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