add ax88772_rst

hc

2024-05-10 9999e48639b3cecb08ffb37358bcba3b48161b29

 kernel/fs/ceph/export.c
..
..
@@ -22,18 +22,77 @@
22
22
 	u64 ino, parent_ino;
23
23
 } __attribute__ ((packed));
24
24
 
25
+/*
26
+ * fh for snapped inode
27
+ */
28
+struct ceph_nfs_snapfh {
29
+	u64 ino;
30
+	u64 snapid;
31
+	u64 parent_ino;
32
+	u32 hash;
33
+} __attribute__ ((packed));
34
+
35
+static int ceph_encode_snapfh(struct inode *inode, u32 *rawfh, int *max_len,
36
+			      struct inode *parent_inode)
37
+{
38
+	static const int snap_handle_length =
39
+		sizeof(struct ceph_nfs_snapfh) >> 2;
40
+	struct ceph_nfs_snapfh *sfh = (void *)rawfh;
41
+	u64 snapid = ceph_snap(inode);
42
+	int ret;
43
+	bool no_parent = true;
44
+
45
+	if (*max_len < snap_handle_length) {
46
+		*max_len = snap_handle_length;
47
+		ret = FILEID_INVALID;
48
+		goto out;
49
+	}
50
+
51
+	ret =  -EINVAL;
52
+	if (snapid != CEPH_SNAPDIR) {
53
+		struct inode *dir;
54
+		struct dentry *dentry = d_find_alias(inode);
55
+		if (!dentry)
56
+			goto out;
57
+
58
+		rcu_read_lock();
59
+		dir = d_inode_rcu(dentry->d_parent);
60
+		if (ceph_snap(dir) != CEPH_SNAPDIR) {
61
+			sfh->parent_ino = ceph_ino(dir);
62
+			sfh->hash = ceph_dentry_hash(dir, dentry);
63
+			no_parent = false;
64
+		}
65
+		rcu_read_unlock();
66
+		dput(dentry);
67
+	}
68
+
69
+	if (no_parent) {
70
+		if (!S_ISDIR(inode->i_mode))
71
+			goto out;
72
+		sfh->parent_ino = sfh->ino;
73
+		sfh->hash = 0;
74
+	}
75
+	sfh->ino = ceph_ino(inode);
76
+	sfh->snapid = snapid;
77
+
78
+	*max_len = snap_handle_length;
79
+	ret = FILEID_BTRFS_WITH_PARENT;
80
+out:
81
+	dout("encode_snapfh %llx.%llx ret=%d\n", ceph_vinop(inode), ret);
82
+	return ret;
83
+}
84
+
25
85
 static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
26
86
 			  struct inode *parent_inode)
27
87
 {
88
+	static const int handle_length =
89
+		sizeof(struct ceph_nfs_fh) >> 2;
90
+	static const int connected_handle_length =
91
+		sizeof(struct ceph_nfs_confh) >> 2;
28
92
 	int type;
29
-	struct ceph_nfs_fh *fh = (void *)rawfh;
30
-	struct ceph_nfs_confh *cfh = (void *)rawfh;
31
-	int connected_handle_length = sizeof(*cfh)/4;
32
-	int handle_length = sizeof(*fh)/4;
33
93
 
34
-	/* don't re-export snaps */
35
94
 	if (ceph_snap(inode) != CEPH_NOSNAP)
36
-		return -EINVAL;
95
+		return ceph_encode_snapfh(inode, rawfh, max_len, parent_inode);
37
96
 
38
97
 	if (parent_inode && (*max_len < connected_handle_length)) {
39
98
 		*max_len = connected_handle_length;
..
..
@@ -44,6 +103,7 @@
44
103
 	}
45
104
 
46
105
 	if (parent_inode) {
106
+		struct ceph_nfs_confh *cfh = (void *)rawfh;
47
107
 		dout("encode_fh %llx with parent %llx\n",
48
108
 		     ceph_ino(inode), ceph_ino(parent_inode));
49
109
 		cfh->ino = ceph_ino(inode);
..
..
@@ -51,6 +111,7 @@
51
111
 		*max_len = connected_handle_length;
52
112
 		type = FILEID_INO32_GEN_PARENT;
53
113
 	} else {
114
+		struct ceph_nfs_fh *fh = (void *)rawfh;
54
115
 		dout("encode_fh %llx\n", ceph_ino(inode));
55
116
 		fh->ino = ceph_ino(inode);
56
117
 		*max_len = handle_length;
..
..
@@ -59,7 +120,7 @@
59
120
 	return type;
60
121
 }
61
122
 
62
-static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
123
+static struct inode *__lookup_inode(struct super_block *sb, u64 ino)
63
124
 {
64
125
 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
65
126
 	struct inode *inode;
..
..
@@ -68,6 +129,10 @@
68
129
 
69
130
 	vino.ino = ino;
70
131
 	vino.snap = CEPH_NOSNAP;
132
+
133
+	if (ceph_vino_is_reserved(vino))
134
+		return ERR_PTR(-ESTALE);
135
+
71
136
 	inode = ceph_find_inode(sb, vino);
72
137
 	if (!inode) {
73
138
 		struct ceph_mds_request *req;
..
..
@@ -81,7 +146,7 @@
81
146
 		mask = CEPH_STAT_CAP_INODE;
82
147
 		if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
83
148
 			mask |= CEPH_CAP_XATTR_SHARED;
84
-		req->r_args.getattr.mask = cpu_to_le32(mask);
149
+		req->r_args.lookupino.mask = cpu_to_le32(mask);
85
150
 
86
151
 		req->r_ino1 = vino;
87
152
 		req->r_num_caps = 1;
..
..
@@ -91,14 +156,125 @@
91
156
 			ihold(inode);
92
157
 		ceph_mdsc_put_request(req);
93
158
 		if (!inode)
94
-			return ERR_PTR(-ESTALE);
95
-		if (inode->i_nlink == 0) {
96
-			iput(inode);
97
-			return ERR_PTR(-ESTALE);
159
+			return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
160
+	}
161
+	return inode;
162
+}
163
+
164
+struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
165
+{
166
+	struct inode *inode = __lookup_inode(sb, ino);
167
+	if (IS_ERR(inode))
168
+		return inode;
169
+	if (inode->i_nlink == 0) {
170
+		iput(inode);
171
+		return ERR_PTR(-ESTALE);
172
+	}
173
+	return inode;
174
+}
175
+
176
+static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
177
+{
178
+	struct inode *inode = __lookup_inode(sb, ino);
179
+	int err;
180
+
181
+	if (IS_ERR(inode))
182
+		return ERR_CAST(inode);
183
+	/* We need LINK caps to reliably check i_nlink */
184
+	err = ceph_do_getattr(inode, CEPH_CAP_LINK_SHARED, false);
185
+	if (err) {
186
+		iput(inode);
187
+		return ERR_PTR(err);
188
+	}
189
+	/* -ESTALE if inode as been unlinked and no file is open */
190
+	if ((inode->i_nlink == 0) && (atomic_read(&inode->i_count) == 1)) {
191
+		iput(inode);
192
+		return ERR_PTR(-ESTALE);
193
+	}
194
+	return d_obtain_alias(inode);
195
+}
196
+
197
+static struct dentry *__snapfh_to_dentry(struct super_block *sb,
198
+					  struct ceph_nfs_snapfh *sfh,
199
+					  bool want_parent)
200
+{
201
+	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
202
+	struct ceph_mds_request *req;
203
+	struct inode *inode;
204
+	struct ceph_vino vino;
205
+	int mask;
206
+	int err;
207
+	bool unlinked = false;
208
+
209
+	if (want_parent) {
210
+		vino.ino = sfh->parent_ino;
211
+		if (sfh->snapid == CEPH_SNAPDIR)
212
+			vino.snap = CEPH_NOSNAP;
213
+		else if (sfh->ino == sfh->parent_ino)
214
+			vino.snap = CEPH_SNAPDIR;
215
+		else
216
+			vino.snap = sfh->snapid;
217
+	} else {
218
+		vino.ino = sfh->ino;
219
+		vino.snap = sfh->snapid;
220
+	}
221
+
222
+	if (ceph_vino_is_reserved(vino))
223
+		return ERR_PTR(-ESTALE);
224
+
225
+	inode = ceph_find_inode(sb, vino);
226
+	if (inode)
227
+		return d_obtain_alias(inode);
228
+
229
+	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
230
+				       USE_ANY_MDS);
231
+	if (IS_ERR(req))
232
+		return ERR_CAST(req);
233
+
234
+	mask = CEPH_STAT_CAP_INODE;
235
+	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
236
+		mask |= CEPH_CAP_XATTR_SHARED;
237
+	req->r_args.lookupino.mask = cpu_to_le32(mask);
238
+	if (vino.snap < CEPH_NOSNAP) {
239
+		req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
240
+		if (!want_parent && sfh->ino != sfh->parent_ino) {
241
+			req->r_args.lookupino.parent =
242
+					cpu_to_le64(sfh->parent_ino);
243
+			req->r_args.lookupino.hash =
244
+					cpu_to_le32(sfh->hash);
98
245
 		}
99
246
 	}
100
247
 
101
-	return d_obtain_alias(inode);
248
+	req->r_ino1 = vino;
249
+	req->r_num_caps = 1;
250
+	err = ceph_mdsc_do_request(mdsc, NULL, req);
251
+	inode = req->r_target_inode;
252
+	if (inode) {
253
+		if (vino.snap == CEPH_SNAPDIR) {
254
+			if (inode->i_nlink == 0)
255
+				unlinked = true;
256
+			inode = ceph_get_snapdir(inode);
257
+		} else if (ceph_snap(inode) == vino.snap) {
258
+			ihold(inode);
259
+		} else {
260
+			/* mds does not support lookup snapped inode */
261
+			err = -EOPNOTSUPP;
262
+			inode = NULL;
263
+		}
264
+	}
265
+	ceph_mdsc_put_request(req);
266
+
267
+	if (want_parent) {
268
+		dout("snapfh_to_parent %llx.%llx\n err=%d\n",
269
+		     vino.ino, vino.snap, err);
270
+	} else {
271
+		dout("snapfh_to_dentry %llx.%llx parent %llx hash %x err=%d",
272
+		      vino.ino, vino.snap, sfh->parent_ino, sfh->hash, err);
273
+	}
274
+	if (!inode)
275
+		return ERR_PTR(-ESTALE);
276
+	/* see comments in ceph_get_parent() */
277
+	return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
102
278
 }
103
279
 
104
280
 /*
..
..
@@ -109,6 +285,11 @@
109
285
 					int fh_len, int fh_type)
110
286
 {
111
287
 	struct ceph_nfs_fh *fh = (void *)fid->raw;
288
+
289
+	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
290
+		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
291
+		return __snapfh_to_dentry(sb, sfh, false);
292
+	}
112
293
 
113
294
 	if (fh_type != FILEID_INO32_GEN  &&
114
295
 	    fh_type != FILEID_INO32_GEN_PARENT)
..
..
@@ -168,13 +349,49 @@
168
349
 
169
350
 static struct dentry *ceph_get_parent(struct dentry *child)
170
351
 {
171
-	/* don't re-export snaps */
172
-	if (ceph_snap(d_inode(child)) != CEPH_NOSNAP)
173
-		return ERR_PTR(-EINVAL);
352
+	struct inode *inode = d_inode(child);
353
+	struct dentry *dn;
174
354
 
175
-	dout("get_parent %p ino %llx.%llx\n",
176
-	     child, ceph_vinop(d_inode(child)));
177
-	return __get_parent(child->d_sb, child, 0);
355
+	if (ceph_snap(inode) != CEPH_NOSNAP) {
356
+		struct inode* dir;
357
+		bool unlinked = false;
358
+		/* do not support non-directory */
359
+		if (!d_is_dir(child)) {
360
+			dn = ERR_PTR(-EINVAL);
361
+			goto out;
362
+		}
363
+		dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
364
+		if (IS_ERR(dir)) {
365
+			dn = ERR_CAST(dir);
366
+			goto out;
367
+		}
368
+		/* There can be multiple paths to access snapped inode.
369
+		 * For simplicity, treat snapdir of head inode as parent */
370
+		if (ceph_snap(inode) != CEPH_SNAPDIR) {
371
+			struct inode *snapdir = ceph_get_snapdir(dir);
372
+			if (dir->i_nlink == 0)
373
+				unlinked = true;
374
+			iput(dir);
375
+			if (IS_ERR(snapdir)) {
376
+				dn = ERR_CAST(snapdir);
377
+				goto out;
378
+			}
379
+			dir = snapdir;
380
+		}
381
+		/* If directory has already been deleted, futher get_parent
382
+		 * will fail. Do not mark snapdir dentry as disconnected,
383
+		 * this prevent exportfs from doing futher get_parent. */
384
+		if (unlinked)
385
+			dn = d_obtain_root(dir);
386
+		else
387
+			dn = d_obtain_alias(dir);
388
+	} else {
389
+		dn = __get_parent(child->d_sb, child, 0);
390
+	}
391
+out:
392
+	dout("get_parent %p ino %llx.%llx err=%ld\n",
393
+	     child, ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
394
+	return dn;
178
395
 }
179
396
 
180
397
 /*
..
..
@@ -186,6 +403,11 @@
186
403
 {
187
404
 	struct ceph_nfs_confh *cfh = (void *)fid->raw;
188
405
 	struct dentry *dentry;
406
+
407
+	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
408
+		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
409
+		return __snapfh_to_dentry(sb, sfh, true);
410
+	}
189
411
 
190
412
 	if (fh_type != FILEID_INO32_GEN_PARENT)
191
413
 		return NULL;
..
..
@@ -199,14 +421,115 @@
199
421
 	return dentry;
200
422
 }
201
423
 
424
+static int __get_snap_name(struct dentry *parent, char *name,
425
+			   struct dentry *child)
426
+{
427
+	struct inode *inode = d_inode(child);
428
+	struct inode *dir = d_inode(parent);
429
+	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
430
+	struct ceph_mds_request *req = NULL;
431
+	char *last_name = NULL;
432
+	unsigned next_offset = 2;
433
+	int err = -EINVAL;
434
+
435
+	if (ceph_ino(inode) != ceph_ino(dir))
436
+		goto out;
437
+	if (ceph_snap(inode) == CEPH_SNAPDIR) {
438
+		if (ceph_snap(dir) == CEPH_NOSNAP) {
439
+			strcpy(name, fsc->mount_options->snapdir_name);
440
+			err = 0;
441
+		}
442
+		goto out;
443
+	}
444
+	if (ceph_snap(dir) != CEPH_SNAPDIR)
445
+		goto out;
446
+
447
+	while (1) {
448
+		struct ceph_mds_reply_info_parsed *rinfo;
449
+		struct ceph_mds_reply_dir_entry *rde;
450
+		int i;
451
+
452
+		req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
453
+					       USE_AUTH_MDS);
454
+		if (IS_ERR(req)) {
455
+			err = PTR_ERR(req);
456
+			req = NULL;
457
+			goto out;
458
+		}
459
+		err = ceph_alloc_readdir_reply_buffer(req, inode);
460
+		if (err)
461
+			goto out;
462
+
463
+		req->r_direct_mode = USE_AUTH_MDS;
464
+		req->r_readdir_offset = next_offset;
465
+		req->r_args.readdir.flags =
466
+				cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
467
+		if (last_name) {
468
+			req->r_path2 = last_name;
469
+			last_name = NULL;
470
+		}
471
+
472
+		req->r_inode = dir;
473
+		ihold(dir);
474
+		req->r_dentry = dget(parent);
475
+
476
+		inode_lock(dir);
477
+		err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
478
+		inode_unlock(dir);
479
+
480
+		if (err < 0)
481
+			goto out;
482
+
483
+		rinfo = &req->r_reply_info;
484
+		for (i = 0; i < rinfo->dir_nr; i++) {
485
+			rde = rinfo->dir_entries + i;
486
+			BUG_ON(!rde->inode.in);
487
+			if (ceph_snap(inode) ==
488
+			    le64_to_cpu(rde->inode.in->snapid)) {
489
+				memcpy(name, rde->name, rde->name_len);
490
+				name[rde->name_len] = '\0';
491
+				err = 0;
492
+				goto out;
493
+			}
494
+		}
495
+
496
+		if (rinfo->dir_end)
497
+			break;
498
+
499
+		BUG_ON(rinfo->dir_nr <= 0);
500
+		rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
501
+		next_offset += rinfo->dir_nr;
502
+		last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
503
+		if (!last_name) {
504
+			err = -ENOMEM;
505
+			goto out;
506
+		}
507
+
508
+		ceph_mdsc_put_request(req);
509
+		req = NULL;
510
+	}
511
+	err = -ENOENT;
512
+out:
513
+	if (req)
514
+		ceph_mdsc_put_request(req);
515
+	kfree(last_name);
516
+	dout("get_snap_name %p ino %llx.%llx err=%d\n",
517
+	     child, ceph_vinop(inode), err);
518
+	return err;
519
+}
520
+
202
521
 static int ceph_get_name(struct dentry *parent, char *name,
203
522
 			 struct dentry *child)
204
523
 {
205
524
 	struct ceph_mds_client *mdsc;
206
525
 	struct ceph_mds_request *req;
526
+	struct inode *inode = d_inode(child);
207
527
 	int err;
208
528
 
209
-	mdsc = ceph_inode_to_client(d_inode(child))->mdsc;
529
+	if (ceph_snap(inode) != CEPH_NOSNAP)
530
+		return __get_snap_name(parent, name, child);
531
+
532
+	mdsc = ceph_inode_to_client(inode)->mdsc;
210
533
 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
211
534
 				       USE_ANY_MDS);
212
535
 	if (IS_ERR(req))
..
..
@@ -214,8 +537,8 @@
214
537
 
215
538
 	inode_lock(d_inode(parent));
216
539
 
217
-	req->r_inode = d_inode(child);
218
-	ihold(d_inode(child));
540
+	req->r_inode = inode;
541
+	ihold(inode);
219
542
 	req->r_ino2 = ceph_vino(d_inode(parent));
220
543
 	req->r_parent = d_inode(parent);
221
544
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
..
..
@@ -229,10 +552,10 @@
229
552
 		memcpy(name, rinfo->dname, rinfo->dname_len);
230
553
 		name[rinfo->dname_len] = 0;
231
554
 		dout("get_name %p ino %llx.%llx name %s\n",
232
-		     child, ceph_vinop(d_inode(child)), name);
555
+		     child, ceph_vinop(inode), name);
233
556
 	} else {
234
557
 		dout("get_name %p ino %llx.%llx err %d\n",
235
-		     child, ceph_vinop(d_inode(child)), err);
558
+		     child, ceph_vinop(inode), err);
236
559
 	}
237
560
 
238
561
 	ceph_mdsc_put_request(req);