xref: /linux/fs/ceph/xattr.c (revision 1f2367a39f17bd553a75e179a747f9b257bc9478) 
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 #include <linux/ceph/pagelist.h>
4 
5 #include "super.h"
6 #include "mds_client.h"
7 
8 #include <linux/ceph/decode.h>
9 
10 #include <linux/xattr.h>
11 #include <linux/posix_acl_xattr.h>
12 #include <linux/slab.h>
13 
14 #define XATTR_CEPH_PREFIX "ceph."
15 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
16 
17 static int __remove_xattr(struct ceph_inode_info *ci,
18 			  struct ceph_inode_xattr *xattr);
19 
20 static const struct xattr_handler ceph_other_xattr_handler;
21 
22 /*
23  * List of handlers for synthetic system.* attributes. Other
24  * attributes are handled directly.
25  */
26 const struct xattr_handler *ceph_xattr_handlers[] = {
27 #ifdef CONFIG_CEPH_FS_POSIX_ACL
28 	&posix_acl_access_xattr_handler,
29 	&posix_acl_default_xattr_handler,
30 #endif
31 	&ceph_other_xattr_handler,
32 	NULL,
33 };
34 
35 static bool ceph_is_valid_xattr(const char *name)
36 {
37 	return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
38 	       !strncmp(name, XATTR_SECURITY_PREFIX,
39 			XATTR_SECURITY_PREFIX_LEN) ||
40 	       !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
41 	       !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
42 }
43 
44 /*
45  * These define virtual xattrs exposing the recursive directory
46  * statistics and layout metadata.
47  */
48 struct ceph_vxattr {
49 	char *name;
50 	size_t name_size;	/* strlen(name) + 1 (for '\0') */
51 	size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
52 			      size_t size);
53 	bool (*exists_cb)(struct ceph_inode_info *ci);
54 	unsigned int flags;
55 };
56 
57 #define VXATTR_FLAG_READONLY		(1<<0)
58 #define VXATTR_FLAG_HIDDEN		(1<<1)
59 #define VXATTR_FLAG_RSTAT		(1<<2)
60 
61 /* layouts */
62 
63 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
64 {
65 	struct ceph_file_layout *fl = &ci->i_layout;
66 	return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
67 		fl->object_size > 0 || fl->pool_id >= 0 ||
68 		rcu_dereference_raw(fl->pool_ns) != NULL);
69 }
70 
71 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
72 				   size_t size)
73 {
74 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
75 	struct ceph_osd_client *osdc = &fsc->client->osdc;
76 	struct ceph_string *pool_ns;
77 	s64 pool = ci->i_layout.pool_id;
78 	const char *pool_name;
79 	const char *ns_field = " pool_namespace=";
80 	char buf[128];
81 	size_t len, total_len = 0;
82 	int ret;
83 
84 	pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
85 
86 	dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
87 	down_read(&osdc->lock);
88 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
89 	if (pool_name) {
90 		len = snprintf(buf, sizeof(buf),
91 		"stripe_unit=%u stripe_count=%u object_size=%u pool=",
92 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
93 	        ci->i_layout.object_size);
94 		total_len = len + strlen(pool_name);
95 	} else {
96 		len = snprintf(buf, sizeof(buf),
97 		"stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
98 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
99 	        ci->i_layout.object_size, (unsigned long long)pool);
100 		total_len = len;
101 	}
102 
103 	if (pool_ns)
104 		total_len += strlen(ns_field) + pool_ns->len;
105 
106 	if (!size) {
107 		ret = total_len;
108 	} else if (total_len > size) {
109 		ret = -ERANGE;
110 	} else {
111 		memcpy(val, buf, len);
112 		ret = len;
113 		if (pool_name) {
114 			len = strlen(pool_name);
115 			memcpy(val + ret, pool_name, len);
116 			ret += len;
117 		}
118 		if (pool_ns) {
119 			len = strlen(ns_field);
120 			memcpy(val + ret, ns_field, len);
121 			ret += len;
122 			memcpy(val + ret, pool_ns->str, pool_ns->len);
123 			ret += pool_ns->len;
124 		}
125 	}
126 	up_read(&osdc->lock);
127 	ceph_put_string(pool_ns);
128 	return ret;
129 }
130 
131 static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
132 					       char *val, size_t size)
133 {
134 	return snprintf(val, size, "%u", ci->i_layout.stripe_unit);
135 }
136 
137 static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
138 						char *val, size_t size)
139 {
140 	return snprintf(val, size, "%u", ci->i_layout.stripe_count);
141 }
142 
143 static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
144 					       char *val, size_t size)
145 {
146 	return snprintf(val, size, "%u", ci->i_layout.object_size);
147 }
148 
149 static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
150 					char *val, size_t size)
151 {
152 	int ret;
153 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
154 	struct ceph_osd_client *osdc = &fsc->client->osdc;
155 	s64 pool = ci->i_layout.pool_id;
156 	const char *pool_name;
157 
158 	down_read(&osdc->lock);
159 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
160 	if (pool_name)
161 		ret = snprintf(val, size, "%s", pool_name);
162 	else
163 		ret = snprintf(val, size, "%lld", (unsigned long long)pool);
164 	up_read(&osdc->lock);
165 	return ret;
166 }
167 
168 static size_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
169 						  char *val, size_t size)
170 {
171 	int ret = 0;
172 	struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
173 	if (ns) {
174 		ret = snprintf(val, size, "%.*s", (int)ns->len, ns->str);
175 		ceph_put_string(ns);
176 	}
177 	return ret;
178 }
179 
180 /* directories */
181 
182 static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
183 					size_t size)
184 {
185 	return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
186 }
187 
188 static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
189 				      size_t size)
190 {
191 	return snprintf(val, size, "%lld", ci->i_files);
192 }
193 
194 static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
195 					size_t size)
196 {
197 	return snprintf(val, size, "%lld", ci->i_subdirs);
198 }
199 
200 static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
201 					 size_t size)
202 {
203 	return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
204 }
205 
206 static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
207 				       size_t size)
208 {
209 	return snprintf(val, size, "%lld", ci->i_rfiles);
210 }
211 
212 static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
213 					 size_t size)
214 {
215 	return snprintf(val, size, "%lld", ci->i_rsubdirs);
216 }
217 
218 static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
219 				       size_t size)
220 {
221 	return snprintf(val, size, "%lld", ci->i_rbytes);
222 }
223 
224 static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
225 				       size_t size)
226 {
227 	return snprintf(val, size, "%lld.09%ld", ci->i_rctime.tv_sec,
228 			ci->i_rctime.tv_nsec);
229 }
230 
231 /* dir pin */
232 static bool ceph_vxattrcb_dir_pin_exists(struct ceph_inode_info *ci)
233 {
234 	return ci->i_dir_pin != -ENODATA;
235 }
236 
237 static size_t ceph_vxattrcb_dir_pin(struct ceph_inode_info *ci, char *val,
238                                     size_t size)
239 {
240 	return snprintf(val, size, "%d", (int)ci->i_dir_pin);
241 }
242 
243 /* quotas */
244 static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
245 {
246 	bool ret = false;
247 	spin_lock(&ci->i_ceph_lock);
248 	if ((ci->i_max_files || ci->i_max_bytes) &&
249 	    ci->i_vino.snap == CEPH_NOSNAP &&
250 	    ci->i_snap_realm &&
251 	    ci->i_snap_realm->ino == ci->i_vino.ino)
252 		ret = true;
253 	spin_unlock(&ci->i_ceph_lock);
254 	return ret;
255 }
256 
257 static size_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
258 				  size_t size)
259 {
260 	return snprintf(val, size, "max_bytes=%llu max_files=%llu",
261 			ci->i_max_bytes, ci->i_max_files);
262 }
263 
264 static size_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
265 					    char *val, size_t size)
266 {
267 	return snprintf(val, size, "%llu", ci->i_max_bytes);
268 }
269 
270 static size_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
271 					    char *val, size_t size)
272 {
273 	return snprintf(val, size, "%llu", ci->i_max_files);
274 }
275 
276 #define CEPH_XATTR_NAME(_type, _name)	XATTR_CEPH_PREFIX #_type "." #_name
277 #define CEPH_XATTR_NAME2(_type, _name, _name2)	\
278 	XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
279 
280 #define XATTR_NAME_CEPH(_type, _name, _flags)				\
281 	{								\
282 		.name = CEPH_XATTR_NAME(_type, _name),			\
283 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
284 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
285 		.exists_cb = NULL,					\
286 		.flags = (VXATTR_FLAG_READONLY | _flags),		\
287 	}
288 #define XATTR_RSTAT_FIELD(_type, _name)			\
289 	XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
290 #define XATTR_LAYOUT_FIELD(_type, _name, _field)			\
291 	{								\
292 		.name = CEPH_XATTR_NAME2(_type, _name, _field),	\
293 		.name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
294 		.getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
295 		.exists_cb = ceph_vxattrcb_layout_exists,	\
296 		.flags = VXATTR_FLAG_HIDDEN,			\
297 	}
298 #define XATTR_QUOTA_FIELD(_type, _name)					\
299 	{								\
300 		.name = CEPH_XATTR_NAME(_type, _name),			\
301 		.name_size = sizeof(CEPH_XATTR_NAME(_type, _name)),	\
302 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
303 		.exists_cb = ceph_vxattrcb_quota_exists,		\
304 		.flags = VXATTR_FLAG_HIDDEN,				\
305 	}
306 
307 static struct ceph_vxattr ceph_dir_vxattrs[] = {
308 	{
309 		.name = "ceph.dir.layout",
310 		.name_size = sizeof("ceph.dir.layout"),
311 		.getxattr_cb = ceph_vxattrcb_layout,
312 		.exists_cb = ceph_vxattrcb_layout_exists,
313 		.flags = VXATTR_FLAG_HIDDEN,
314 	},
315 	XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
316 	XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
317 	XATTR_LAYOUT_FIELD(dir, layout, object_size),
318 	XATTR_LAYOUT_FIELD(dir, layout, pool),
319 	XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
320 	XATTR_NAME_CEPH(dir, entries, 0),
321 	XATTR_NAME_CEPH(dir, files, 0),
322 	XATTR_NAME_CEPH(dir, subdirs, 0),
323 	XATTR_RSTAT_FIELD(dir, rentries),
324 	XATTR_RSTAT_FIELD(dir, rfiles),
325 	XATTR_RSTAT_FIELD(dir, rsubdirs),
326 	XATTR_RSTAT_FIELD(dir, rbytes),
327 	XATTR_RSTAT_FIELD(dir, rctime),
328 	{
329 		.name = "ceph.dir.pin",
330 		.name_size = sizeof("ceph.dir_pin"),
331 		.getxattr_cb = ceph_vxattrcb_dir_pin,
332 		.exists_cb = ceph_vxattrcb_dir_pin_exists,
333 		.flags = VXATTR_FLAG_HIDDEN,
334 	},
335 	{
336 		.name = "ceph.quota",
337 		.name_size = sizeof("ceph.quota"),
338 		.getxattr_cb = ceph_vxattrcb_quota,
339 		.exists_cb = ceph_vxattrcb_quota_exists,
340 		.flags = VXATTR_FLAG_HIDDEN,
341 	},
342 	XATTR_QUOTA_FIELD(quota, max_bytes),
343 	XATTR_QUOTA_FIELD(quota, max_files),
344 	{ .name = NULL, 0 }	/* Required table terminator */
345 };
346 static size_t ceph_dir_vxattrs_name_size;	/* total size of all names */
347 
348 /* files */
349 
350 static struct ceph_vxattr ceph_file_vxattrs[] = {
351 	{
352 		.name = "ceph.file.layout",
353 		.name_size = sizeof("ceph.file.layout"),
354 		.getxattr_cb = ceph_vxattrcb_layout,
355 		.exists_cb = ceph_vxattrcb_layout_exists,
356 		.flags = VXATTR_FLAG_HIDDEN,
357 	},
358 	XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
359 	XATTR_LAYOUT_FIELD(file, layout, stripe_count),
360 	XATTR_LAYOUT_FIELD(file, layout, object_size),
361 	XATTR_LAYOUT_FIELD(file, layout, pool),
362 	XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
363 	{ .name = NULL, 0 }	/* Required table terminator */
364 };
365 static size_t ceph_file_vxattrs_name_size;	/* total size of all names */
366 
367 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
368 {
369 	if (S_ISDIR(inode->i_mode))
370 		return ceph_dir_vxattrs;
371 	else if (S_ISREG(inode->i_mode))
372 		return ceph_file_vxattrs;
373 	return NULL;
374 }
375 
376 static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
377 {
378 	if (vxattrs == ceph_dir_vxattrs)
379 		return ceph_dir_vxattrs_name_size;
380 	if (vxattrs == ceph_file_vxattrs)
381 		return ceph_file_vxattrs_name_size;
382 	BUG_ON(vxattrs);
383 	return 0;
384 }
385 
386 /*
387  * Compute the aggregate size (including terminating '\0') of all
388  * virtual extended attribute names in the given vxattr table.
389  */
390 static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
391 {
392 	struct ceph_vxattr *vxattr;
393 	size_t size = 0;
394 
395 	for (vxattr = vxattrs; vxattr->name; vxattr++) {
396 		if (!(vxattr->flags & VXATTR_FLAG_HIDDEN))
397 			size += vxattr->name_size;
398 	}
399 
400 	return size;
401 }
402 
403 /* Routines called at initialization and exit time */
404 
405 void __init ceph_xattr_init(void)
406 {
407 	ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
408 	ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
409 }
410 
411 void ceph_xattr_exit(void)
412 {
413 	ceph_dir_vxattrs_name_size = 0;
414 	ceph_file_vxattrs_name_size = 0;
415 }
416 
417 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
418 						const char *name)
419 {
420 	struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
421 
422 	if (vxattr) {
423 		while (vxattr->name) {
424 			if (!strcmp(vxattr->name, name))
425 				return vxattr;
426 			vxattr++;
427 		}
428 	}
429 
430 	return NULL;
431 }
432 
433 static int __set_xattr(struct ceph_inode_info *ci,
434 			   const char *name, int name_len,
435 			   const char *val, int val_len,
436 			   int flags, int update_xattr,
437 			   struct ceph_inode_xattr **newxattr)
438 {
439 	struct rb_node **p;
440 	struct rb_node *parent = NULL;
441 	struct ceph_inode_xattr *xattr = NULL;
442 	int c;
443 	int new = 0;
444 
445 	p = &ci->i_xattrs.index.rb_node;
446 	while (*p) {
447 		parent = *p;
448 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
449 		c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
450 		if (c < 0)
451 			p = &(*p)->rb_left;
452 		else if (c > 0)
453 			p = &(*p)->rb_right;
454 		else {
455 			if (name_len == xattr->name_len)
456 				break;
457 			else if (name_len < xattr->name_len)
458 				p = &(*p)->rb_left;
459 			else
460 				p = &(*p)->rb_right;
461 		}
462 		xattr = NULL;
463 	}
464 
465 	if (update_xattr) {
466 		int err = 0;
467 
468 		if (xattr && (flags & XATTR_CREATE))
469 			err = -EEXIST;
470 		else if (!xattr && (flags & XATTR_REPLACE))
471 			err = -ENODATA;
472 		if (err) {
473 			kfree(name);
474 			kfree(val);
475 			kfree(*newxattr);
476 			return err;
477 		}
478 		if (update_xattr < 0) {
479 			if (xattr)
480 				__remove_xattr(ci, xattr);
481 			kfree(name);
482 			kfree(*newxattr);
483 			return 0;
484 		}
485 	}
486 
487 	if (!xattr) {
488 		new = 1;
489 		xattr = *newxattr;
490 		xattr->name = name;
491 		xattr->name_len = name_len;
492 		xattr->should_free_name = update_xattr;
493 
494 		ci->i_xattrs.count++;
495 		dout("__set_xattr count=%d\n", ci->i_xattrs.count);
496 	} else {
497 		kfree(*newxattr);
498 		*newxattr = NULL;
499 		if (xattr->should_free_val)
500 			kfree((void *)xattr->val);
501 
502 		if (update_xattr) {
503 			kfree((void *)name);
504 			name = xattr->name;
505 		}
506 		ci->i_xattrs.names_size -= xattr->name_len;
507 		ci->i_xattrs.vals_size -= xattr->val_len;
508 	}
509 	ci->i_xattrs.names_size += name_len;
510 	ci->i_xattrs.vals_size += val_len;
511 	if (val)
512 		xattr->val = val;
513 	else
514 		xattr->val = "";
515 
516 	xattr->val_len = val_len;
517 	xattr->dirty = update_xattr;
518 	xattr->should_free_val = (val && update_xattr);
519 
520 	if (new) {
521 		rb_link_node(&xattr->node, parent, p);
522 		rb_insert_color(&xattr->node, &ci->i_xattrs.index);
523 		dout("__set_xattr_val p=%p\n", p);
524 	}
525 
526 	dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
527 	     ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
528 
529 	return 0;
530 }
531 
532 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
533 			   const char *name)
534 {
535 	struct rb_node **p;
536 	struct rb_node *parent = NULL;
537 	struct ceph_inode_xattr *xattr = NULL;
538 	int name_len = strlen(name);
539 	int c;
540 
541 	p = &ci->i_xattrs.index.rb_node;
542 	while (*p) {
543 		parent = *p;
544 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
545 		c = strncmp(name, xattr->name, xattr->name_len);
546 		if (c == 0 && name_len > xattr->name_len)
547 			c = 1;
548 		if (c < 0)
549 			p = &(*p)->rb_left;
550 		else if (c > 0)
551 			p = &(*p)->rb_right;
552 		else {
553 			dout("__get_xattr %s: found %.*s\n", name,
554 			     xattr->val_len, xattr->val);
555 			return xattr;
556 		}
557 	}
558 
559 	dout("__get_xattr %s: not found\n", name);
560 
561 	return NULL;
562 }
563 
564 static void __free_xattr(struct ceph_inode_xattr *xattr)
565 {
566 	BUG_ON(!xattr);
567 
568 	if (xattr->should_free_name)
569 		kfree((void *)xattr->name);
570 	if (xattr->should_free_val)
571 		kfree((void *)xattr->val);
572 
573 	kfree(xattr);
574 }
575 
576 static int __remove_xattr(struct ceph_inode_info *ci,
577 			  struct ceph_inode_xattr *xattr)
578 {
579 	if (!xattr)
580 		return -ENODATA;
581 
582 	rb_erase(&xattr->node, &ci->i_xattrs.index);
583 
584 	if (xattr->should_free_name)
585 		kfree((void *)xattr->name);
586 	if (xattr->should_free_val)
587 		kfree((void *)xattr->val);
588 
589 	ci->i_xattrs.names_size -= xattr->name_len;
590 	ci->i_xattrs.vals_size -= xattr->val_len;
591 	ci->i_xattrs.count--;
592 	kfree(xattr);
593 
594 	return 0;
595 }
596 
597 static char *__copy_xattr_names(struct ceph_inode_info *ci,
598 				char *dest)
599 {
600 	struct rb_node *p;
601 	struct ceph_inode_xattr *xattr = NULL;
602 
603 	p = rb_first(&ci->i_xattrs.index);
604 	dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
605 
606 	while (p) {
607 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
608 		memcpy(dest, xattr->name, xattr->name_len);
609 		dest[xattr->name_len] = '\0';
610 
611 		dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
612 		     xattr->name_len, ci->i_xattrs.names_size);
613 
614 		dest += xattr->name_len + 1;
615 		p = rb_next(p);
616 	}
617 
618 	return dest;
619 }
620 
621 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
622 {
623 	struct rb_node *p, *tmp;
624 	struct ceph_inode_xattr *xattr = NULL;
625 
626 	p = rb_first(&ci->i_xattrs.index);
627 
628 	dout("__ceph_destroy_xattrs p=%p\n", p);
629 
630 	while (p) {
631 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
632 		tmp = p;
633 		p = rb_next(tmp);
634 		dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
635 		     xattr->name_len, xattr->name);
636 		rb_erase(tmp, &ci->i_xattrs.index);
637 
638 		__free_xattr(xattr);
639 	}
640 
641 	ci->i_xattrs.names_size = 0;
642 	ci->i_xattrs.vals_size = 0;
643 	ci->i_xattrs.index_version = 0;
644 	ci->i_xattrs.count = 0;
645 	ci->i_xattrs.index = RB_ROOT;
646 }
647 
648 static int __build_xattrs(struct inode *inode)
649 	__releases(ci->i_ceph_lock)
650 	__acquires(ci->i_ceph_lock)
651 {
652 	u32 namelen;
653 	u32 numattr = 0;
654 	void *p, *end;
655 	u32 len;
656 	const char *name, *val;
657 	struct ceph_inode_info *ci = ceph_inode(inode);
658 	int xattr_version;
659 	struct ceph_inode_xattr **xattrs = NULL;
660 	int err = 0;
661 	int i;
662 
663 	dout("__build_xattrs() len=%d\n",
664 	     ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
665 
666 	if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
667 		return 0; /* already built */
668 
669 	__ceph_destroy_xattrs(ci);
670 
671 start:
672 	/* updated internal xattr rb tree */
673 	if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
674 		p = ci->i_xattrs.blob->vec.iov_base;
675 		end = p + ci->i_xattrs.blob->vec.iov_len;
676 		ceph_decode_32_safe(&p, end, numattr, bad);
677 		xattr_version = ci->i_xattrs.version;
678 		spin_unlock(&ci->i_ceph_lock);
679 
680 		xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
681 				 GFP_NOFS);
682 		err = -ENOMEM;
683 		if (!xattrs)
684 			goto bad_lock;
685 
686 		for (i = 0; i < numattr; i++) {
687 			xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
688 					    GFP_NOFS);
689 			if (!xattrs[i])
690 				goto bad_lock;
691 		}
692 
693 		spin_lock(&ci->i_ceph_lock);
694 		if (ci->i_xattrs.version != xattr_version) {
695 			/* lost a race, retry */
696 			for (i = 0; i < numattr; i++)
697 				kfree(xattrs[i]);
698 			kfree(xattrs);
699 			xattrs = NULL;
700 			goto start;
701 		}
702 		err = -EIO;
703 		while (numattr--) {
704 			ceph_decode_32_safe(&p, end, len, bad);
705 			namelen = len;
706 			name = p;
707 			p += len;
708 			ceph_decode_32_safe(&p, end, len, bad);
709 			val = p;
710 			p += len;
711 
712 			err = __set_xattr(ci, name, namelen, val, len,
713 					  0, 0, &xattrs[numattr]);
714 
715 			if (err < 0)
716 				goto bad;
717 		}
718 		kfree(xattrs);
719 	}
720 	ci->i_xattrs.index_version = ci->i_xattrs.version;
721 	ci->i_xattrs.dirty = false;
722 
723 	return err;
724 bad_lock:
725 	spin_lock(&ci->i_ceph_lock);
726 bad:
727 	if (xattrs) {
728 		for (i = 0; i < numattr; i++)
729 			kfree(xattrs[i]);
730 		kfree(xattrs);
731 	}
732 	ci->i_xattrs.names_size = 0;
733 	return err;
734 }
735 
736 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
737 				    int val_size)
738 {
739 	/*
740 	 * 4 bytes for the length, and additional 4 bytes per each xattr name,
741 	 * 4 bytes per each value
742 	 */
743 	int size = 4 + ci->i_xattrs.count*(4 + 4) +
744 			     ci->i_xattrs.names_size +
745 			     ci->i_xattrs.vals_size;
746 	dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
747 	     ci->i_xattrs.count, ci->i_xattrs.names_size,
748 	     ci->i_xattrs.vals_size);
749 
750 	if (name_size)
751 		size += 4 + 4 + name_size + val_size;
752 
753 	return size;
754 }
755 
756 /*
757  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
758  * and swap into place.
759  */
760 void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
761 {
762 	struct rb_node *p;
763 	struct ceph_inode_xattr *xattr = NULL;
764 	void *dest;
765 
766 	dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
767 	if (ci->i_xattrs.dirty) {
768 		int need = __get_required_blob_size(ci, 0, 0);
769 
770 		BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
771 
772 		p = rb_first(&ci->i_xattrs.index);
773 		dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
774 
775 		ceph_encode_32(&dest, ci->i_xattrs.count);
776 		while (p) {
777 			xattr = rb_entry(p, struct ceph_inode_xattr, node);
778 
779 			ceph_encode_32(&dest, xattr->name_len);
780 			memcpy(dest, xattr->name, xattr->name_len);
781 			dest += xattr->name_len;
782 			ceph_encode_32(&dest, xattr->val_len);
783 			memcpy(dest, xattr->val, xattr->val_len);
784 			dest += xattr->val_len;
785 
786 			p = rb_next(p);
787 		}
788 
789 		/* adjust buffer len; it may be larger than we need */
790 		ci->i_xattrs.prealloc_blob->vec.iov_len =
791 			dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
792 
793 		if (ci->i_xattrs.blob)
794 			ceph_buffer_put(ci->i_xattrs.blob);
795 		ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
796 		ci->i_xattrs.prealloc_blob = NULL;
797 		ci->i_xattrs.dirty = false;
798 		ci->i_xattrs.version++;
799 	}
800 }
801 
802 static inline int __get_request_mask(struct inode *in) {
803 	struct ceph_mds_request *req = current->journal_info;
804 	int mask = 0;
805 	if (req && req->r_target_inode == in) {
806 		if (req->r_op == CEPH_MDS_OP_LOOKUP ||
807 		    req->r_op == CEPH_MDS_OP_LOOKUPINO ||
808 		    req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
809 		    req->r_op == CEPH_MDS_OP_GETATTR) {
810 			mask = le32_to_cpu(req->r_args.getattr.mask);
811 		} else if (req->r_op == CEPH_MDS_OP_OPEN ||
812 			   req->r_op == CEPH_MDS_OP_CREATE) {
813 			mask = le32_to_cpu(req->r_args.open.mask);
814 		}
815 	}
816 	return mask;
817 }
818 
819 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
820 		      size_t size)
821 {
822 	struct ceph_inode_info *ci = ceph_inode(inode);
823 	struct ceph_inode_xattr *xattr;
824 	struct ceph_vxattr *vxattr = NULL;
825 	int req_mask;
826 	int err;
827 
828 	/* let's see if a virtual xattr was requested */
829 	vxattr = ceph_match_vxattr(inode, name);
830 	if (vxattr) {
831 		int mask = 0;
832 		if (vxattr->flags & VXATTR_FLAG_RSTAT)
833 			mask |= CEPH_STAT_RSTAT;
834 		err = ceph_do_getattr(inode, mask, true);
835 		if (err)
836 			return err;
837 		err = -ENODATA;
838 		if (!(vxattr->exists_cb && !vxattr->exists_cb(ci)))
839 			err = vxattr->getxattr_cb(ci, value, size);
840 		return err;
841 	}
842 
843 	req_mask = __get_request_mask(inode);
844 
845 	spin_lock(&ci->i_ceph_lock);
846 	dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
847 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
848 
849 	if (ci->i_xattrs.version == 0 ||
850 	    !((req_mask & CEPH_CAP_XATTR_SHARED) ||
851 	      __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
852 		spin_unlock(&ci->i_ceph_lock);
853 
854 		/* security module gets xattr while filling trace */
855 		if (current->journal_info) {
856 			pr_warn_ratelimited("sync getxattr %p "
857 					    "during filling trace\n", inode);
858 			return -EBUSY;
859 		}
860 
861 		/* get xattrs from mds (if we don't already have them) */
862 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
863 		if (err)
864 			return err;
865 		spin_lock(&ci->i_ceph_lock);
866 	}
867 
868 	err = __build_xattrs(inode);
869 	if (err < 0)
870 		goto out;
871 
872 	err = -ENODATA;  /* == ENOATTR */
873 	xattr = __get_xattr(ci, name);
874 	if (!xattr)
875 		goto out;
876 
877 	err = -ERANGE;
878 	if (size && size < xattr->val_len)
879 		goto out;
880 
881 	err = xattr->val_len;
882 	if (size == 0)
883 		goto out;
884 
885 	memcpy(value, xattr->val, xattr->val_len);
886 
887 	if (current->journal_info &&
888 	    !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
889 		ci->i_ceph_flags |= CEPH_I_SEC_INITED;
890 out:
891 	spin_unlock(&ci->i_ceph_lock);
892 	return err;
893 }
894 
895 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
896 {
897 	struct inode *inode = d_inode(dentry);
898 	struct ceph_inode_info *ci = ceph_inode(inode);
899 	struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
900 	u32 vir_namelen = 0;
901 	u32 namelen;
902 	int err;
903 	u32 len;
904 	int i;
905 
906 	spin_lock(&ci->i_ceph_lock);
907 	dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
908 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
909 
910 	if (ci->i_xattrs.version == 0 ||
911 	    !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
912 		spin_unlock(&ci->i_ceph_lock);
913 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
914 		if (err)
915 			return err;
916 		spin_lock(&ci->i_ceph_lock);
917 	}
918 
919 	err = __build_xattrs(inode);
920 	if (err < 0)
921 		goto out;
922 	/*
923 	 * Start with virtual dir xattr names (if any) (including
924 	 * terminating '\0' characters for each).
925 	 */
926 	vir_namelen = ceph_vxattrs_name_size(vxattrs);
927 
928 	/* adding 1 byte per each variable due to the null termination */
929 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
930 	err = -ERANGE;
931 	if (size && vir_namelen + namelen > size)
932 		goto out;
933 
934 	err = namelen + vir_namelen;
935 	if (size == 0)
936 		goto out;
937 
938 	names = __copy_xattr_names(ci, names);
939 
940 	/* virtual xattr names, too */
941 	err = namelen;
942 	if (vxattrs) {
943 		for (i = 0; vxattrs[i].name; i++) {
944 			if (!(vxattrs[i].flags & VXATTR_FLAG_HIDDEN) &&
945 			    !(vxattrs[i].exists_cb &&
946 			      !vxattrs[i].exists_cb(ci))) {
947 				len = sprintf(names, "%s", vxattrs[i].name);
948 				names += len + 1;
949 				err += len + 1;
950 			}
951 		}
952 	}
953 
954 out:
955 	spin_unlock(&ci->i_ceph_lock);
956 	return err;
957 }
958 
959 static int ceph_sync_setxattr(struct inode *inode, const char *name,
960 			      const char *value, size_t size, int flags)
961 {
962 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
963 	struct ceph_inode_info *ci = ceph_inode(inode);
964 	struct ceph_mds_request *req;
965 	struct ceph_mds_client *mdsc = fsc->mdsc;
966 	struct ceph_pagelist *pagelist = NULL;
967 	int op = CEPH_MDS_OP_SETXATTR;
968 	int err;
969 
970 	if (size > 0) {
971 		/* copy value into pagelist */
972 		pagelist = ceph_pagelist_alloc(GFP_NOFS);
973 		if (!pagelist)
974 			return -ENOMEM;
975 
976 		err = ceph_pagelist_append(pagelist, value, size);
977 		if (err)
978 			goto out;
979 	} else if (!value) {
980 		if (flags & CEPH_XATTR_REPLACE)
981 			op = CEPH_MDS_OP_RMXATTR;
982 		else
983 			flags |= CEPH_XATTR_REMOVE;
984 	}
985 
986 	dout("setxattr value=%.*s\n", (int)size, value);
987 
988 	/* do request */
989 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
990 	if (IS_ERR(req)) {
991 		err = PTR_ERR(req);
992 		goto out;
993 	}
994 
995 	req->r_path2 = kstrdup(name, GFP_NOFS);
996 	if (!req->r_path2) {
997 		ceph_mdsc_put_request(req);
998 		err = -ENOMEM;
999 		goto out;
1000 	}
1001 
1002 	if (op == CEPH_MDS_OP_SETXATTR) {
1003 		req->r_args.setxattr.flags = cpu_to_le32(flags);
1004 		req->r_pagelist = pagelist;
1005 		pagelist = NULL;
1006 	}
1007 
1008 	req->r_inode = inode;
1009 	ihold(inode);
1010 	req->r_num_caps = 1;
1011 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1012 
1013 	dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
1014 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1015 	ceph_mdsc_put_request(req);
1016 	dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
1017 
1018 out:
1019 	if (pagelist)
1020 		ceph_pagelist_release(pagelist);
1021 	return err;
1022 }
1023 
1024 int __ceph_setxattr(struct inode *inode, const char *name,
1025 			const void *value, size_t size, int flags)
1026 {
1027 	struct ceph_vxattr *vxattr;
1028 	struct ceph_inode_info *ci = ceph_inode(inode);
1029 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1030 	struct ceph_cap_flush *prealloc_cf = NULL;
1031 	int issued;
1032 	int err;
1033 	int dirty = 0;
1034 	int name_len = strlen(name);
1035 	int val_len = size;
1036 	char *newname = NULL;
1037 	char *newval = NULL;
1038 	struct ceph_inode_xattr *xattr = NULL;
1039 	int required_blob_size;
1040 	bool check_realm = false;
1041 	bool lock_snap_rwsem = false;
1042 
1043 	if (ceph_snap(inode) != CEPH_NOSNAP)
1044 		return -EROFS;
1045 
1046 	vxattr = ceph_match_vxattr(inode, name);
1047 	if (vxattr) {
1048 		if (vxattr->flags & VXATTR_FLAG_READONLY)
1049 			return -EOPNOTSUPP;
1050 		if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1051 			check_realm = true;
1052 	}
1053 
1054 	/* pass any unhandled ceph.* xattrs through to the MDS */
1055 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1056 		goto do_sync_unlocked;
1057 
1058 	/* preallocate memory for xattr name, value, index node */
1059 	err = -ENOMEM;
1060 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
1061 	if (!newname)
1062 		goto out;
1063 
1064 	if (val_len) {
1065 		newval = kmemdup(value, val_len, GFP_NOFS);
1066 		if (!newval)
1067 			goto out;
1068 	}
1069 
1070 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1071 	if (!xattr)
1072 		goto out;
1073 
1074 	prealloc_cf = ceph_alloc_cap_flush();
1075 	if (!prealloc_cf)
1076 		goto out;
1077 
1078 	spin_lock(&ci->i_ceph_lock);
1079 retry:
1080 	issued = __ceph_caps_issued(ci, NULL);
1081 	if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1082 		goto do_sync;
1083 
1084 	if (!lock_snap_rwsem && !ci->i_head_snapc) {
1085 		lock_snap_rwsem = true;
1086 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
1087 			spin_unlock(&ci->i_ceph_lock);
1088 			down_read(&mdsc->snap_rwsem);
1089 			spin_lock(&ci->i_ceph_lock);
1090 			goto retry;
1091 		}
1092 	}
1093 
1094 	dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1095 	__build_xattrs(inode);
1096 
1097 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1098 
1099 	if (!ci->i_xattrs.prealloc_blob ||
1100 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1101 		struct ceph_buffer *blob;
1102 
1103 		spin_unlock(&ci->i_ceph_lock);
1104 		dout(" preaallocating new blob size=%d\n", required_blob_size);
1105 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1106 		if (!blob)
1107 			goto do_sync_unlocked;
1108 		spin_lock(&ci->i_ceph_lock);
1109 		if (ci->i_xattrs.prealloc_blob)
1110 			ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1111 		ci->i_xattrs.prealloc_blob = blob;
1112 		goto retry;
1113 	}
1114 
1115 	err = __set_xattr(ci, newname, name_len, newval, val_len,
1116 			  flags, value ? 1 : -1, &xattr);
1117 
1118 	if (!err) {
1119 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1120 					       &prealloc_cf);
1121 		ci->i_xattrs.dirty = true;
1122 		inode->i_ctime = current_time(inode);
1123 	}
1124 
1125 	spin_unlock(&ci->i_ceph_lock);
1126 	if (lock_snap_rwsem)
1127 		up_read(&mdsc->snap_rwsem);
1128 	if (dirty)
1129 		__mark_inode_dirty(inode, dirty);
1130 	ceph_free_cap_flush(prealloc_cf);
1131 	return err;
1132 
1133 do_sync:
1134 	spin_unlock(&ci->i_ceph_lock);
1135 do_sync_unlocked:
1136 	if (lock_snap_rwsem)
1137 		up_read(&mdsc->snap_rwsem);
1138 
1139 	/* security module set xattr while filling trace */
1140 	if (current->journal_info) {
1141 		pr_warn_ratelimited("sync setxattr %p "
1142 				    "during filling trace\n", inode);
1143 		err = -EBUSY;
1144 	} else {
1145 		err = ceph_sync_setxattr(inode, name, value, size, flags);
1146 		if (err >= 0 && check_realm) {
1147 			/* check if snaprealm was created for quota inode */
1148 			spin_lock(&ci->i_ceph_lock);
1149 			if ((ci->i_max_files || ci->i_max_bytes) &&
1150 			    !(ci->i_snap_realm &&
1151 			      ci->i_snap_realm->ino == ci->i_vino.ino))
1152 				err = -EOPNOTSUPP;
1153 			spin_unlock(&ci->i_ceph_lock);
1154 		}
1155 	}
1156 out:
1157 	ceph_free_cap_flush(prealloc_cf);
1158 	kfree(newname);
1159 	kfree(newval);
1160 	kfree(xattr);
1161 	return err;
1162 }
1163 
1164 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1165 				  struct dentry *dentry, struct inode *inode,
1166 				  const char *name, void *value, size_t size)
1167 {
1168 	if (!ceph_is_valid_xattr(name))
1169 		return -EOPNOTSUPP;
1170 	return __ceph_getxattr(inode, name, value, size);
1171 }
1172 
1173 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1174 				  struct dentry *unused, struct inode *inode,
1175 				  const char *name, const void *value,
1176 				  size_t size, int flags)
1177 {
1178 	if (!ceph_is_valid_xattr(name))
1179 		return -EOPNOTSUPP;
1180 	return __ceph_setxattr(inode, name, value, size, flags);
1181 }
1182 
1183 static const struct xattr_handler ceph_other_xattr_handler = {
1184 	.prefix = "",  /* match any name => handlers called with full name */
1185 	.get = ceph_get_xattr_handler,
1186 	.set = ceph_set_xattr_handler,
1187 };
1188 
1189 #ifdef CONFIG_SECURITY
1190 bool ceph_security_xattr_wanted(struct inode *in)
1191 {
1192 	return in->i_security != NULL;
1193 }
1194 
1195 bool ceph_security_xattr_deadlock(struct inode *in)
1196 {
1197 	struct ceph_inode_info *ci;
1198 	bool ret;
1199 	if (!in->i_security)
1200 		return false;
1201 	ci = ceph_inode(in);
1202 	spin_lock(&ci->i_ceph_lock);
1203 	ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1204 	      !(ci->i_xattrs.version > 0 &&
1205 		__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1206 	spin_unlock(&ci->i_ceph_lock);
1207 	return ret;
1208 }
1209 #endif
1210