode) &&
1026 			    ci->i_files == 0 && ci->i_subdirs == 0 &&
1027 			    (info_caps & CEPH_CAP_FILE_SHARED) &&
1028 			    (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1029 			    !__ceph_dir_is_complete(ci)) {
1030 				dout(" marking %p complete (empty)\n", inode);
1031 				i_size_write(inode, 0);
1032 				__ceph_dir_set_complete(ci,
1033 					atomic64_read(&ci->i_release_count),
1034 					atomic64_read(&ci->i_ordered_count));
1035 			}
1036 
1037 			wake = true;
1038 		} else {
1039 			dout(" %p got snap_caps %s\n", inode,
1040 			     ceph_cap_string(info_caps));
1041 			ci->i_snap_caps |= info_caps;
1042 		}
1043 	}
1044 
1045 	if (iinfo->inline_version > 0 &&
1046 	    iinfo->inline_version >= ci->i_inline_version) {
1047 		int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1048 		ci->i_inline_version = iinfo->inline_version;
1049 		if (ci->i_inline_version != CEPH_INLINE_NONE &&
1050 		    (locked_page || (info_caps & cache_caps)))
1051 			fill_inline = true;
1052 	}
1053 
1054 	if (cap_fmode >= 0) {
1055 		if (!info_caps)
1056 			pr_warn("mds issued no caps on %llx.%llx\n",
1057 				ceph_vinop(inode));
1058 		__ceph_touch_fmode(ci, mdsc, cap_fmode);
1059 	}
1060 
1061 	spin_unlock(&ci->i_ceph_lock);
1062 
1063 	if (fill_inline)
1064 		ceph_fill_inline_data(inode, locked_page,
1065 				      iinfo->inline_data, iinfo->inline_len);
1066 
1067 	if (wake)
1068 		wake_up_all(&ci->i_cap_wq);
1069 
1070 	/* queue truncate if we saw i_size decrease */
1071 	if (queue_trunc)
1072 		ceph_queue_vmtruncate(inode);
1073 
1074 	/* populate frag tree */
1075 	if (S_ISDIR(inode->i_mode))
1076 		ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1077 
1078 	/* update delegation info? */
1079 	if (dirinfo)
1080 		ceph_fill_dirfrag(inode, dirinfo);
1081 
1082 	err = 0;
1083 out:
1084 	if (new_cap)
1085 		ceph_put_cap(mdsc, new_cap);
1086 	ceph_buffer_put(old_blob);
1087 	ceph_buffer_put(xattr_blob);
1088 	ceph_put_string(pool_ns);
1089 	return err;
1090 }
1091 
1092 /*
1093  * caller should hold session s_mutex and dentry->d_lock.
1094  */
1095 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1096 				  struct ceph_mds_reply_lease *lease,
1097 				  struct ceph_mds_session *session,
1098 				  unsigned long from_time,
1099 				  struct ceph_mds_session **old_lease_session)
1100 {
1101 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1102 	unsigned mask = le16_to_cpu(lease->mask);
1103 	long unsigned duration = le32_to_cpu(lease->duration_ms);
1104 	long unsigned ttl = from_time + (duration * HZ) / 1000;
1105 	long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1106 
1107 	dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1108 	     dentry, duration, ttl);
1109 
1110 	/* only track leases on regular dentries */
1111 	if (ceph_snap(dir) != CEPH_NOSNAP)
1112 		return;
1113 
1114 	if (mask & CEPH_LEASE_PRIMARY_LINK)
1115 		di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1116 	else
1117 		di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1118 
1119 	di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1120 	if (!(mask & CEPH_LEASE_VALID)) {
1121 		__ceph_dentry_dir_lease_touch(di);
1122 		return;
1123 	}
1124 
1125 	if (di->lease_gen == session->s_cap_gen &&
1126 	    time_before(ttl, di->time))
1127 		return;  /* we already have a newer lease. */
1128 
1129 	if (di->lease_session && di->lease_session != session) {
1130 		*old_lease_session = di->lease_session;
1131 		di->lease_session = NULL;
1132 	}
1133 
1134 	if (!di->lease_session)
1135 		di->lease_session = ceph_get_mds_session(session);
1136 	di->lease_gen = session->s_cap_gen;
1137 	di->lease_seq = le32_to_cpu(lease->seq);
1138 	di->lease_renew_after = half_ttl;
1139 	di->lease_renew_from = 0;
1140 	di->time = ttl;
1141 
1142 	__ceph_dentry_lease_touch(di);
1143 }
1144 
1145 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1146 					struct ceph_mds_reply_lease *lease,
1147 					struct ceph_mds_session *session,
1148 					unsigned long from_time)
1149 {
1150 	struct ceph_mds_session *old_lease_session = NULL;
1151 	spin_lock(&dentry->d_lock);
1152 	__update_dentry_lease(dir, dentry, lease, session, from_time,
1153 			      &old_lease_session);
1154 	spin_unlock(&dentry->d_lock);
1155 	if (old_lease_session)
1156 		ceph_put_mds_session(old_lease_session);
1157 }
1158 
1159 /*
1160  * update dentry lease without having parent inode locked
1161  */
1162 static void update_dentry_lease_careful(struct dentry *dentry,
1163 					struct ceph_mds_reply_lease *lease,
1164 					struct ceph_mds_session *session,
1165 					unsigned long from_time,
1166 					char *dname, u32 dname_len,
1167 					struct ceph_vino *pdvino,
1168 					struct ceph_vino *ptvino)
1169 
1170 {
1171 	struct inode *dir;
1172 	struct ceph_mds_session *old_lease_session = NULL;
1173 
1174 	spin_lock(&dentry->d_lock);
1175 	/* make sure dentry's name matches target */
1176 	if (dentry->d_name.len != dname_len ||
1177 	    memcmp(dentry->d_name.name, dname, dname_len))
1178 		goto out_unlock;
1179 
1180 	dir = d_inode(dentry->d_parent);
1181 	/* make sure parent matches dvino */
1182 	if (!ceph_ino_compare(dir, pdvino))
1183 		goto out_unlock;
1184 
1185 	/* make sure dentry's inode matches target. NULL ptvino means that
1186 	 * we expect a negative dentry */
1187 	if (ptvino) {
1188 		if (d_really_is_negative(dentry))
1189 			goto out_unlock;
1190 		if (!ceph_ino_compare(d_inode(dentry), ptvino))
1191 			goto out_unlock;
1192 	} else {
1193 		if (d_really_is_positive(dentry))
1194 			goto out_unlock;
1195 	}
1196 
1197 	__update_dentry_lease(dir, dentry, lease, session,
1198 			      from_time, &old_lease_session);
1199 out_unlock:
1200 	spin_unlock(&dentry->d_lock);
1201 	if (old_lease_session)
1202 		ceph_put_mds_session(old_lease_session);
1203 }
1204 
1205 /*
1206  * splice a dentry to an inode.
1207  * caller must hold directory i_mutex for this to be safe.
1208  */
1209 static int splice_dentry(struct dentry **pdn, struct inode *in)
1210 {
1211 	struct dentry *dn = *pdn;
1212 	struct dentry *realdn;
1213 
1214 	BUG_ON(d_inode(dn));
1215 
1216 	if (S_ISDIR(in->i_mode)) {
1217 		/* If inode is directory, d_splice_alias() below will remove
1218 		 * 'realdn' from its origin parent. We need to ensure that
1219 		 * origin parent's readdir cache will not reference 'realdn'
1220 		 */
1221 		realdn = d_find_any_alias(in);
1222 		if (realdn) {
1223 			struct ceph_dentry_info *di = ceph_dentry(realdn);
1224 			spin_lock(&realdn->d_lock);
1225 
1226 			realdn->d_op->d_prune(realdn);
1227 
1228 			di->time = jiffies;
1229 			di->lease_shared_gen = 0;
1230 			di->offset = 0;
1231 
1232 			spin_unlock(&realdn->d_lock);
1233 			dput(realdn);
1234 		}
1235 	}
1236 
1237 	/* dn must be unhashed */
1238 	if (!d_unhashed(dn))
1239 		d_drop(dn);
1240 	realdn = d_splice_alias(in, dn);
1241 	if (IS_ERR(realdn)) {
1242 		pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1243 		       PTR_ERR(realdn), dn, in, ceph_vinop(in));
1244 		return PTR_ERR(realdn);
1245 	}
1246 
1247 	if (realdn) {
1248 		dout("dn %p (%d) spliced with %p (%d) "
1249 		     "inode %p ino %llx.%llx\n",
1250 		     dn, d_count(dn),
1251 		     realdn, d_count(realdn),
1252 		     d_inode(realdn), ceph_vinop(d_inode(realdn)));
1253 		dput(dn);
1254 		*pdn = realdn;
1255 	} else {
1256 		BUG_ON(!ceph_dentry(dn));
1257 		dout("dn %p attached to %p ino %llx.%llx\n",
1258 		     dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1259 	}
1260 	return 0;
1261 }
1262 
1263 /*
1264  * Incorporate results into the local cache.  This is either just
1265  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1266  * after a lookup).
1267  *
1268  * A reply may contain
1269  *         a directory inode along with a dentry.
1270  *  and/or a target inode
1271  *
1272  * Called with snap_rwsem (read).
1273  */
1274 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1275 {
1276 	struct ceph_mds_session *session = req->r_session;
1277 	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1278 	struct inode *in = NULL;
1279 	struct ceph_vino tvino, dvino;
1280 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1281 	int err = 0;
1282 
1283 	dout("fill_trace %p is_dentry %d is_target %d\n", req,
1284 	     rinfo->head->is_dentry, rinfo->head->is_target);
1285 
1286 	if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1287 		dout("fill_trace reply is empty!\n");
1288 		if (rinfo->head->result == 0 && req->r_parent)
1289 			ceph_invalidate_dir_request(req);
1290 		return 0;
1291 	}
1292 
1293 	if (rinfo->head->is_dentry) {
1294 		struct inode *dir = req->r_parent;
1295 
1296 		if (dir) {
1297 			err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1298 					      rinfo->dirfrag, session, -1,
1299 					      &req->r_caps_reservation);
1300 			if (err < 0)
1301 				goto done;
1302 		} else {
1303 			WARN_ON_ONCE(1);
1304 		}
1305 
1306 		if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1307 		    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1308 		    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1309 			struct qstr dname;
1310 			struct dentry *dn, *parent;
1311 
1312 			BUG_ON(!rinfo->head->is_target);
1313 			BUG_ON(req->r_dentry);
1314 
1315 			parent = d_find_any_alias(dir);
1316 			BUG_ON(!parent);
1317 
1318 			dname.name = rinfo->dname;
1319 			dname.len = rinfo->dname_len;
1320 			dname.hash = full_name_hash(parent, dname.name, dname.len);
1321 			tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1322 			tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1323 retry_lookup:
1324 			dn = d_lookup(parent, &dname);
1325 			dout("d_lookup on parent=%p name=%.*s got %p\n",
1326 			     parent, dname.len, dname.name, dn);
1327 
1328 			if (!dn) {
1329 				dn = d_alloc(parent, &dname);
1330 				dout("d_alloc %p '%.*s' = %p\n", parent,
1331 				     dname.len, dname.name, dn);
1332 				if (!dn) {
1333 					dput(parent);
1334 					err = -ENOMEM;
1335 					goto done;
1336 				}
1337 				err = 0;
1338 			} else if (d_really_is_positive(dn) &&
1339 				   (ceph_ino(d_inode(dn)) != tvino.ino ||
1340 				    ceph_snap(d_inode(dn)) != tvino.snap)) {
1341 				dout(" dn %p points to wrong inode %p\n",
1342 				     dn, d_inode(dn));
1343 				ceph_dir_clear_ordered(dir);
1344 				d_delete(dn);
1345 				dput(dn);
1346 				goto retry_lookup;
1347 			}
1348 
1349 			req->r_dentry = dn;
1350 			dput(parent);
1351 		}
1352 	}
1353 
1354 	if (rinfo->head->is_target) {
1355 		/* Should be filled in by handle_reply */
1356 		BUG_ON(!req->r_target_inode);
1357 
1358 		in = req->r_target_inode;
1359 		err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1360 				NULL, session,
1361 				(!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1362 				 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1363 				 rinfo->head->result == 0) ?  req->r_fmode : -1,
1364 				&req->r_caps_reservation);
1365 		if (err < 0) {
1366 			pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1367 				in, ceph_vinop(in));
1368 			req->r_target_inode = NULL;
1369 			if (in->i_state & I_NEW)
1370 				discard_new_inode(in);
1371 			else
1372 				iput(in);
1373 			goto done;
1374 		}
1375 		if (in->i_state & I_NEW)
1376 			unlock_new_inode(in);
1377 	}
1378 
1379 	/*
1380 	 * ignore null lease/binding on snapdir ENOENT, or else we
1381 	 * will have trouble splicing in the virtual snapdir later
1382 	 */
1383 	if (rinfo->head->is_dentry &&
1384             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1385 	    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1386 	    (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1387 					       fsc->mount_options->snapdir_name,
1388 					       req->r_dentry->d_name.len))) {
1389 		/*
1390 		 * lookup link rename   : null -> possibly existing inode
1391 		 * mknod symlink mkdir  : null -> new inode
1392 		 * unlink               : linked -> null
1393 		 */
1394 		struct inode *dir = req->r_parent;
1395 		struct dentry *dn = req->r_dentry;
1396 		bool have_dir_cap, have_lease;
1397 
1398 		BUG_ON(!dn);
1399 		BUG_ON(!dir);
1400 		BUG_ON(d_inode(dn->d_parent) != dir);
1401 
1402 		dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1403 		dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1404 
1405 		BUG_ON(ceph_ino(dir) != dvino.ino);
1406 		BUG_ON(ceph_snap(dir) != dvino.snap);
1407 
1408 		/* do we have a lease on the whole dir? */
1409 		have_dir_cap =
1410 			(le32_to_cpu(rinfo->diri.in->cap.caps) &
1411 			 CEPH_CAP_FILE_SHARED);
1412 
1413 		/* do we have a dn lease? */
1414 		have_lease = have_dir_cap ||
1415 			le32_to_cpu(rinfo->dlease->duration_ms);
1416 		if (!have_lease)
1417 			dout("fill_trace  no dentry lease or dir cap\n");
1418 
1419 		/* rename? */
1420 		if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1421 			struct inode *olddir = req->r_old_dentry_dir;
1422 			BUG_ON(!olddir);
1423 
1424 			dout(" src %p '%pd' dst %p '%pd'\n",
1425 			     req->r_old_dentry,
1426 			     req->r_old_dentry,
1427 			     dn, dn);
1428 			dout("fill_trace doing d_move %p -> %p\n",
1429 			     req->r_old_dentry, dn);
1430 
1431 			/* d_move screws up sibling dentries' offsets */
1432 			ceph_dir_clear_ordered(dir);
1433 			ceph_dir_clear_ordered(olddir);
1434 
1435 			d_move(req->r_old_dentry, dn);
1436 			dout(" src %p '%pd' dst %p '%pd'\n",
1437 			     req->r_old_dentry,
1438 			     req->r_old_dentry,
1439 			     dn, dn);
1440 
1441 			/* ensure target dentry is invalidated, despite
1442 			   rehashing bug in vfs_rename_dir */
1443 			ceph_invalidate_dentry_lease(dn);
1444 
1445 			dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1446 			     ceph_dentry(req->r_old_dentry)->offset);
1447 
1448 			/* swap r_dentry and r_old_dentry in case that
1449 			 * splice_dentry() gets called later. This is safe
1450 			 * because no other place will use them */
1451 			req->r_dentry = req->r_old_dentry;
1452 			req->r_old_dentry = dn;
1453 			dn = req->r_dentry;
1454 		}
1455 
1456 		/* null dentry? */
1457 		if (!rinfo->head->is_target) {
1458 			dout("fill_trace null dentry\n");
1459 			if (d_really_is_positive(dn)) {
1460 				dout("d_delete %p\n", dn);
1461 				ceph_dir_clear_ordered(dir);
1462 				d_delete(dn);
1463 			} else if (have_lease) {
1464 				if (d_unhashed(dn))
1465 					d_add(dn, NULL);
1466 				update_dentry_lease(dir, dn,
1467 						    rinfo->dlease, session,
1468 						    req->r_request_started);
1469 			}
1470 			goto done;
1471 		}
1472 
1473 		/* attach proper inode */
1474 		if (d_really_is_negative(dn)) {
1475 			ceph_dir_clear_ordered(dir);
1476 			ihold(in);
1477 			err = splice_dentry(&req->r_dentry, in);
1478 			if (err < 0)
1479 				goto done;
1480 			dn = req->r_dentry;  /* may have spliced */
1481 		} else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1482 			dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1483 			     dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1484 			     ceph_vinop(in));
1485 			d_invalidate(dn);
1486 			have_lease = false;
1487 		}
1488 
1489 		if (have_lease) {
1490 			update_dentry_lease(dir, dn,
1491 					    rinfo->dlease, session,
1492 					    req->r_request_started);
1493 		}
1494 		dout(" final dn %p\n", dn);
1495 	} else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1496 		    req->r_op == CEPH_MDS_OP_MKSNAP) &&
1497 	           test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1498 		   !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1499 		struct inode *dir = req->r_parent;
1500 
1501 		/* fill out a snapdir LOOKUPSNAP dentry */
1502 		BUG_ON(!dir);
1503 		BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1504 		BUG_ON(!req->r_dentry);
1505 		dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1506 		ceph_dir_clear_ordered(dir);
1507 		ihold(in);
1508 		err = splice_dentry(&req->r_dentry, in);
1509 		if (err < 0)
1510 			goto done;
1511 	} else if (rinfo->head->is_dentry && req->r_dentry) {
1512 		/* parent inode is not locked, be carefull */
1513 		struct ceph_vino *ptvino = NULL;
1514 		dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1515 		dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1516 		if (rinfo->head->is_target) {
1517 			tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1518 			tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1519 			ptvino = &tvino;
1520 		}
1521 		update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1522 					    session, req->r_request_started,
1523 					    rinfo->dname, rinfo->dname_len,
1524 					    &dvino, ptvino);
1525 	}
1526 done:
1527 	dout("fill_trace done err=%d\n", err);
1528 	return err;
1529 }
1530 
1531 /*
1532  * Prepopulate our cache with readdir results, leases, etc.
1533  */
1534 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1535 					   struct ceph_mds_session *session)
1536 {
1537 	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1538 	int i, err = 0;
1539 
1540 	for (i = 0; i < rinfo->dir_nr; i++) {
1541 		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1542 		struct ceph_vino vino;
1543 		struct inode *in;
1544 		int rc;
1545 
1546 		vino.ino = le64_to_cpu(rde->inode.in->ino);
1547 		vino.snap = le64_to_cpu(rde->inode.in->snapid);
1548 
1549 		in = ceph_get_inode(req->r_dentry->d_sb, vino);
1550 		if (IS_ERR(in)) {
1551 			err = PTR_ERR(in);
1552 			dout("new_inode badness got %d\n", err);
1553 			continue;
1554 		}
1555 		rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1556 				     -1, &req->r_caps_reservation);
1557 		if (rc < 0) {
1558 			pr_err("ceph_fill_inode badness on %p got %d\n",
1559 			       in, rc);
1560 			err = rc;
1561 			if (in->i_state & I_NEW) {
1562 				ihold(in);
1563 				discard_new_inode(in);
1564 			}
1565 		} else if (in->i_state & I_NEW) {
1566 			unlock_new_inode(in);
1567 		}
1568 
1569 		/* avoid calling iput_final() in mds dispatch threads */
1570 		ceph_async_iput(in);
1571 	}
1572 
1573 	return err;
1574 }
1575 
1576 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1577 {
1578 	if (ctl->page) {
1579 		kunmap(ctl->page);
1580 		put_page(ctl->page);
1581 		ctl->page = NULL;
1582 	}
1583 }
1584 
1585 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1586 			      struct ceph_readdir_cache_control *ctl,
1587 			      struct ceph_mds_request *req)
1588 {
1589 	struct ceph_inode_info *ci = ceph_inode(dir);
1590 	unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1591 	unsigned idx = ctl->index % nsize;
1592 	pgoff_t pgoff = ctl->index / nsize;
1593 
1594 	if (!ctl->page || pgoff != page_index(ctl->page)) {
1595 		ceph_readdir_cache_release(ctl);
1596 		if (idx == 0)
1597 			ctl->page = grab_cache_page(&dir->i_data, pgoff);
1598 		else
1599 			ctl->page = find_lock_page(&dir->i_data, pgoff);
1600 		if (!ctl->page) {
1601 			ctl->index = -1;
1602 			return idx == 0 ? -ENOMEM : 0;
1603 		}
1604 		/* reading/filling the cache are serialized by
1605 		 * i_mutex, no need to use page lock */
1606 		unlock_page(ctl->page);
1607 		ctl->dentries = kmap(ctl->page);
1608 		if (idx == 0)
1609 			memset(ctl->dentries, 0, PAGE_SIZE);
1610 	}
1611 
1612 	if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1613 	    req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1614 		dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1615 		ctl->dentries[idx] = dn;
1616 		ctl->index++;
1617 	} else {
1618 		dout("disable readdir cache\n");
1619 		ctl->index = -1;
1620 	}
1621 	return 0;
1622 }
1623 
1624 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1625 			     struct ceph_mds_session *session)
1626 {
1627 	struct dentry *parent = req->r_dentry;
1628 	struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1629 	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1630 	struct qstr dname;
1631 	struct dentry *dn;
1632 	struct inode *in;
1633 	int err = 0, skipped = 0, ret, i;
1634 	u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1635 	u32 last_hash = 0;
1636 	u32 fpos_offset;
1637 	struct ceph_readdir_cache_control cache_ctl = {};
1638 
1639 	if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1640 		return readdir_prepopulate_inodes_only(req, session);
1641 
1642 	if (rinfo->hash_order) {
1643 		if (req->r_path2) {
1644 			last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1645 						  req->r_path2,
1646 						  strlen(req->r_path2));
1647 			last_hash = ceph_frag_value(last_hash);
1648 		} else if (rinfo->offset_hash) {
1649 			/* mds understands offset_hash */
1650 			WARN_ON_ONCE(req->r_readdir_offset != 2);
1651 			last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1652 		}
1653 	}
1654 
1655 	if (rinfo->dir_dir &&
1656 	    le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1657 		dout("readdir_prepopulate got new frag %x -> %x\n",
1658 		     frag, le32_to_cpu(rinfo->dir_dir->frag));
1659 		frag = le32_to_cpu(rinfo->dir_dir->frag);
1660 		if (!rinfo->hash_order)
1661 			req->r_readdir_offset = 2;
1662 	}
1663 
1664 	if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1665 		dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1666 		     rinfo->dir_nr, parent);
1667 	} else {
1668 		dout("readdir_prepopulate %d items under dn %p\n",
1669 		     rinfo->dir_nr, parent);
1670 		if (rinfo->dir_dir)
1671 			ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1672 
1673 		if (ceph_frag_is_leftmost(frag) &&
1674 		    req->r_readdir_offset == 2 &&
1675 		    !(rinfo->hash_order && last_hash)) {
1676 			/* note dir version at start of readdir so we can
1677 			 * tell if any dentries get dropped */
1678 			req->r_dir_release_cnt =
1679 				atomic64_read(&ci->i_release_count);
1680 			req->r_dir_ordered_cnt =
1681 				atomic64_read(&ci->i_ordered_count);
1682 			req->r_readdir_cache_idx = 0;
1683 		}
1684 	}
1685 
1686 	cache_ctl.index = req->r_readdir_cache_idx;
1687 	fpos_offset = req->r_readdir_offset;
1688 
1689 	/* FIXME: release caps/leases if error occurs */
1690 	for (i = 0; i < rinfo->dir_nr; i++) {
1691 		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1692 		struct ceph_vino tvino;
1693 
1694 		dname.name = rde->name;
1695 		dname.len = rde->name_len;
1696 		dname.hash = full_name_hash(parent, dname.name, dname.len);
1697 
1698 		tvino.ino = le64_to_cpu(rde->inode.in->ino);
1699 		tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1700 
1701 		if (rinfo->hash_order) {
1702 			u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1703 						 rde->name, rde->name_len);
1704 			hash = ceph_frag_value(hash);
1705 			if (hash != last_hash)
1706 				fpos_offset = 2;
1707 			last_hash = hash;
1708 			rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1709 		} else {
1710 			rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1711 		}
1712 
1713 retry_lookup:
1714 		dn = d_lookup(parent, &dname);
1715 		dout("d_lookup on parent=%p name=%.*s got %p\n",
1716 		     parent, dname.len, dname.name, dn);
1717 
1718 		if (!dn) {
1719 			dn = d_alloc(parent, &dname);
1720 			dout("d_alloc %p '%.*s' = %p\n", parent,
1721 			     dname.len, dname.name, dn);
1722 			if (!dn) {
1723 				dout("d_alloc badness\n");
1724 				err = -ENOMEM;
1725 				goto out;
1726 			}
1727 		} else if (d_really_is_positive(dn) &&
1728 			   (ceph_ino(d_inode(dn)) != tvino.ino ||
1729 			    ceph_snap(d_inode(dn)) != tvino.snap)) {
1730 			struct ceph_dentry_info *di = ceph_dentry(dn);
1731 			dout(" dn %p points to wrong inode %p\n",
1732 			     dn, d_inode(dn));
1733 
1734 			spin_lock(&dn->d_lock);
1735 			if (di->offset > 0 &&
1736 			    di->lease_shared_gen ==
1737 			    atomic_read(&ci->i_shared_gen)) {
1738 				__ceph_dir_clear_ordered(ci);
1739 				di->offset = 0;
1740 			}
1741 			spin_unlock(&dn->d_lock);
1742 
1743 			d_delete(dn);
1744 			dput(dn);
1745 			goto retry_lookup;
1746 		}
1747 
1748 		/* inode */
1749 		if (d_really_is_positive(dn)) {
1750 			in = d_inode(dn);
1751 		} else {
1752 			in = ceph_get_inode(parent->d_sb, tvino);
1753 			if (IS_ERR(in)) {
1754 				dout("new_inode badness\n");
1755 				d_drop(dn);
1756 				dput(dn);
1757 				err = PTR_ERR(in);
1758 				goto out;
1759 			}
1760 		}
1761 
1762 		ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1763 				      -1, &req->r_caps_reservation);
1764 		if (ret < 0) {
1765 			pr_err("ceph_fill_inode badness on %p\n", in);
1766 			if (d_really_is_negative(dn)) {
1767 				/* avoid calling iput_final() in mds
1768 				 * dispatch threads */
1769 				if (in->i_state & I_NEW) {
1770 					ihold(in);
1771 					discard_new_inode(in);
1772 				}
1773 				ceph_async_iput(in);
1774 			}
1775 			d_drop(dn);
1776 			err = ret;
1777 			goto next_item;
1778 		}
1779 		if (in->i_state & I_NEW)
1780 			unlock_new_inode(in);
1781 
1782 		if (d_really_is_negative(dn)) {
1783 			if (ceph_security_xattr_deadlock(in)) {
1784 				dout(" skip splicing dn %p to inode %p"
1785 				     " (security xattr deadlock)\n", dn, in);
1786 				ceph_async_iput(in);
1787 				skipped++;
1788 				goto next_item;
1789 			}
1790 
1791 			err = splice_dentry(&dn, in);
1792 			if (err < 0)
1793 				goto next_item;
1794 		}
1795 
1796 		ceph_dentry(dn)->offset = rde->offset;
1797 
1798 		update_dentry_lease(d_inode(parent), dn,
1799 				    rde->lease, req->r_session,
1800 				    req->r_request_started);
1801 
1802 		if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1803 			ret = fill_readdir_cache(d_inode(parent), dn,
1804 						 &cache_ctl, req);
1805 			if (ret < 0)
1806 				err = ret;
1807 		}
1808 next_item:
1809 		dput(dn);
1810 	}
1811 out:
1812 	if (err == 0 && skipped == 0) {
1813 		set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1814 		req->r_readdir_cache_idx = cache_ctl.index;
1815 	}
1816 	ceph_readdir_cache_release(&cache_ctl);
1817 	dout("readdir_prepopulate done\n");
1818 	return err;
1819 }
1820 
1821 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1822 {
1823 	struct ceph_inode_info *ci = ceph_inode(inode);
1824 	bool ret;
1825 
1826 	spin_lock(&ci->i_ceph_lock);
1827 	dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
1828 	i_size_write(inode, size);
1829 	inode->i_blocks = calc_inode_blocks(size);
1830 
1831 	ret = __ceph_should_report_size(ci);
1832 
1833 	spin_unlock(&ci->i_ceph_lock);
1834 	return ret;
1835 }
1836 
1837 /*
1838  * Put reference to inode, but avoid calling iput_final() in current thread.
1839  * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1840  * some contexts.
1841  */
1842 void ceph_async_iput(struct inode *inode)
1843 {
1844 	if (!inode)
1845 		return;
1846 	for (;;) {
1847 		if (atomic_add_unless(&inode->i_count, -1, 1))
1848 			break;
1849 		if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1850 			       &ceph_inode(inode)->i_work))
1851 			break;
1852 		/* queue work failed, i_count must be at least 2 */
1853 	}
1854 }
1855 
1856 void ceph_queue_inode_work(struct inode *inode, int work_bit)
1857 {
1858 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1859 	struct ceph_inode_info *ci = ceph_inode(inode);
1860 	set_bit(work_bit, &ci->i_work_mask);
1861 
1862 	ihold(inode);
1863 	if (queue_work(fsc->inode_wq, &ci->i_work)) {
1864 		dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
1865 	} else {
1866 		dout("queue_inode_work %p already queued, mask=%lx\n",
1867 		     inode, ci->i_work_mask);
1868 		iput(inode);
1869 	}
1870 }
1871 
1872 static void ceph_do_invalidate_pages(struct inode *inode)
1873 {
1874 	struct ceph_inode_info *ci = ceph_inode(inode);
1875 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1876 	u32 orig_gen;
1877 	int check = 0;
1878 
1879 	mutex_lock(&ci->i_truncate_mutex);
1880 
1881 	if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1882 		pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1883 				    inode, ceph_ino(inode));
1884 		mapping_set_error(inode->i_mapping, -EIO);
1885 		truncate_pagecache(inode, 0);
1886 		mutex_unlock(&ci->i_truncate_mutex);
1887 		goto out;
1888 	}
1889 
1890 	spin_lock(&ci->i_ceph_lock);
1891 	dout("invalidate_pages %p gen %d revoking %d\n", inode,
1892 	     ci->i_rdcache_gen, ci->i_rdcache_revoking);
1893 	if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1894 		if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1895 			check = 1;
1896 		spin_unlock(&ci->i_ceph_lock);
1897 		mutex_unlock(&ci->i_truncate_mutex);
1898 		goto out;
1899 	}
1900 	orig_gen = ci->i_rdcache_gen;
1901 	spin_unlock(&ci->i_ceph_lock);
1902 
1903 	ceph_fscache_invalidate(inode);
1904 	if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1905 		pr_err("invalidate_pages %p fails\n", inode);
1906 	}
1907 
1908 	spin_lock(&ci->i_ceph_lock);
1909 	if (orig_gen == ci->i_rdcache_gen &&
1910 	    orig_gen == ci->i_rdcache_revoking) {
1911 		dout("invalidate_pages %p gen %d successful\n", inode,
1912 		     ci->i_rdcache_gen);
1913 		ci->i_rdcache_revoking--;
1914 		check = 1;
1915 	} else {
1916 		dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1917 		     inode, orig_gen, ci->i_rdcache_gen,
1918 		     ci->i_rdcache_revoking);
1919 		if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1920 			check = 1;
1921 	}
1922 	spin_unlock(&ci->i_ceph_lock);
1923 	mutex_unlock(&ci->i_truncate_mutex);
1924 out:
1925 	if (check)
1926 		ceph_check_caps(ci, 0, NULL);
1927 }
1928 
1929 /*
1930  * Make sure any pending truncation is applied before doing anything
1931  * that may depend on it.
1932  */
1933 void __ceph_do_pending_vmtruncate(struct inode *inode)
1934 {
1935 	struct ceph_inode_info *ci = ceph_inode(inode);
1936 	u64 to;
1937 	int wrbuffer_refs, finish = 0;
1938 
1939 	mutex_lock(&ci->i_truncate_mutex);
1940 retry:
1941 	spin_lock(&ci->i_ceph_lock);
1942 	if (ci->i_truncate_pending == 0) {
1943 		dout("__do_pending_vmtruncate %p none pending\n", inode);
1944 		spin_unlock(&ci->i_ceph_lock);
1945 		mutex_unlock(&ci->i_truncate_mutex);
1946 		return;
1947 	}
1948 
1949 	/*
1950 	 * make sure any dirty snapped pages are flushed before we
1951 	 * possibly truncate them.. so write AND block!
1952 	 */
1953 	if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1954 		spin_unlock(&ci->i_ceph_lock);
1955 		dout("__do_pending_vmtruncate %p flushing snaps first\n",
1956 		     inode);
1957 		filemap_write_and_wait_range(&inode->i_data, 0,
1958 					     inode->i_sb->s_maxbytes);
1959 		goto retry;
1960 	}
1961 
1962 	/* there should be no reader or writer */
1963 	WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1964 
1965 	to = ci->i_truncate_size;
1966 	wrbuffer_refs = ci->i_wrbuffer_ref;
1967 	dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1968 	     ci->i_truncate_pending, to);
1969 	spin_unlock(&ci->i_ceph_lock);
1970 
1971 	truncate_pagecache(inode, to);
1972 
1973 	spin_lock(&ci->i_ceph_lock);
1974 	if (to == ci->i_truncate_size) {
1975 		ci->i_truncate_pending = 0;
1976 		finish = 1;
1977 	}
1978 	spin_unlock(&ci->i_ceph_lock);
1979 	if (!finish)
1980 		goto retry;
1981 
1982 	mutex_unlock(&ci->i_truncate_mutex);
1983 
1984 	if (wrbuffer_refs == 0)
1985 		ceph_check_caps(ci, 0, NULL);
1986 
1987 	wake_up_all(&ci->i_cap_wq);
1988 }
1989 
1990 static void ceph_inode_work(struct work_struct *work)
1991 {
1992 	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1993 						 i_work);
1994 	struct inode *inode = &ci->vfs_inode;
1995 
1996 	if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1997 		dout("writeback %p\n", inode);
1998 		filemap_fdatawrite(&inode->i_data);
1999 	}
2000 	if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2001 		ceph_do_invalidate_pages(inode);
2002 
2003 	if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2004 		__ceph_do_pending_vmtruncate(inode);
2005 
2006 	if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
2007 		ceph_check_caps(ci, 0, NULL);
2008 
2009 	if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
2010 		ceph_flush_snaps(ci, NULL);
2011 
2012 	iput(inode);
2013 }
2014 
2015 /*
2016  * symlinks
2017  */
2018 static const struct inode_operations ceph_symlink_iops = {
2019 	.get_link = simple_get_link,
2020 	.setattr = ceph_setattr,
2021 	.getattr = ceph_getattr,
2022 	.listxattr = ceph_listxattr,
2023 };
2024 
2025 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2026 {
2027 	struct ceph_inode_info *ci = ceph_inode(inode);
2028 	unsigned int ia_valid = attr->ia_valid;
2029 	struct ceph_mds_request *req;
2030 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2031 	struct ceph_cap_flush *prealloc_cf;
2032 	int issued;
2033 	int release = 0, dirtied = 0;
2034 	int mask = 0;
2035 	int err = 0;
2036 	int inode_dirty_flags = 0;
2037 	bool lock_snap_rwsem = false;
2038 
2039 	prealloc_cf = ceph_alloc_cap_flush();
2040 	if (!prealloc_cf)
2041 		return -ENOMEM;
2042 
2043 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2044 				       USE_AUTH_MDS);
2045 	if (IS_ERR(req)) {
2046 		ceph_free_cap_flush(prealloc_cf);
2047 		return PTR_ERR(req);
2048 	}
2049 
2050 	spin_lock(&ci->i_ceph_lock);
2051 	issued = __ceph_caps_issued(ci, NULL);
2052 
2053 	if (!ci->i_head_snapc &&
2054 	    (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2055 		lock_snap_rwsem = true;
2056 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
2057 			spin_unlock(&ci->i_ceph_lock);
2058 			down_read(&mdsc->snap_rwsem);
2059 			spin_lock(&ci->i_ceph_lock);
2060 			issued = __ceph_caps_issued(ci, NULL);
2061 		}
2062 	}
2063 
2064 	dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2065 
2066 	if (ia_valid & ATTR_UID) {
2067 		dout("setattr %p uid %d -> %d\n", inode,
2068 		     from_kuid(&init_user_ns, inode->i_uid),
2069 		     from_kuid(&init_user_ns, attr->ia_uid));
2070 		if (issued & CEPH_CAP_AUTH_EXCL) {
2071 			inode->i_uid = attr->ia_uid;
2072 			dirtied |= CEPH_CAP_AUTH_EXCL;
2073 		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2074 			   !uid_eq(attr->ia_uid, inode->i_uid)) {
2075 			req->r_args.setattr.uid = cpu_to_le32(
2076 				from_kuid(&init_user_ns, attr->ia_uid));
2077 			mask |= CEPH_SETATTR_UID;
2078 			release |= CEPH_CAP_AUTH_SHARED;
2079 		}
2080 	}
2081 	if (ia_valid & ATTR_GID) {
2082 		dout("setattr %p gid %d -> %d\n", inode,
2083 		     from_kgid(&init_user_ns, inode->i_gid),
2084 		     from_kgid(&init_user_ns, attr->ia_gid));
2085 		if (issued & CEPH_CAP_AUTH_EXCL) {
2086 			inode->i_gid = attr->ia_gid;
2087 			dirtied |= CEPH_CAP_AUTH_EXCL;
2088 		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2089 			   !gid_eq(attr->ia_gid, inode->i_gid)) {
2090 			req->r_args.setattr.gid = cpu_to_le32(
2091 				from_kgid(&init_user_ns, attr->ia_gid));
2092 			mask |= CEPH_SETATTR_GID;
2093 			release |= CEPH_CAP_AUTH_SHARED;
2094 		}
2095 	}
2096 	if (ia_valid & ATTR_MODE) {
2097 		dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2098 		     attr->ia_mode);
2099 		if (issued & CEPH_CAP_AUTH_EXCL) {
2100 			inode->i_mode = attr->ia_mode;
2101 			dirtied |= CEPH_CAP_AUTH_EXCL;
2102 		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2103 			   attr->ia_mode != inode->i_mode) {
2104 			inode->i_mode = attr->ia_mode;
2105 			req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2106 			mask |= CEPH_SETATTR_MODE;
2107 			release |= CEPH_CAP_AUTH_SHARED;
2108 		}
2109 	}
2110 
2111 	if (ia_valid & ATTR_ATIME) {
2112 		dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2113 		     inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2114 		     attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2115 		if (issued & CEPH_CAP_FILE_EXCL) {
2116 			ci->i_time_warp_seq++;
2117 			inode->i_atime = attr->ia_atime;
2118 			dirtied |= CEPH_CAP_FILE_EXCL;
2119 		} else if ((issued & CEPH_CAP_FILE_WR) &&
2120 			   timespec64_compare(&inode->i_atime,
2121 					    &attr->ia_atime) < 0) {
2122 			inode->i_atime = attr->ia_atime;
2123 			dirtied |= CEPH_CAP_FILE_WR;
2124 		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2125 			   !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2126 			ceph_encode_timespec64(&req->r_args.setattr.atime,
2127 					       &attr->ia_atime);
2128 			mask |= CEPH_SETATTR_ATIME;
2129 			release |= CEPH_CAP_FILE_SHARED |
2130 				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2131 		}
2132 	}
2133 	if (ia_valid & ATTR_SIZE) {
2134 		loff_t isize = i_size_read(inode);
2135 
2136 		dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2137 		if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size > isize) {
2138 			i_size_write(inode, attr->ia_size);
2139 			inode->i_blocks = calc_inode_blocks(attr->ia_size);
2140 			ci->i_reported_size = attr->ia_size;
2141 			dirtied |= CEPH_CAP_FILE_EXCL;
2142 			ia_valid |= ATTR_MTIME;
2143 		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2144 			   attr->ia_size != isize) {
2145 			req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2146 			req->r_args.setattr.old_size = cpu_to_le64(isize);
2147 			mask |= CEPH_SETATTR_SIZE;
2148 			release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2149 				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2150 		}
2151 	}
2152 	if (ia_valid & ATTR_MTIME) {
2153 		dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2154 		     inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2155 		     attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2156 		if (issued & CEPH_CAP_FILE_EXCL) {
2157 			ci->i_time_warp_seq++;
2158 			inode->i_mtime = attr->ia_mtime;
2159 			dirtied |= CEPH_CAP_FILE_EXCL;
2160 		} else if ((issued & CEPH_CAP_FILE_WR) &&
2161 			   timespec64_compare(&inode->i_mtime,
2162 					    &attr->ia_mtime) < 0) {
2163 			inode->i_mtime = attr->ia_mtime;
2164 			dirtied |= CEPH_CAP_FILE_WR;
2165 		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2166 			   !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2167 			ceph_encode_timespec64(&req->r_args.setattr.mtime,
2168 					       &attr->ia_mtime);
2169 			mask |= CEPH_SETATTR_MTIME;
2170 			release |= CEPH_CAP_FILE_SHARED |
2171 				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2172 		}
2173 	}
2174 
2175 	/* these do nothing */
2176 	if (ia_valid & ATTR_CTIME) {
2177 		bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2178 					 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2179 		dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2180 		     inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2181 		     attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2182 		     only ? "ctime only" : "ignored");
2183 		if (only) {
2184 			/*
2185 			 * if kernel wants to dirty ctime but nothing else,
2186 			 * we need to choose a cap to dirty under, or do
2187 			 * a almost-no-op setattr
2188 			 */
2189 			if (issued & CEPH_CAP_AUTH_EXCL)
2190 				dirtied |= CEPH_CAP_AUTH_EXCL;
2191 			else if (issued & CEPH_CAP_FILE_EXCL)
2192 				dirtied |= CEPH_CAP_FILE_EXCL;
2193 			else if (issued & CEPH_CAP_XATTR_EXCL)
2194 				dirtied |= CEPH_CAP_XATTR_EXCL;
2195 			else
2196 				mask |= CEPH_SETATTR_CTIME;
2197 		}
2198 	}
2199 	if (ia_valid & ATTR_FILE)
2200 		dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2201 
2202 	if (dirtied) {
2203 		inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2204 							   &prealloc_cf);
2205 		inode->i_ctime = attr->ia_ctime;
2206 	}
2207 
2208 	release &= issued;
2209 	spin_unlock(&ci->i_ceph_lock);
2210 	if (lock_snap_rwsem)
2211 		up_read(&mdsc->snap_rwsem);
2212 
2213 	if (inode_dirty_flags)
2214 		__mark_inode_dirty(inode, inode_dirty_flags);
2215 
2216 
2217 	if (mask) {
2218 		req->r_inode = inode;
2219 		ihold(inode);
2220 		req->r_inode_drop = release;
2221 		req->r_args.setattr.mask = cpu_to_le32(mask);
2222 		req->r_num_caps = 1;
2223 		req->r_stamp = attr->ia_ctime;
2224 		err = ceph_mdsc_do_request(mdsc, NULL, req);
2225 	}
2226 	dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2227 	     ceph_cap_string(dirtied), mask);
2228 
2229 	ceph_mdsc_put_request(req);
2230 	ceph_free_cap_flush(prealloc_cf);
2231 
2232 	if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2233 		__ceph_do_pending_vmtruncate(inode);
2234 
2235 	return err;
2236 }
2237 
2238 /*
2239  * setattr
2240  */
2241 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
2242 		 struct iattr *attr)
2243 {
2244 	struct inode *inode = d_inode(dentry);
2245 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2246 	int err;
2247 
2248 	if (ceph_snap(inode) != CEPH_NOSNAP)
2249 		return -EROFS;
2250 
2251 	err = setattr_prepare(&init_user_ns, dentry, attr);
2252 	if (err != 0)
2253 		return err;
2254 
2255 	if ((attr->ia_valid & ATTR_SIZE) &&
2256 	    attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2257 		return -EFBIG;
2258 
2259 	if ((attr->ia_valid & ATTR_SIZE) &&
2260 	    ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2261 		return -EDQUOT;
2262 
2263 	err = __ceph_setattr(inode, attr);
2264 
2265 	if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2266 		err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode);
2267 
2268 	return err;
2269 }
2270 
2271 /*
2272  * Verify that we have a lease on the given mask.  If not,
2273  * do a getattr against an mds.
2274  */
2275 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2276 		      int mask, bool force)
2277 {
2278 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2279 	struct ceph_mds_client *mdsc = fsc->mdsc;
2280 	struct ceph_mds_request *req;
2281 	int mode;
2282 	int err;
2283 
2284 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
2285 		dout("do_getattr inode %p SNAPDIR\n", inode);
2286 		return 0;
2287 	}
2288 
2289 	dout("do_getattr inode %p mask %s mode 0%o\n",
2290 	     inode, ceph_cap_string(mask), inode->i_mode);
2291 	if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2292 			return 0;
2293 
2294 	mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2295 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2296 	if (IS_ERR(req))
2297 		return PTR_ERR(req);
2298 	req->r_inode = inode;
2299 	ihold(inode);
2300 	req->r_num_caps = 1;
2301 	req->r_args.getattr.mask = cpu_to_le32(mask);
2302 	req->r_locked_page = locked_page;
2303 	err = ceph_mdsc_do_request(mdsc, NULL, req);
2304 	if (locked_page && err == 0) {
2305 		u64 inline_version = req->r_reply_info.targeti.inline_version;
2306 		if (inline_version == 0) {
2307 			/* the reply is supposed to contain inline data */
2308 			err = -EINVAL;
2309 		} else if (inline_version == CEPH_INLINE_NONE) {
2310 			err = -ENODATA;
2311 		} else {
2312 			err = req->r_reply_info.targeti.inline_len;
2313 		}
2314 	}
2315 	ceph_mdsc_put_request(req);
2316 	dout("do_getattr result=%d\n", err);
2317 	return err;
2318 }
2319 
2320 
2321 /*
2322  * Check inode permissions.  We verify we have a valid value for
2323  * the AUTH cap, then call the generic handler.
2324  */
2325 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode,
2326 		    int mask)
2327 {
2328 	int err;
2329 
2330 	if (mask & MAY_NOT_BLOCK)
2331 		return -ECHILD;
2332 
2333 	err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2334 
2335 	if (!err)
2336 		err = generic_permission(&init_user_ns, inode, mask);
2337 	return err;
2338 }
2339 
2340 /* Craft a mask of needed caps given a set of requested statx attrs. */
2341 static int statx_to_caps(u32 want, umode_t mode)
2342 {
2343 	int mask = 0;
2344 
2345 	if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2346 		mask |= CEPH_CAP_AUTH_SHARED;
2347 
2348 	if (want & (STATX_NLINK|STATX_CTIME)) {
2349 		/*
2350 		 * The link count for directories depends on inode->i_subdirs,
2351 		 * and that is only updated when Fs caps are held.
2352 		 */
2353 		if (S_ISDIR(mode))
2354 			mask |= CEPH_CAP_FILE_SHARED;
2355 		else
2356 			mask |= CEPH_CAP_LINK_SHARED;
2357 	}
2358 
2359 	if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2360 		    STATX_BLOCKS))
2361 		mask |= CEPH_CAP_FILE_SHARED;
2362 
2363 	if (want & (STATX_CTIME))
2364 		mask |= CEPH_CAP_XATTR_SHARED;
2365 
2366 	return mask;
2367 }
2368 
2369 /*
2370  * Get all the attributes. If we have sufficient caps for the requested attrs,
2371  * then we can avoid talking to the MDS at all.
2372  */
2373 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path,
2374 		 struct kstat *stat, u32 request_mask, unsigned int flags)
2375 {
2376 	struct inode *inode = d_inode(path->dentry);
2377 	struct ceph_inode_info *ci = ceph_inode(inode);
2378 	u32 valid_mask = STATX_BASIC_STATS;
2379 	int err = 0;
2380 
2381 	/* Skip the getattr altogether if we're asked not to sync */
2382 	if (!(flags & AT_STATX_DONT_SYNC)) {
2383 		err = ceph_do_getattr(inode,
2384 				statx_to_caps(request_mask, inode->i_mode),
2385 				flags & AT_STATX_FORCE_SYNC);
2386 		if (err)
2387 			return err;
2388 	}
2389 
2390 	generic_fillattr(&init_user_ns, inode, stat);
2391 	stat->ino = ceph_present_inode(inode);
2392 
2393 	/*
2394 	 * btime on newly-allocated inodes is 0, so if this is still set to
2395 	 * that, then assume that it's not valid.
2396 	 */
2397 	if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2398 		stat->btime = ci->i_btime;
2399 		valid_mask |= STATX_BTIME;
2400 	}
2401 
2402 	if (ceph_snap(inode) == CEPH_NOSNAP)
2403 		stat->dev = inode->i_sb->s_dev;
2404 	else
2405 		stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2406 
2407 	if (S_ISDIR(inode->i_mode)) {
2408 		if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2409 					RBYTES))
2410 			stat->size = ci->i_rbytes;
2411 		else
2412 			stat->size = ci->i_files + ci->i_subdirs;
2413 		stat->blocks = 0;
2414 		stat->blksize = 65536;
2415 		/*
2416 		 * Some applications rely on the number of st_nlink
2417 		 * value on directories to be either 0 (if unlinked)
2418 		 * or 2 + number of subdirectories.
2419 		 */
2420 		if (stat->nlink == 1)
2421 			/* '.' + '..' + subdirs */
2422 			stat->nlink = 1 + 1 + ci->i_subdirs;
2423 	}
2424 
2425 	stat->result_mask = request_mask & valid_mask;
2426 	return err;
2427 }
2428