xref: /openbmc/linux/fs/ceph/inode.c (revision 295fc4aa7de4b72cfd764b75a238f79b9433e3ec)
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 #include <linux/fscrypt.h>
18 
19 #include "super.h"
20 #include "mds_client.h"
21 #include "cache.h"
22 #include "crypto.h"
23 #include <linux/ceph/decode.h>
24 
25 /*
26  * Ceph inode operations
27  *
28  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
29  * setattr, etc.), xattr helpers, and helpers for assimilating
30  * metadata returned by the MDS into our cache.
31  *
32  * Also define helpers for doing asynchronous writeback, invalidation,
33  * and truncation for the benefit of those who can't afford to block
34  * (typically because they are in the message handler path).
35  */
36 
37 static const struct inode_operations ceph_symlink_iops;
38 static const struct inode_operations ceph_encrypted_symlink_iops;
39 
40 static void ceph_inode_work(struct work_struct *work);
41 
42 /*
43  * find or create an inode, given the ceph ino number
44  */
45 static int ceph_set_ino_cb(struct inode *inode, void *data)
46 {
47 	struct ceph_inode_info *ci = ceph_inode(inode);
48 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
49 
50 	ci->i_vino = *(struct ceph_vino *)data;
51 	inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
52 	inode_set_iversion_raw(inode, 0);
53 	percpu_counter_inc(&mdsc->metric.total_inodes);
54 
55 	return 0;
56 }
57 
58 /**
59  * ceph_new_inode - allocate a new inode in advance of an expected create
60  * @dir: parent directory for new inode
61  * @dentry: dentry that may eventually point to new inode
62  * @mode: mode of new inode
63  * @as_ctx: pointer to inherited security context
64  *
65  * Allocate a new inode in advance of an operation to create a new inode.
66  * This allocates the inode and sets up the acl_sec_ctx with appropriate
67  * info for the new inode.
68  *
69  * Returns a pointer to the new inode or an ERR_PTR.
70  */
71 struct inode *ceph_new_inode(struct inode *dir, struct dentry *dentry,
72 			     umode_t *mode, struct ceph_acl_sec_ctx *as_ctx)
73 {
74 	int err;
75 	struct inode *inode;
76 
77 	inode = new_inode(dir->i_sb);
78 	if (!inode)
79 		return ERR_PTR(-ENOMEM);
80 
81 	if (!S_ISLNK(*mode)) {
82 		err = ceph_pre_init_acls(dir, mode, as_ctx);
83 		if (err < 0)
84 			goto out_err;
85 	}
86 
87 	inode->i_state = 0;
88 	inode->i_mode = *mode;
89 
90 	err = ceph_security_init_secctx(dentry, *mode, as_ctx);
91 	if (err < 0)
92 		goto out_err;
93 
94 	/*
95 	 * We'll skip setting fscrypt context for snapshots, leaving that for
96 	 * the handle_reply().
97 	 */
98 	if (ceph_snap(dir) != CEPH_SNAPDIR) {
99 		err = ceph_fscrypt_prepare_context(dir, inode, as_ctx);
100 		if (err)
101 			goto out_err;
102 	}
103 
104 	return inode;
105 out_err:
106 	iput(inode);
107 	return ERR_PTR(err);
108 }
109 
110 void ceph_as_ctx_to_req(struct ceph_mds_request *req,
111 			struct ceph_acl_sec_ctx *as_ctx)
112 {
113 	if (as_ctx->pagelist) {
114 		req->r_pagelist = as_ctx->pagelist;
115 		as_ctx->pagelist = NULL;
116 	}
117 	ceph_fscrypt_as_ctx_to_req(req, as_ctx);
118 }
119 
120 /**
121  * ceph_get_inode - find or create/hash a new inode
122  * @sb: superblock to search and allocate in
123  * @vino: vino to search for
124  * @newino: optional new inode to insert if one isn't found (may be NULL)
125  *
126  * Search for or insert a new inode into the hash for the given vino, and
127  * return a reference to it. If new is non-NULL, its reference is consumed.
128  */
129 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino,
130 			     struct inode *newino)
131 {
132 	struct inode *inode;
133 
134 	if (ceph_vino_is_reserved(vino))
135 		return ERR_PTR(-EREMOTEIO);
136 
137 	if (newino) {
138 		inode = inode_insert5(newino, (unsigned long)vino.ino,
139 				      ceph_ino_compare, ceph_set_ino_cb, &vino);
140 		if (inode != newino)
141 			iput(newino);
142 	} else {
143 		inode = iget5_locked(sb, (unsigned long)vino.ino,
144 				     ceph_ino_compare, ceph_set_ino_cb, &vino);
145 	}
146 
147 	if (!inode) {
148 		dout("No inode found for %llx.%llx\n", vino.ino, vino.snap);
149 		return ERR_PTR(-ENOMEM);
150 	}
151 
152 	dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
153 	     ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
154 	return inode;
155 }
156 
157 /*
158  * get/constuct snapdir inode for a given directory
159  */
160 struct inode *ceph_get_snapdir(struct inode *parent)
161 {
162 	struct ceph_vino vino = {
163 		.ino = ceph_ino(parent),
164 		.snap = CEPH_SNAPDIR,
165 	};
166 	struct inode *inode = ceph_get_inode(parent->i_sb, vino, NULL);
167 	struct ceph_inode_info *ci = ceph_inode(inode);
168 	int ret = -ENOTDIR;
169 
170 	if (IS_ERR(inode))
171 		return inode;
172 
173 	if (!S_ISDIR(parent->i_mode)) {
174 		pr_warn_once("bad snapdir parent type (mode=0%o)\n",
175 			     parent->i_mode);
176 		goto err;
177 	}
178 
179 	if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
180 		pr_warn_once("bad snapdir inode type (mode=0%o)\n",
181 			     inode->i_mode);
182 		goto err;
183 	}
184 
185 	inode->i_mode = parent->i_mode;
186 	inode->i_uid = parent->i_uid;
187 	inode->i_gid = parent->i_gid;
188 	inode->i_mtime = parent->i_mtime;
189 	inode->i_ctime = parent->i_ctime;
190 	inode->i_atime = parent->i_atime;
191 	ci->i_rbytes = 0;
192 	ci->i_btime = ceph_inode(parent)->i_btime;
193 
194 #ifdef CONFIG_FS_ENCRYPTION
195 	/* if encrypted, just borrow fscrypt_auth from parent */
196 	if (IS_ENCRYPTED(parent)) {
197 		struct ceph_inode_info *pci = ceph_inode(parent);
198 
199 		ci->fscrypt_auth = kmemdup(pci->fscrypt_auth,
200 					   pci->fscrypt_auth_len,
201 					   GFP_KERNEL);
202 		if (ci->fscrypt_auth) {
203 			inode->i_flags |= S_ENCRYPTED;
204 			ci->fscrypt_auth_len = pci->fscrypt_auth_len;
205 		} else {
206 			dout("Failed to alloc snapdir fscrypt_auth\n");
207 			ret = -ENOMEM;
208 			goto err;
209 		}
210 	}
211 #endif
212 	if (inode->i_state & I_NEW) {
213 		inode->i_op = &ceph_snapdir_iops;
214 		inode->i_fop = &ceph_snapdir_fops;
215 		ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
216 		unlock_new_inode(inode);
217 	}
218 
219 	return inode;
220 err:
221 	if ((inode->i_state & I_NEW))
222 		discard_new_inode(inode);
223 	else
224 		iput(inode);
225 	return ERR_PTR(ret);
226 }
227 
228 const struct inode_operations ceph_file_iops = {
229 	.permission = ceph_permission,
230 	.setattr = ceph_setattr,
231 	.getattr = ceph_getattr,
232 	.listxattr = ceph_listxattr,
233 	.get_inode_acl = ceph_get_acl,
234 	.set_acl = ceph_set_acl,
235 };
236 
237 
238 /*
239  * We use a 'frag tree' to keep track of the MDS's directory fragments
240  * for a given inode (usually there is just a single fragment).  We
241  * need to know when a child frag is delegated to a new MDS, or when
242  * it is flagged as replicated, so we can direct our requests
243  * accordingly.
244  */
245 
246 /*
247  * find/create a frag in the tree
248  */
249 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
250 						    u32 f)
251 {
252 	struct rb_node **p;
253 	struct rb_node *parent = NULL;
254 	struct ceph_inode_frag *frag;
255 	int c;
256 
257 	p = &ci->i_fragtree.rb_node;
258 	while (*p) {
259 		parent = *p;
260 		frag = rb_entry(parent, struct ceph_inode_frag, node);
261 		c = ceph_frag_compare(f, frag->frag);
262 		if (c < 0)
263 			p = &(*p)->rb_left;
264 		else if (c > 0)
265 			p = &(*p)->rb_right;
266 		else
267 			return frag;
268 	}
269 
270 	frag = kmalloc(sizeof(*frag), GFP_NOFS);
271 	if (!frag)
272 		return ERR_PTR(-ENOMEM);
273 
274 	frag->frag = f;
275 	frag->split_by = 0;
276 	frag->mds = -1;
277 	frag->ndist = 0;
278 
279 	rb_link_node(&frag->node, parent, p);
280 	rb_insert_color(&frag->node, &ci->i_fragtree);
281 
282 	dout("get_or_create_frag added %llx.%llx frag %x\n",
283 	     ceph_vinop(&ci->netfs.inode), f);
284 	return frag;
285 }
286 
287 /*
288  * find a specific frag @f
289  */
290 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
291 {
292 	struct rb_node *n = ci->i_fragtree.rb_node;
293 
294 	while (n) {
295 		struct ceph_inode_frag *frag =
296 			rb_entry(n, struct ceph_inode_frag, node);
297 		int c = ceph_frag_compare(f, frag->frag);
298 		if (c < 0)
299 			n = n->rb_left;
300 		else if (c > 0)
301 			n = n->rb_right;
302 		else
303 			return frag;
304 	}
305 	return NULL;
306 }
307 
308 /*
309  * Choose frag containing the given value @v.  If @pfrag is
310  * specified, copy the frag delegation info to the caller if
311  * it is present.
312  */
313 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
314 			      struct ceph_inode_frag *pfrag, int *found)
315 {
316 	u32 t = ceph_frag_make(0, 0);
317 	struct ceph_inode_frag *frag;
318 	unsigned nway, i;
319 	u32 n;
320 
321 	if (found)
322 		*found = 0;
323 
324 	while (1) {
325 		WARN_ON(!ceph_frag_contains_value(t, v));
326 		frag = __ceph_find_frag(ci, t);
327 		if (!frag)
328 			break; /* t is a leaf */
329 		if (frag->split_by == 0) {
330 			if (pfrag)
331 				memcpy(pfrag, frag, sizeof(*pfrag));
332 			if (found)
333 				*found = 1;
334 			break;
335 		}
336 
337 		/* choose child */
338 		nway = 1 << frag->split_by;
339 		dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
340 		     frag->split_by, nway);
341 		for (i = 0; i < nway; i++) {
342 			n = ceph_frag_make_child(t, frag->split_by, i);
343 			if (ceph_frag_contains_value(n, v)) {
344 				t = n;
345 				break;
346 			}
347 		}
348 		BUG_ON(i == nway);
349 	}
350 	dout("choose_frag(%x) = %x\n", v, t);
351 
352 	return t;
353 }
354 
355 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
356 		     struct ceph_inode_frag *pfrag, int *found)
357 {
358 	u32 ret;
359 	mutex_lock(&ci->i_fragtree_mutex);
360 	ret = __ceph_choose_frag(ci, v, pfrag, found);
361 	mutex_unlock(&ci->i_fragtree_mutex);
362 	return ret;
363 }
364 
365 /*
366  * Process dirfrag (delegation) info from the mds.  Include leaf
367  * fragment in tree ONLY if ndist > 0.  Otherwise, only
368  * branches/splits are included in i_fragtree)
369  */
370 static int ceph_fill_dirfrag(struct inode *inode,
371 			     struct ceph_mds_reply_dirfrag *dirinfo)
372 {
373 	struct ceph_inode_info *ci = ceph_inode(inode);
374 	struct ceph_inode_frag *frag;
375 	u32 id = le32_to_cpu(dirinfo->frag);
376 	int mds = le32_to_cpu(dirinfo->auth);
377 	int ndist = le32_to_cpu(dirinfo->ndist);
378 	int diri_auth = -1;
379 	int i;
380 	int err = 0;
381 
382 	spin_lock(&ci->i_ceph_lock);
383 	if (ci->i_auth_cap)
384 		diri_auth = ci->i_auth_cap->mds;
385 	spin_unlock(&ci->i_ceph_lock);
386 
387 	if (mds == -1) /* CDIR_AUTH_PARENT */
388 		mds = diri_auth;
389 
390 	mutex_lock(&ci->i_fragtree_mutex);
391 	if (ndist == 0 && mds == diri_auth) {
392 		/* no delegation info needed. */
393 		frag = __ceph_find_frag(ci, id);
394 		if (!frag)
395 			goto out;
396 		if (frag->split_by == 0) {
397 			/* tree leaf, remove */
398 			dout("fill_dirfrag removed %llx.%llx frag %x"
399 			     " (no ref)\n", ceph_vinop(inode), id);
400 			rb_erase(&frag->node, &ci->i_fragtree);
401 			kfree(frag);
402 		} else {
403 			/* tree branch, keep and clear */
404 			dout("fill_dirfrag cleared %llx.%llx frag %x"
405 			     " referral\n", ceph_vinop(inode), id);
406 			frag->mds = -1;
407 			frag->ndist = 0;
408 		}
409 		goto out;
410 	}
411 
412 
413 	/* find/add this frag to store mds delegation info */
414 	frag = __get_or_create_frag(ci, id);
415 	if (IS_ERR(frag)) {
416 		/* this is not the end of the world; we can continue
417 		   with bad/inaccurate delegation info */
418 		pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
419 		       ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
420 		err = -ENOMEM;
421 		goto out;
422 	}
423 
424 	frag->mds = mds;
425 	frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
426 	for (i = 0; i < frag->ndist; i++)
427 		frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
428 	dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
429 	     ceph_vinop(inode), frag->frag, frag->ndist);
430 
431 out:
432 	mutex_unlock(&ci->i_fragtree_mutex);
433 	return err;
434 }
435 
436 static int frag_tree_split_cmp(const void *l, const void *r)
437 {
438 	struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
439 	struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
440 	return ceph_frag_compare(le32_to_cpu(ls->frag),
441 				 le32_to_cpu(rs->frag));
442 }
443 
444 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
445 {
446 	if (!frag)
447 		return f == ceph_frag_make(0, 0);
448 	if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
449 		return false;
450 	return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
451 }
452 
453 static int ceph_fill_fragtree(struct inode *inode,
454 			      struct ceph_frag_tree_head *fragtree,
455 			      struct ceph_mds_reply_dirfrag *dirinfo)
456 {
457 	struct ceph_inode_info *ci = ceph_inode(inode);
458 	struct ceph_inode_frag *frag, *prev_frag = NULL;
459 	struct rb_node *rb_node;
460 	unsigned i, split_by, nsplits;
461 	u32 id;
462 	bool update = false;
463 
464 	mutex_lock(&ci->i_fragtree_mutex);
465 	nsplits = le32_to_cpu(fragtree->nsplits);
466 	if (nsplits != ci->i_fragtree_nsplits) {
467 		update = true;
468 	} else if (nsplits) {
469 		i = get_random_u32_below(nsplits);
470 		id = le32_to_cpu(fragtree->splits[i].frag);
471 		if (!__ceph_find_frag(ci, id))
472 			update = true;
473 	} else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
474 		rb_node = rb_first(&ci->i_fragtree);
475 		frag = rb_entry(rb_node, struct ceph_inode_frag, node);
476 		if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
477 			update = true;
478 	}
479 	if (!update && dirinfo) {
480 		id = le32_to_cpu(dirinfo->frag);
481 		if (id != __ceph_choose_frag(ci, id, NULL, NULL))
482 			update = true;
483 	}
484 	if (!update)
485 		goto out_unlock;
486 
487 	if (nsplits > 1) {
488 		sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
489 		     frag_tree_split_cmp, NULL);
490 	}
491 
492 	dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
493 	rb_node = rb_first(&ci->i_fragtree);
494 	for (i = 0; i < nsplits; i++) {
495 		id = le32_to_cpu(fragtree->splits[i].frag);
496 		split_by = le32_to_cpu(fragtree->splits[i].by);
497 		if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
498 			pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
499 			       "frag %x split by %d\n", ceph_vinop(inode),
500 			       i, nsplits, id, split_by);
501 			continue;
502 		}
503 		frag = NULL;
504 		while (rb_node) {
505 			frag = rb_entry(rb_node, struct ceph_inode_frag, node);
506 			if (ceph_frag_compare(frag->frag, id) >= 0) {
507 				if (frag->frag != id)
508 					frag = NULL;
509 				else
510 					rb_node = rb_next(rb_node);
511 				break;
512 			}
513 			rb_node = rb_next(rb_node);
514 			/* delete stale split/leaf node */
515 			if (frag->split_by > 0 ||
516 			    !is_frag_child(frag->frag, prev_frag)) {
517 				rb_erase(&frag->node, &ci->i_fragtree);
518 				if (frag->split_by > 0)
519 					ci->i_fragtree_nsplits--;
520 				kfree(frag);
521 			}
522 			frag = NULL;
523 		}
524 		if (!frag) {
525 			frag = __get_or_create_frag(ci, id);
526 			if (IS_ERR(frag))
527 				continue;
528 		}
529 		if (frag->split_by == 0)
530 			ci->i_fragtree_nsplits++;
531 		frag->split_by = split_by;
532 		dout(" frag %x split by %d\n", frag->frag, frag->split_by);
533 		prev_frag = frag;
534 	}
535 	while (rb_node) {
536 		frag = rb_entry(rb_node, struct ceph_inode_frag, node);
537 		rb_node = rb_next(rb_node);
538 		/* delete stale split/leaf node */
539 		if (frag->split_by > 0 ||
540 		    !is_frag_child(frag->frag, prev_frag)) {
541 			rb_erase(&frag->node, &ci->i_fragtree);
542 			if (frag->split_by > 0)
543 				ci->i_fragtree_nsplits--;
544 			kfree(frag);
545 		}
546 	}
547 out_unlock:
548 	mutex_unlock(&ci->i_fragtree_mutex);
549 	return 0;
550 }
551 
552 /*
553  * initialize a newly allocated inode.
554  */
555 struct inode *ceph_alloc_inode(struct super_block *sb)
556 {
557 	struct ceph_inode_info *ci;
558 	int i;
559 
560 	ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS);
561 	if (!ci)
562 		return NULL;
563 
564 	dout("alloc_inode %p\n", &ci->netfs.inode);
565 
566 	/* Set parameters for the netfs library */
567 	netfs_inode_init(&ci->netfs, &ceph_netfs_ops);
568 
569 	spin_lock_init(&ci->i_ceph_lock);
570 
571 	ci->i_version = 0;
572 	ci->i_inline_version = 0;
573 	ci->i_time_warp_seq = 0;
574 	ci->i_ceph_flags = 0;
575 	atomic64_set(&ci->i_ordered_count, 1);
576 	atomic64_set(&ci->i_release_count, 1);
577 	atomic64_set(&ci->i_complete_seq[0], 0);
578 	atomic64_set(&ci->i_complete_seq[1], 0);
579 	ci->i_symlink = NULL;
580 
581 	ci->i_max_bytes = 0;
582 	ci->i_max_files = 0;
583 
584 	memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
585 	memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
586 	RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
587 
588 	ci->i_fragtree = RB_ROOT;
589 	mutex_init(&ci->i_fragtree_mutex);
590 
591 	ci->i_xattrs.blob = NULL;
592 	ci->i_xattrs.prealloc_blob = NULL;
593 	ci->i_xattrs.dirty = false;
594 	ci->i_xattrs.index = RB_ROOT;
595 	ci->i_xattrs.count = 0;
596 	ci->i_xattrs.names_size = 0;
597 	ci->i_xattrs.vals_size = 0;
598 	ci->i_xattrs.version = 0;
599 	ci->i_xattrs.index_version = 0;
600 
601 	ci->i_caps = RB_ROOT;
602 	ci->i_auth_cap = NULL;
603 	ci->i_dirty_caps = 0;
604 	ci->i_flushing_caps = 0;
605 	INIT_LIST_HEAD(&ci->i_dirty_item);
606 	INIT_LIST_HEAD(&ci->i_flushing_item);
607 	ci->i_prealloc_cap_flush = NULL;
608 	INIT_LIST_HEAD(&ci->i_cap_flush_list);
609 	init_waitqueue_head(&ci->i_cap_wq);
610 	ci->i_hold_caps_max = 0;
611 	INIT_LIST_HEAD(&ci->i_cap_delay_list);
612 	INIT_LIST_HEAD(&ci->i_cap_snaps);
613 	ci->i_head_snapc = NULL;
614 	ci->i_snap_caps = 0;
615 
616 	ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
617 	for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
618 		ci->i_nr_by_mode[i] = 0;
619 
620 	mutex_init(&ci->i_truncate_mutex);
621 	ci->i_truncate_seq = 0;
622 	ci->i_truncate_size = 0;
623 	ci->i_truncate_pending = 0;
624 	ci->i_truncate_pagecache_size = 0;
625 
626 	ci->i_max_size = 0;
627 	ci->i_reported_size = 0;
628 	ci->i_wanted_max_size = 0;
629 	ci->i_requested_max_size = 0;
630 
631 	ci->i_pin_ref = 0;
632 	ci->i_rd_ref = 0;
633 	ci->i_rdcache_ref = 0;
634 	ci->i_wr_ref = 0;
635 	ci->i_wb_ref = 0;
636 	ci->i_fx_ref = 0;
637 	ci->i_wrbuffer_ref = 0;
638 	ci->i_wrbuffer_ref_head = 0;
639 	atomic_set(&ci->i_filelock_ref, 0);
640 	atomic_set(&ci->i_shared_gen, 1);
641 	ci->i_rdcache_gen = 0;
642 	ci->i_rdcache_revoking = 0;
643 
644 	INIT_LIST_HEAD(&ci->i_unsafe_dirops);
645 	INIT_LIST_HEAD(&ci->i_unsafe_iops);
646 	spin_lock_init(&ci->i_unsafe_lock);
647 
648 	ci->i_snap_realm = NULL;
649 	INIT_LIST_HEAD(&ci->i_snap_realm_item);
650 	INIT_LIST_HEAD(&ci->i_snap_flush_item);
651 
652 	INIT_WORK(&ci->i_work, ceph_inode_work);
653 	ci->i_work_mask = 0;
654 	memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
655 #ifdef CONFIG_FS_ENCRYPTION
656 	ci->fscrypt_auth = NULL;
657 	ci->fscrypt_auth_len = 0;
658 #endif
659 	return &ci->netfs.inode;
660 }
661 
662 void ceph_free_inode(struct inode *inode)
663 {
664 	struct ceph_inode_info *ci = ceph_inode(inode);
665 
666 	kfree(ci->i_symlink);
667 #ifdef CONFIG_FS_ENCRYPTION
668 	kfree(ci->fscrypt_auth);
669 #endif
670 	fscrypt_free_inode(inode);
671 	kmem_cache_free(ceph_inode_cachep, ci);
672 }
673 
674 void ceph_evict_inode(struct inode *inode)
675 {
676 	struct ceph_inode_info *ci = ceph_inode(inode);
677 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
678 	struct ceph_inode_frag *frag;
679 	struct rb_node *n;
680 
681 	dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
682 
683 	percpu_counter_dec(&mdsc->metric.total_inodes);
684 
685 	truncate_inode_pages_final(&inode->i_data);
686 	if (inode->i_state & I_PINNING_FSCACHE_WB)
687 		ceph_fscache_unuse_cookie(inode, true);
688 	clear_inode(inode);
689 
690 	ceph_fscache_unregister_inode_cookie(ci);
691 	fscrypt_put_encryption_info(inode);
692 
693 	__ceph_remove_caps(ci);
694 
695 	if (__ceph_has_quota(ci, QUOTA_GET_ANY))
696 		ceph_adjust_quota_realms_count(inode, false);
697 
698 	/*
699 	 * we may still have a snap_realm reference if there are stray
700 	 * caps in i_snap_caps.
701 	 */
702 	if (ci->i_snap_realm) {
703 		if (ceph_snap(inode) == CEPH_NOSNAP) {
704 			dout(" dropping residual ref to snap realm %p\n",
705 			     ci->i_snap_realm);
706 			ceph_change_snap_realm(inode, NULL);
707 		} else {
708 			ceph_put_snapid_map(mdsc, ci->i_snapid_map);
709 			ci->i_snap_realm = NULL;
710 		}
711 	}
712 
713 	while ((n = rb_first(&ci->i_fragtree)) != NULL) {
714 		frag = rb_entry(n, struct ceph_inode_frag, node);
715 		rb_erase(n, &ci->i_fragtree);
716 		kfree(frag);
717 	}
718 	ci->i_fragtree_nsplits = 0;
719 
720 	__ceph_destroy_xattrs(ci);
721 	if (ci->i_xattrs.blob)
722 		ceph_buffer_put(ci->i_xattrs.blob);
723 	if (ci->i_xattrs.prealloc_blob)
724 		ceph_buffer_put(ci->i_xattrs.prealloc_blob);
725 
726 	ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
727 	ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
728 }
729 
730 static inline blkcnt_t calc_inode_blocks(u64 size)
731 {
732 	return (size + (1<<9) - 1) >> 9;
733 }
734 
735 /*
736  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
737  * careful because either the client or MDS may have more up to date
738  * info, depending on which capabilities are held, and whether
739  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
740  * and size are monotonically increasing, except when utimes() or
741  * truncate() increments the corresponding _seq values.)
742  */
743 int ceph_fill_file_size(struct inode *inode, int issued,
744 			u32 truncate_seq, u64 truncate_size, u64 size)
745 {
746 	struct ceph_inode_info *ci = ceph_inode(inode);
747 	int queue_trunc = 0;
748 	loff_t isize = i_size_read(inode);
749 
750 	if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
751 	    (truncate_seq == ci->i_truncate_seq && size > isize)) {
752 		dout("size %lld -> %llu\n", isize, size);
753 		if (size > 0 && S_ISDIR(inode->i_mode)) {
754 			pr_err("fill_file_size non-zero size for directory\n");
755 			size = 0;
756 		}
757 		i_size_write(inode, size);
758 		inode->i_blocks = calc_inode_blocks(size);
759 		/*
760 		 * If we're expanding, then we should be able to just update
761 		 * the existing cookie.
762 		 */
763 		if (size > isize)
764 			ceph_fscache_update(inode);
765 		ci->i_reported_size = size;
766 		if (truncate_seq != ci->i_truncate_seq) {
767 			dout("%s truncate_seq %u -> %u\n", __func__,
768 			     ci->i_truncate_seq, truncate_seq);
769 			ci->i_truncate_seq = truncate_seq;
770 
771 			/* the MDS should have revoked these caps */
772 			WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
773 					       CEPH_CAP_FILE_RD |
774 					       CEPH_CAP_FILE_WR |
775 					       CEPH_CAP_FILE_LAZYIO));
776 			/*
777 			 * If we hold relevant caps, or in the case where we're
778 			 * not the only client referencing this file and we
779 			 * don't hold those caps, then we need to check whether
780 			 * the file is either opened or mmaped
781 			 */
782 			if ((issued & (CEPH_CAP_FILE_CACHE|
783 				       CEPH_CAP_FILE_BUFFER)) ||
784 			    mapping_mapped(inode->i_mapping) ||
785 			    __ceph_is_file_opened(ci)) {
786 				ci->i_truncate_pending++;
787 				queue_trunc = 1;
788 			}
789 		}
790 	}
791 
792 	/*
793 	 * It's possible that the new sizes of the two consecutive
794 	 * size truncations will be in the same fscrypt last block,
795 	 * and we need to truncate the corresponding page caches
796 	 * anyway.
797 	 */
798 	if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0) {
799 		dout("%s truncate_size %lld -> %llu, encrypted %d\n", __func__,
800 		     ci->i_truncate_size, truncate_size, !!IS_ENCRYPTED(inode));
801 
802 		ci->i_truncate_size = truncate_size;
803 
804 		if (IS_ENCRYPTED(inode)) {
805 			dout("%s truncate_pagecache_size %lld -> %llu\n",
806 			     __func__, ci->i_truncate_pagecache_size, size);
807 			ci->i_truncate_pagecache_size = size;
808 		} else {
809 			ci->i_truncate_pagecache_size = truncate_size;
810 		}
811 	}
812 	return queue_trunc;
813 }
814 
815 void ceph_fill_file_time(struct inode *inode, int issued,
816 			 u64 time_warp_seq, struct timespec64 *ctime,
817 			 struct timespec64 *mtime, struct timespec64 *atime)
818 {
819 	struct ceph_inode_info *ci = ceph_inode(inode);
820 	int warn = 0;
821 
822 	if (issued & (CEPH_CAP_FILE_EXCL|
823 		      CEPH_CAP_FILE_WR|
824 		      CEPH_CAP_FILE_BUFFER|
825 		      CEPH_CAP_AUTH_EXCL|
826 		      CEPH_CAP_XATTR_EXCL)) {
827 		if (ci->i_version == 0 ||
828 		    timespec64_compare(ctime, &inode->i_ctime) > 0) {
829 			dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
830 			     inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
831 			     ctime->tv_sec, ctime->tv_nsec);
832 			inode->i_ctime = *ctime;
833 		}
834 		if (ci->i_version == 0 ||
835 		    ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
836 			/* the MDS did a utimes() */
837 			dout("mtime %lld.%09ld -> %lld.%09ld "
838 			     "tw %d -> %d\n",
839 			     inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
840 			     mtime->tv_sec, mtime->tv_nsec,
841 			     ci->i_time_warp_seq, (int)time_warp_seq);
842 
843 			inode->i_mtime = *mtime;
844 			inode->i_atime = *atime;
845 			ci->i_time_warp_seq = time_warp_seq;
846 		} else if (time_warp_seq == ci->i_time_warp_seq) {
847 			/* nobody did utimes(); take the max */
848 			if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
849 				dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
850 				     inode->i_mtime.tv_sec,
851 				     inode->i_mtime.tv_nsec,
852 				     mtime->tv_sec, mtime->tv_nsec);
853 				inode->i_mtime = *mtime;
854 			}
855 			if (timespec64_compare(atime, &inode->i_atime) > 0) {
856 				dout("atime %lld.%09ld -> %lld.%09ld inc\n",
857 				     inode->i_atime.tv_sec,
858 				     inode->i_atime.tv_nsec,
859 				     atime->tv_sec, atime->tv_nsec);
860 				inode->i_atime = *atime;
861 			}
862 		} else if (issued & CEPH_CAP_FILE_EXCL) {
863 			/* we did a utimes(); ignore mds values */
864 		} else {
865 			warn = 1;
866 		}
867 	} else {
868 		/* we have no write|excl caps; whatever the MDS says is true */
869 		if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
870 			inode->i_ctime = *ctime;
871 			inode->i_mtime = *mtime;
872 			inode->i_atime = *atime;
873 			ci->i_time_warp_seq = time_warp_seq;
874 		} else {
875 			warn = 1;
876 		}
877 	}
878 	if (warn) /* time_warp_seq shouldn't go backwards */
879 		dout("%p mds time_warp_seq %llu < %u\n",
880 		     inode, time_warp_seq, ci->i_time_warp_seq);
881 }
882 
883 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
884 static int decode_encrypted_symlink(const char *encsym, int enclen, u8 **decsym)
885 {
886 	int declen;
887 	u8 *sym;
888 
889 	sym = kmalloc(enclen + 1, GFP_NOFS);
890 	if (!sym)
891 		return -ENOMEM;
892 
893 	declen = ceph_base64_decode(encsym, enclen, sym);
894 	if (declen < 0) {
895 		pr_err("%s: can't decode symlink (%d). Content: %.*s\n",
896 		       __func__, declen, enclen, encsym);
897 		kfree(sym);
898 		return -EIO;
899 	}
900 	sym[declen + 1] = '\0';
901 	*decsym = sym;
902 	return declen;
903 }
904 #else
905 static int decode_encrypted_symlink(const char *encsym, int symlen, u8 **decsym)
906 {
907 	return -EOPNOTSUPP;
908 }
909 #endif
910 
911 /*
912  * Populate an inode based on info from mds.  May be called on new or
913  * existing inodes.
914  */
915 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
916 		    struct ceph_mds_reply_info_in *iinfo,
917 		    struct ceph_mds_reply_dirfrag *dirinfo,
918 		    struct ceph_mds_session *session, int cap_fmode,
919 		    struct ceph_cap_reservation *caps_reservation)
920 {
921 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
922 	struct ceph_mds_reply_inode *info = iinfo->in;
923 	struct ceph_inode_info *ci = ceph_inode(inode);
924 	int issued, new_issued, info_caps;
925 	struct timespec64 mtime, atime, ctime;
926 	struct ceph_buffer *xattr_blob = NULL;
927 	struct ceph_buffer *old_blob = NULL;
928 	struct ceph_string *pool_ns = NULL;
929 	struct ceph_cap *new_cap = NULL;
930 	int err = 0;
931 	bool wake = false;
932 	bool queue_trunc = false;
933 	bool new_version = false;
934 	bool fill_inline = false;
935 	umode_t mode = le32_to_cpu(info->mode);
936 	dev_t rdev = le32_to_cpu(info->rdev);
937 
938 	lockdep_assert_held(&mdsc->snap_rwsem);
939 
940 	dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
941 	     inode, ceph_vinop(inode), le64_to_cpu(info->version),
942 	     ci->i_version);
943 
944 	/* Once I_NEW is cleared, we can't change type or dev numbers */
945 	if (inode->i_state & I_NEW) {
946 		inode->i_mode = mode;
947 	} else {
948 		if (inode_wrong_type(inode, mode)) {
949 			pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
950 				     ceph_vinop(inode), inode->i_mode, mode);
951 			return -ESTALE;
952 		}
953 
954 		if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
955 			pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
956 				     ceph_vinop(inode), MAJOR(inode->i_rdev),
957 				     MINOR(inode->i_rdev), MAJOR(rdev),
958 				     MINOR(rdev));
959 			return -ESTALE;
960 		}
961 	}
962 
963 	info_caps = le32_to_cpu(info->cap.caps);
964 
965 	/* prealloc new cap struct */
966 	if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
967 		new_cap = ceph_get_cap(mdsc, caps_reservation);
968 		if (!new_cap)
969 			return -ENOMEM;
970 	}
971 
972 	/*
973 	 * prealloc xattr data, if it looks like we'll need it.  only
974 	 * if len > 4 (meaning there are actually xattrs; the first 4
975 	 * bytes are the xattr count).
976 	 */
977 	if (iinfo->xattr_len > 4) {
978 		xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
979 		if (!xattr_blob)
980 			pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
981 			       iinfo->xattr_len);
982 	}
983 
984 	if (iinfo->pool_ns_len > 0)
985 		pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
986 						     iinfo->pool_ns_len);
987 
988 	if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
989 		ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
990 
991 	spin_lock(&ci->i_ceph_lock);
992 
993 	/*
994 	 * provided version will be odd if inode value is projected,
995 	 * even if stable.  skip the update if we have newer stable
996 	 * info (ours>=theirs, e.g. due to racing mds replies), unless
997 	 * we are getting projected (unstable) info (in which case the
998 	 * version is odd, and we want ours>theirs).
999 	 *   us   them
1000 	 *   2    2     skip
1001 	 *   3    2     skip
1002 	 *   3    3     update
1003 	 */
1004 	if (ci->i_version == 0 ||
1005 	    ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1006 	     le64_to_cpu(info->version) > (ci->i_version & ~1)))
1007 		new_version = true;
1008 
1009 	/* Update change_attribute */
1010 	inode_set_max_iversion_raw(inode, iinfo->change_attr);
1011 
1012 	__ceph_caps_issued(ci, &issued);
1013 	issued |= __ceph_caps_dirty(ci);
1014 	new_issued = ~issued & info_caps;
1015 
1016 	__ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
1017 
1018 #ifdef CONFIG_FS_ENCRYPTION
1019 	if (iinfo->fscrypt_auth_len &&
1020 	    ((inode->i_state & I_NEW) || (ci->fscrypt_auth_len == 0))) {
1021 		kfree(ci->fscrypt_auth);
1022 		ci->fscrypt_auth_len = iinfo->fscrypt_auth_len;
1023 		ci->fscrypt_auth = iinfo->fscrypt_auth;
1024 		iinfo->fscrypt_auth = NULL;
1025 		iinfo->fscrypt_auth_len = 0;
1026 		inode_set_flags(inode, S_ENCRYPTED, S_ENCRYPTED);
1027 	}
1028 #endif
1029 
1030 	if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
1031 	    (issued & CEPH_CAP_AUTH_EXCL) == 0) {
1032 		inode->i_mode = mode;
1033 		inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
1034 		inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
1035 		dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
1036 		     from_kuid(&init_user_ns, inode->i_uid),
1037 		     from_kgid(&init_user_ns, inode->i_gid));
1038 		ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
1039 		ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
1040 	}
1041 
1042 	/* directories have fl_stripe_unit set to zero */
1043 	if (IS_ENCRYPTED(inode))
1044 		inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
1045 	else if (le32_to_cpu(info->layout.fl_stripe_unit))
1046 		inode->i_blkbits =
1047 			fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
1048 	else
1049 		inode->i_blkbits = CEPH_BLOCK_SHIFT;
1050 
1051 	if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
1052 	    (issued & CEPH_CAP_LINK_EXCL) == 0)
1053 		set_nlink(inode, le32_to_cpu(info->nlink));
1054 
1055 	if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
1056 		/* be careful with mtime, atime, size */
1057 		ceph_decode_timespec64(&atime, &info->atime);
1058 		ceph_decode_timespec64(&mtime, &info->mtime);
1059 		ceph_decode_timespec64(&ctime, &info->ctime);
1060 		ceph_fill_file_time(inode, issued,
1061 				le32_to_cpu(info->time_warp_seq),
1062 				&ctime, &mtime, &atime);
1063 	}
1064 
1065 	if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
1066 		ci->i_files = le64_to_cpu(info->files);
1067 		ci->i_subdirs = le64_to_cpu(info->subdirs);
1068 	}
1069 
1070 	if (new_version ||
1071 	    (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
1072 		u64 size = le64_to_cpu(info->size);
1073 		s64 old_pool = ci->i_layout.pool_id;
1074 		struct ceph_string *old_ns;
1075 
1076 		ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
1077 		old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
1078 					lockdep_is_held(&ci->i_ceph_lock));
1079 		rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
1080 
1081 		if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
1082 			ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
1083 
1084 		pool_ns = old_ns;
1085 
1086 		if (IS_ENCRYPTED(inode) && size &&
1087 		    iinfo->fscrypt_file_len == sizeof(__le64)) {
1088 			u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
1089 
1090 			if (size == round_up(fsize, CEPH_FSCRYPT_BLOCK_SIZE)) {
1091 				size = fsize;
1092 			} else {
1093 				pr_warn("fscrypt size mismatch: size=%llu fscrypt_file=%llu, discarding fscrypt_file size.\n",
1094 					info->size, size);
1095 			}
1096 		}
1097 
1098 		queue_trunc = ceph_fill_file_size(inode, issued,
1099 					le32_to_cpu(info->truncate_seq),
1100 					le64_to_cpu(info->truncate_size),
1101 					size);
1102 		/* only update max_size on auth cap */
1103 		if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1104 		    ci->i_max_size != le64_to_cpu(info->max_size)) {
1105 			dout("max_size %lld -> %llu\n", ci->i_max_size,
1106 					le64_to_cpu(info->max_size));
1107 			ci->i_max_size = le64_to_cpu(info->max_size);
1108 		}
1109 	}
1110 
1111 	/* layout and rstat are not tracked by capability, update them if
1112 	 * the inode info is from auth mds */
1113 	if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
1114 		if (S_ISDIR(inode->i_mode)) {
1115 			ci->i_dir_layout = iinfo->dir_layout;
1116 			ci->i_rbytes = le64_to_cpu(info->rbytes);
1117 			ci->i_rfiles = le64_to_cpu(info->rfiles);
1118 			ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
1119 			ci->i_dir_pin = iinfo->dir_pin;
1120 			ci->i_rsnaps = iinfo->rsnaps;
1121 			ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
1122 		}
1123 	}
1124 
1125 	/* xattrs */
1126 	/* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
1127 	if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
1128 	    le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
1129 		if (ci->i_xattrs.blob)
1130 			old_blob = ci->i_xattrs.blob;
1131 		ci->i_xattrs.blob = xattr_blob;
1132 		if (xattr_blob)
1133 			memcpy(ci->i_xattrs.blob->vec.iov_base,
1134 			       iinfo->xattr_data, iinfo->xattr_len);
1135 		ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
1136 		ceph_forget_all_cached_acls(inode);
1137 		ceph_security_invalidate_secctx(inode);
1138 		xattr_blob = NULL;
1139 	}
1140 
1141 	/* finally update i_version */
1142 	if (le64_to_cpu(info->version) > ci->i_version)
1143 		ci->i_version = le64_to_cpu(info->version);
1144 
1145 	inode->i_mapping->a_ops = &ceph_aops;
1146 
1147 	switch (inode->i_mode & S_IFMT) {
1148 	case S_IFIFO:
1149 	case S_IFBLK:
1150 	case S_IFCHR:
1151 	case S_IFSOCK:
1152 		inode->i_blkbits = PAGE_SHIFT;
1153 		init_special_inode(inode, inode->i_mode, rdev);
1154 		inode->i_op = &ceph_file_iops;
1155 		break;
1156 	case S_IFREG:
1157 		inode->i_op = &ceph_file_iops;
1158 		inode->i_fop = &ceph_file_fops;
1159 		break;
1160 	case S_IFLNK:
1161 		if (!ci->i_symlink) {
1162 			u32 symlen = iinfo->symlink_len;
1163 			char *sym;
1164 
1165 			spin_unlock(&ci->i_ceph_lock);
1166 
1167 			if (IS_ENCRYPTED(inode)) {
1168 				if (symlen != i_size_read(inode))
1169 					pr_err("%s %llx.%llx BAD symlink size %lld\n",
1170 						__func__, ceph_vinop(inode),
1171 						i_size_read(inode));
1172 
1173 				err = decode_encrypted_symlink(iinfo->symlink,
1174 							       symlen, (u8 **)&sym);
1175 				if (err < 0) {
1176 					pr_err("%s decoding encrypted symlink failed: %d\n",
1177 						__func__, err);
1178 					goto out;
1179 				}
1180 				symlen = err;
1181 				i_size_write(inode, symlen);
1182 				inode->i_blocks = calc_inode_blocks(symlen);
1183 			} else {
1184 				if (symlen != i_size_read(inode)) {
1185 					pr_err("%s %llx.%llx BAD symlink size %lld\n",
1186 						__func__, ceph_vinop(inode),
1187 						i_size_read(inode));
1188 					i_size_write(inode, symlen);
1189 					inode->i_blocks = calc_inode_blocks(symlen);
1190 				}
1191 
1192 				err = -ENOMEM;
1193 				sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
1194 				if (!sym)
1195 					goto out;
1196 			}
1197 
1198 			spin_lock(&ci->i_ceph_lock);
1199 			if (!ci->i_symlink)
1200 				ci->i_symlink = sym;
1201 			else
1202 				kfree(sym); /* lost a race */
1203 		}
1204 
1205 		if (IS_ENCRYPTED(inode)) {
1206 			/*
1207 			 * Encrypted symlinks need to be decrypted before we can
1208 			 * cache their targets in i_link. Don't touch it here.
1209 			 */
1210 			inode->i_op = &ceph_encrypted_symlink_iops;
1211 		} else {
1212 			inode->i_link = ci->i_symlink;
1213 			inode->i_op = &ceph_symlink_iops;
1214 		}
1215 		break;
1216 	case S_IFDIR:
1217 		inode->i_op = &ceph_dir_iops;
1218 		inode->i_fop = &ceph_dir_fops;
1219 		break;
1220 	default:
1221 		pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1222 		       ceph_vinop(inode), inode->i_mode);
1223 	}
1224 
1225 	/* were we issued a capability? */
1226 	if (info_caps) {
1227 		if (ceph_snap(inode) == CEPH_NOSNAP) {
1228 			ceph_add_cap(inode, session,
1229 				     le64_to_cpu(info->cap.cap_id),
1230 				     info_caps,
1231 				     le32_to_cpu(info->cap.wanted),
1232 				     le32_to_cpu(info->cap.seq),
1233 				     le32_to_cpu(info->cap.mseq),
1234 				     le64_to_cpu(info->cap.realm),
1235 				     info->cap.flags, &new_cap);
1236 
1237 			/* set dir completion flag? */
1238 			if (S_ISDIR(inode->i_mode) &&
1239 			    ci->i_files == 0 && ci->i_subdirs == 0 &&
1240 			    (info_caps & CEPH_CAP_FILE_SHARED) &&
1241 			    (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1242 			    !__ceph_dir_is_complete(ci)) {
1243 				dout(" marking %p complete (empty)\n", inode);
1244 				i_size_write(inode, 0);
1245 				__ceph_dir_set_complete(ci,
1246 					atomic64_read(&ci->i_release_count),
1247 					atomic64_read(&ci->i_ordered_count));
1248 			}
1249 
1250 			wake = true;
1251 		} else {
1252 			dout(" %p got snap_caps %s\n", inode,
1253 			     ceph_cap_string(info_caps));
1254 			ci->i_snap_caps |= info_caps;
1255 		}
1256 	}
1257 
1258 	if (iinfo->inline_version > 0 &&
1259 	    iinfo->inline_version >= ci->i_inline_version) {
1260 		int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1261 		ci->i_inline_version = iinfo->inline_version;
1262 		if (ceph_has_inline_data(ci) &&
1263 		    (locked_page || (info_caps & cache_caps)))
1264 			fill_inline = true;
1265 	}
1266 
1267 	if (cap_fmode >= 0) {
1268 		if (!info_caps)
1269 			pr_warn("mds issued no caps on %llx.%llx\n",
1270 				ceph_vinop(inode));
1271 		__ceph_touch_fmode(ci, mdsc, cap_fmode);
1272 	}
1273 
1274 	spin_unlock(&ci->i_ceph_lock);
1275 
1276 	ceph_fscache_register_inode_cookie(inode);
1277 
1278 	if (fill_inline)
1279 		ceph_fill_inline_data(inode, locked_page,
1280 				      iinfo->inline_data, iinfo->inline_len);
1281 
1282 	if (wake)
1283 		wake_up_all(&ci->i_cap_wq);
1284 
1285 	/* queue truncate if we saw i_size decrease */
1286 	if (queue_trunc)
1287 		ceph_queue_vmtruncate(inode);
1288 
1289 	/* populate frag tree */
1290 	if (S_ISDIR(inode->i_mode))
1291 		ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1292 
1293 	/* update delegation info? */
1294 	if (dirinfo)
1295 		ceph_fill_dirfrag(inode, dirinfo);
1296 
1297 	err = 0;
1298 out:
1299 	if (new_cap)
1300 		ceph_put_cap(mdsc, new_cap);
1301 	ceph_buffer_put(old_blob);
1302 	ceph_buffer_put(xattr_blob);
1303 	ceph_put_string(pool_ns);
1304 	return err;
1305 }
1306 
1307 /*
1308  * caller should hold session s_mutex and dentry->d_lock.
1309  */
1310 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1311 				  struct ceph_mds_reply_lease *lease,
1312 				  struct ceph_mds_session *session,
1313 				  unsigned long from_time,
1314 				  struct ceph_mds_session **old_lease_session)
1315 {
1316 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1317 	unsigned mask = le16_to_cpu(lease->mask);
1318 	long unsigned duration = le32_to_cpu(lease->duration_ms);
1319 	long unsigned ttl = from_time + (duration * HZ) / 1000;
1320 	long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1321 
1322 	dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1323 	     dentry, duration, ttl);
1324 
1325 	/* only track leases on regular dentries */
1326 	if (ceph_snap(dir) != CEPH_NOSNAP)
1327 		return;
1328 
1329 	if (mask & CEPH_LEASE_PRIMARY_LINK)
1330 		di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1331 	else
1332 		di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1333 
1334 	di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1335 	if (!(mask & CEPH_LEASE_VALID)) {
1336 		__ceph_dentry_dir_lease_touch(di);
1337 		return;
1338 	}
1339 
1340 	if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1341 	    time_before(ttl, di->time))
1342 		return;  /* we already have a newer lease. */
1343 
1344 	if (di->lease_session && di->lease_session != session) {
1345 		*old_lease_session = di->lease_session;
1346 		di->lease_session = NULL;
1347 	}
1348 
1349 	if (!di->lease_session)
1350 		di->lease_session = ceph_get_mds_session(session);
1351 	di->lease_gen = atomic_read(&session->s_cap_gen);
1352 	di->lease_seq = le32_to_cpu(lease->seq);
1353 	di->lease_renew_after = half_ttl;
1354 	di->lease_renew_from = 0;
1355 	di->time = ttl;
1356 
1357 	__ceph_dentry_lease_touch(di);
1358 }
1359 
1360 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1361 					struct ceph_mds_reply_lease *lease,
1362 					struct ceph_mds_session *session,
1363 					unsigned long from_time)
1364 {
1365 	struct ceph_mds_session *old_lease_session = NULL;
1366 	spin_lock(&dentry->d_lock);
1367 	__update_dentry_lease(dir, dentry, lease, session, from_time,
1368 			      &old_lease_session);
1369 	spin_unlock(&dentry->d_lock);
1370 	ceph_put_mds_session(old_lease_session);
1371 }
1372 
1373 /*
1374  * update dentry lease without having parent inode locked
1375  */
1376 static void update_dentry_lease_careful(struct dentry *dentry,
1377 					struct ceph_mds_reply_lease *lease,
1378 					struct ceph_mds_session *session,
1379 					unsigned long from_time,
1380 					char *dname, u32 dname_len,
1381 					struct ceph_vino *pdvino,
1382 					struct ceph_vino *ptvino)
1383 
1384 {
1385 	struct inode *dir;
1386 	struct ceph_mds_session *old_lease_session = NULL;
1387 
1388 	spin_lock(&dentry->d_lock);
1389 	/* make sure dentry's name matches target */
1390 	if (dentry->d_name.len != dname_len ||
1391 	    memcmp(dentry->d_name.name, dname, dname_len))
1392 		goto out_unlock;
1393 
1394 	dir = d_inode(dentry->d_parent);
1395 	/* make sure parent matches dvino */
1396 	if (!ceph_ino_compare(dir, pdvino))
1397 		goto out_unlock;
1398 
1399 	/* make sure dentry's inode matches target. NULL ptvino means that
1400 	 * we expect a negative dentry */
1401 	if (ptvino) {
1402 		if (d_really_is_negative(dentry))
1403 			goto out_unlock;
1404 		if (!ceph_ino_compare(d_inode(dentry), ptvino))
1405 			goto out_unlock;
1406 	} else {
1407 		if (d_really_is_positive(dentry))
1408 			goto out_unlock;
1409 	}
1410 
1411 	__update_dentry_lease(dir, dentry, lease, session,
1412 			      from_time, &old_lease_session);
1413 out_unlock:
1414 	spin_unlock(&dentry->d_lock);
1415 	ceph_put_mds_session(old_lease_session);
1416 }
1417 
1418 /*
1419  * splice a dentry to an inode.
1420  * caller must hold directory i_rwsem for this to be safe.
1421  */
1422 static int splice_dentry(struct dentry **pdn, struct inode *in)
1423 {
1424 	struct dentry *dn = *pdn;
1425 	struct dentry *realdn;
1426 
1427 	BUG_ON(d_inode(dn));
1428 
1429 	if (S_ISDIR(in->i_mode)) {
1430 		/* If inode is directory, d_splice_alias() below will remove
1431 		 * 'realdn' from its origin parent. We need to ensure that
1432 		 * origin parent's readdir cache will not reference 'realdn'
1433 		 */
1434 		realdn = d_find_any_alias(in);
1435 		if (realdn) {
1436 			struct ceph_dentry_info *di = ceph_dentry(realdn);
1437 			spin_lock(&realdn->d_lock);
1438 
1439 			realdn->d_op->d_prune(realdn);
1440 
1441 			di->time = jiffies;
1442 			di->lease_shared_gen = 0;
1443 			di->offset = 0;
1444 
1445 			spin_unlock(&realdn->d_lock);
1446 			dput(realdn);
1447 		}
1448 	}
1449 
1450 	/* dn must be unhashed */
1451 	if (!d_unhashed(dn))
1452 		d_drop(dn);
1453 	realdn = d_splice_alias(in, dn);
1454 	if (IS_ERR(realdn)) {
1455 		pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1456 		       PTR_ERR(realdn), dn, in, ceph_vinop(in));
1457 		return PTR_ERR(realdn);
1458 	}
1459 
1460 	if (realdn) {
1461 		dout("dn %p (%d) spliced with %p (%d) "
1462 		     "inode %p ino %llx.%llx\n",
1463 		     dn, d_count(dn),
1464 		     realdn, d_count(realdn),
1465 		     d_inode(realdn), ceph_vinop(d_inode(realdn)));
1466 		dput(dn);
1467 		*pdn = realdn;
1468 	} else {
1469 		BUG_ON(!ceph_dentry(dn));
1470 		dout("dn %p attached to %p ino %llx.%llx\n",
1471 		     dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1472 	}
1473 	return 0;
1474 }
1475 
1476 /*
1477  * Incorporate results into the local cache.  This is either just
1478  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1479  * after a lookup).
1480  *
1481  * A reply may contain
1482  *         a directory inode along with a dentry.
1483  *  and/or a target inode
1484  *
1485  * Called with snap_rwsem (read).
1486  */
1487 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1488 {
1489 	struct ceph_mds_session *session = req->r_session;
1490 	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1491 	struct inode *in = NULL;
1492 	struct ceph_vino tvino, dvino;
1493 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1494 	int err = 0;
1495 
1496 	dout("fill_trace %p is_dentry %d is_target %d\n", req,
1497 	     rinfo->head->is_dentry, rinfo->head->is_target);
1498 
1499 	if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1500 		dout("fill_trace reply is empty!\n");
1501 		if (rinfo->head->result == 0 && req->r_parent)
1502 			ceph_invalidate_dir_request(req);
1503 		return 0;
1504 	}
1505 
1506 	if (rinfo->head->is_dentry) {
1507 		struct inode *dir = req->r_parent;
1508 
1509 		if (dir) {
1510 			err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1511 					      rinfo->dirfrag, session, -1,
1512 					      &req->r_caps_reservation);
1513 			if (err < 0)
1514 				goto done;
1515 		} else {
1516 			WARN_ON_ONCE(1);
1517 		}
1518 
1519 		if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1520 		    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1521 		    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1522 			bool is_nokey = false;
1523 			struct qstr dname;
1524 			struct dentry *dn, *parent;
1525 			struct fscrypt_str oname = FSTR_INIT(NULL, 0);
1526 			struct ceph_fname fname = { .dir	= dir,
1527 						    .name	= rinfo->dname,
1528 						    .ctext	= rinfo->altname,
1529 						    .name_len	= rinfo->dname_len,
1530 						    .ctext_len	= rinfo->altname_len };
1531 
1532 			BUG_ON(!rinfo->head->is_target);
1533 			BUG_ON(req->r_dentry);
1534 
1535 			parent = d_find_any_alias(dir);
1536 			BUG_ON(!parent);
1537 
1538 			err = ceph_fname_alloc_buffer(dir, &oname);
1539 			if (err < 0) {
1540 				dput(parent);
1541 				goto done;
1542 			}
1543 
1544 			err = ceph_fname_to_usr(&fname, NULL, &oname, &is_nokey);
1545 			if (err < 0) {
1546 				dput(parent);
1547 				ceph_fname_free_buffer(dir, &oname);
1548 				goto done;
1549 			}
1550 			dname.name = oname.name;
1551 			dname.len = oname.len;
1552 			dname.hash = full_name_hash(parent, dname.name, dname.len);
1553 			tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1554 			tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1555 retry_lookup:
1556 			dn = d_lookup(parent, &dname);
1557 			dout("d_lookup on parent=%p name=%.*s got %p\n",
1558 			     parent, dname.len, dname.name, dn);
1559 
1560 			if (!dn) {
1561 				dn = d_alloc(parent, &dname);
1562 				dout("d_alloc %p '%.*s' = %p\n", parent,
1563 				     dname.len, dname.name, dn);
1564 				if (!dn) {
1565 					dput(parent);
1566 					ceph_fname_free_buffer(dir, &oname);
1567 					err = -ENOMEM;
1568 					goto done;
1569 				}
1570 				if (is_nokey) {
1571 					spin_lock(&dn->d_lock);
1572 					dn->d_flags |= DCACHE_NOKEY_NAME;
1573 					spin_unlock(&dn->d_lock);
1574 				}
1575 				err = 0;
1576 			} else if (d_really_is_positive(dn) &&
1577 				   (ceph_ino(d_inode(dn)) != tvino.ino ||
1578 				    ceph_snap(d_inode(dn)) != tvino.snap)) {
1579 				dout(" dn %p points to wrong inode %p\n",
1580 				     dn, d_inode(dn));
1581 				ceph_dir_clear_ordered(dir);
1582 				d_delete(dn);
1583 				dput(dn);
1584 				goto retry_lookup;
1585 			}
1586 			ceph_fname_free_buffer(dir, &oname);
1587 
1588 			req->r_dentry = dn;
1589 			dput(parent);
1590 		}
1591 	}
1592 
1593 	if (rinfo->head->is_target) {
1594 		/* Should be filled in by handle_reply */
1595 		BUG_ON(!req->r_target_inode);
1596 
1597 		in = req->r_target_inode;
1598 		err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1599 				NULL, session,
1600 				(!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1601 				 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1602 				 rinfo->head->result == 0) ?  req->r_fmode : -1,
1603 				&req->r_caps_reservation);
1604 		if (err < 0) {
1605 			pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1606 				in, ceph_vinop(in));
1607 			req->r_target_inode = NULL;
1608 			if (in->i_state & I_NEW)
1609 				discard_new_inode(in);
1610 			else
1611 				iput(in);
1612 			goto done;
1613 		}
1614 		if (in->i_state & I_NEW)
1615 			unlock_new_inode(in);
1616 	}
1617 
1618 	/*
1619 	 * ignore null lease/binding on snapdir ENOENT, or else we
1620 	 * will have trouble splicing in the virtual snapdir later
1621 	 */
1622 	if (rinfo->head->is_dentry &&
1623             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1624 	    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1625 	    (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1626 					       fsc->mount_options->snapdir_name,
1627 					       req->r_dentry->d_name.len))) {
1628 		/*
1629 		 * lookup link rename   : null -> possibly existing inode
1630 		 * mknod symlink mkdir  : null -> new inode
1631 		 * unlink               : linked -> null
1632 		 */
1633 		struct inode *dir = req->r_parent;
1634 		struct dentry *dn = req->r_dentry;
1635 		bool have_dir_cap, have_lease;
1636 
1637 		BUG_ON(!dn);
1638 		BUG_ON(!dir);
1639 		BUG_ON(d_inode(dn->d_parent) != dir);
1640 
1641 		dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1642 		dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1643 
1644 		BUG_ON(ceph_ino(dir) != dvino.ino);
1645 		BUG_ON(ceph_snap(dir) != dvino.snap);
1646 
1647 		/* do we have a lease on the whole dir? */
1648 		have_dir_cap =
1649 			(le32_to_cpu(rinfo->diri.in->cap.caps) &
1650 			 CEPH_CAP_FILE_SHARED);
1651 
1652 		/* do we have a dn lease? */
1653 		have_lease = have_dir_cap ||
1654 			le32_to_cpu(rinfo->dlease->duration_ms);
1655 		if (!have_lease)
1656 			dout("fill_trace  no dentry lease or dir cap\n");
1657 
1658 		/* rename? */
1659 		if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1660 			struct inode *olddir = req->r_old_dentry_dir;
1661 			BUG_ON(!olddir);
1662 
1663 			dout(" src %p '%pd' dst %p '%pd'\n",
1664 			     req->r_old_dentry,
1665 			     req->r_old_dentry,
1666 			     dn, dn);
1667 			dout("fill_trace doing d_move %p -> %p\n",
1668 			     req->r_old_dentry, dn);
1669 
1670 			/* d_move screws up sibling dentries' offsets */
1671 			ceph_dir_clear_ordered(dir);
1672 			ceph_dir_clear_ordered(olddir);
1673 
1674 			d_move(req->r_old_dentry, dn);
1675 			dout(" src %p '%pd' dst %p '%pd'\n",
1676 			     req->r_old_dentry,
1677 			     req->r_old_dentry,
1678 			     dn, dn);
1679 
1680 			/* ensure target dentry is invalidated, despite
1681 			   rehashing bug in vfs_rename_dir */
1682 			ceph_invalidate_dentry_lease(dn);
1683 
1684 			dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1685 			     ceph_dentry(req->r_old_dentry)->offset);
1686 
1687 			/* swap r_dentry and r_old_dentry in case that
1688 			 * splice_dentry() gets called later. This is safe
1689 			 * because no other place will use them */
1690 			req->r_dentry = req->r_old_dentry;
1691 			req->r_old_dentry = dn;
1692 			dn = req->r_dentry;
1693 		}
1694 
1695 		/* null dentry? */
1696 		if (!rinfo->head->is_target) {
1697 			dout("fill_trace null dentry\n");
1698 			if (d_really_is_positive(dn)) {
1699 				dout("d_delete %p\n", dn);
1700 				ceph_dir_clear_ordered(dir);
1701 				d_delete(dn);
1702 			} else if (have_lease) {
1703 				if (d_unhashed(dn))
1704 					d_add(dn, NULL);
1705 			}
1706 
1707 			if (!d_unhashed(dn) && have_lease)
1708 				update_dentry_lease(dir, dn,
1709 						    rinfo->dlease, session,
1710 						    req->r_request_started);
1711 			goto done;
1712 		}
1713 
1714 		/* attach proper inode */
1715 		if (d_really_is_negative(dn)) {
1716 			ceph_dir_clear_ordered(dir);
1717 			ihold(in);
1718 			err = splice_dentry(&req->r_dentry, in);
1719 			if (err < 0)
1720 				goto done;
1721 			dn = req->r_dentry;  /* may have spliced */
1722 		} else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1723 			dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1724 			     dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1725 			     ceph_vinop(in));
1726 			d_invalidate(dn);
1727 			have_lease = false;
1728 		}
1729 
1730 		if (have_lease) {
1731 			update_dentry_lease(dir, dn,
1732 					    rinfo->dlease, session,
1733 					    req->r_request_started);
1734 		}
1735 		dout(" final dn %p\n", dn);
1736 	} else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1737 		    req->r_op == CEPH_MDS_OP_MKSNAP) &&
1738 	           test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1739 		   !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1740 		struct inode *dir = req->r_parent;
1741 
1742 		/* fill out a snapdir LOOKUPSNAP dentry */
1743 		BUG_ON(!dir);
1744 		BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1745 		BUG_ON(!req->r_dentry);
1746 		dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1747 		ceph_dir_clear_ordered(dir);
1748 		ihold(in);
1749 		err = splice_dentry(&req->r_dentry, in);
1750 		if (err < 0)
1751 			goto done;
1752 	} else if (rinfo->head->is_dentry && req->r_dentry) {
1753 		/* parent inode is not locked, be carefull */
1754 		struct ceph_vino *ptvino = NULL;
1755 		dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1756 		dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1757 		if (rinfo->head->is_target) {
1758 			tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1759 			tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1760 			ptvino = &tvino;
1761 		}
1762 		update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1763 					    session, req->r_request_started,
1764 					    rinfo->dname, rinfo->dname_len,
1765 					    &dvino, ptvino);
1766 	}
1767 done:
1768 	dout("fill_trace done err=%d\n", err);
1769 	return err;
1770 }
1771 
1772 /*
1773  * Prepopulate our cache with readdir results, leases, etc.
1774  */
1775 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1776 					   struct ceph_mds_session *session)
1777 {
1778 	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1779 	int i, err = 0;
1780 
1781 	for (i = 0; i < rinfo->dir_nr; i++) {
1782 		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1783 		struct ceph_vino vino;
1784 		struct inode *in;
1785 		int rc;
1786 
1787 		vino.ino = le64_to_cpu(rde->inode.in->ino);
1788 		vino.snap = le64_to_cpu(rde->inode.in->snapid);
1789 
1790 		in = ceph_get_inode(req->r_dentry->d_sb, vino, NULL);
1791 		if (IS_ERR(in)) {
1792 			err = PTR_ERR(in);
1793 			dout("new_inode badness got %d\n", err);
1794 			continue;
1795 		}
1796 		rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1797 				     -1, &req->r_caps_reservation);
1798 		if (rc < 0) {
1799 			pr_err("ceph_fill_inode badness on %p got %d\n",
1800 			       in, rc);
1801 			err = rc;
1802 			if (in->i_state & I_NEW) {
1803 				ihold(in);
1804 				discard_new_inode(in);
1805 			}
1806 		} else if (in->i_state & I_NEW) {
1807 			unlock_new_inode(in);
1808 		}
1809 
1810 		iput(in);
1811 	}
1812 
1813 	return err;
1814 }
1815 
1816 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1817 {
1818 	if (ctl->page) {
1819 		kunmap(ctl->page);
1820 		put_page(ctl->page);
1821 		ctl->page = NULL;
1822 	}
1823 }
1824 
1825 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1826 			      struct ceph_readdir_cache_control *ctl,
1827 			      struct ceph_mds_request *req)
1828 {
1829 	struct ceph_inode_info *ci = ceph_inode(dir);
1830 	unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1831 	unsigned idx = ctl->index % nsize;
1832 	pgoff_t pgoff = ctl->index / nsize;
1833 
1834 	if (!ctl->page || pgoff != page_index(ctl->page)) {
1835 		ceph_readdir_cache_release(ctl);
1836 		if (idx == 0)
1837 			ctl->page = grab_cache_page(&dir->i_data, pgoff);
1838 		else
1839 			ctl->page = find_lock_page(&dir->i_data, pgoff);
1840 		if (!ctl->page) {
1841 			ctl->index = -1;
1842 			return idx == 0 ? -ENOMEM : 0;
1843 		}
1844 		/* reading/filling the cache are serialized by
1845 		 * i_rwsem, no need to use page lock */
1846 		unlock_page(ctl->page);
1847 		ctl->dentries = kmap(ctl->page);
1848 		if (idx == 0)
1849 			memset(ctl->dentries, 0, PAGE_SIZE);
1850 	}
1851 
1852 	if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1853 	    req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1854 		dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1855 		ctl->dentries[idx] = dn;
1856 		ctl->index++;
1857 	} else {
1858 		dout("disable readdir cache\n");
1859 		ctl->index = -1;
1860 	}
1861 	return 0;
1862 }
1863 
1864 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1865 			     struct ceph_mds_session *session)
1866 {
1867 	struct dentry *parent = req->r_dentry;
1868 	struct inode *inode = d_inode(parent);
1869 	struct ceph_inode_info *ci = ceph_inode(inode);
1870 	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1871 	struct qstr dname;
1872 	struct dentry *dn;
1873 	struct inode *in;
1874 	int err = 0, skipped = 0, ret, i;
1875 	u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1876 	u32 last_hash = 0;
1877 	u32 fpos_offset;
1878 	struct ceph_readdir_cache_control cache_ctl = {};
1879 
1880 	if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1881 		return readdir_prepopulate_inodes_only(req, session);
1882 
1883 	if (rinfo->hash_order) {
1884 		if (req->r_path2) {
1885 			last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1886 						  req->r_path2,
1887 						  strlen(req->r_path2));
1888 			last_hash = ceph_frag_value(last_hash);
1889 		} else if (rinfo->offset_hash) {
1890 			/* mds understands offset_hash */
1891 			WARN_ON_ONCE(req->r_readdir_offset != 2);
1892 			last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1893 		}
1894 	}
1895 
1896 	if (rinfo->dir_dir &&
1897 	    le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1898 		dout("readdir_prepopulate got new frag %x -> %x\n",
1899 		     frag, le32_to_cpu(rinfo->dir_dir->frag));
1900 		frag = le32_to_cpu(rinfo->dir_dir->frag);
1901 		if (!rinfo->hash_order)
1902 			req->r_readdir_offset = 2;
1903 	}
1904 
1905 	if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1906 		dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1907 		     rinfo->dir_nr, parent);
1908 	} else {
1909 		dout("readdir_prepopulate %d items under dn %p\n",
1910 		     rinfo->dir_nr, parent);
1911 		if (rinfo->dir_dir)
1912 			ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1913 
1914 		if (ceph_frag_is_leftmost(frag) &&
1915 		    req->r_readdir_offset == 2 &&
1916 		    !(rinfo->hash_order && last_hash)) {
1917 			/* note dir version at start of readdir so we can
1918 			 * tell if any dentries get dropped */
1919 			req->r_dir_release_cnt =
1920 				atomic64_read(&ci->i_release_count);
1921 			req->r_dir_ordered_cnt =
1922 				atomic64_read(&ci->i_ordered_count);
1923 			req->r_readdir_cache_idx = 0;
1924 		}
1925 	}
1926 
1927 	cache_ctl.index = req->r_readdir_cache_idx;
1928 	fpos_offset = req->r_readdir_offset;
1929 
1930 	/* FIXME: release caps/leases if error occurs */
1931 	for (i = 0; i < rinfo->dir_nr; i++) {
1932 		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1933 		struct ceph_vino tvino;
1934 
1935 		dname.name = rde->name;
1936 		dname.len = rde->name_len;
1937 		dname.hash = full_name_hash(parent, dname.name, dname.len);
1938 
1939 		tvino.ino = le64_to_cpu(rde->inode.in->ino);
1940 		tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1941 
1942 		if (rinfo->hash_order) {
1943 			u32 hash = ceph_frag_value(rde->raw_hash);
1944 			if (hash != last_hash)
1945 				fpos_offset = 2;
1946 			last_hash = hash;
1947 			rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1948 		} else {
1949 			rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1950 		}
1951 
1952 retry_lookup:
1953 		dn = d_lookup(parent, &dname);
1954 		dout("d_lookup on parent=%p name=%.*s got %p\n",
1955 		     parent, dname.len, dname.name, dn);
1956 
1957 		if (!dn) {
1958 			dn = d_alloc(parent, &dname);
1959 			dout("d_alloc %p '%.*s' = %p\n", parent,
1960 			     dname.len, dname.name, dn);
1961 			if (!dn) {
1962 				dout("d_alloc badness\n");
1963 				err = -ENOMEM;
1964 				goto out;
1965 			}
1966 			if (rde->is_nokey) {
1967 				spin_lock(&dn->d_lock);
1968 				dn->d_flags |= DCACHE_NOKEY_NAME;
1969 				spin_unlock(&dn->d_lock);
1970 			}
1971 		} else if (d_really_is_positive(dn) &&
1972 			   (ceph_ino(d_inode(dn)) != tvino.ino ||
1973 			    ceph_snap(d_inode(dn)) != tvino.snap)) {
1974 			struct ceph_dentry_info *di = ceph_dentry(dn);
1975 			dout(" dn %p points to wrong inode %p\n",
1976 			     dn, d_inode(dn));
1977 
1978 			spin_lock(&dn->d_lock);
1979 			if (di->offset > 0 &&
1980 			    di->lease_shared_gen ==
1981 			    atomic_read(&ci->i_shared_gen)) {
1982 				__ceph_dir_clear_ordered(ci);
1983 				di->offset = 0;
1984 			}
1985 			spin_unlock(&dn->d_lock);
1986 
1987 			d_delete(dn);
1988 			dput(dn);
1989 			goto retry_lookup;
1990 		}
1991 
1992 		/* inode */
1993 		if (d_really_is_positive(dn)) {
1994 			in = d_inode(dn);
1995 		} else {
1996 			in = ceph_get_inode(parent->d_sb, tvino, NULL);
1997 			if (IS_ERR(in)) {
1998 				dout("new_inode badness\n");
1999 				d_drop(dn);
2000 				dput(dn);
2001 				err = PTR_ERR(in);
2002 				goto out;
2003 			}
2004 		}
2005 
2006 		ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
2007 				      -1, &req->r_caps_reservation);
2008 		if (ret < 0) {
2009 			pr_err("ceph_fill_inode badness on %p\n", in);
2010 			if (d_really_is_negative(dn)) {
2011 				if (in->i_state & I_NEW) {
2012 					ihold(in);
2013 					discard_new_inode(in);
2014 				}
2015 				iput(in);
2016 			}
2017 			d_drop(dn);
2018 			err = ret;
2019 			goto next_item;
2020 		}
2021 		if (in->i_state & I_NEW)
2022 			unlock_new_inode(in);
2023 
2024 		if (d_really_is_negative(dn)) {
2025 			if (ceph_security_xattr_deadlock(in)) {
2026 				dout(" skip splicing dn %p to inode %p"
2027 				     " (security xattr deadlock)\n", dn, in);
2028 				iput(in);
2029 				skipped++;
2030 				goto next_item;
2031 			}
2032 
2033 			err = splice_dentry(&dn, in);
2034 			if (err < 0)
2035 				goto next_item;
2036 		}
2037 
2038 		ceph_dentry(dn)->offset = rde->offset;
2039 
2040 		update_dentry_lease(d_inode(parent), dn,
2041 				    rde->lease, req->r_session,
2042 				    req->r_request_started);
2043 
2044 		if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
2045 			ret = fill_readdir_cache(d_inode(parent), dn,
2046 						 &cache_ctl, req);
2047 			if (ret < 0)
2048 				err = ret;
2049 		}
2050 next_item:
2051 		dput(dn);
2052 	}
2053 out:
2054 	if (err == 0 && skipped == 0) {
2055 		set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
2056 		req->r_readdir_cache_idx = cache_ctl.index;
2057 	}
2058 	ceph_readdir_cache_release(&cache_ctl);
2059 	dout("readdir_prepopulate done\n");
2060 	return err;
2061 }
2062 
2063 bool ceph_inode_set_size(struct inode *inode, loff_t size)
2064 {
2065 	struct ceph_inode_info *ci = ceph_inode(inode);
2066 	bool ret;
2067 
2068 	spin_lock(&ci->i_ceph_lock);
2069 	dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
2070 	i_size_write(inode, size);
2071 	ceph_fscache_update(inode);
2072 	inode->i_blocks = calc_inode_blocks(size);
2073 
2074 	ret = __ceph_should_report_size(ci);
2075 
2076 	spin_unlock(&ci->i_ceph_lock);
2077 
2078 	return ret;
2079 }
2080 
2081 void ceph_queue_inode_work(struct inode *inode, int work_bit)
2082 {
2083 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2084 	struct ceph_inode_info *ci = ceph_inode(inode);
2085 	set_bit(work_bit, &ci->i_work_mask);
2086 
2087 	ihold(inode);
2088 	if (queue_work(fsc->inode_wq, &ci->i_work)) {
2089 		dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
2090 	} else {
2091 		dout("queue_inode_work %p already queued, mask=%lx\n",
2092 		     inode, ci->i_work_mask);
2093 		iput(inode);
2094 	}
2095 }
2096 
2097 static void ceph_do_invalidate_pages(struct inode *inode)
2098 {
2099 	struct ceph_inode_info *ci = ceph_inode(inode);
2100 	u32 orig_gen;
2101 	int check = 0;
2102 
2103 	ceph_fscache_invalidate(inode, false);
2104 
2105 	mutex_lock(&ci->i_truncate_mutex);
2106 
2107 	if (ceph_inode_is_shutdown(inode)) {
2108 		pr_warn_ratelimited("%s: inode %llx.%llx is shut down\n",
2109 				    __func__, ceph_vinop(inode));
2110 		mapping_set_error(inode->i_mapping, -EIO);
2111 		truncate_pagecache(inode, 0);
2112 		mutex_unlock(&ci->i_truncate_mutex);
2113 		goto out;
2114 	}
2115 
2116 	spin_lock(&ci->i_ceph_lock);
2117 	dout("invalidate_pages %p gen %d revoking %d\n", inode,
2118 	     ci->i_rdcache_gen, ci->i_rdcache_revoking);
2119 	if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
2120 		if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2121 			check = 1;
2122 		spin_unlock(&ci->i_ceph_lock);
2123 		mutex_unlock(&ci->i_truncate_mutex);
2124 		goto out;
2125 	}
2126 	orig_gen = ci->i_rdcache_gen;
2127 	spin_unlock(&ci->i_ceph_lock);
2128 
2129 	if (invalidate_inode_pages2(inode->i_mapping) < 0) {
2130 		pr_err("invalidate_inode_pages2 %llx.%llx failed\n",
2131 		       ceph_vinop(inode));
2132 	}
2133 
2134 	spin_lock(&ci->i_ceph_lock);
2135 	if (orig_gen == ci->i_rdcache_gen &&
2136 	    orig_gen == ci->i_rdcache_revoking) {
2137 		dout("invalidate_pages %p gen %d successful\n", inode,
2138 		     ci->i_rdcache_gen);
2139 		ci->i_rdcache_revoking--;
2140 		check = 1;
2141 	} else {
2142 		dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
2143 		     inode, orig_gen, ci->i_rdcache_gen,
2144 		     ci->i_rdcache_revoking);
2145 		if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2146 			check = 1;
2147 	}
2148 	spin_unlock(&ci->i_ceph_lock);
2149 	mutex_unlock(&ci->i_truncate_mutex);
2150 out:
2151 	if (check)
2152 		ceph_check_caps(ci, 0);
2153 }
2154 
2155 /*
2156  * Make sure any pending truncation is applied before doing anything
2157  * that may depend on it.
2158  */
2159 void __ceph_do_pending_vmtruncate(struct inode *inode)
2160 {
2161 	struct ceph_inode_info *ci = ceph_inode(inode);
2162 	u64 to;
2163 	int wrbuffer_refs, finish = 0;
2164 
2165 	mutex_lock(&ci->i_truncate_mutex);
2166 retry:
2167 	spin_lock(&ci->i_ceph_lock);
2168 	if (ci->i_truncate_pending == 0) {
2169 		dout("%s %p none pending\n", __func__, inode);
2170 		spin_unlock(&ci->i_ceph_lock);
2171 		mutex_unlock(&ci->i_truncate_mutex);
2172 		return;
2173 	}
2174 
2175 	/*
2176 	 * make sure any dirty snapped pages are flushed before we
2177 	 * possibly truncate them.. so write AND block!
2178 	 */
2179 	if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
2180 		spin_unlock(&ci->i_ceph_lock);
2181 		dout("%s %p flushing snaps first\n", __func__, inode);
2182 		filemap_write_and_wait_range(&inode->i_data, 0,
2183 					     inode->i_sb->s_maxbytes);
2184 		goto retry;
2185 	}
2186 
2187 	/* there should be no reader or writer */
2188 	WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
2189 
2190 	to = ci->i_truncate_pagecache_size;
2191 	wrbuffer_refs = ci->i_wrbuffer_ref;
2192 	dout("%s %p (%d) to %lld\n", __func__, inode,
2193 	     ci->i_truncate_pending, to);
2194 	spin_unlock(&ci->i_ceph_lock);
2195 
2196 	ceph_fscache_resize(inode, to);
2197 	truncate_pagecache(inode, to);
2198 
2199 	spin_lock(&ci->i_ceph_lock);
2200 	if (to == ci->i_truncate_pagecache_size) {
2201 		ci->i_truncate_pending = 0;
2202 		finish = 1;
2203 	}
2204 	spin_unlock(&ci->i_ceph_lock);
2205 	if (!finish)
2206 		goto retry;
2207 
2208 	mutex_unlock(&ci->i_truncate_mutex);
2209 
2210 	if (wrbuffer_refs == 0)
2211 		ceph_check_caps(ci, 0);
2212 
2213 	wake_up_all(&ci->i_cap_wq);
2214 }
2215 
2216 static void ceph_inode_work(struct work_struct *work)
2217 {
2218 	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
2219 						 i_work);
2220 	struct inode *inode = &ci->netfs.inode;
2221 
2222 	if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2223 		dout("writeback %p\n", inode);
2224 		filemap_fdatawrite(&inode->i_data);
2225 	}
2226 	if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2227 		ceph_do_invalidate_pages(inode);
2228 
2229 	if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2230 		__ceph_do_pending_vmtruncate(inode);
2231 
2232 	if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
2233 		ceph_check_caps(ci, 0);
2234 
2235 	if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
2236 		ceph_flush_snaps(ci, NULL);
2237 
2238 	iput(inode);
2239 }
2240 
2241 static const char *ceph_encrypted_get_link(struct dentry *dentry,
2242 					   struct inode *inode,
2243 					   struct delayed_call *done)
2244 {
2245 	struct ceph_inode_info *ci = ceph_inode(inode);
2246 
2247 	if (!dentry)
2248 		return ERR_PTR(-ECHILD);
2249 
2250 	return fscrypt_get_symlink(inode, ci->i_symlink, i_size_read(inode),
2251 				   done);
2252 }
2253 
2254 static int ceph_encrypted_symlink_getattr(struct mnt_idmap *idmap,
2255 					  const struct path *path,
2256 					  struct kstat *stat, u32 request_mask,
2257 					  unsigned int query_flags)
2258 {
2259 	int ret;
2260 
2261 	ret = ceph_getattr(idmap, path, stat, request_mask, query_flags);
2262 	if (ret)
2263 		return ret;
2264 	return fscrypt_symlink_getattr(path, stat);
2265 }
2266 
2267 /*
2268  * symlinks
2269  */
2270 static const struct inode_operations ceph_symlink_iops = {
2271 	.get_link = simple_get_link,
2272 	.setattr = ceph_setattr,
2273 	.getattr = ceph_getattr,
2274 	.listxattr = ceph_listxattr,
2275 };
2276 
2277 static const struct inode_operations ceph_encrypted_symlink_iops = {
2278 	.get_link = ceph_encrypted_get_link,
2279 	.setattr = ceph_setattr,
2280 	.getattr = ceph_encrypted_symlink_getattr,
2281 	.listxattr = ceph_listxattr,
2282 };
2283 
2284 /*
2285  * Transfer the encrypted last block to the MDS and the MDS
2286  * will help update it when truncating a smaller size.
2287  *
2288  * We don't support a PAGE_SIZE that is smaller than the
2289  * CEPH_FSCRYPT_BLOCK_SIZE.
2290  */
2291 static int fill_fscrypt_truncate(struct inode *inode,
2292 				 struct ceph_mds_request *req,
2293 				 struct iattr *attr)
2294 {
2295 	struct ceph_inode_info *ci = ceph_inode(inode);
2296 	int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
2297 	loff_t pos, orig_pos = round_down(attr->ia_size,
2298 					  CEPH_FSCRYPT_BLOCK_SIZE);
2299 	u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
2300 	struct ceph_pagelist *pagelist = NULL;
2301 	struct kvec iov = {0};
2302 	struct iov_iter iter;
2303 	struct page *page = NULL;
2304 	struct ceph_fscrypt_truncate_size_header header;
2305 	int retry_op = 0;
2306 	int len = CEPH_FSCRYPT_BLOCK_SIZE;
2307 	loff_t i_size = i_size_read(inode);
2308 	int got, ret, issued;
2309 	u64 objver;
2310 
2311 	ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
2312 	if (ret < 0)
2313 		return ret;
2314 
2315 	issued = __ceph_caps_issued(ci, NULL);
2316 
2317 	dout("%s size %lld -> %lld got cap refs on %s, issued %s\n", __func__,
2318 	     i_size, attr->ia_size, ceph_cap_string(got),
2319 	     ceph_cap_string(issued));
2320 
2321 	/* Try to writeback the dirty pagecaches */
2322 	if (issued & (CEPH_CAP_FILE_BUFFER)) {
2323 		loff_t lend = orig_pos + CEPH_FSCRYPT_BLOCK_SHIFT - 1;
2324 
2325 		ret = filemap_write_and_wait_range(inode->i_mapping,
2326 						   orig_pos, lend);
2327 		if (ret < 0)
2328 			goto out;
2329 	}
2330 
2331 	page = __page_cache_alloc(GFP_KERNEL);
2332 	if (page == NULL) {
2333 		ret = -ENOMEM;
2334 		goto out;
2335 	}
2336 
2337 	pagelist = ceph_pagelist_alloc(GFP_KERNEL);
2338 	if (!pagelist) {
2339 		ret = -ENOMEM;
2340 		goto out;
2341 	}
2342 
2343 	iov.iov_base = kmap_local_page(page);
2344 	iov.iov_len = len;
2345 	iov_iter_kvec(&iter, READ, &iov, 1, len);
2346 
2347 	pos = orig_pos;
2348 	ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objver);
2349 	if (ret < 0)
2350 		goto out;
2351 
2352 	/* Insert the header first */
2353 	header.ver = 1;
2354 	header.compat = 1;
2355 	header.change_attr = cpu_to_le64(inode_peek_iversion_raw(inode));
2356 
2357 	/*
2358 	 * Always set the block_size to CEPH_FSCRYPT_BLOCK_SIZE,
2359 	 * because in MDS it may need this to do the truncate.
2360 	 */
2361 	header.block_size = cpu_to_le32(CEPH_FSCRYPT_BLOCK_SIZE);
2362 
2363 	/*
2364 	 * If we hit a hole here, we should just skip filling
2365 	 * the fscrypt for the request, because once the fscrypt
2366 	 * is enabled, the file will be split into many blocks
2367 	 * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
2368 	 * has a hole, the hole size should be multiple of block
2369 	 * size.
2370 	 *
2371 	 * If the Rados object doesn't exist, it will be set to 0.
2372 	 */
2373 	if (!objver) {
2374 		dout("%s hit hole, ppos %lld < size %lld\n", __func__,
2375 		     pos, i_size);
2376 
2377 		header.data_len = cpu_to_le32(8 + 8 + 4);
2378 		header.file_offset = 0;
2379 		ret = 0;
2380 	} else {
2381 		header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
2382 		header.file_offset = cpu_to_le64(orig_pos);
2383 
2384 		dout("%s encrypt block boff/bsize %d/%lu\n", __func__,
2385 		     boff, CEPH_FSCRYPT_BLOCK_SIZE);
2386 
2387 		/* truncate and zero out the extra contents for the last block */
2388 		memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
2389 
2390 		/* encrypt the last block */
2391 		ret = ceph_fscrypt_encrypt_block_inplace(inode, page,
2392 						    CEPH_FSCRYPT_BLOCK_SIZE,
2393 						    0, block,
2394 						    GFP_KERNEL);
2395 		if (ret)
2396 			goto out;
2397 	}
2398 
2399 	/* Insert the header */
2400 	ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
2401 	if (ret)
2402 		goto out;
2403 
2404 	if (header.block_size) {
2405 		/* Append the last block contents to pagelist */
2406 		ret = ceph_pagelist_append(pagelist, iov.iov_base,
2407 					   CEPH_FSCRYPT_BLOCK_SIZE);
2408 		if (ret)
2409 			goto out;
2410 	}
2411 	req->r_pagelist = pagelist;
2412 out:
2413 	dout("%s %p size dropping cap refs on %s\n", __func__,
2414 	     inode, ceph_cap_string(got));
2415 	ceph_put_cap_refs(ci, got);
2416 	if (iov.iov_base)
2417 		kunmap_local(iov.iov_base);
2418 	if (page)
2419 		__free_pages(page, 0);
2420 	if (ret && pagelist)
2421 		ceph_pagelist_release(pagelist);
2422 	return ret;
2423 }
2424 
2425 int __ceph_setattr(struct inode *inode, struct iattr *attr,
2426 		   struct ceph_iattr *cia)
2427 {
2428 	struct ceph_inode_info *ci = ceph_inode(inode);
2429 	unsigned int ia_valid = attr->ia_valid;
2430 	struct ceph_mds_request *req;
2431 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2432 	struct ceph_cap_flush *prealloc_cf;
2433 	loff_t isize = i_size_read(inode);
2434 	int issued;
2435 	int release = 0, dirtied = 0;
2436 	int mask = 0;
2437 	int err = 0;
2438 	int inode_dirty_flags = 0;
2439 	bool lock_snap_rwsem = false;
2440 	bool fill_fscrypt;
2441 	int truncate_retry = 20; /* The RMW will take around 50ms */
2442 
2443 retry:
2444 	prealloc_cf = ceph_alloc_cap_flush();
2445 	if (!prealloc_cf)
2446 		return -ENOMEM;
2447 
2448 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2449 				       USE_AUTH_MDS);
2450 	if (IS_ERR(req)) {
2451 		ceph_free_cap_flush(prealloc_cf);
2452 		return PTR_ERR(req);
2453 	}
2454 
2455 	fill_fscrypt = false;
2456 	spin_lock(&ci->i_ceph_lock);
2457 	issued = __ceph_caps_issued(ci, NULL);
2458 
2459 	if (!ci->i_head_snapc &&
2460 	    (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2461 		lock_snap_rwsem = true;
2462 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
2463 			spin_unlock(&ci->i_ceph_lock);
2464 			down_read(&mdsc->snap_rwsem);
2465 			spin_lock(&ci->i_ceph_lock);
2466 			issued = __ceph_caps_issued(ci, NULL);
2467 		}
2468 	}
2469 
2470 	dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2471 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
2472 	if (cia && cia->fscrypt_auth) {
2473 		u32 len = ceph_fscrypt_auth_len(cia->fscrypt_auth);
2474 
2475 		if (len > sizeof(*cia->fscrypt_auth)) {
2476 			err = -EINVAL;
2477 			spin_unlock(&ci->i_ceph_lock);
2478 			goto out;
2479 		}
2480 
2481 		dout("setattr %llx:%llx fscrypt_auth len %u to %u)\n",
2482 			ceph_vinop(inode), ci->fscrypt_auth_len, len);
2483 
2484 		/* It should never be re-set once set */
2485 		WARN_ON_ONCE(ci->fscrypt_auth);
2486 
2487 		if (issued & CEPH_CAP_AUTH_EXCL) {
2488 			dirtied |= CEPH_CAP_AUTH_EXCL;
2489 			kfree(ci->fscrypt_auth);
2490 			ci->fscrypt_auth = (u8 *)cia->fscrypt_auth;
2491 			ci->fscrypt_auth_len = len;
2492 		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2493 			   ci->fscrypt_auth_len != len ||
2494 			   memcmp(ci->fscrypt_auth, cia->fscrypt_auth, len)) {
2495 			req->r_fscrypt_auth = cia->fscrypt_auth;
2496 			mask |= CEPH_SETATTR_FSCRYPT_AUTH;
2497 			release |= CEPH_CAP_AUTH_SHARED;
2498 		}
2499 		cia->fscrypt_auth = NULL;
2500 	}
2501 #else
2502 	if (cia && cia->fscrypt_auth) {
2503 		err = -EINVAL;
2504 		spin_unlock(&ci->i_ceph_lock);
2505 		goto out;
2506 	}
2507 #endif /* CONFIG_FS_ENCRYPTION */
2508 
2509 	if (ia_valid & ATTR_UID) {
2510 		dout("setattr %p uid %d -> %d\n", inode,
2511 		     from_kuid(&init_user_ns, inode->i_uid),
2512 		     from_kuid(&init_user_ns, attr->ia_uid));
2513 		if (issued & CEPH_CAP_AUTH_EXCL) {
2514 			inode->i_uid = attr->ia_uid;
2515 			dirtied |= CEPH_CAP_AUTH_EXCL;
2516 		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2517 			   !uid_eq(attr->ia_uid, inode->i_uid)) {
2518 			req->r_args.setattr.uid = cpu_to_le32(
2519 				from_kuid(&init_user_ns, attr->ia_uid));
2520 			mask |= CEPH_SETATTR_UID;
2521 			release |= CEPH_CAP_AUTH_SHARED;
2522 		}
2523 	}
2524 	if (ia_valid & ATTR_GID) {
2525 		dout("setattr %p gid %d -> %d\n", inode,
2526 		     from_kgid(&init_user_ns, inode->i_gid),
2527 		     from_kgid(&init_user_ns, attr->ia_gid));
2528 		if (issued & CEPH_CAP_AUTH_EXCL) {
2529 			inode->i_gid = attr->ia_gid;
2530 			dirtied |= CEPH_CAP_AUTH_EXCL;
2531 		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2532 			   !gid_eq(attr->ia_gid, inode->i_gid)) {
2533 			req->r_args.setattr.gid = cpu_to_le32(
2534 				from_kgid(&init_user_ns, attr->ia_gid));
2535 			mask |= CEPH_SETATTR_GID;
2536 			release |= CEPH_CAP_AUTH_SHARED;
2537 		}
2538 	}
2539 	if (ia_valid & ATTR_MODE) {
2540 		dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2541 		     attr->ia_mode);
2542 		if (issued & CEPH_CAP_AUTH_EXCL) {
2543 			inode->i_mode = attr->ia_mode;
2544 			dirtied |= CEPH_CAP_AUTH_EXCL;
2545 		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2546 			   attr->ia_mode != inode->i_mode) {
2547 			inode->i_mode = attr->ia_mode;
2548 			req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2549 			mask |= CEPH_SETATTR_MODE;
2550 			release |= CEPH_CAP_AUTH_SHARED;
2551 		}
2552 	}
2553 
2554 	if (ia_valid & ATTR_ATIME) {
2555 		dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2556 		     inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2557 		     attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2558 		if (issued & CEPH_CAP_FILE_EXCL) {
2559 			ci->i_time_warp_seq++;
2560 			inode->i_atime = attr->ia_atime;
2561 			dirtied |= CEPH_CAP_FILE_EXCL;
2562 		} else if ((issued & CEPH_CAP_FILE_WR) &&
2563 			   timespec64_compare(&inode->i_atime,
2564 					    &attr->ia_atime) < 0) {
2565 			inode->i_atime = attr->ia_atime;
2566 			dirtied |= CEPH_CAP_FILE_WR;
2567 		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2568 			   !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2569 			ceph_encode_timespec64(&req->r_args.setattr.atime,
2570 					       &attr->ia_atime);
2571 			mask |= CEPH_SETATTR_ATIME;
2572 			release |= CEPH_CAP_FILE_SHARED |
2573 				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2574 		}
2575 	}
2576 	if (ia_valid & ATTR_SIZE) {
2577 		dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2578 		/*
2579 		 * Only when the new size is smaller and not aligned to
2580 		 * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
2581 		 */
2582 		if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
2583 		    (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
2584 			mask |= CEPH_SETATTR_SIZE;
2585 			release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2586 				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2587 			set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2588 			mask |= CEPH_SETATTR_FSCRYPT_FILE;
2589 			req->r_args.setattr.size =
2590 				cpu_to_le64(round_up(attr->ia_size,
2591 						     CEPH_FSCRYPT_BLOCK_SIZE));
2592 			req->r_args.setattr.old_size =
2593 				cpu_to_le64(round_up(isize,
2594 						     CEPH_FSCRYPT_BLOCK_SIZE));
2595 			req->r_fscrypt_file = attr->ia_size;
2596 			fill_fscrypt = true;
2597 		} else if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
2598 			if (attr->ia_size > isize) {
2599 				i_size_write(inode, attr->ia_size);
2600 				inode->i_blocks = calc_inode_blocks(attr->ia_size);
2601 				ci->i_reported_size = attr->ia_size;
2602 				dirtied |= CEPH_CAP_FILE_EXCL;
2603 				ia_valid |= ATTR_MTIME;
2604 			}
2605 		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2606 			   attr->ia_size != isize) {
2607 			mask |= CEPH_SETATTR_SIZE;
2608 			release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2609 				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2610 			if (IS_ENCRYPTED(inode) && attr->ia_size) {
2611 				set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2612 				mask |= CEPH_SETATTR_FSCRYPT_FILE;
2613 				req->r_args.setattr.size =
2614 					cpu_to_le64(round_up(attr->ia_size,
2615 							     CEPH_FSCRYPT_BLOCK_SIZE));
2616 				req->r_args.setattr.old_size =
2617 					cpu_to_le64(round_up(isize,
2618 							     CEPH_FSCRYPT_BLOCK_SIZE));
2619 				req->r_fscrypt_file = attr->ia_size;
2620 			} else {
2621 				req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2622 				req->r_args.setattr.old_size = cpu_to_le64(isize);
2623 				req->r_fscrypt_file = 0;
2624 			}
2625 		}
2626 	}
2627 	if (ia_valid & ATTR_MTIME) {
2628 		dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2629 		     inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2630 		     attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2631 		if (issued & CEPH_CAP_FILE_EXCL) {
2632 			ci->i_time_warp_seq++;
2633 			inode->i_mtime = attr->ia_mtime;
2634 			dirtied |= CEPH_CAP_FILE_EXCL;
2635 		} else if ((issued & CEPH_CAP_FILE_WR) &&
2636 			   timespec64_compare(&inode->i_mtime,
2637 					    &attr->ia_mtime) < 0) {
2638 			inode->i_mtime = attr->ia_mtime;
2639 			dirtied |= CEPH_CAP_FILE_WR;
2640 		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2641 			   !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2642 			ceph_encode_timespec64(&req->r_args.setattr.mtime,
2643 					       &attr->ia_mtime);
2644 			mask |= CEPH_SETATTR_MTIME;
2645 			release |= CEPH_CAP_FILE_SHARED |
2646 				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2647 		}
2648 	}
2649 
2650 	/* these do nothing */
2651 	if (ia_valid & ATTR_CTIME) {
2652 		bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2653 					 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2654 		dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2655 		     inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2656 		     attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2657 		     only ? "ctime only" : "ignored");
2658 		if (only) {
2659 			/*
2660 			 * if kernel wants to dirty ctime but nothing else,
2661 			 * we need to choose a cap to dirty under, or do
2662 			 * a almost-no-op setattr
2663 			 */
2664 			if (issued & CEPH_CAP_AUTH_EXCL)
2665 				dirtied |= CEPH_CAP_AUTH_EXCL;
2666 			else if (issued & CEPH_CAP_FILE_EXCL)
2667 				dirtied |= CEPH_CAP_FILE_EXCL;
2668 			else if (issued & CEPH_CAP_XATTR_EXCL)
2669 				dirtied |= CEPH_CAP_XATTR_EXCL;
2670 			else
2671 				mask |= CEPH_SETATTR_CTIME;
2672 		}
2673 	}
2674 	if (ia_valid & ATTR_FILE)
2675 		dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2676 
2677 	if (dirtied) {
2678 		inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2679 							   &prealloc_cf);
2680 		inode->i_ctime = attr->ia_ctime;
2681 		inode_inc_iversion_raw(inode);
2682 	}
2683 
2684 	release &= issued;
2685 	spin_unlock(&ci->i_ceph_lock);
2686 	if (lock_snap_rwsem) {
2687 		up_read(&mdsc->snap_rwsem);
2688 		lock_snap_rwsem = false;
2689 	}
2690 
2691 	if (inode_dirty_flags)
2692 		__mark_inode_dirty(inode, inode_dirty_flags);
2693 
2694 	if (mask) {
2695 		req->r_inode = inode;
2696 		ihold(inode);
2697 		req->r_inode_drop = release;
2698 		req->r_args.setattr.mask = cpu_to_le32(mask);
2699 		req->r_num_caps = 1;
2700 		req->r_stamp = attr->ia_ctime;
2701 		if (fill_fscrypt) {
2702 			err = fill_fscrypt_truncate(inode, req, attr);
2703 			if (err)
2704 				goto out;
2705 		}
2706 
2707 		/*
2708 		 * The truncate request will return -EAGAIN when the
2709 		 * last block has been updated just before the MDS
2710 		 * successfully gets the xlock for the FILE lock. To
2711 		 * avoid corrupting the file contents we need to retry
2712 		 * it.
2713 		 */
2714 		err = ceph_mdsc_do_request(mdsc, NULL, req);
2715 		if (err == -EAGAIN && truncate_retry--) {
2716 			dout("setattr %p result=%d (%s locally, %d remote), retry it!\n",
2717 			     inode, err, ceph_cap_string(dirtied), mask);
2718 			ceph_mdsc_put_request(req);
2719 			ceph_free_cap_flush(prealloc_cf);
2720 			goto retry;
2721 		}
2722 	}
2723 out:
2724 	dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2725 	     ceph_cap_string(dirtied), mask);
2726 
2727 	ceph_mdsc_put_request(req);
2728 	ceph_free_cap_flush(prealloc_cf);
2729 
2730 	if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2731 		__ceph_do_pending_vmtruncate(inode);
2732 
2733 	return err;
2734 }
2735 
2736 /*
2737  * setattr
2738  */
2739 int ceph_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
2740 		 struct iattr *attr)
2741 {
2742 	struct inode *inode = d_inode(dentry);
2743 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2744 	int err;
2745 
2746 	if (ceph_snap(inode) != CEPH_NOSNAP)
2747 		return -EROFS;
2748 
2749 	if (ceph_inode_is_shutdown(inode))
2750 		return -ESTALE;
2751 
2752 	err = fscrypt_prepare_setattr(dentry, attr);
2753 	if (err)
2754 		return err;
2755 
2756 	err = setattr_prepare(&nop_mnt_idmap, dentry, attr);
2757 	if (err != 0)
2758 		return err;
2759 
2760 	if ((attr->ia_valid & ATTR_SIZE) &&
2761 	    attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2762 		return -EFBIG;
2763 
2764 	if ((attr->ia_valid & ATTR_SIZE) &&
2765 	    ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2766 		return -EDQUOT;
2767 
2768 	err = __ceph_setattr(inode, attr, NULL);
2769 
2770 	if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2771 		err = posix_acl_chmod(&nop_mnt_idmap, dentry, attr->ia_mode);
2772 
2773 	return err;
2774 }
2775 
2776 int ceph_try_to_choose_auth_mds(struct inode *inode, int mask)
2777 {
2778 	int issued = ceph_caps_issued(ceph_inode(inode));
2779 
2780 	/*
2781 	 * If any 'x' caps is issued we can just choose the auth MDS
2782 	 * instead of the random replica MDSes. Because only when the
2783 	 * Locker is in LOCK_EXEC state will the loner client could
2784 	 * get the 'x' caps. And if we send the getattr requests to
2785 	 * any replica MDS it must auth pin and tries to rdlock from
2786 	 * the auth MDS, and then the auth MDS need to do the Locker
2787 	 * state transition to LOCK_SYNC. And after that the lock state
2788 	 * will change back.
2789 	 *
2790 	 * This cost much when doing the Locker state transition and
2791 	 * usually will need to revoke caps from clients.
2792 	 *
2793 	 * And for the 'Xs' caps for getxattr we will also choose the
2794 	 * auth MDS, because the MDS side code is buggy due to setxattr
2795 	 * won't notify the replica MDSes when the values changed and
2796 	 * the replica MDS will return the old values. Though we will
2797 	 * fix it in MDS code, but this still makes sense for old ceph.
2798 	 */
2799 	if (((mask & CEPH_CAP_ANY_SHARED) && (issued & CEPH_CAP_ANY_EXCL))
2800 	    || (mask & (CEPH_STAT_RSTAT | CEPH_STAT_CAP_XATTR)))
2801 		return USE_AUTH_MDS;
2802 	else
2803 		return USE_ANY_MDS;
2804 }
2805 
2806 /*
2807  * Verify that we have a lease on the given mask.  If not,
2808  * do a getattr against an mds.
2809  */
2810 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2811 		      int mask, bool force)
2812 {
2813 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2814 	struct ceph_mds_client *mdsc = fsc->mdsc;
2815 	struct ceph_mds_request *req;
2816 	int mode;
2817 	int err;
2818 
2819 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
2820 		dout("do_getattr inode %p SNAPDIR\n", inode);
2821 		return 0;
2822 	}
2823 
2824 	dout("do_getattr inode %p mask %s mode 0%o\n",
2825 	     inode, ceph_cap_string(mask), inode->i_mode);
2826 	if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2827 			return 0;
2828 
2829 	mode = ceph_try_to_choose_auth_mds(inode, mask);
2830 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2831 	if (IS_ERR(req))
2832 		return PTR_ERR(req);
2833 	req->r_inode = inode;
2834 	ihold(inode);
2835 	req->r_num_caps = 1;
2836 	req->r_args.getattr.mask = cpu_to_le32(mask);
2837 	req->r_locked_page = locked_page;
2838 	err = ceph_mdsc_do_request(mdsc, NULL, req);
2839 	if (locked_page && err == 0) {
2840 		u64 inline_version = req->r_reply_info.targeti.inline_version;
2841 		if (inline_version == 0) {
2842 			/* the reply is supposed to contain inline data */
2843 			err = -EINVAL;
2844 		} else if (inline_version == CEPH_INLINE_NONE ||
2845 			   inline_version == 1) {
2846 			err = -ENODATA;
2847 		} else {
2848 			err = req->r_reply_info.targeti.inline_len;
2849 		}
2850 	}
2851 	ceph_mdsc_put_request(req);
2852 	dout("do_getattr result=%d\n", err);
2853 	return err;
2854 }
2855 
2856 int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
2857 		      size_t size)
2858 {
2859 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2860 	struct ceph_mds_client *mdsc = fsc->mdsc;
2861 	struct ceph_mds_request *req;
2862 	int mode = USE_AUTH_MDS;
2863 	int err;
2864 	char *xattr_value;
2865 	size_t xattr_value_len;
2866 
2867 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
2868 	if (IS_ERR(req)) {
2869 		err = -ENOMEM;
2870 		goto out;
2871 	}
2872 
2873 	req->r_feature_needed = CEPHFS_FEATURE_OP_GETVXATTR;
2874 	req->r_path2 = kstrdup(name, GFP_NOFS);
2875 	if (!req->r_path2) {
2876 		err = -ENOMEM;
2877 		goto put;
2878 	}
2879 
2880 	ihold(inode);
2881 	req->r_inode = inode;
2882 	err = ceph_mdsc_do_request(mdsc, NULL, req);
2883 	if (err < 0)
2884 		goto put;
2885 
2886 	xattr_value = req->r_reply_info.xattr_info.xattr_value;
2887 	xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
2888 
2889 	dout("do_getvxattr xattr_value_len:%zu, size:%zu\n", xattr_value_len, size);
2890 
2891 	err = (int)xattr_value_len;
2892 	if (size == 0)
2893 		goto put;
2894 
2895 	if (xattr_value_len > size) {
2896 		err = -ERANGE;
2897 		goto put;
2898 	}
2899 
2900 	memcpy(value, xattr_value, xattr_value_len);
2901 put:
2902 	ceph_mdsc_put_request(req);
2903 out:
2904 	dout("do_getvxattr result=%d\n", err);
2905 	return err;
2906 }
2907 
2908 
2909 /*
2910  * Check inode permissions.  We verify we have a valid value for
2911  * the AUTH cap, then call the generic handler.
2912  */
2913 int ceph_permission(struct mnt_idmap *idmap, struct inode *inode,
2914 		    int mask)
2915 {
2916 	int err;
2917 
2918 	if (mask & MAY_NOT_BLOCK)
2919 		return -ECHILD;
2920 
2921 	err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2922 
2923 	if (!err)
2924 		err = generic_permission(&nop_mnt_idmap, inode, mask);
2925 	return err;
2926 }
2927 
2928 /* Craft a mask of needed caps given a set of requested statx attrs. */
2929 static int statx_to_caps(u32 want, umode_t mode)
2930 {
2931 	int mask = 0;
2932 
2933 	if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME|STATX_CHANGE_COOKIE))
2934 		mask |= CEPH_CAP_AUTH_SHARED;
2935 
2936 	if (want & (STATX_NLINK|STATX_CTIME|STATX_CHANGE_COOKIE)) {
2937 		/*
2938 		 * The link count for directories depends on inode->i_subdirs,
2939 		 * and that is only updated when Fs caps are held.
2940 		 */
2941 		if (S_ISDIR(mode))
2942 			mask |= CEPH_CAP_FILE_SHARED;
2943 		else
2944 			mask |= CEPH_CAP_LINK_SHARED;
2945 	}
2946 
2947 	if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|STATX_BLOCKS|STATX_CHANGE_COOKIE))
2948 		mask |= CEPH_CAP_FILE_SHARED;
2949 
2950 	if (want & (STATX_CTIME|STATX_CHANGE_COOKIE))
2951 		mask |= CEPH_CAP_XATTR_SHARED;
2952 
2953 	return mask;
2954 }
2955 
2956 /*
2957  * Get all the attributes. If we have sufficient caps for the requested attrs,
2958  * then we can avoid talking to the MDS at all.
2959  */
2960 int ceph_getattr(struct mnt_idmap *idmap, const struct path *path,
2961 		 struct kstat *stat, u32 request_mask, unsigned int flags)
2962 {
2963 	struct inode *inode = d_inode(path->dentry);
2964 	struct super_block *sb = inode->i_sb;
2965 	struct ceph_inode_info *ci = ceph_inode(inode);
2966 	u32 valid_mask = STATX_BASIC_STATS;
2967 	int err = 0;
2968 
2969 	if (ceph_inode_is_shutdown(inode))
2970 		return -ESTALE;
2971 
2972 	/* Skip the getattr altogether if we're asked not to sync */
2973 	if ((flags & AT_STATX_SYNC_TYPE) != AT_STATX_DONT_SYNC) {
2974 		err = ceph_do_getattr(inode,
2975 				statx_to_caps(request_mask, inode->i_mode),
2976 				flags & AT_STATX_FORCE_SYNC);
2977 		if (err)
2978 			return err;
2979 	}
2980 
2981 	generic_fillattr(&nop_mnt_idmap, inode, stat);
2982 	stat->ino = ceph_present_inode(inode);
2983 
2984 	/*
2985 	 * btime on newly-allocated inodes is 0, so if this is still set to
2986 	 * that, then assume that it's not valid.
2987 	 */
2988 	if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2989 		stat->btime = ci->i_btime;
2990 		valid_mask |= STATX_BTIME;
2991 	}
2992 
2993 	if (request_mask & STATX_CHANGE_COOKIE) {
2994 		stat->change_cookie = inode_peek_iversion_raw(inode);
2995 		valid_mask |= STATX_CHANGE_COOKIE;
2996 	}
2997 
2998 	if (ceph_snap(inode) == CEPH_NOSNAP)
2999 		stat->dev = sb->s_dev;
3000 	else
3001 		stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
3002 
3003 	if (S_ISDIR(inode->i_mode)) {
3004 		if (ceph_test_mount_opt(ceph_sb_to_client(sb), RBYTES)) {
3005 			stat->size = ci->i_rbytes;
3006 		} else if (ceph_snap(inode) == CEPH_SNAPDIR) {
3007 			struct ceph_inode_info *pci;
3008 			struct ceph_snap_realm *realm;
3009 			struct inode *parent;
3010 
3011 			parent = ceph_lookup_inode(sb, ceph_ino(inode));
3012 			if (IS_ERR(parent))
3013 				return PTR_ERR(parent);
3014 
3015 			pci = ceph_inode(parent);
3016 			spin_lock(&pci->i_ceph_lock);
3017 			realm = pci->i_snap_realm;
3018 			if (realm)
3019 				stat->size = realm->num_snaps;
3020 			else
3021 				stat->size = 0;
3022 			spin_unlock(&pci->i_ceph_lock);
3023 			iput(parent);
3024 		} else {
3025 			stat->size = ci->i_files + ci->i_subdirs;
3026 		}
3027 		stat->blocks = 0;
3028 		stat->blksize = 65536;
3029 		/*
3030 		 * Some applications rely on the number of st_nlink
3031 		 * value on directories to be either 0 (if unlinked)
3032 		 * or 2 + number of subdirectories.
3033 		 */
3034 		if (stat->nlink == 1)
3035 			/* '.' + '..' + subdirs */
3036 			stat->nlink = 1 + 1 + ci->i_subdirs;
3037 	}
3038 
3039 	stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
3040 	if (IS_ENCRYPTED(inode))
3041 		stat->attributes |= STATX_ATTR_ENCRYPTED;
3042 	stat->attributes_mask |= (STATX_ATTR_CHANGE_MONOTONIC |
3043 				  STATX_ATTR_ENCRYPTED);
3044 
3045 	stat->result_mask = request_mask & valid_mask;
3046 	return err;
3047 }
3048 
3049 void ceph_inode_shutdown(struct inode *inode)
3050 {
3051 	struct ceph_inode_info *ci = ceph_inode(inode);
3052 	struct rb_node *p;
3053 	int iputs = 0;
3054 	bool invalidate = false;
3055 
3056 	spin_lock(&ci->i_ceph_lock);
3057 	ci->i_ceph_flags |= CEPH_I_SHUTDOWN;
3058 	p = rb_first(&ci->i_caps);
3059 	while (p) {
3060 		struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
3061 
3062 		p = rb_next(p);
3063 		iputs += ceph_purge_inode_cap(inode, cap, &invalidate);
3064 	}
3065 	spin_unlock(&ci->i_ceph_lock);
3066 
3067 	if (invalidate)
3068 		ceph_queue_invalidate(inode);
3069 	while (iputs--)
3070 		iput(inode);
3071 }
3072