xref: /linux/fs/smb/server/vfs_cache.c (revision 5a7f4d6d8e7fc9c3b67412f1b8e5b56c9aec21af)
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 			list_move_tail(&fp->node, &dispose);
585 		}
586 	}
587 	up_write(&ci->m_lock);
588 
589 	/*
590 	 * Drop our lookup reference before closing so the last __ksmbd_close_fd()
591 	 * can drop m_count to zero and unlink the delete-on-close file.  The
592 	 * collected handles still hold references, so ci stays valid until they
593 	 * are closed below.
594 	 */
595 	ksmbd_inode_put(ci);
596 
597 	while (!list_empty(&dispose)) {
598 		fp = list_first_entry(&dispose, struct ksmbd_file, node);
599 		list_del_init(&fp->node);
600 		__ksmbd_close_fd(NULL, fp);
601 		closed = true;
602 	}
603 
604 	return closed;
605 }
606 
607 static struct ksmbd_file *ksmbd_fp_get(struct ksmbd_file *fp)
608 {
609 	if (fp->f_state != FP_INITED)
610 		return NULL;
611 
612 	if (!atomic_inc_not_zero(&fp->refcount))
613 		return NULL;
614 	return fp;
615 }
616 
617 static struct ksmbd_file *__ksmbd_lookup_fd(struct ksmbd_file_table *ft,
618 					    u64 id)
619 {
620 	struct ksmbd_file *fp;
621 
622 	if (!has_file_id(id))
623 		return NULL;
624 
625 	read_lock(&ft->lock);
626 	fp = idr_find(ft->idr, id);
627 	if (fp)
628 		fp = ksmbd_fp_get(fp);
629 	read_unlock(&ft->lock);
630 	return fp;
631 }
632 
633 static void __put_fd_final(struct ksmbd_work *work, struct ksmbd_file *fp)
634 {
635 	/*
636 	 * Detached durable fp -- session_fd_check() cleared fp->conn at
637 	 * preserve, so this fp is no longer tracked by any conn's
638 	 * stats.open_files_count.  This happens when
639 	 * ksmbd_scavenger_dispose_dh() hands the final close off to an
640 	 * m_fp_list walker (e.g. ksmbd_lookup_fd_inode()) whose work->conn
641 	 * is unrelated to the conn that originally opened the handle; close
642 	 * via the NULL-ft path so we do not underflow that unrelated
643 	 * counter.
644 	 */
645 	if (!fp->conn) {
646 		__ksmbd_close_fd(NULL, fp);
647 		return;
648 	}
649 	__ksmbd_close_fd(&work->sess->file_table, fp);
650 	atomic_dec(&work->conn->stats.open_files_count);
651 }
652 
653 static void set_close_state_blocked_works(struct ksmbd_file *fp)
654 {
655 	struct ksmbd_work *cancel_work;
656 
657 	spin_lock(&fp->f_lock);
658 	list_for_each_entry(cancel_work, &fp->blocked_works,
659 				 fp_entry) {
660 		cancel_work->state = KSMBD_WORK_CLOSED;
661 		cancel_work->cancel_fn(cancel_work->cancel_argv);
662 	}
663 	spin_unlock(&fp->f_lock);
664 }
665 
666 int ksmbd_close_fd(struct ksmbd_work *work, u64 id)
667 {
668 	struct ksmbd_file	*fp;
669 	struct ksmbd_file_table	*ft;
670 	bool closed = false;
671 
672 	if (!has_file_id(id))
673 		return 0;
674 
675 	ft = &work->sess->file_table;
676 	write_lock(&ft->lock);
677 	fp = idr_find(ft->idr, id);
678 	if (fp) {
679 		set_close_state_blocked_works(fp);
680 
681 		if (fp->f_state != FP_INITED)
682 			fp = NULL;
683 		else {
684 			fp->f_state = FP_CLOSED;
685 			closed = true;
686 			if (!atomic_dec_and_test(&fp->refcount))
687 				fp = NULL;
688 		}
689 	}
690 	write_unlock(&ft->lock);
691 
692 	if (!fp)
693 		return closed ? 0 : -EINVAL;
694 
695 	__put_fd_final(work, fp);
696 	return 0;
697 }
698 
699 void ksmbd_fd_put(struct ksmbd_work *work, struct ksmbd_file *fp)
700 {
701 	if (!fp)
702 		return;
703 
704 	if (!atomic_dec_and_test(&fp->refcount))
705 		return;
706 	__put_fd_final(work, fp);
707 }
708 
709 static bool __sanity_check(struct ksmbd_tree_connect *tcon, struct ksmbd_file *fp)
710 {
711 	if (!fp)
712 		return false;
713 	if (fp->tcon != tcon)
714 		return false;
715 	return true;
716 }
717 
718 struct ksmbd_file *ksmbd_lookup_foreign_fd(struct ksmbd_work *work, u64 id)
719 {
720 	return __ksmbd_lookup_fd(&work->sess->file_table, id);
721 }
722 
723 struct ksmbd_file *ksmbd_lookup_fd_fast(struct ksmbd_work *work, u64 id)
724 {
725 	struct ksmbd_file *fp = __ksmbd_lookup_fd(&work->sess->file_table, id);
726 
727 	if (__sanity_check(work->tcon, fp))
728 		return fp;
729 
730 	ksmbd_fd_put(work, fp);
731 	return NULL;
732 }
733 
734 struct ksmbd_file *ksmbd_lookup_fd_slow(struct ksmbd_work *work, u64 id,
735 					u64 pid)
736 {
737 	struct ksmbd_file *fp;
738 
739 	if (!has_file_id(id)) {
740 		id = work->compound_fid;
741 		pid = work->compound_pfid;
742 	}
743 
744 	fp = __ksmbd_lookup_fd(&work->sess->file_table, id);
745 	if (!__sanity_check(work->tcon, fp)) {
746 		ksmbd_fd_put(work, fp);
747 		return NULL;
748 	}
749 	if (fp->persistent_id != pid) {
750 		ksmbd_fd_put(work, fp);
751 		return NULL;
752 	}
753 	return fp;
754 }
755 
756 struct ksmbd_file *ksmbd_lookup_global_fd(unsigned long long id)
757 {
758 	return __ksmbd_lookup_fd(&global_ft, id);
759 }
760 
761 struct ksmbd_file *ksmbd_lookup_durable_fd(unsigned long long id)
762 {
763 	struct ksmbd_file *fp;
764 
765 	fp = __ksmbd_lookup_fd(&global_ft, id);
766 	if (fp && (fp->durable_reconnect_disabled ||
767 		   fp->conn ||
768 		   (fp->durable_scavenger_timeout &&
769 		    (fp->durable_scavenger_timeout <
770 		     jiffies_to_msecs(jiffies))))) {
771 		ksmbd_put_durable_fd(fp);
772 		fp = NULL;
773 	}
774 
775 	return fp;
776 }
777 
778 void ksmbd_put_durable_fd(struct ksmbd_file *fp)
779 {
780 	if (!atomic_dec_and_test(&fp->refcount))
781 		return;
782 
783 	__ksmbd_close_fd(NULL, fp);
784 }
785 
786 bool ksmbd_has_other_active_fd(struct ksmbd_file *fp)
787 {
788 	struct ksmbd_file *lfp;
789 	struct ksmbd_inode *ci = fp->f_ci;
790 	bool ret = false;
791 
792 	down_read(&ci->m_lock);
793 	list_for_each_entry(lfp, &ci->m_fp_list, node) {
794 		if (lfp == fp)
795 			continue;
796 
797 		if (lfp->f_state == FP_INITED &&
798 		    (READ_ONCE(lfp->conn) || READ_ONCE(lfp->tcon))) {
799 			ret = true;
800 			break;
801 		}
802 	}
803 	up_read(&ci->m_lock);
804 
805 	return ret;
806 }
807 
808 static struct ksmbd_file *ksmbd_lookup_fd_app_instance_id(char *app_instance_id)
809 {
810 	struct ksmbd_file *fp = NULL;
811 	unsigned int id;
812 
813 	if (!memchr_inv(app_instance_id, 0, SMB2_CREATE_GUID_SIZE))
814 		return NULL;
815 
816 	read_lock(&global_ft.lock);
817 	idr_for_each_entry(global_ft.idr, fp, id) {
818 		if (!memcmp(fp->app_instance_id, app_instance_id,
819 			    SMB2_CREATE_GUID_SIZE)) {
820 			fp = ksmbd_fp_get(fp);
821 			break;
822 		}
823 	}
824 	read_unlock(&global_ft.lock);
825 
826 	return fp;
827 }
828 
829 int ksmbd_close_fd_app_instance_id(char *app_instance_id)
830 {
831 	struct ksmbd_file_table *ft;
832 	struct ksmbd_file *fp;
833 	struct oplock_info *opinfo;
834 	int n_to_drop = 0;
835 
836 	fp = ksmbd_lookup_fd_app_instance_id(app_instance_id);
837 	if (!fp)
838 		return 0;
839 
840 	opinfo = opinfo_get(fp);
841 	if (!opinfo || !opinfo->sess)
842 		goto out;
843 
844 	ft = &opinfo->sess->file_table;
845 	write_lock(&ft->lock);
846 	if (fp->f_state == FP_INITED) {
847 		if (has_file_id(fp->volatile_id)) {
848 			idr_remove(ft->idr, fp->volatile_id);
849 			fp->volatile_id = KSMBD_NO_FID;
850 		}
851 		n_to_drop = ksmbd_mark_fp_closed(fp);
852 	}
853 	write_unlock(&ft->lock);
854 	opinfo_put(opinfo);
855 	opinfo = NULL;
856 
857 	if (!n_to_drop)
858 		goto out;
859 
860 	down_write(&fp->f_ci->m_lock);
861 	list_del_init(&fp->node);
862 	up_write(&fp->f_ci->m_lock);
863 
864 	if (atomic_sub_and_test(n_to_drop, &fp->refcount)) {
865 		if (fp->conn)
866 			atomic_dec(&fp->conn->stats.open_files_count);
867 		__ksmbd_close_fd(NULL, fp);
868 	}
869 	return 0;
870 
871 out:
872 	if (opinfo)
873 		opinfo_put(opinfo);
874 	ksmbd_put_durable_fd(fp);
875 	return 0;
876 }
877 
878 int ksmbd_invalidate_durable_fd(unsigned long long id)
879 {
880 	struct ksmbd_file *fp;
881 
882 	fp = ksmbd_lookup_global_fd(id);
883 	if (!fp)
884 		return -ENOENT;
885 
886 	fp->durable_reconnect_disabled = true;
887 
888 	if (fp->conn) {
889 		ksmbd_put_durable_fd(fp);
890 		return -ENOENT;
891 	}
892 
893 	fp->durable_timeout = 1;
894 	fp->durable_scavenger_timeout = jiffies_to_msecs(jiffies);
895 	ksmbd_put_durable_fd(fp);
896 	if (waitqueue_active(&dh_wq))
897 		wake_up(&dh_wq);
898 
899 	return -ENOENT;
900 }
901 
902 struct ksmbd_file *ksmbd_lookup_fd_cguid(char *cguid)
903 {
904 	struct ksmbd_file	*fp = NULL;
905 	unsigned int		id;
906 
907 	read_lock(&global_ft.lock);
908 	idr_for_each_entry(global_ft.idr, fp, id) {
909 		if (!memcmp(fp->create_guid,
910 			    cguid,
911 			    SMB2_CREATE_GUID_SIZE)) {
912 			fp = ksmbd_fp_get(fp);
913 			break;
914 		}
915 	}
916 	read_unlock(&global_ft.lock);
917 
918 	return fp;
919 }
920 
921 struct ksmbd_file *ksmbd_lookup_fd_inode(struct dentry *dentry)
922 {
923 	struct ksmbd_file	*lfp;
924 	struct ksmbd_inode	*ci;
925 	struct inode		*inode = d_inode(dentry);
926 
927 	read_lock(&inode_hash_lock);
928 	ci = __ksmbd_inode_lookup(dentry);
929 	read_unlock(&inode_hash_lock);
930 	if (!ci)
931 		return NULL;
932 
933 	down_read(&ci->m_lock);
934 	list_for_each_entry(lfp, &ci->m_fp_list, node) {
935 		if (inode == file_inode(lfp->filp)) {
936 			lfp = ksmbd_fp_get(lfp);
937 			up_read(&ci->m_lock);
938 			ksmbd_inode_put(ci);
939 			return lfp;
940 		}
941 	}
942 	up_read(&ci->m_lock);
943 	ksmbd_inode_put(ci);
944 	return NULL;
945 }
946 
947 bool ksmbd_has_open_files(struct dentry *dentry)
948 {
949 	struct ksmbd_file *fp;
950 	unsigned int id;
951 	bool ret = false;
952 
953 	read_lock(&global_ft.lock);
954 	idr_for_each_entry(global_ft.idr, fp, id) {
955 		struct dentry *fp_dentry = fp->filp->f_path.dentry;
956 
957 		if (fp->f_state != FP_INITED)
958 			continue;
959 		if (fp_dentry == dentry)
960 			continue;
961 		if (is_subdir(fp_dentry, dentry)) {
962 			ret = true;
963 			break;
964 		}
965 	}
966 	read_unlock(&global_ft.lock);
967 
968 	return ret;
969 }
970 
971 #define OPEN_ID_TYPE_VOLATILE_ID	(0)
972 #define OPEN_ID_TYPE_PERSISTENT_ID	(1)
973 
974 static void __open_id_set(struct ksmbd_file *fp, u64 id, int type)
975 {
976 	if (type == OPEN_ID_TYPE_VOLATILE_ID)
977 		fp->volatile_id = id;
978 	if (type == OPEN_ID_TYPE_PERSISTENT_ID)
979 		fp->persistent_id = id;
980 }
981 
982 static int __open_id(struct ksmbd_file_table *ft, struct ksmbd_file *fp,
983 		     int type)
984 {
985 	u64			id = 0;
986 	int			ret;
987 
988 	if (type == OPEN_ID_TYPE_VOLATILE_ID && fd_limit_depleted()) {
989 		__open_id_set(fp, KSMBD_NO_FID, type);
990 		return -EMFILE;
991 	}
992 
993 	idr_preload(KSMBD_DEFAULT_GFP);
994 	write_lock(&ft->lock);
995 	ret = idr_alloc_cyclic(ft->idr, fp, KSMBD_START_FID, INT_MAX - 1,
996 			       GFP_NOWAIT);
997 	if (ret >= 0) {
998 		id = ret;
999 		ret = 0;
1000 	} else {
1001 		id = KSMBD_NO_FID;
1002 		fd_limit_close();
1003 	}
1004 
1005 	__open_id_set(fp, id, type);
1006 	write_unlock(&ft->lock);
1007 	idr_preload_end();
1008 	return ret;
1009 }
1010 
1011 unsigned int ksmbd_open_durable_fd(struct ksmbd_file *fp)
1012 {
1013 	__open_id(&global_ft, fp, OPEN_ID_TYPE_PERSISTENT_ID);
1014 	return fp->persistent_id;
1015 }
1016 
1017 struct ksmbd_file *ksmbd_open_fd(struct ksmbd_work *work, struct file *filp)
1018 {
1019 	struct ksmbd_file *fp;
1020 	int ret;
1021 
1022 	fp = kmem_cache_zalloc(filp_cache, KSMBD_DEFAULT_GFP);
1023 	if (!fp) {
1024 		pr_err("Failed to allocate memory\n");
1025 		return ERR_PTR(-ENOMEM);
1026 	}
1027 
1028 	INIT_LIST_HEAD(&fp->blocked_works);
1029 	INIT_LIST_HEAD(&fp->node);
1030 	INIT_LIST_HEAD(&fp->lock_list);
1031 	spin_lock_init(&fp->f_lock);
1032 	mutex_init(&fp->readdir_lock);
1033 	atomic_set(&fp->refcount, 1);
1034 
1035 	fp->filp		= filp;
1036 	/*
1037 	 * fp owns a strong reference on fp->conn for as long as fp->conn is
1038 	 * non-NULL, so session_fd_check() and __ksmbd_close_fd() never
1039 	 * dereference a dangling pointer.  Paired with ksmbd_conn_put() in
1040 	 * session_fd_check() (durable preserve), in __ksmbd_close_fd()
1041 	 * (final close), and on the error paths below.
1042 	 */
1043 	fp->conn		= ksmbd_conn_get(work->conn);
1044 	fp->tcon		= work->tcon;
1045 	fp->volatile_id		= KSMBD_NO_FID;
1046 	fp->persistent_id	= KSMBD_NO_FID;
1047 	fp->f_state		= FP_NEW;
1048 	fp->f_ci		= ksmbd_inode_get(fp);
1049 
1050 	if (!fp->f_ci) {
1051 		ret = -ENOMEM;
1052 		goto err_out;
1053 	}
1054 
1055 	ret = __open_id(&work->sess->file_table, fp, OPEN_ID_TYPE_VOLATILE_ID);
1056 	if (ret) {
1057 		ksmbd_inode_put(fp->f_ci);
1058 		goto err_out;
1059 	}
1060 
1061 	atomic_inc(&work->conn->stats.open_files_count);
1062 	return fp;
1063 
1064 err_out:
1065 	/* fp->conn was set and refcounted before every branch here. */
1066 	ksmbd_conn_put(fp->conn);
1067 	kmem_cache_free(filp_cache, fp);
1068 	return ERR_PTR(ret);
1069 }
1070 
1071 /**
1072  * ksmbd_update_fstate() - update an fp state under the file-table lock
1073  * @ft: file table that publishes @fp's volatile id
1074  * @fp: file pointer to update
1075  * @state: new state
1076  *
1077  * Return: 0 on success.  The FP_NEW -> FP_INITED transition is special:
1078  * -ENOENT if teardown already unpublished @fp by advancing the state or
1079  * clearing the volatile id.  Other state updates preserve the historical
1080  * fire-and-forget behavior.
1081  */
1082 int ksmbd_update_fstate(struct ksmbd_file_table *ft, struct ksmbd_file *fp,
1083 			unsigned int state)
1084 {
1085 	int ret;
1086 
1087 	if (!fp)
1088 		return -ENOENT;
1089 
1090 	write_lock(&ft->lock);
1091 	if (state == FP_INITED &&
1092 	    (fp->f_state != FP_NEW || !has_file_id(fp->volatile_id))) {
1093 		ret = -ENOENT;
1094 	} else {
1095 		fp->f_state = state;
1096 		ret = 0;
1097 	}
1098 	write_unlock(&ft->lock);
1099 
1100 	return ret;
1101 }
1102 
1103 /*
1104  * ksmbd_mark_fp_closed() - mark fp closed under ft->lock and return how many
1105  * refs the teardown path owns.
1106  *
1107  * FP_INITED has a normal idr-owned reference, so teardown owns both that
1108  * reference and the transient lookup reference.  FP_NEW is still owned by the
1109  * in-flight opener/reopener, which will drop the original reference after
1110  * ksmbd_update_fstate(..., FP_INITED) observes the cleared volatile id.
1111  * FP_CLOSED on entry means an earlier ksmbd_close_fd() already consumed the
1112  * idr-owned ref.
1113  */
1114 static int ksmbd_mark_fp_closed(struct ksmbd_file *fp)
1115 {
1116 	if (fp->f_state == FP_INITED) {
1117 		set_close_state_blocked_works(fp);
1118 		fp->f_state = FP_CLOSED;
1119 		return 2;
1120 	}
1121 
1122 	return 1;
1123 }
1124 
1125 static int
1126 __close_file_table_ids(struct ksmbd_session *sess,
1127 		       struct ksmbd_tree_connect *tcon,
1128 		       bool (*skip)(struct ksmbd_tree_connect *tcon,
1129 				    struct ksmbd_file *fp,
1130 				    struct ksmbd_user *user),
1131 		       bool skip_preserves_fp)
1132 {
1133 	struct ksmbd_file_table *ft = &sess->file_table;
1134 	struct ksmbd_file *fp;
1135 	unsigned int id = 0;
1136 	int num = 0;
1137 
1138 	while (1) {
1139 		int n_to_drop;
1140 
1141 		write_lock(&ft->lock);
1142 		fp = idr_get_next(ft->idr, &id);
1143 		if (!fp) {
1144 			write_unlock(&ft->lock);
1145 			break;
1146 		}
1147 		if (!atomic_inc_not_zero(&fp->refcount)) {
1148 			id++;
1149 			write_unlock(&ft->lock);
1150 			continue;
1151 		}
1152 
1153 		if (skip_preserves_fp) {
1154 			/*
1155 			 * Session teardown: skip() is session_fd_check(),
1156 			 * which may sleep and mutates fp->conn / fp->tcon /
1157 			 * fp->volatile_id when it chooses to preserve fp
1158 			 * for durable reconnect.  Unpublish fp from the
1159 			 * session idr here, under ft->lock, so that
1160 			 * __ksmbd_lookup_fd() through this session cannot
1161 			 * grant a new ksmbd_fp_get() reference to an fp
1162 			 * whose fields are about to be rewritten outside
1163 			 * the lock.  Durable reconnect still reaches fp via
1164 			 * global_ft.
1165 			 */
1166 			idr_remove(ft->idr, id);
1167 			fp->durable_volatile_id = fp->volatile_id;
1168 			fp->volatile_id = KSMBD_NO_FID;
1169 			write_unlock(&ft->lock);
1170 
1171 			if (skip(tcon, fp, sess->user)) {
1172 				/*
1173 				 * session_fd_check() has converted fp to
1174 				 * durable-preserve state and cleared its
1175 				 * per-conn fields.  fp is already unpublished
1176 				 * above; the original idr-owned ref keeps it
1177 				 * alive for the durable scavenger.  Drop only
1178 				 * the transient ref.  atomic_dec() is safe --
1179 				 * atomic_inc_not_zero() succeeded on a
1180 				 * positive value and we added one more, so
1181 				 * refcount cannot be zero here.
1182 				 */
1183 				atomic_dec(&fp->refcount);
1184 				id++;
1185 				continue;
1186 			}
1187 
1188 			/*
1189 			 * Keep the close-state decision under the same lock
1190 			 * observed by ksmbd_update_fstate(), which is how an
1191 			 * in-flight FP_NEW opener learns that teardown has
1192 			 * cleared its volatile id.
1193 			 */
1194 			write_lock(&ft->lock);
1195 			n_to_drop = ksmbd_mark_fp_closed(fp);
1196 			write_unlock(&ft->lock);
1197 		} else {
1198 			/*
1199 			 * Tree teardown: skip() is tree_conn_fd_check(), a
1200 			 * cheap pointer compare that doesn't sleep and has
1201 			 * no side effects, so keep the skip decision plus
1202 			 * the unpublish-and-mark-closed sequence atomic
1203 			 * under ft->lock.  fps belonging to other tree
1204 			 * connects (skip() == true) stay fully published in
1205 			 * the session idr with no lock window.
1206 			 */
1207 			if (skip(tcon, fp, sess->user)) {
1208 				atomic_dec(&fp->refcount);
1209 				write_unlock(&ft->lock);
1210 				id++;
1211 				continue;
1212 			}
1213 			idr_remove(ft->idr, id);
1214 			fp->volatile_id = KSMBD_NO_FID;
1215 			n_to_drop = ksmbd_mark_fp_closed(fp);
1216 			write_unlock(&ft->lock);
1217 		}
1218 
1219 		/*
1220 		 * fp->volatile_id is already cleared to prevent stale idr
1221 		 * removal from a deferred final close.  Remove fp from
1222 		 * m_fp_list here because __ksmbd_remove_fd() will skip the
1223 		 * list unlink when volatile_id is KSMBD_NO_FID.
1224 		 */
1225 		down_write(&fp->f_ci->m_lock);
1226 		list_del_init(&fp->node);
1227 		up_write(&fp->f_ci->m_lock);
1228 
1229 		/*
1230 		 * Drop the references this iteration owns:
1231 		 *
1232 		 *   n_to_drop == 2: we observed FP_INITED and committed
1233 		 *     the FP_CLOSED transition ourselves, so we own the
1234 		 *     transient (+1) and the still-intact idr-owned ref.
1235 		 *
1236 		 *   n_to_drop == 1: either a prior ksmbd_close_fd()
1237 		 *     already consumed the idr-owned ref, or fp was still
1238 		 *     FP_NEW and the in-flight opener/reopener must keep
1239 		 *     the original reference until ksmbd_update_fstate()
1240 		 *     observes the cleared volatile id.
1241 		 *
1242 		 * If we end up as the final putter, finalize fp and
1243 		 * account the open_files_count decrement via the caller's
1244 		 * atomic_sub(num, ...).  Otherwise the remaining user's
1245 		 * ksmbd_fd_put() reaches __put_fd_final(), which does its
1246 		 * own atomic_dec(&open_files_count), so we must not count
1247 		 * this fp here -- doing so would double-decrement the
1248 		 * connection-wide counter.
1249 		 */
1250 		if (atomic_sub_and_test(n_to_drop, &fp->refcount)) {
1251 			__ksmbd_close_fd(NULL, fp);
1252 			num++;
1253 		}
1254 		id++;
1255 	}
1256 
1257 	return num;
1258 }
1259 
1260 static inline bool is_reconnectable(struct ksmbd_file *fp)
1261 {
1262 	struct oplock_info *opinfo = opinfo_get(fp);
1263 	bool reconn = false;
1264 
1265 	if (!opinfo)
1266 		return false;
1267 
1268 	if (opinfo->op_state != OPLOCK_STATE_NONE) {
1269 		opinfo_put(opinfo);
1270 		return false;
1271 	}
1272 
1273 	if (fp->is_resilient || fp->is_persistent)
1274 		reconn = true;
1275 	else if (fp->is_durable && opinfo->is_lease &&
1276 		 opinfo->o_lease->state & SMB2_LEASE_HANDLE_CACHING_LE)
1277 		reconn = true;
1278 
1279 	else if (fp->is_durable && opinfo->level == SMB2_OPLOCK_LEVEL_BATCH)
1280 		reconn = true;
1281 
1282 	opinfo_put(opinfo);
1283 	return reconn;
1284 }
1285 
1286 static bool tree_conn_fd_check(struct ksmbd_tree_connect *tcon,
1287 			       struct ksmbd_file *fp,
1288 			       struct ksmbd_user *user)
1289 {
1290 	return fp->tcon != tcon;
1291 }
1292 
1293 static bool ksmbd_durable_scavenger_alive(void)
1294 {
1295 	if (!durable_scavenger_running)
1296 		return false;
1297 
1298 	if (kthread_should_stop())
1299 		return false;
1300 
1301 	if (idr_is_empty(global_ft.idr))
1302 		return false;
1303 
1304 	return true;
1305 }
1306 
1307 static void ksmbd_scavenger_dispose_dh(struct ksmbd_file *fp)
1308 {
1309 	/*
1310 	 * Durable-preserved fp can remain linked on f_ci->m_fp_list for
1311 	 * share-mode checks.  Unlink it before final close; fp->node is not
1312 	 * available as a scavenger-private list node because re-adding it to
1313 	 * another list corrupts m_fp_list.
1314 	 */
1315 	down_write(&fp->f_ci->m_lock);
1316 	list_del_init(&fp->node);
1317 	up_write(&fp->f_ci->m_lock);
1318 
1319 	/*
1320 	 * Drop both the durable lifetime reference and the transient reference
1321 	 * taken by the scavenger under global_ft.lock.  If a concurrent
1322 	 * ksmbd_lookup_fd_inode() (or any other m_fp_list walker) snatched fp
1323 	 * before the unlink above, that holder owns the final close via
1324 	 * ksmbd_fd_put() -> __ksmbd_close_fd().  Otherwise the scavenger is
1325 	 * the last putter and finalises fp here.
1326 	 */
1327 	if (atomic_sub_and_test(2, &fp->refcount))
1328 		__ksmbd_close_fd(NULL, fp);
1329 }
1330 
1331 static int ksmbd_durable_scavenger(void *dummy)
1332 {
1333 	struct ksmbd_file *fp = NULL;
1334 	struct ksmbd_file *expired_fp;
1335 	unsigned int id;
1336 	unsigned int min_timeout = 1;
1337 	bool found_fp_timeout;
1338 	unsigned long remaining_jiffies;
1339 
1340 	__module_get(THIS_MODULE);
1341 
1342 	set_freezable();
1343 	while (ksmbd_durable_scavenger_alive()) {
1344 		if (try_to_freeze())
1345 			continue;
1346 
1347 		remaining_jiffies = wait_event_interruptible_timeout(dh_wq,
1348 				   ksmbd_durable_scavenger_alive() == false,
1349 				   __msecs_to_jiffies(min_timeout));
1350 		if ((long)remaining_jiffies > 0)
1351 			min_timeout = jiffies_to_msecs(remaining_jiffies);
1352 		else
1353 			min_timeout = DURABLE_HANDLE_MAX_TIMEOUT;
1354 
1355 		do {
1356 			expired_fp = NULL;
1357 			found_fp_timeout = false;
1358 
1359 			write_lock(&global_ft.lock);
1360 			idr_for_each_entry(global_ft.idr, fp, id) {
1361 				unsigned long durable_timeout;
1362 
1363 				if (!fp->durable_timeout)
1364 					continue;
1365 
1366 				if (atomic_read(&fp->refcount) > 1 ||
1367 				    fp->conn)
1368 					continue;
1369 
1370 				found_fp_timeout = true;
1371 				if (fp->durable_scavenger_timeout <=
1372 				    jiffies_to_msecs(jiffies)) {
1373 					__ksmbd_remove_durable_fd(fp);
1374 					/*
1375 					 * Take a transient reference so fp
1376 					 * cannot be freed by an in-flight
1377 					 * ksmbd_lookup_fd_inode() that found
1378 					 * it through f_ci->m_fp_list while we
1379 					 * drop global_ft.lock and reach the
1380 					 * m_fp_list unlink in
1381 					 * ksmbd_scavenger_dispose_dh().
1382 					 */
1383 					atomic_inc(&fp->refcount);
1384 					expired_fp = fp;
1385 					break;
1386 				}
1387 
1388 				durable_timeout =
1389 					fp->durable_scavenger_timeout -
1390 						jiffies_to_msecs(jiffies);
1391 
1392 				if (min_timeout > durable_timeout)
1393 					min_timeout = durable_timeout;
1394 			}
1395 			write_unlock(&global_ft.lock);
1396 
1397 			if (expired_fp)
1398 				ksmbd_scavenger_dispose_dh(expired_fp);
1399 		} while (expired_fp);
1400 
1401 		if (found_fp_timeout == false)
1402 			break;
1403 	}
1404 
1405 	durable_scavenger_running = false;
1406 
1407 	module_put(THIS_MODULE);
1408 
1409 	return 0;
1410 }
1411 
1412 void ksmbd_launch_ksmbd_durable_scavenger(void)
1413 {
1414 	if (!(server_conf.flags & KSMBD_GLOBAL_FLAG_DURABLE_HANDLE))
1415 		return;
1416 
1417 	mutex_lock(&durable_scavenger_lock);
1418 	if (durable_scavenger_running == true) {
1419 		mutex_unlock(&durable_scavenger_lock);
1420 		return;
1421 	}
1422 
1423 	durable_scavenger_running = true;
1424 
1425 	server_conf.dh_task = kthread_run(ksmbd_durable_scavenger,
1426 				     (void *)NULL, "ksmbd-durable-scavenger");
1427 	if (IS_ERR(server_conf.dh_task))
1428 		pr_err("cannot start conn thread, err : %ld\n",
1429 		       PTR_ERR(server_conf.dh_task));
1430 	mutex_unlock(&durable_scavenger_lock);
1431 }
1432 
1433 void ksmbd_stop_durable_scavenger(void)
1434 {
1435 	if (!(server_conf.flags & KSMBD_GLOBAL_FLAG_DURABLE_HANDLE))
1436 		return;
1437 
1438 	mutex_lock(&durable_scavenger_lock);
1439 	if (!durable_scavenger_running) {
1440 		mutex_unlock(&durable_scavenger_lock);
1441 		return;
1442 	}
1443 
1444 	durable_scavenger_running = false;
1445 	if (waitqueue_active(&dh_wq))
1446 		wake_up(&dh_wq);
1447 	mutex_unlock(&durable_scavenger_lock);
1448 	kthread_stop(server_conf.dh_task);
1449 }
1450 
1451 /*
1452  * ksmbd_vfs_copy_durable_owner - Copy owner info for durable reconnect
1453  * @fp: ksmbd file pointer to store owner info
1454  * @user: user pointer to copy from
1455  *
1456  * This function binds the current user's identity to the file handle
1457  * to satisfy MS-SMB2 Step 8 (SecurityContext matching) during reconnect.
1458  *
1459  * Return: 0 on success, or negative error code on failure
1460  */
1461 static int ksmbd_vfs_copy_durable_owner(struct ksmbd_file *fp,
1462 		struct ksmbd_user *user)
1463 {
1464 	if (!user)
1465 		return -EINVAL;
1466 
1467 	/* Duplicate the user name to ensure identity persistence */
1468 	fp->owner.name = kstrdup(user->name, GFP_KERNEL);
1469 	if (!fp->owner.name)
1470 		return -ENOMEM;
1471 
1472 	fp->owner.uid = user->uid;
1473 	fp->owner.gid = user->gid;
1474 
1475 	return 0;
1476 }
1477 
1478 /**
1479  * ksmbd_vfs_compare_durable_owner - Verify if the requester is original owner
1480  * @fp: existing ksmbd file pointer
1481  * @user: user pointer of the reconnect requester
1482  *
1483  * Compares the UID, GID, and name of the current requester against the
1484  * original owner stored in the file handle.
1485  *
1486  * Return: true if the user matches, false otherwise
1487  */
1488 bool ksmbd_vfs_compare_durable_owner(struct ksmbd_file *fp,
1489 		struct ksmbd_user *user)
1490 {
1491 	if (!user || !fp->owner.name)
1492 		return false;
1493 
1494 	/* Check if the UID and GID match first (fast path) */
1495 	if (fp->owner.uid != user->uid || fp->owner.gid != user->gid)
1496 		return false;
1497 
1498 	/* Validate the account name to ensure the same SecurityContext */
1499 	if (strcmp(fp->owner.name, user->name))
1500 		return false;
1501 
1502 	return true;
1503 }
1504 
1505 static bool session_fd_check(struct ksmbd_tree_connect *tcon,
1506 			     struct ksmbd_file *fp, struct ksmbd_user *user)
1507 {
1508 	struct ksmbd_inode *ci;
1509 	struct oplock_info *op;
1510 	struct ksmbd_conn *conn;
1511 	struct ksmbd_lock *smb_lock, *tmp_lock;
1512 
1513 	if (!is_reconnectable(fp))
1514 		return false;
1515 
1516 	if (fp->f_state != FP_INITED)
1517 		return false;
1518 
1519 	if (WARN_ON_ONCE(!fp->conn))
1520 		return false;
1521 
1522 	if (ksmbd_vfs_copy_durable_owner(fp, user))
1523 		return false;
1524 
1525 	/*
1526 	 * fp owns a strong reference on fp->conn (taken in ksmbd_open_fd()
1527 	 * / ksmbd_reopen_durable_fd()), so conn stays valid for the whole
1528 	 * body of this function regardless of any op->conn puts below.
1529 	 */
1530 	conn = fp->conn;
1531 	ci = fp->f_ci;
1532 	down_write(&ci->m_lock);
1533 	list_for_each_entry_rcu(op, &ci->m_op_list, op_entry) {
1534 		if (op->conn != conn)
1535 			continue;
1536 		ksmbd_conn_put(op->conn);
1537 		op->conn = NULL;
1538 	}
1539 	up_write(&ci->m_lock);
1540 
1541 	list_for_each_entry_safe(smb_lock, tmp_lock, &fp->lock_list, flist) {
1542 		struct ksmbd_conn *lock_conn = smb_lock->conn;
1543 
1544 		if (!lock_conn)
1545 			continue;
1546 		spin_lock(&lock_conn->llist_lock);
1547 		list_del_init(&smb_lock->clist);
1548 		smb_lock->conn = NULL;
1549 		spin_unlock(&lock_conn->llist_lock);
1550 		ksmbd_conn_put(lock_conn);
1551 	}
1552 
1553 	fp->conn = NULL;
1554 	fp->tcon = NULL;
1555 	fp->volatile_id = KSMBD_NO_FID;
1556 
1557 	if (fp->durable_timeout)
1558 		fp->durable_scavenger_timeout =
1559 			jiffies_to_msecs(jiffies) + fp->durable_timeout;
1560 
1561 	/* Drop fp's own reference on conn. */
1562 	ksmbd_conn_put(conn);
1563 	return true;
1564 }
1565 
1566 void ksmbd_close_tree_conn_fds(struct ksmbd_work *work)
1567 {
1568 	int num = __close_file_table_ids(work->sess,
1569 					 work->tcon,
1570 					 tree_conn_fd_check,
1571 					 false);
1572 
1573 	atomic_sub(num, &work->conn->stats.open_files_count);
1574 }
1575 
1576 void ksmbd_close_session_fds(struct ksmbd_work *work)
1577 {
1578 	int num = __close_file_table_ids(work->sess,
1579 					 work->tcon,
1580 					 session_fd_check,
1581 					 true);
1582 
1583 	atomic_sub(num, &work->conn->stats.open_files_count);
1584 }
1585 
1586 int ksmbd_init_global_file_table(void)
1587 {
1588 	create_proc_files();
1589 	return ksmbd_init_file_table(&global_ft);
1590 }
1591 
1592 void ksmbd_free_global_file_table(void)
1593 {
1594 	struct ksmbd_file	*fp = NULL;
1595 	unsigned int		id;
1596 
1597 	idr_for_each_entry(global_ft.idr, fp, id) {
1598 		ksmbd_remove_durable_fd(fp);
1599 		__ksmbd_close_fd(NULL, fp);
1600 	}
1601 
1602 	idr_destroy(global_ft.idr);
1603 	kfree(global_ft.idr);
1604 }
1605 
1606 int ksmbd_validate_name_reconnect(struct ksmbd_share_config *share,
1607 				  struct ksmbd_file *fp, char *name)
1608 {
1609 	char *pathname, *ab_pathname;
1610 	int ret = 0;
1611 
1612 	pathname = kmalloc(PATH_MAX, KSMBD_DEFAULT_GFP);
1613 	if (!pathname)
1614 		return -EACCES;
1615 
1616 	ab_pathname = d_path(&fp->filp->f_path, pathname, PATH_MAX);
1617 	if (IS_ERR(ab_pathname)) {
1618 		kfree(pathname);
1619 		return -EACCES;
1620 	}
1621 
1622 	if (name && strcmp(&ab_pathname[share->path_sz + 1], name)) {
1623 		ksmbd_debug(SMB, "invalid name reconnect %s\n", name);
1624 		ret = -EINVAL;
1625 	}
1626 
1627 	kfree(pathname);
1628 
1629 	return ret;
1630 }
1631 
1632 int ksmbd_reopen_durable_fd(struct ksmbd_work *work, struct ksmbd_file *fp)
1633 {
1634 	struct ksmbd_inode *ci;
1635 	struct oplock_info *op;
1636 	struct ksmbd_conn *conn = work->conn;
1637 	struct ksmbd_lock *smb_lock;
1638 	unsigned int old_f_state;
1639 
1640 	write_lock(&global_ft.lock);
1641 	if (!fp->is_durable || fp->conn || fp->tcon) {
1642 		write_unlock(&global_ft.lock);
1643 		pr_err("Invalid durable fd [%p:%p]\n", fp->conn, fp->tcon);
1644 		return -EBADF;
1645 	}
1646 
1647 	if (has_file_id(fp->volatile_id)) {
1648 		write_unlock(&global_ft.lock);
1649 		pr_err("Still in use durable fd: %llu\n", fp->volatile_id);
1650 		return -EBADF;
1651 	}
1652 
1653 	/*
1654 	 * Initialize fp's connection binding before publishing fp into the
1655 	 * session's file table.  If __open_id() is ordered first, a
1656 	 * concurrent teardown that iterates the table can observe a valid
1657 	 * volatile_id with fp->conn == NULL and preserve a
1658 	 * partially-initialized fp.  fp owns a strong reference on the new
1659 	 * conn (see ksmbd_open_fd()); undo it on __open_id() failure.
1660 	 */
1661 	fp->conn = ksmbd_conn_get(conn);
1662 	fp->tcon = work->tcon;
1663 	write_unlock(&global_ft.lock);
1664 
1665 	old_f_state = fp->f_state;
1666 	fp->f_state = FP_NEW;
1667 
1668 	__open_id(&work->sess->file_table, fp, OPEN_ID_TYPE_VOLATILE_ID);
1669 	if (!has_file_id(fp->volatile_id)) {
1670 		write_lock(&global_ft.lock);
1671 		fp->conn = NULL;
1672 		fp->tcon = NULL;
1673 		write_unlock(&global_ft.lock);
1674 		ksmbd_conn_put(conn);
1675 		fp->f_state = old_f_state;
1676 		return -EBADF;
1677 	}
1678 
1679 	list_for_each_entry(smb_lock, &fp->lock_list, flist) {
1680 		smb_lock->conn = ksmbd_conn_get(conn);
1681 		spin_lock(&conn->llist_lock);
1682 		list_add_tail(&smb_lock->clist, &conn->lock_list);
1683 		spin_unlock(&conn->llist_lock);
1684 	}
1685 
1686 	ci = fp->f_ci;
1687 	down_write(&ci->m_lock);
1688 	list_for_each_entry_rcu(op, &ci->m_op_list, op_entry) {
1689 		if (op->conn)
1690 			continue;
1691 		op->conn = ksmbd_conn_get(fp->conn);
1692 	}
1693 	up_write(&ci->m_lock);
1694 
1695 	fp->owner.uid = fp->owner.gid = 0;
1696 	kfree(fp->owner.name);
1697 	fp->owner.name = NULL;
1698 
1699 	return 0;
1700 }
1701 
1702 int ksmbd_init_file_table(struct ksmbd_file_table *ft)
1703 {
1704 	ft->idr = kzalloc_obj(struct idr, KSMBD_DEFAULT_GFP);
1705 	if (!ft->idr)
1706 		return -ENOMEM;
1707 
1708 	idr_init(ft->idr);
1709 	rwlock_init(&ft->lock);
1710 	return 0;
1711 }
1712 
1713 void ksmbd_destroy_file_table(struct ksmbd_session *sess)
1714 {
1715 	struct ksmbd_file_table *ft = &sess->file_table;
1716 
1717 	if (!ft->idr)
1718 		return;
1719 
1720 	__close_file_table_ids(sess, NULL, session_fd_check, true);
1721 	idr_destroy(ft->idr);
1722 	kfree(ft->idr);
1723 	ft->idr = NULL;
1724 }
1725 
1726 int ksmbd_init_file_cache(void)
1727 {
1728 	filp_cache = kmem_cache_create("ksmbd_file_cache",
1729 				       sizeof(struct ksmbd_file), 0,
1730 				       SLAB_HWCACHE_ALIGN, NULL);
1731 	if (!filp_cache)
1732 		goto out;
1733 
1734 	init_waitqueue_head(&dh_wq);
1735 
1736 	return 0;
1737 
1738 out:
1739 	pr_err("failed to allocate file cache\n");
1740 	return -ENOMEM;
1741 }
1742 
1743 void ksmbd_exit_file_cache(void)
1744 {
1745 	kmem_cache_destroy(filp_cache);
1746 }
1747