xref: /linux/fs/smb/server/vfs_cache.c (revision a10ea943356b9d70c5616a0a06f6fa97cfdaccb1)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2016 Namjae Jeon <linkinjeon@kernel.org>
4  * Copyright (C) 2019 Samsung Electronics Co., Ltd.
5  */
6 
7 #include <linux/fs.h>
8 #include <linux/filelock.h>
9 #include <linux/slab.h>
10 #include <linux/vmalloc.h>
11 #include <linux/kthread.h>
12 #include <linux/freezer.h>
13 #include <linux/dcache.h>
14 
15 #include "glob.h"
16 #include "vfs_cache.h"
17 #include "oplock.h"
18 #include "vfs.h"
19 #include "connection.h"
20 #include "misc.h"
21 #include "mgmt/tree_connect.h"
22 #include "mgmt/user_session.h"
23 #include "mgmt/user_config.h"
24 #include "smb_common.h"
25 #include "server.h"
26 #include "smb2pdu.h"
27 
28 #define S_DEL_PENDING			1
29 #define S_DEL_ON_CLS			2
30 #define S_DEL_ON_CLS_STREAM		8
31 
32 static unsigned int inode_hash_mask __read_mostly;
33 static unsigned int inode_hash_shift __read_mostly;
34 static struct hlist_head *inode_hashtable __read_mostly;
35 static DEFINE_RWLOCK(inode_hash_lock);
36 
37 static struct ksmbd_file_table global_ft;
38 static atomic_long_t fd_limit;
39 static struct kmem_cache *filp_cache;
40 
41 static int ksmbd_mark_fp_closed(struct ksmbd_file *fp);
42 
43 #define OPLOCK_NONE      0
44 #define OPLOCK_EXCLUSIVE 1
45 #define OPLOCK_BATCH     2
46 #define OPLOCK_READ      3  /* level 2 oplock */
47 
48 #ifdef CONFIG_PROC_FS
49 
50 static const struct ksmbd_const_name ksmbd_lease_const_names[] = {
51 	{le32_to_cpu(SMB2_LEASE_NONE_LE), "LEASE_NONE"},
52 	{le32_to_cpu(SMB2_LEASE_READ_CACHING_LE), "LEASE_R"},
53 	{le32_to_cpu(SMB2_LEASE_HANDLE_CACHING_LE), "LEASE_H"},
54 	{le32_to_cpu(SMB2_LEASE_WRITE_CACHING_LE), "LEASE_W"},
55 	{le32_to_cpu(SMB2_LEASE_READ_CACHING_LE |
56 		     SMB2_LEASE_HANDLE_CACHING_LE), "LEASE_RH"},
57 	{le32_to_cpu(SMB2_LEASE_READ_CACHING_LE |
58 		     SMB2_LEASE_WRITE_CACHING_LE), "LEASE_RW"},
59 	{le32_to_cpu(SMB2_LEASE_HANDLE_CACHING_LE |
60 		     SMB2_LEASE_WRITE_CACHING_LE), "LEASE_WH"},
61 	{le32_to_cpu(SMB2_LEASE_READ_CACHING_LE |
62 		     SMB2_LEASE_HANDLE_CACHING_LE |
63 		     SMB2_LEASE_WRITE_CACHING_LE), "LEASE_RWH"},
64 };
65 
66 static const struct ksmbd_const_name ksmbd_oplock_const_names[] = {
67 	{SMB2_OPLOCK_LEVEL_NONE, "OPLOCK_NONE"},
68 	{SMB2_OPLOCK_LEVEL_II, "OPLOCK_II"},
69 	{SMB2_OPLOCK_LEVEL_EXCLUSIVE, "OPLOCK_EXECL"},
70 	{SMB2_OPLOCK_LEVEL_BATCH, "OPLOCK_BATCH"},
71 };
72 
73 static int proc_show_files(struct seq_file *m, void *v)
74 {
75 	struct ksmbd_file *fp = NULL;
76 	unsigned int id;
77 	struct oplock_info *opinfo;
78 
79 	seq_printf(m, "#%-10s %-10s %-10s %-10s %-15s %-10s %-10s %s\n",
80 		   "<tree id>", "<pid>", "<vid>", "<refcnt>",
81 		   "<oplock>", "<daccess>", "<saccess>",
82 		   "<name>");
83 
84 	read_lock(&global_ft.lock);
85 	idr_for_each_entry(global_ft.idr, fp, id) {
86 		seq_printf(m, "%#-10x %#-10llx %#-10llx %#-10x",
87 			   fp->tcon ? fp->tcon->id : 0,
88 			   fp->persistent_id,
89 			   fp->volatile_id,
90 			   atomic_read(&fp->refcount));
91 
92 		rcu_read_lock();
93 		opinfo = rcu_dereference(fp->f_opinfo);
94 		if (opinfo) {
95 			const struct ksmbd_const_name *const_names;
96 			int count;
97 			unsigned int level;
98 
99 			if (opinfo->is_lease) {
100 				const_names = ksmbd_lease_const_names;
101 				count = ARRAY_SIZE(ksmbd_lease_const_names);
102 				level = le32_to_cpu(opinfo->o_lease->state);
103 			} else {
104 				const_names = ksmbd_oplock_const_names;
105 				count = ARRAY_SIZE(ksmbd_oplock_const_names);
106 				level = opinfo->level;
107 			}
108 			rcu_read_unlock();
109 			ksmbd_proc_show_const_name(m, " %-15s",
110 						   const_names, count, level);
111 		} else {
112 			rcu_read_unlock();
113 			seq_printf(m, " %-15s", " ");
114 		}
115 
116 		seq_printf(m, " %#010x %#010x %s\n",
117 			   le32_to_cpu(fp->daccess),
118 			   le32_to_cpu(fp->saccess),
119 			   fp->filp->f_path.dentry->d_name.name);
120 	}
121 	read_unlock(&global_ft.lock);
122 	return 0;
123 }
124 
125 static int create_proc_files(void)
126 {
127 	ksmbd_proc_create("files", proc_show_files, NULL);
128 	return 0;
129 }
130 #else
131 static int create_proc_files(void) { return 0; }
132 #endif
133 
134 static bool durable_scavenger_running;
135 static DEFINE_MUTEX(durable_scavenger_lock);
136 static wait_queue_head_t dh_wq;
137 
138 void ksmbd_set_fd_limit(unsigned long limit)
139 {
140 	limit = min(limit, get_max_files());
141 	atomic_long_set(&fd_limit, limit);
142 }
143 
144 static bool fd_limit_depleted(void)
145 {
146 	long v = atomic_long_dec_return(&fd_limit);
147 
148 	if (v >= 0)
149 		return false;
150 	atomic_long_inc(&fd_limit);
151 	return true;
152 }
153 
154 static void fd_limit_close(void)
155 {
156 	atomic_long_inc(&fd_limit);
157 }
158 
159 /*
160  * INODE hash
161  */
162 
163 static unsigned long inode_hash(struct super_block *sb, unsigned long hashval)
164 {
165 	unsigned long tmp;
166 
167 	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
168 		L1_CACHE_BYTES;
169 	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> inode_hash_shift);
170 	return tmp & inode_hash_mask;
171 }
172 
173 static struct ksmbd_inode *__ksmbd_inode_lookup(struct dentry *de)
174 {
175 	struct hlist_head *head = inode_hashtable +
176 		inode_hash(d_inode(de)->i_sb, (unsigned long)de);
177 	struct ksmbd_inode *ci = NULL, *ret_ci = NULL;
178 
179 	hlist_for_each_entry(ci, head, m_hash) {
180 		if (ci->m_de == de) {
181 			if (atomic_inc_not_zero(&ci->m_count))
182 				ret_ci = ci;
183 			break;
184 		}
185 	}
186 	return ret_ci;
187 }
188 
189 static struct ksmbd_inode *ksmbd_inode_lookup(struct ksmbd_file *fp)
190 {
191 	return __ksmbd_inode_lookup(fp->filp->f_path.dentry);
192 }
193 
194 struct ksmbd_inode *ksmbd_inode_lookup_lock(struct dentry *d)
195 {
196 	struct ksmbd_inode *ci;
197 
198 	read_lock(&inode_hash_lock);
199 	ci = __ksmbd_inode_lookup(d);
200 	read_unlock(&inode_hash_lock);
201 
202 	return ci;
203 }
204 
205 int ksmbd_query_inode_status(struct dentry *dentry)
206 {
207 	struct ksmbd_inode *ci;
208 	int ret = KSMBD_INODE_STATUS_UNKNOWN;
209 
210 	read_lock(&inode_hash_lock);
211 	ci = __ksmbd_inode_lookup(dentry);
212 	read_unlock(&inode_hash_lock);
213 	if (!ci)
214 		return ret;
215 
216 	down_read(&ci->m_lock);
217 	if (ci->m_flags & S_DEL_PENDING)
218 		ret = KSMBD_INODE_STATUS_PENDING_DELETE;
219 	else
220 		ret = KSMBD_INODE_STATUS_OK;
221 	up_read(&ci->m_lock);
222 
223 	ksmbd_inode_put(ci);
224 	return ret;
225 }
226 
227 bool ksmbd_inode_pending_delete(struct ksmbd_file *fp)
228 {
229 	struct ksmbd_inode *ci = fp->f_ci;
230 	int ret;
231 
232 	down_read(&ci->m_lock);
233 	ret = (ci->m_flags & S_DEL_PENDING);
234 	up_read(&ci->m_lock);
235 
236 	return ret;
237 }
238 
239 void ksmbd_set_inode_pending_delete(struct ksmbd_file *fp)
240 {
241 	struct ksmbd_inode *ci = fp->f_ci;
242 
243 	down_write(&ci->m_lock);
244 	ci->m_flags |= S_DEL_PENDING;
245 	up_write(&ci->m_lock);
246 }
247 
248 void ksmbd_clear_inode_pending_delete(struct ksmbd_file *fp)
249 {
250 	struct ksmbd_inode *ci = fp->f_ci;
251 
252 	down_write(&ci->m_lock);
253 	ci->m_flags &= ~S_DEL_PENDING;
254 	up_write(&ci->m_lock);
255 }
256 
257 bool ksmbd_has_stream_without_delete_share(struct ksmbd_file *fp)
258 {
259 	struct ksmbd_file *prev_fp;
260 	struct ksmbd_inode *ci = fp->f_ci;
261 	bool ret = false;
262 
263 	if (ksmbd_stream_fd(fp))
264 		return false;
265 
266 	down_read(&ci->m_lock);
267 	list_for_each_entry(prev_fp, &ci->m_fp_list, node) {
268 		if (prev_fp == fp || !ksmbd_stream_fd(prev_fp))
269 			continue;
270 
271 		if (file_inode(fp->filp) != file_inode(prev_fp->filp))
272 			continue;
273 
274 		if (!(prev_fp->saccess & FILE_SHARE_DELETE_LE)) {
275 			ret = true;
276 			break;
277 		}
278 	}
279 	up_read(&ci->m_lock);
280 
281 	return ret;
282 }
283 
284 void ksmbd_fd_set_delete_on_close(struct ksmbd_file *fp,
285 				  int file_info)
286 {
287 	struct ksmbd_inode *ci = fp->f_ci;
288 
289 	down_write(&ci->m_lock);
290 	if (ksmbd_stream_fd(fp))
291 		ci->m_flags |= S_DEL_ON_CLS_STREAM;
292 	else
293 		ci->m_flags |= S_DEL_ON_CLS;
294 	up_write(&ci->m_lock);
295 }
296 
297 static void ksmbd_inode_hash(struct ksmbd_inode *ci)
298 {
299 	struct hlist_head *b = inode_hashtable +
300 		inode_hash(d_inode(ci->m_de)->i_sb, (unsigned long)ci->m_de);
301 
302 	hlist_add_head(&ci->m_hash, b);
303 }
304 
305 static void ksmbd_inode_unhash(struct ksmbd_inode *ci)
306 {
307 	write_lock(&inode_hash_lock);
308 	hlist_del_init(&ci->m_hash);
309 	write_unlock(&inode_hash_lock);
310 }
311 
312 static int ksmbd_inode_init(struct ksmbd_inode *ci, struct ksmbd_file *fp)
313 {
314 	atomic_set(&ci->m_count, 1);
315 	atomic_set(&ci->op_count, 0);
316 	atomic_set(&ci->sop_count, 0);
317 	ci->m_flags = 0;
318 	ci->m_fattr = 0;
319 	INIT_LIST_HEAD(&ci->m_fp_list);
320 	INIT_LIST_HEAD(&ci->m_op_list);
321 	init_rwsem(&ci->m_lock);
322 	ci->m_de = fp->filp->f_path.dentry;
323 	return 0;
324 }
325 
326 static struct ksmbd_inode *ksmbd_inode_get(struct ksmbd_file *fp)
327 {
328 	struct ksmbd_inode *ci, *tmpci;
329 	int rc;
330 
331 	read_lock(&inode_hash_lock);
332 	ci = ksmbd_inode_lookup(fp);
333 	read_unlock(&inode_hash_lock);
334 	if (ci)
335 		return ci;
336 
337 	ci = kmalloc_obj(struct ksmbd_inode, KSMBD_DEFAULT_GFP);
338 	if (!ci)
339 		return NULL;
340 
341 	rc = ksmbd_inode_init(ci, fp);
342 	if (rc) {
343 		pr_err("inode initialized failed\n");
344 		kfree(ci);
345 		return NULL;
346 	}
347 
348 	write_lock(&inode_hash_lock);
349 	tmpci = ksmbd_inode_lookup(fp);
350 	if (!tmpci) {
351 		ksmbd_inode_hash(ci);
352 	} else {
353 		kfree(ci);
354 		ci = tmpci;
355 	}
356 	write_unlock(&inode_hash_lock);
357 	return ci;
358 }
359 
360 static void ksmbd_inode_free(struct ksmbd_inode *ci)
361 {
362 	ksmbd_inode_unhash(ci);
363 	kfree(ci);
364 }
365 
366 void ksmbd_inode_put(struct ksmbd_inode *ci)
367 {
368 	if (atomic_dec_and_test(&ci->m_count))
369 		ksmbd_inode_free(ci);
370 }
371 
372 int __init ksmbd_inode_hash_init(void)
373 {
374 	unsigned int loop;
375 	unsigned long numentries = 16384;
376 	unsigned long bucketsize = sizeof(struct hlist_head);
377 	unsigned long size;
378 
379 	inode_hash_shift = ilog2(numentries);
380 	inode_hash_mask = (1 << inode_hash_shift) - 1;
381 
382 	size = bucketsize << inode_hash_shift;
383 
384 	/* init master fp hash table */
385 	inode_hashtable = vmalloc(size);
386 	if (!inode_hashtable)
387 		return -ENOMEM;
388 
389 	for (loop = 0; loop < (1U << inode_hash_shift); loop++)
390 		INIT_HLIST_HEAD(&inode_hashtable[loop]);
391 	return 0;
392 }
393 
394 void ksmbd_release_inode_hash(void)
395 {
396 	vfree(inode_hashtable);
397 }
398 
399 static void __ksmbd_inode_close(struct ksmbd_file *fp)
400 {
401 	struct ksmbd_inode *ci = fp->f_ci;
402 	int err;
403 	struct file *filp;
404 
405 	filp = fp->filp;
406 
407 	if (ksmbd_stream_fd(fp)) {
408 		bool remove_stream_xattr = false;
409 
410 		down_write(&ci->m_lock);
411 		if (ci->m_flags & S_DEL_ON_CLS_STREAM) {
412 			ci->m_flags &= ~S_DEL_ON_CLS_STREAM;
413 			remove_stream_xattr = true;
414 		}
415 		up_write(&ci->m_lock);
416 
417 		if (remove_stream_xattr) {
418 			const struct cred *saved_cred;
419 
420 			saved_cred = override_creds(filp->f_cred);
421 			err = ksmbd_vfs_remove_xattr(file_mnt_idmap(filp),
422 						     &filp->f_path,
423 						     fp->stream.name,
424 						     true);
425 			revert_creds(saved_cred);
426 			if (err)
427 				pr_err("remove xattr failed : %s\n",
428 				       fp->stream.name);
429 		}
430 	}
431 
432 	down_write(&ci->m_lock);
433 	/* Promote S_DEL_ON_CLS to S_DEL_PENDING when close */
434 	if (ci->m_flags & S_DEL_ON_CLS) {
435 		ci->m_flags &= ~S_DEL_ON_CLS;
436 		ci->m_flags |= S_DEL_PENDING;
437 	}
438 	up_write(&ci->m_lock);
439 
440 	if (atomic_dec_and_test(&ci->m_count)) {
441 		bool do_unlink = false;
442 
443 		down_write(&ci->m_lock);
444 		if (ci->m_flags & S_DEL_PENDING) {
445 			ci->m_flags &= ~S_DEL_PENDING;
446 			do_unlink = true;
447 		}
448 		up_write(&ci->m_lock);
449 
450 		if (do_unlink)
451 			ksmbd_vfs_unlink(filp);
452 
453 		ksmbd_inode_free(ci);
454 	}
455 }
456 
457 static void __ksmbd_remove_durable_fd(struct ksmbd_file *fp)
458 {
459 	if (!has_file_id(fp->persistent_id))
460 		return;
461 
462 	idr_remove(global_ft.idr, fp->persistent_id);
463 	/*
464 	 * Clear persistent_id so a later __ksmbd_close_fd() that runs from a
465 	 * delayed putter (e.g. when a concurrent ksmbd_lookup_fd_inode()
466 	 * walker held the final reference) does not re-issue idr_remove() on
467 	 * an id that idr_alloc_cyclic() may have already handed out to a new
468 	 * durable handle.
469 	 */
470 	fp->persistent_id = KSMBD_NO_FID;
471 }
472 
473 static void ksmbd_remove_durable_fd(struct ksmbd_file *fp)
474 {
475 	write_lock(&global_ft.lock);
476 	__ksmbd_remove_durable_fd(fp);
477 	write_unlock(&global_ft.lock);
478 	if (waitqueue_active(&dh_wq))
479 		wake_up(&dh_wq);
480 }
481 
482 static void __ksmbd_remove_fd(struct ksmbd_file_table *ft, struct ksmbd_file *fp)
483 {
484 	down_write(&fp->f_ci->m_lock);
485 	list_del_init(&fp->node);
486 	up_write(&fp->f_ci->m_lock);
487 
488 	if (!has_file_id(fp->volatile_id))
489 		return;
490 
491 	write_lock(&ft->lock);
492 	idr_remove(ft->idr, fp->volatile_id);
493 	write_unlock(&ft->lock);
494 }
495 
496 static void __ksmbd_close_fd(struct ksmbd_file_table *ft, struct ksmbd_file *fp)
497 {
498 	struct file *filp;
499 	struct ksmbd_lock *smb_lock, *tmp_lock;
500 
501 	fd_limit_close();
502 	ksmbd_remove_durable_fd(fp);
503 	if (ft)
504 		__ksmbd_remove_fd(ft, fp);
505 
506 	close_id_del_oplock(fp);
507 	filp = fp->filp;
508 
509 	__ksmbd_inode_close(fp);
510 	if (!IS_ERR_OR_NULL(filp))
511 		fput(filp);
512 
513 	/* because the reference count of fp is 0, it is guaranteed that
514 	 * there are not accesses to fp->lock_list.
515 	 */
516 	list_for_each_entry_safe(smb_lock, tmp_lock, &fp->lock_list, flist) {
517 		struct ksmbd_conn *conn = smb_lock->conn;
518 
519 		if (conn) {
520 			spin_lock(&conn->llist_lock);
521 			list_del_init(&smb_lock->clist);
522 			smb_lock->conn = NULL;
523 			spin_unlock(&conn->llist_lock);
524 			ksmbd_conn_put(conn);
525 		}
526 
527 		list_del(&smb_lock->flist);
528 		locks_free_lock(smb_lock->fl);
529 		kfree(smb_lock);
530 	}
531 
532 	/*
533 	 * Drop fp's strong reference on conn (taken in ksmbd_open_fd() /
534 	 * ksmbd_reopen_durable_fd()).  Durable fps that reached the
535 	 * scavenger have already had fp->conn cleared by session_fd_check(),
536 	 * in which case there is nothing to drop here.
537 	 */
538 	if (fp->conn) {
539 		ksmbd_conn_put(fp->conn);
540 		fp->conn = NULL;
541 	}
542 
543 	if (ksmbd_stream_fd(fp))
544 		kfree(fp->stream.name);
545 	kfree(fp->owner.name);
546 
547 	kmem_cache_free(filp_cache, fp);
548 }
549 
550 /**
551  * ksmbd_close_disconnected_durable_delete_on_close() - drop a delete-on-close
552  *	file kept present only by disconnected durable handles
553  * @dentry:	dentry of the file being opened
554  *
555  * A durable handle opened with delete-on-close is preserved across a
556  * disconnect so it can be reclaimed by a durable reconnect.  When a new
557  * (non-reconnect) open arrives for the same name instead, the disconnected
558  * handle has to give way.  Close such handles so their delete-on-close is
559  * applied and the file is removed once the last handle is gone, letting the
560  * new open create a fresh file.
561  *
562  * The caller's inode reference is dropped before closing so that the final
563  * close can promote S_DEL_ON_CLS to S_DEL_PENDING and unlink the file.
564  *
565  * Return:	true if a disconnected durable handle was closed.
566  */
567 bool ksmbd_close_disconnected_durable_delete_on_close(struct dentry *dentry)
568 {
569 	struct ksmbd_inode *ci;
570 	struct ksmbd_file *fp, *tmp;
571 	LIST_HEAD(dispose);
572 	bool closed = false;
573 
574 	ci = ksmbd_inode_lookup_lock(dentry);
575 	if (!ci)
576 		return false;
577 
578 	down_write(&ci->m_lock);
579 	if (ci->m_flags & (S_DEL_ON_CLS | S_DEL_ON_CLS_STREAM | S_DEL_PENDING)) {
580 		list_for_each_entry_safe(fp, tmp, &ci->m_fp_list, node) {
581 			if (fp->conn || !fp->is_durable ||
582 			    fp->f_state != FP_INITED)
583 				continue;
584 
585 			/*
586 			 * Claim the close before unlinking fp from m_fp_list.
587 			 * refcount == 1 means only the durable lifetime ref is
588 			 * left. Add a transient ref so final close can drop both.
589 			 */
590 			write_lock(&global_ft.lock);
591 			if (atomic_read(&fp->refcount) == 1) {
592 				atomic_inc(&fp->refcount);
593 				__ksmbd_remove_durable_fd(fp);
594 				ksmbd_mark_fp_closed(fp);
595 				list_move_tail(&fp->node, &dispose);
596 			}
597 			write_unlock(&global_ft.lock);
598 		}
599 	}
600 	up_write(&ci->m_lock);
601 
602 	/*
603 	 * Drop our lookup reference before closing so the last __ksmbd_close_fd()
604 	 * can drop m_count to zero and unlink the delete-on-close file.  The
605 	 * collected handles still hold the transient reference taken above, so
606 	 * ci stays valid until they are closed below.
607 	 */
608 	ksmbd_inode_put(ci);
609 
610 	while (!list_empty(&dispose)) {
611 		fp = list_first_entry(&dispose, struct ksmbd_file, node);
612 		list_del_init(&fp->node);
613 		if (atomic_sub_and_test(2, &fp->refcount)) {
614 			__ksmbd_close_fd(NULL, fp);
615 			closed = true;
616 		}
617 	}
618 
619 	return closed;
620 }
621 
622 static struct ksmbd_file *ksmbd_fp_get(struct ksmbd_file *fp)
623 {
624 	if (fp->f_state != FP_INITED)
625 		return NULL;
626 
627 	if (!atomic_inc_not_zero(&fp->refcount))
628 		return NULL;
629 	return fp;
630 }
631 
632 static struct ksmbd_file *__ksmbd_lookup_fd(struct ksmbd_file_table *ft,
633 					    u64 id)
634 {
635 	struct ksmbd_file *fp;
636 
637 	if (!has_file_id(id))
638 		return NULL;
639 
640 	read_lock(&ft->lock);
641 	fp = idr_find(ft->idr, id);
642 	if (fp)
643 		fp = ksmbd_fp_get(fp);
644 	read_unlock(&ft->lock);
645 	return fp;
646 }
647 
648 static void __put_fd_final(struct ksmbd_work *work, struct ksmbd_file *fp)
649 {
650 	/*
651 	 * Detached durable fp -- session_fd_check() cleared fp->conn at
652 	 * preserve, so this fp is no longer tracked by any conn's
653 	 * stats.open_files_count.  This happens when
654 	 * ksmbd_scavenger_dispose_dh() hands the final close off to an
655 	 * m_fp_list walker (e.g. ksmbd_lookup_fd_inode()) whose work->conn
656 	 * is unrelated to the conn that originally opened the handle; close
657 	 * via the NULL-ft path so we do not underflow that unrelated
658 	 * counter.
659 	 */
660 	if (!fp->conn) {
661 		__ksmbd_close_fd(NULL, fp);
662 		return;
663 	}
664 	__ksmbd_close_fd(&work->sess->file_table, fp);
665 	atomic_dec(&work->conn->stats.open_files_count);
666 }
667 
668 static void set_close_state_blocked_works(struct ksmbd_file *fp)
669 {
670 	struct ksmbd_work *cancel_work;
671 
672 	spin_lock(&fp->f_lock);
673 	list_for_each_entry(cancel_work, &fp->blocked_works,
674 				 fp_entry) {
675 		cancel_work->state = KSMBD_WORK_CLOSED;
676 		cancel_work->cancel_fn(cancel_work->cancel_argv);
677 	}
678 	spin_unlock(&fp->f_lock);
679 }
680 
681 int ksmbd_close_fd(struct ksmbd_work *work, u64 id)
682 {
683 	struct ksmbd_file	*fp;
684 	struct ksmbd_file_table	*ft;
685 	bool closed = false;
686 
687 	if (!has_file_id(id))
688 		return 0;
689 
690 	ft = &work->sess->file_table;
691 	write_lock(&ft->lock);
692 	fp = idr_find(ft->idr, id);
693 	if (fp) {
694 		set_close_state_blocked_works(fp);
695 
696 		if (fp->f_state != FP_INITED)
697 			fp = NULL;
698 		else {
699 			fp->f_state = FP_CLOSED;
700 			closed = true;
701 			if (!atomic_dec_and_test(&fp->refcount))
702 				fp = NULL;
703 		}
704 	}
705 	write_unlock(&ft->lock);
706 
707 	if (!fp)
708 		return closed ? 0 : -EINVAL;
709 
710 	__put_fd_final(work, fp);
711 	return 0;
712 }
713 
714 void ksmbd_fd_put(struct ksmbd_work *work, struct ksmbd_file *fp)
715 {
716 	if (!fp)
717 		return;
718 
719 	if (!atomic_dec_and_test(&fp->refcount))
720 		return;
721 	__put_fd_final(work, fp);
722 }
723 
724 static bool __sanity_check(struct ksmbd_tree_connect *tcon, struct ksmbd_file *fp)
725 {
726 	if (!fp)
727 		return false;
728 	if (fp->tcon != tcon)
729 		return false;
730 	return true;
731 }
732 
733 struct ksmbd_file *ksmbd_lookup_foreign_fd(struct ksmbd_work *work, u64 id)
734 {
735 	return __ksmbd_lookup_fd(&work->sess->file_table, id);
736 }
737 
738 struct ksmbd_file *ksmbd_lookup_fd_fast(struct ksmbd_work *work, u64 id)
739 {
740 	struct ksmbd_file *fp = __ksmbd_lookup_fd(&work->sess->file_table, id);
741 
742 	if (__sanity_check(work->tcon, fp))
743 		return fp;
744 
745 	ksmbd_fd_put(work, fp);
746 	return NULL;
747 }
748 
749 struct ksmbd_file *ksmbd_lookup_fd_slow(struct ksmbd_work *work, u64 id,
750 					u64 pid)
751 {
752 	struct ksmbd_file *fp;
753 
754 	if (!has_file_id(id)) {
755 		id = work->compound_fid;
756 		pid = work->compound_pfid;
757 	}
758 
759 	fp = __ksmbd_lookup_fd(&work->sess->file_table, id);
760 	if (!__sanity_check(work->tcon, fp)) {
761 		ksmbd_fd_put(work, fp);
762 		return NULL;
763 	}
764 	if (fp->persistent_id != pid) {
765 		ksmbd_fd_put(work, fp);
766 		return NULL;
767 	}
768 	return fp;
769 }
770 
771 struct ksmbd_file *ksmbd_lookup_global_fd(unsigned long long id)
772 {
773 	return __ksmbd_lookup_fd(&global_ft, id);
774 }
775 
776 struct ksmbd_file *ksmbd_lookup_durable_fd(unsigned long long id)
777 {
778 	struct ksmbd_file *fp;
779 
780 	fp = __ksmbd_lookup_fd(&global_ft, id);
781 	if (fp && (fp->durable_reconnect_disabled ||
782 		   fp->conn ||
783 		   (fp->durable_scavenger_timeout &&
784 		    (fp->durable_scavenger_timeout <
785 		     jiffies_to_msecs(jiffies))))) {
786 		ksmbd_put_durable_fd(fp);
787 		fp = NULL;
788 	}
789 
790 	return fp;
791 }
792 
793 void ksmbd_put_durable_fd(struct ksmbd_file *fp)
794 {
795 	if (!atomic_dec_and_test(&fp->refcount))
796 		return;
797 
798 	__ksmbd_close_fd(NULL, fp);
799 }
800 
801 bool ksmbd_has_other_active_fd(struct ksmbd_file *fp)
802 {
803 	struct ksmbd_file *lfp;
804 	struct ksmbd_inode *ci = fp->f_ci;
805 	bool ret = false;
806 
807 	down_read(&ci->m_lock);
808 	list_for_each_entry(lfp, &ci->m_fp_list, node) {
809 		if (lfp == fp)
810 			continue;
811 
812 		if (lfp->f_state == FP_INITED &&
813 		    (READ_ONCE(lfp->conn) || READ_ONCE(lfp->tcon))) {
814 			ret = true;
815 			break;
816 		}
817 	}
818 	up_read(&ci->m_lock);
819 
820 	return ret;
821 }
822 
823 static struct ksmbd_file *ksmbd_lookup_fd_app_instance_id(char *app_instance_id)
824 {
825 	struct ksmbd_file *fp = NULL;
826 	unsigned int id;
827 
828 	if (!memchr_inv(app_instance_id, 0, SMB2_CREATE_GUID_SIZE))
829 		return NULL;
830 
831 	read_lock(&global_ft.lock);
832 	idr_for_each_entry(global_ft.idr, fp, id) {
833 		if (!memcmp(fp->app_instance_id, app_instance_id,
834 			    SMB2_CREATE_GUID_SIZE)) {
835 			fp = ksmbd_fp_get(fp);
836 			break;
837 		}
838 	}
839 	read_unlock(&global_ft.lock);
840 
841 	return fp;
842 }
843 
844 int ksmbd_close_fd_app_instance_id(char *app_instance_id)
845 {
846 	struct ksmbd_file_table *ft;
847 	struct ksmbd_file *fp;
848 	struct oplock_info *opinfo;
849 	int n_to_drop = 0;
850 
851 	fp = ksmbd_lookup_fd_app_instance_id(app_instance_id);
852 	if (!fp)
853 		return 0;
854 
855 	opinfo = opinfo_get(fp);
856 	if (!opinfo)
857 		goto out;
858 
859 	down_read(&fp->f_ci->m_lock);
860 	if (!opinfo->conn) {
861 		up_read(&fp->f_ci->m_lock);
862 		goto out;
863 	}
864 
865 	ft = &opinfo->sess->file_table;
866 	write_lock(&ft->lock);
867 	if (fp->f_state == FP_INITED && has_file_id(fp->volatile_id)) {
868 		idr_remove(ft->idr, fp->volatile_id);
869 		fp->volatile_id = KSMBD_NO_FID;
870 		n_to_drop = ksmbd_mark_fp_closed(fp);
871 	}
872 	write_unlock(&ft->lock);
873 	up_read(&fp->f_ci->m_lock);
874 	opinfo_put(opinfo);
875 	opinfo = NULL;
876 
877 	if (!n_to_drop)
878 		goto out;
879 
880 	down_write(&fp->f_ci->m_lock);
881 	list_del_init(&fp->node);
882 	up_write(&fp->f_ci->m_lock);
883 
884 	if (atomic_sub_and_test(n_to_drop, &fp->refcount)) {
885 		if (fp->conn)
886 			atomic_dec(&fp->conn->stats.open_files_count);
887 		__ksmbd_close_fd(NULL, fp);
888 	}
889 	return 0;
890 
891 out:
892 	if (opinfo)
893 		opinfo_put(opinfo);
894 	ksmbd_put_durable_fd(fp);
895 	return 0;
896 }
897 
898 int ksmbd_invalidate_durable_fd(unsigned long long id)
899 {
900 	struct ksmbd_file *fp;
901 
902 	fp = ksmbd_lookup_global_fd(id);
903 	if (!fp)
904 		return -ENOENT;
905 
906 	fp->durable_reconnect_disabled = true;
907 
908 	if (fp->conn) {
909 		ksmbd_put_durable_fd(fp);
910 		return -ENOENT;
911 	}
912 
913 	fp->durable_timeout = 1;
914 	fp->durable_scavenger_timeout = jiffies_to_msecs(jiffies);
915 	ksmbd_put_durable_fd(fp);
916 	if (waitqueue_active(&dh_wq))
917 		wake_up(&dh_wq);
918 
919 	return -ENOENT;
920 }
921 
922 struct ksmbd_file *ksmbd_lookup_fd_cguid(char *cguid)
923 {
924 	struct ksmbd_file	*fp = NULL;
925 	unsigned int		id;
926 
927 	read_lock(&global_ft.lock);
928 	idr_for_each_entry(global_ft.idr, fp, id) {
929 		if (!memcmp(fp->create_guid,
930 			    cguid,
931 			    SMB2_CREATE_GUID_SIZE)) {
932 			fp = ksmbd_fp_get(fp);
933 			break;
934 		}
935 	}
936 	read_unlock(&global_ft.lock);
937 
938 	return fp;
939 }
940 
941 struct ksmbd_file *ksmbd_lookup_fd_inode(struct dentry *dentry)
942 {
943 	struct ksmbd_file	*lfp;
944 	struct ksmbd_inode	*ci;
945 	struct inode		*inode = d_inode(dentry);
946 
947 	read_lock(&inode_hash_lock);
948 	ci = __ksmbd_inode_lookup(dentry);
949 	read_unlock(&inode_hash_lock);
950 	if (!ci)
951 		return NULL;
952 
953 	down_read(&ci->m_lock);
954 	list_for_each_entry(lfp, &ci->m_fp_list, node) {
955 		if (inode == file_inode(lfp->filp)) {
956 			lfp = ksmbd_fp_get(lfp);
957 			up_read(&ci->m_lock);
958 			ksmbd_inode_put(ci);
959 			return lfp;
960 		}
961 	}
962 	up_read(&ci->m_lock);
963 	ksmbd_inode_put(ci);
964 	return NULL;
965 }
966 
967 bool ksmbd_has_open_files(struct dentry *dentry)
968 {
969 	struct ksmbd_file *fp;
970 	unsigned int id;
971 	bool ret = false;
972 
973 	read_lock(&global_ft.lock);
974 	idr_for_each_entry(global_ft.idr, fp, id) {
975 		struct dentry *fp_dentry = fp->filp->f_path.dentry;
976 
977 		if (fp->f_state != FP_INITED)
978 			continue;
979 		if (fp_dentry == dentry)
980 			continue;
981 		if (is_subdir(fp_dentry, dentry)) {
982 			ret = true;
983 			break;
984 		}
985 	}
986 	read_unlock(&global_ft.lock);
987 
988 	return ret;
989 }
990 
991 #define OPEN_ID_TYPE_VOLATILE_ID	(0)
992 #define OPEN_ID_TYPE_PERSISTENT_ID	(1)
993 
994 static void __open_id_set(struct ksmbd_file *fp, u64 id, int type)
995 {
996 	if (type == OPEN_ID_TYPE_VOLATILE_ID)
997 		fp->volatile_id = id;
998 	if (type == OPEN_ID_TYPE_PERSISTENT_ID)
999 		fp->persistent_id = id;
1000 }
1001 
1002 static int __open_id(struct ksmbd_file_table *ft, struct ksmbd_file *fp,
1003 		     int type)
1004 {
1005 	u64			id = 0;
1006 	int			ret;
1007 
1008 	if (type == OPEN_ID_TYPE_VOLATILE_ID && fd_limit_depleted()) {
1009 		__open_id_set(fp, KSMBD_NO_FID, type);
1010 		return -EMFILE;
1011 	}
1012 
1013 	idr_preload(KSMBD_DEFAULT_GFP);
1014 	write_lock(&ft->lock);
1015 	ret = idr_alloc_cyclic(ft->idr, fp, KSMBD_START_FID, INT_MAX - 1,
1016 			       GFP_NOWAIT);
1017 	if (ret >= 0) {
1018 		id = ret;
1019 		ret = 0;
1020 	} else {
1021 		id = KSMBD_NO_FID;
1022 		fd_limit_close();
1023 	}
1024 
1025 	__open_id_set(fp, id, type);
1026 	write_unlock(&ft->lock);
1027 	idr_preload_end();
1028 	return ret;
1029 }
1030 
1031 unsigned int ksmbd_open_durable_fd(struct ksmbd_file *fp)
1032 {
1033 	__open_id(&global_ft, fp, OPEN_ID_TYPE_PERSISTENT_ID);
1034 	return fp->persistent_id;
1035 }
1036 
1037 struct ksmbd_file *ksmbd_open_fd(struct ksmbd_work *work, struct file *filp)
1038 {
1039 	struct ksmbd_file *fp;
1040 	int ret;
1041 
1042 	fp = kmem_cache_zalloc(filp_cache, KSMBD_DEFAULT_GFP);
1043 	if (!fp) {
1044 		pr_err("Failed to allocate memory\n");
1045 		return ERR_PTR(-ENOMEM);
1046 	}
1047 
1048 	INIT_LIST_HEAD(&fp->blocked_works);
1049 	INIT_LIST_HEAD(&fp->node);
1050 	INIT_LIST_HEAD(&fp->lock_list);
1051 	spin_lock_init(&fp->f_lock);
1052 	mutex_init(&fp->readdir_lock);
1053 	atomic_set(&fp->refcount, 1);
1054 
1055 	fp->filp		= filp;
1056 	/*
1057 	 * fp owns a strong reference on fp->conn for as long as fp->conn is
1058 	 * non-NULL, so session_fd_check() and __ksmbd_close_fd() never
1059 	 * dereference a dangling pointer.  Paired with ksmbd_conn_put() in
1060 	 * session_fd_check() (durable preserve), in __ksmbd_close_fd()
1061 	 * (final close), and on the error paths below.
1062 	 */
1063 	fp->conn		= ksmbd_conn_get(work->conn);
1064 	fp->tcon		= work->tcon;
1065 	fp->volatile_id		= KSMBD_NO_FID;
1066 	fp->persistent_id	= KSMBD_NO_FID;
1067 	fp->f_state		= FP_NEW;
1068 	fp->f_ci		= ksmbd_inode_get(fp);
1069 
1070 	if (!fp->f_ci) {
1071 		ret = -ENOMEM;
1072 		goto err_out;
1073 	}
1074 
1075 	ret = __open_id(&work->sess->file_table, fp, OPEN_ID_TYPE_VOLATILE_ID);
1076 	if (ret) {
1077 		ksmbd_inode_put(fp->f_ci);
1078 		goto err_out;
1079 	}
1080 
1081 	atomic_inc(&work->conn->stats.open_files_count);
1082 	return fp;
1083 
1084 err_out:
1085 	/* fp->conn was set and refcounted before every branch here. */
1086 	ksmbd_conn_put(fp->conn);
1087 	kmem_cache_free(filp_cache, fp);
1088 	return ERR_PTR(ret);
1089 }
1090 
1091 /**
1092  * ksmbd_update_fstate() - update an fp state under the file-table lock
1093  * @ft: file table that publishes @fp's volatile id
1094  * @fp: file pointer to update
1095  * @state: new state
1096  *
1097  * Return: 0 on success.  The FP_NEW -> FP_INITED transition is special:
1098  * -ENOENT if teardown already unpublished @fp by advancing the state or
1099  * clearing the volatile id.  Other state updates preserve the historical
1100  * fire-and-forget behavior.
1101  */
1102 int ksmbd_update_fstate(struct ksmbd_file_table *ft, struct ksmbd_file *fp,
1103 			unsigned int state)
1104 {
1105 	int ret;
1106 
1107 	if (!fp)
1108 		return -ENOENT;
1109 
1110 	write_lock(&ft->lock);
1111 	if (state == FP_INITED &&
1112 	    (fp->f_state != FP_NEW || !has_file_id(fp->volatile_id))) {
1113 		ret = -ENOENT;
1114 	} else {
1115 		fp->f_state = state;
1116 		ret = 0;
1117 	}
1118 	write_unlock(&ft->lock);
1119 
1120 	return ret;
1121 }
1122 
1123 /*
1124  * ksmbd_mark_fp_closed() - mark fp closed under ft->lock and return how many
1125  * refs the teardown path owns.
1126  *
1127  * FP_INITED has a normal idr-owned reference, so teardown owns both that
1128  * reference and the transient lookup reference.  FP_NEW is still owned by the
1129  * in-flight opener/reopener, which will drop the original reference after
1130  * ksmbd_update_fstate(..., FP_INITED) observes the cleared volatile id.
1131  * FP_CLOSED on entry means an earlier ksmbd_close_fd() already consumed the
1132  * idr-owned ref.
1133  */
1134 static int ksmbd_mark_fp_closed(struct ksmbd_file *fp)
1135 {
1136 	if (fp->f_state == FP_INITED) {
1137 		set_close_state_blocked_works(fp);
1138 		fp->f_state = FP_CLOSED;
1139 		return 2;
1140 	}
1141 
1142 	return 1;
1143 }
1144 
1145 static int
1146 __close_file_table_ids(struct ksmbd_session *sess,
1147 		       struct ksmbd_tree_connect *tcon,
1148 		       bool (*skip)(struct ksmbd_tree_connect *tcon,
1149 				    struct ksmbd_file *fp,
1150 				    struct ksmbd_user *user),
1151 		       bool skip_preserves_fp)
1152 {
1153 	struct ksmbd_file_table *ft = &sess->file_table;
1154 	struct ksmbd_file *fp;
1155 	unsigned int id = 0;
1156 	int num = 0;
1157 
1158 	while (1) {
1159 		int n_to_drop;
1160 
1161 		write_lock(&ft->lock);
1162 		fp = idr_get_next(ft->idr, &id);
1163 		if (!fp) {
1164 			write_unlock(&ft->lock);
1165 			break;
1166 		}
1167 		if (!atomic_inc_not_zero(&fp->refcount)) {
1168 			id++;
1169 			write_unlock(&ft->lock);
1170 			continue;
1171 		}
1172 
1173 		if (skip_preserves_fp) {
1174 			/*
1175 			 * Session teardown: skip() is session_fd_check(),
1176 			 * which may sleep and mutates fp->conn / fp->tcon /
1177 			 * fp->volatile_id when it chooses to preserve fp
1178 			 * for durable reconnect.  Unpublish fp from the
1179 			 * session idr here, under ft->lock, so that
1180 			 * __ksmbd_lookup_fd() through this session cannot
1181 			 * grant a new ksmbd_fp_get() reference to an fp
1182 			 * whose fields are about to be rewritten outside
1183 			 * the lock.  Durable reconnect still reaches fp via
1184 			 * global_ft.
1185 			 */
1186 			idr_remove(ft->idr, id);
1187 			fp->durable_volatile_id = fp->volatile_id;
1188 			fp->volatile_id = KSMBD_NO_FID;
1189 			write_unlock(&ft->lock);
1190 
1191 			if (skip(tcon, fp, sess->user)) {
1192 				/*
1193 				 * session_fd_check() has converted fp to
1194 				 * durable-preserve state and cleared its
1195 				 * per-conn fields.  fp is already unpublished
1196 				 * above; the original idr-owned ref keeps it
1197 				 * alive for the durable scavenger.  Drop only
1198 				 * the transient ref.  atomic_dec() is safe --
1199 				 * atomic_inc_not_zero() succeeded on a
1200 				 * positive value and we added one more, so
1201 				 * refcount cannot be zero here.
1202 				 */
1203 				atomic_dec(&fp->refcount);
1204 				id++;
1205 				continue;
1206 			}
1207 
1208 			/*
1209 			 * Keep the close-state decision under the same lock
1210 			 * observed by ksmbd_update_fstate(), which is how an
1211 			 * in-flight FP_NEW opener learns that teardown has
1212 			 * cleared its volatile id.
1213 			 */
1214 			write_lock(&ft->lock);
1215 			n_to_drop = ksmbd_mark_fp_closed(fp);
1216 			write_unlock(&ft->lock);
1217 		} else {
1218 			/*
1219 			 * Tree teardown: skip() is tree_conn_fd_check(), a
1220 			 * cheap pointer compare that doesn't sleep and has
1221 			 * no side effects, so keep the skip decision plus
1222 			 * the unpublish-and-mark-closed sequence atomic
1223 			 * under ft->lock.  fps belonging to other tree
1224 			 * connects (skip() == true) stay fully published in
1225 			 * the session idr with no lock window.
1226 			 */
1227 			if (skip(tcon, fp, sess->user)) {
1228 				atomic_dec(&fp->refcount);
1229 				write_unlock(&ft->lock);
1230 				id++;
1231 				continue;
1232 			}
1233 			idr_remove(ft->idr, id);
1234 			fp->volatile_id = KSMBD_NO_FID;
1235 			n_to_drop = ksmbd_mark_fp_closed(fp);
1236 			write_unlock(&ft->lock);
1237 		}
1238 
1239 		/*
1240 		 * fp->volatile_id is already cleared to prevent stale idr
1241 		 * removal from a deferred final close.  Remove fp from
1242 		 * m_fp_list here because __ksmbd_remove_fd() will skip the
1243 		 * list unlink when volatile_id is KSMBD_NO_FID.
1244 		 */
1245 		down_write(&fp->f_ci->m_lock);
1246 		list_del_init(&fp->node);
1247 		up_write(&fp->f_ci->m_lock);
1248 
1249 		/*
1250 		 * Drop the references this iteration owns:
1251 		 *
1252 		 *   n_to_drop == 2: we observed FP_INITED and committed
1253 		 *     the FP_CLOSED transition ourselves, so we own the
1254 		 *     transient (+1) and the still-intact idr-owned ref.
1255 		 *
1256 		 *   n_to_drop == 1: either a prior ksmbd_close_fd()
1257 		 *     already consumed the idr-owned ref, or fp was still
1258 		 *     FP_NEW and the in-flight opener/reopener must keep
1259 		 *     the original reference until ksmbd_update_fstate()
1260 		 *     observes the cleared volatile id.
1261 		 *
1262 		 * If we end up as the final putter, finalize fp and
1263 		 * account the open_files_count decrement via the caller's
1264 		 * atomic_sub(num, ...).  Otherwise the remaining user's
1265 		 * ksmbd_fd_put() reaches __put_fd_final(), which does its
1266 		 * own atomic_dec(&open_files_count), so we must not count
1267 		 * this fp here -- doing so would double-decrement the
1268 		 * connection-wide counter.
1269 		 */
1270 		if (atomic_sub_and_test(n_to_drop, &fp->refcount)) {
1271 			__ksmbd_close_fd(NULL, fp);
1272 			num++;
1273 		}
1274 		id++;
1275 	}
1276 
1277 	return num;
1278 }
1279 
1280 static inline bool is_reconnectable(struct ksmbd_file *fp)
1281 {
1282 	struct oplock_info *opinfo = opinfo_get(fp);
1283 	bool reconn = false;
1284 
1285 	if (!opinfo)
1286 		return false;
1287 
1288 	if (opinfo->op_state != OPLOCK_STATE_NONE) {
1289 		opinfo_put(opinfo);
1290 		return false;
1291 	}
1292 
1293 	if (fp->is_resilient || fp->is_persistent)
1294 		reconn = true;
1295 	else if (fp->is_durable && opinfo->is_lease &&
1296 		 opinfo->o_lease->state & SMB2_LEASE_HANDLE_CACHING_LE)
1297 		reconn = true;
1298 
1299 	else if (fp->is_durable && opinfo->level == SMB2_OPLOCK_LEVEL_BATCH)
1300 		reconn = true;
1301 
1302 	opinfo_put(opinfo);
1303 	return reconn;
1304 }
1305 
1306 static bool tree_conn_fd_check(struct ksmbd_tree_connect *tcon,
1307 			       struct ksmbd_file *fp,
1308 			       struct ksmbd_user *user)
1309 {
1310 	return fp->tcon != tcon;
1311 }
1312 
1313 static bool ksmbd_durable_scavenger_alive(void)
1314 {
1315 	if (!durable_scavenger_running)
1316 		return false;
1317 
1318 	if (kthread_should_stop())
1319 		return false;
1320 
1321 	if (idr_is_empty(global_ft.idr))
1322 		return false;
1323 
1324 	return true;
1325 }
1326 
1327 static void ksmbd_scavenger_dispose_dh(struct ksmbd_file *fp)
1328 {
1329 	/*
1330 	 * Durable-preserved fp can remain linked on f_ci->m_fp_list for
1331 	 * share-mode checks.  Unlink it before final close; fp->node is not
1332 	 * available as a scavenger-private list node because re-adding it to
1333 	 * another list corrupts m_fp_list.
1334 	 */
1335 	down_write(&fp->f_ci->m_lock);
1336 	list_del_init(&fp->node);
1337 	up_write(&fp->f_ci->m_lock);
1338 
1339 	/*
1340 	 * Drop both the durable lifetime reference and the transient reference
1341 	 * taken by the scavenger under global_ft.lock.  If a concurrent
1342 	 * ksmbd_lookup_fd_inode() (or any other m_fp_list walker) snatched fp
1343 	 * before the unlink above, that holder owns the final close via
1344 	 * ksmbd_fd_put() -> __ksmbd_close_fd().  Otherwise the scavenger is
1345 	 * the last putter and finalises fp here.
1346 	 */
1347 	if (atomic_sub_and_test(2, &fp->refcount))
1348 		__ksmbd_close_fd(NULL, fp);
1349 }
1350 
1351 static int ksmbd_durable_scavenger(void *dummy)
1352 {
1353 	struct ksmbd_file *fp = NULL;
1354 	struct ksmbd_file *expired_fp;
1355 	unsigned int id;
1356 	unsigned int min_timeout = 1;
1357 	bool found_fp_timeout;
1358 	unsigned long remaining_jiffies;
1359 
1360 	__module_get(THIS_MODULE);
1361 
1362 	set_freezable();
1363 	while (ksmbd_durable_scavenger_alive()) {
1364 		if (try_to_freeze())
1365 			continue;
1366 
1367 		remaining_jiffies = wait_event_interruptible_timeout(dh_wq,
1368 				   ksmbd_durable_scavenger_alive() == false,
1369 				   __msecs_to_jiffies(min_timeout));
1370 		if ((long)remaining_jiffies > 0)
1371 			min_timeout = jiffies_to_msecs(remaining_jiffies);
1372 		else
1373 			min_timeout = DURABLE_HANDLE_MAX_TIMEOUT;
1374 
1375 		do {
1376 			expired_fp = NULL;
1377 			found_fp_timeout = false;
1378 
1379 			write_lock(&global_ft.lock);
1380 			idr_for_each_entry(global_ft.idr, fp, id) {
1381 				unsigned long durable_timeout;
1382 
1383 				if (!fp->durable_timeout)
1384 					continue;
1385 
1386 				if (atomic_read(&fp->refcount) > 1 ||
1387 				    fp->conn)
1388 					continue;
1389 
1390 				found_fp_timeout = true;
1391 				if (fp->durable_scavenger_timeout <=
1392 				    jiffies_to_msecs(jiffies)) {
1393 					__ksmbd_remove_durable_fd(fp);
1394 					/*
1395 					 * Take a transient reference so fp
1396 					 * cannot be freed by an in-flight
1397 					 * ksmbd_lookup_fd_inode() that found
1398 					 * it through f_ci->m_fp_list while we
1399 					 * drop global_ft.lock and reach the
1400 					 * m_fp_list unlink in
1401 					 * ksmbd_scavenger_dispose_dh().
1402 					 */
1403 					atomic_inc(&fp->refcount);
1404 					expired_fp = fp;
1405 					break;
1406 				}
1407 
1408 				durable_timeout =
1409 					fp->durable_scavenger_timeout -
1410 						jiffies_to_msecs(jiffies);
1411 
1412 				if (min_timeout > durable_timeout)
1413 					min_timeout = durable_timeout;
1414 			}
1415 			write_unlock(&global_ft.lock);
1416 
1417 			if (expired_fp)
1418 				ksmbd_scavenger_dispose_dh(expired_fp);
1419 		} while (expired_fp);
1420 
1421 		if (found_fp_timeout == false)
1422 			break;
1423 	}
1424 
1425 	durable_scavenger_running = false;
1426 
1427 	module_put(THIS_MODULE);
1428 
1429 	return 0;
1430 }
1431 
1432 void ksmbd_launch_ksmbd_durable_scavenger(void)
1433 {
1434 	if (!(server_conf.flags & KSMBD_GLOBAL_FLAG_DURABLE_HANDLE))
1435 		return;
1436 
1437 	mutex_lock(&durable_scavenger_lock);
1438 	if (durable_scavenger_running == true) {
1439 		mutex_unlock(&durable_scavenger_lock);
1440 		return;
1441 	}
1442 
1443 	durable_scavenger_running = true;
1444 
1445 	server_conf.dh_task = kthread_run(ksmbd_durable_scavenger,
1446 				     (void *)NULL, "ksmbd-durable-scavenger");
1447 	if (IS_ERR(server_conf.dh_task))
1448 		pr_err("cannot start conn thread, err : %ld\n",
1449 		       PTR_ERR(server_conf.dh_task));
1450 	mutex_unlock(&durable_scavenger_lock);
1451 }
1452 
1453 void ksmbd_stop_durable_scavenger(void)
1454 {
1455 	if (!(server_conf.flags & KSMBD_GLOBAL_FLAG_DURABLE_HANDLE))
1456 		return;
1457 
1458 	mutex_lock(&durable_scavenger_lock);
1459 	if (!durable_scavenger_running) {
1460 		mutex_unlock(&durable_scavenger_lock);
1461 		return;
1462 	}
1463 
1464 	durable_scavenger_running = false;
1465 	if (waitqueue_active(&dh_wq))
1466 		wake_up(&dh_wq);
1467 	mutex_unlock(&durable_scavenger_lock);
1468 	kthread_stop(server_conf.dh_task);
1469 }
1470 
1471 /*
1472  * ksmbd_vfs_copy_durable_owner - Copy owner info for durable reconnect
1473  * @fp: ksmbd file pointer to store owner info
1474  * @user: user pointer to copy from
1475  *
1476  * This function binds the current user's identity to the file handle
1477  * to satisfy MS-SMB2 Step 8 (SecurityContext matching) during reconnect.
1478  *
1479  * Return: 0 on success, or negative error code on failure
1480  */
1481 static int ksmbd_vfs_copy_durable_owner(struct ksmbd_file *fp,
1482 		struct ksmbd_user *user)
1483 {
1484 	char *name;
1485 
1486 	if (!user)
1487 		return -EINVAL;
1488 
1489 	/* Duplicate the user name to ensure identity persistence */
1490 	name = kstrdup(user->name, GFP_KERNEL);
1491 	if (!name)
1492 		return -ENOMEM;
1493 
1494 	spin_lock(&fp->f_lock);
1495 	fp->owner.uid = user->uid;
1496 	fp->owner.gid = user->gid;
1497 	fp->owner.name = name;
1498 	spin_unlock(&fp->f_lock);
1499 
1500 	return 0;
1501 }
1502 
1503 /**
1504  * ksmbd_vfs_compare_durable_owner - Verify if the requester is original owner
1505  * @fp: existing ksmbd file pointer
1506  * @user: user pointer of the reconnect requester
1507  *
1508  * Compares the UID, GID, and name of the current requester against the
1509  * original owner stored in the file handle.
1510  *
1511  * Return: true if the user matches, false otherwise
1512  */
1513 bool ksmbd_vfs_compare_durable_owner(struct ksmbd_file *fp,
1514 		struct ksmbd_user *user)
1515 {
1516 	bool ret = false;
1517 
1518 	if (!user)
1519 		return false;
1520 
1521 	spin_lock(&fp->f_lock);
1522 	if (!fp->owner.name)
1523 		goto out;
1524 
1525 	/* Check if the UID and GID match first (fast path) */
1526 	if (fp->owner.uid != user->uid || fp->owner.gid != user->gid)
1527 		goto out;
1528 
1529 	/* Validate the account name to ensure the same SecurityContext */
1530 	ret = (strcmp(fp->owner.name, user->name) == 0);
1531 out:
1532 	spin_unlock(&fp->f_lock);
1533 	return ret;
1534 }
1535 
1536 static bool session_fd_check(struct ksmbd_tree_connect *tcon,
1537 			     struct ksmbd_file *fp, struct ksmbd_user *user)
1538 {
1539 	struct ksmbd_inode *ci;
1540 	struct oplock_info *op;
1541 	struct ksmbd_conn *conn;
1542 	struct ksmbd_lock *smb_lock, *tmp_lock;
1543 
1544 	if (!is_reconnectable(fp))
1545 		return false;
1546 
1547 	if (fp->f_state != FP_INITED)
1548 		return false;
1549 
1550 	if (WARN_ON_ONCE(!fp->conn))
1551 		return false;
1552 
1553 	if (ksmbd_vfs_copy_durable_owner(fp, user))
1554 		return false;
1555 
1556 	/*
1557 	 * fp owns a strong reference on fp->conn (taken in ksmbd_open_fd()
1558 	 * / ksmbd_reopen_durable_fd()), so conn stays valid for the whole
1559 	 * body of this function regardless of any op->conn puts below.
1560 	 */
1561 	conn = fp->conn;
1562 	ci = fp->f_ci;
1563 	down_write(&ci->m_lock);
1564 	list_for_each_entry_rcu(op, &ci->m_op_list, op_entry,
1565 				lockdep_is_held(&ci->m_lock)) {
1566 		if (op->conn != conn)
1567 			continue;
1568 		ksmbd_conn_put(op->conn);
1569 		op->conn = NULL;
1570 		op->sess = NULL;
1571 	}
1572 	up_write(&ci->m_lock);
1573 
1574 	list_for_each_entry_safe(smb_lock, tmp_lock, &fp->lock_list, flist) {
1575 		struct ksmbd_conn *lock_conn = smb_lock->conn;
1576 
1577 		if (!lock_conn)
1578 			continue;
1579 		spin_lock(&lock_conn->llist_lock);
1580 		list_del_init(&smb_lock->clist);
1581 		smb_lock->conn = NULL;
1582 		spin_unlock(&lock_conn->llist_lock);
1583 		ksmbd_conn_put(lock_conn);
1584 	}
1585 
1586 	fp->conn = NULL;
1587 	fp->tcon = NULL;
1588 	fp->volatile_id = KSMBD_NO_FID;
1589 
1590 	if (fp->durable_timeout)
1591 		fp->durable_scavenger_timeout =
1592 			jiffies_to_msecs(jiffies) + fp->durable_timeout;
1593 
1594 	/* Drop fp's own reference on conn. */
1595 	ksmbd_conn_put(conn);
1596 	return true;
1597 }
1598 
1599 void ksmbd_close_tree_conn_fds(struct ksmbd_work *work)
1600 {
1601 	int num = __close_file_table_ids(work->sess,
1602 					 work->tcon,
1603 					 tree_conn_fd_check,
1604 					 false);
1605 
1606 	atomic_sub(num, &work->conn->stats.open_files_count);
1607 }
1608 
1609 void ksmbd_close_session_fds(struct ksmbd_work *work)
1610 {
1611 	int num = __close_file_table_ids(work->sess,
1612 					 work->tcon,
1613 					 session_fd_check,
1614 					 true);
1615 
1616 	atomic_sub(num, &work->conn->stats.open_files_count);
1617 }
1618 
1619 int ksmbd_init_global_file_table(void)
1620 {
1621 	create_proc_files();
1622 	return ksmbd_init_file_table(&global_ft);
1623 }
1624 
1625 void ksmbd_free_global_file_table(void)
1626 {
1627 	struct ksmbd_file	*fp = NULL;
1628 	unsigned int		id;
1629 
1630 	idr_for_each_entry(global_ft.idr, fp, id) {
1631 		ksmbd_remove_durable_fd(fp);
1632 		__ksmbd_close_fd(NULL, fp);
1633 	}
1634 
1635 	idr_destroy(global_ft.idr);
1636 	kfree(global_ft.idr);
1637 }
1638 
1639 int ksmbd_validate_name_reconnect(struct ksmbd_share_config *share,
1640 				  struct ksmbd_file *fp, char *name)
1641 {
1642 	char *pathname, *ab_pathname;
1643 	int ret = 0;
1644 
1645 	pathname = kmalloc(PATH_MAX, KSMBD_DEFAULT_GFP);
1646 	if (!pathname)
1647 		return -EACCES;
1648 
1649 	ab_pathname = d_path(&fp->filp->f_path, pathname, PATH_MAX);
1650 	if (IS_ERR(ab_pathname)) {
1651 		kfree(pathname);
1652 		return -EACCES;
1653 	}
1654 
1655 	if (name && strcmp(&ab_pathname[share->path_sz + 1], name)) {
1656 		ksmbd_debug(SMB, "invalid name reconnect %s\n", name);
1657 		ret = -EINVAL;
1658 	}
1659 
1660 	kfree(pathname);
1661 
1662 	return ret;
1663 }
1664 
1665 int ksmbd_reopen_durable_fd(struct ksmbd_work *work, struct ksmbd_file *fp)
1666 {
1667 	struct ksmbd_inode *ci;
1668 	struct oplock_info *op;
1669 	struct ksmbd_conn *conn = work->conn;
1670 	struct ksmbd_lock *smb_lock;
1671 	unsigned int old_f_state;
1672 
1673 	write_lock(&global_ft.lock);
1674 	if (!fp->is_durable || fp->conn || fp->tcon) {
1675 		write_unlock(&global_ft.lock);
1676 		pr_err("Invalid durable fd [%p:%p]\n", fp->conn, fp->tcon);
1677 		return -EBADF;
1678 	}
1679 
1680 	if (has_file_id(fp->volatile_id)) {
1681 		write_unlock(&global_ft.lock);
1682 		pr_err("Still in use durable fd: %llu\n", fp->volatile_id);
1683 		return -EBADF;
1684 	}
1685 
1686 	/*
1687 	 * Initialize fp's connection binding before publishing fp into the
1688 	 * session's file table.  If __open_id() is ordered first, a
1689 	 * concurrent teardown that iterates the table can observe a valid
1690 	 * volatile_id with fp->conn == NULL and preserve a
1691 	 * partially-initialized fp.  fp owns a strong reference on the new
1692 	 * conn (see ksmbd_open_fd()); undo it on __open_id() failure.
1693 	 */
1694 	fp->conn = ksmbd_conn_get(conn);
1695 	fp->tcon = work->tcon;
1696 	write_unlock(&global_ft.lock);
1697 
1698 	old_f_state = fp->f_state;
1699 	fp->f_state = FP_NEW;
1700 
1701 	__open_id(&work->sess->file_table, fp, OPEN_ID_TYPE_VOLATILE_ID);
1702 	if (!has_file_id(fp->volatile_id)) {
1703 		write_lock(&global_ft.lock);
1704 		fp->conn = NULL;
1705 		fp->tcon = NULL;
1706 		write_unlock(&global_ft.lock);
1707 		ksmbd_conn_put(conn);
1708 		fp->f_state = old_f_state;
1709 		return -EBADF;
1710 	}
1711 
1712 	list_for_each_entry(smb_lock, &fp->lock_list, flist) {
1713 		smb_lock->conn = ksmbd_conn_get(conn);
1714 		spin_lock(&conn->llist_lock);
1715 		list_add_tail(&smb_lock->clist, &conn->lock_list);
1716 		spin_unlock(&conn->llist_lock);
1717 	}
1718 
1719 	ci = fp->f_ci;
1720 	down_write(&ci->m_lock);
1721 	list_for_each_entry_rcu(op, &ci->m_op_list, op_entry,
1722 				lockdep_is_held(&ci->m_lock)) {
1723 		if (op->conn)
1724 			continue;
1725 		op->conn = ksmbd_conn_get(fp->conn);
1726 		op->sess = work->sess;
1727 	}
1728 	up_write(&ci->m_lock);
1729 
1730 	spin_lock(&fp->f_lock);
1731 	fp->owner.uid = fp->owner.gid = 0;
1732 	kfree(fp->owner.name);
1733 	fp->owner.name = NULL;
1734 	spin_unlock(&fp->f_lock);
1735 
1736 	return 0;
1737 }
1738 
1739 int ksmbd_init_file_table(struct ksmbd_file_table *ft)
1740 {
1741 	ft->idr = kzalloc_obj(struct idr, KSMBD_DEFAULT_GFP);
1742 	if (!ft->idr)
1743 		return -ENOMEM;
1744 
1745 	idr_init(ft->idr);
1746 	rwlock_init(&ft->lock);
1747 	return 0;
1748 }
1749 
1750 void ksmbd_destroy_file_table(struct ksmbd_session *sess)
1751 {
1752 	struct ksmbd_file_table *ft = &sess->file_table;
1753 
1754 	if (!ft->idr)
1755 		return;
1756 
1757 	__close_file_table_ids(sess, NULL, session_fd_check, true);
1758 	idr_destroy(ft->idr);
1759 	kfree(ft->idr);
1760 	ft->idr = NULL;
1761 }
1762 
1763 int ksmbd_init_file_cache(void)
1764 {
1765 	filp_cache = kmem_cache_create("ksmbd_file_cache",
1766 				       sizeof(struct ksmbd_file), 0,
1767 				       SLAB_HWCACHE_ALIGN, NULL);
1768 	if (!filp_cache)
1769 		goto out;
1770 
1771 	init_waitqueue_head(&dh_wq);
1772 
1773 	return 0;
1774 
1775 out:
1776 	pr_err("failed to allocate file cache\n");
1777 	return -ENOMEM;
1778 }
1779 
1780 void ksmbd_exit_file_cache(void)
1781 {
1782 	kmem_cache_destroy(filp_cache);
1783 }
1784