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