xref: /linux/fs/ceph/xattr.c (revision d0fde6aae2bacdc024fff43461ba0f325375fa97)
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/security.h>
12 #include <linux/posix_acl_xattr.h>
13 #include <linux/slab.h>
14 
15 #define XATTR_CEPH_PREFIX "ceph."
16 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
17 
18 static int __remove_xattr(struct ceph_inode_info *ci,
19 			  struct ceph_inode_xattr *xattr);
20 
21 static bool ceph_is_valid_xattr(const char *name)
22 {
23 	return !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) ||
24 	       !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
25 	       !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
26 	       !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
27 }
28 
29 /*
30  * These define virtual xattrs exposing the recursive directory
31  * statistics and layout metadata.
32  */
33 struct ceph_vxattr {
34 	char *name;
35 	size_t name_size;	/* strlen(name) + 1 (for '\0') */
36 	ssize_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
37 			       size_t size);
38 	bool (*exists_cb)(struct ceph_inode_info *ci);
39 	unsigned int flags;
40 };
41 
42 #define VXATTR_FLAG_READONLY		(1<<0)
43 #define VXATTR_FLAG_HIDDEN		(1<<1)
44 #define VXATTR_FLAG_RSTAT		(1<<2)
45 #define VXATTR_FLAG_DIRSTAT		(1<<3)
46 
47 /* layouts */
48 
49 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
50 {
51 	struct ceph_file_layout *fl = &ci->i_layout;
52 	return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
53 		fl->object_size > 0 || fl->pool_id >= 0 ||
54 		rcu_dereference_raw(fl->pool_ns) != NULL);
55 }
56 
57 static ssize_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
58 				    size_t size)
59 {
60 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
61 	struct ceph_client *cl = fsc->client;
62 	struct ceph_osd_client *osdc = &fsc->client->osdc;
63 	struct ceph_string *pool_ns;
64 	s64 pool = ci->i_layout.pool_id;
65 	const char *pool_name;
66 	const char *ns_field = " pool_namespace=";
67 	char buf[128];
68 	size_t len, total_len = 0;
69 	ssize_t ret;
70 
71 	pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
72 
73 	doutc(cl, "%p\n", &ci->netfs.inode);
74 	down_read(&osdc->lock);
75 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
76 	if (pool_name) {
77 		len = snprintf(buf, sizeof(buf),
78 		"stripe_unit=%u stripe_count=%u object_size=%u pool=",
79 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
80 	        ci->i_layout.object_size);
81 		total_len = len + strlen(pool_name);
82 	} else {
83 		len = snprintf(buf, sizeof(buf),
84 		"stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
85 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
86 		ci->i_layout.object_size, pool);
87 		total_len = len;
88 	}
89 
90 	if (pool_ns)
91 		total_len += strlen(ns_field) + pool_ns->len;
92 
93 	ret = total_len;
94 	if (size >= total_len) {
95 		memcpy(val, buf, len);
96 		ret = len;
97 		if (pool_name) {
98 			len = strlen(pool_name);
99 			memcpy(val + ret, pool_name, len);
100 			ret += len;
101 		}
102 		if (pool_ns) {
103 			len = strlen(ns_field);
104 			memcpy(val + ret, ns_field, len);
105 			ret += len;
106 			memcpy(val + ret, pool_ns->str, pool_ns->len);
107 			ret += pool_ns->len;
108 		}
109 	}
110 	up_read(&osdc->lock);
111 	ceph_put_string(pool_ns);
112 	return ret;
113 }
114 
115 /*
116  * The convention with strings in xattrs is that they should not be NULL
117  * terminated, since we're returning the length with them. snprintf always
118  * NULL terminates however, so call it on a temporary buffer and then memcpy
119  * the result into place.
120  */
121 static __printf(3, 4)
122 int ceph_fmt_xattr(char *val, size_t size, const char *fmt, ...)
123 {
124 	int ret;
125 	va_list args;
126 	char buf[96]; /* NB: reevaluate size if new vxattrs are added */
127 
128 	va_start(args, fmt);
129 	ret = vsnprintf(buf, size ? sizeof(buf) : 0, fmt, args);
130 	va_end(args);
131 
132 	/* Sanity check */
133 	if (size && ret + 1 > sizeof(buf)) {
134 		WARN_ONCE(true, "Returned length too big (%d)", ret);
135 		return -E2BIG;
136 	}
137 
138 	if (ret <= size)
139 		memcpy(val, buf, ret);
140 	return ret;
141 }
142 
143 static ssize_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
144 						char *val, size_t size)
145 {
146 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_unit);
147 }
148 
149 static ssize_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
150 						 char *val, size_t size)
151 {
152 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_count);
153 }
154 
155 static ssize_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
156 						char *val, size_t size)
157 {
158 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.object_size);
159 }
160 
161 static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
162 					 char *val, size_t size)
163 {
164 	ssize_t ret;
165 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
166 	struct ceph_osd_client *osdc = &fsc->client->osdc;
167 	s64 pool = ci->i_layout.pool_id;
168 	const char *pool_name;
169 
170 	down_read(&osdc->lock);
171 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
172 	if (pool_name) {
173 		ret = strlen(pool_name);
174 		if (ret <= size)
175 			memcpy(val, pool_name, ret);
176 	} else {
177 		ret = ceph_fmt_xattr(val, size, "%lld", pool);
178 	}
179 	up_read(&osdc->lock);
180 	return ret;
181 }
182 
183 static ssize_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
184 						   char *val, size_t size)
185 {
186 	ssize_t ret = 0;
187 	struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
188 
189 	if (ns) {
190 		ret = ns->len;
191 		if (ret <= size)
192 			memcpy(val, ns->str, ret);
193 		ceph_put_string(ns);
194 	}
195 	return ret;
196 }
197 
198 /* directories */
199 
200 static ssize_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
201 					 size_t size)
202 {
203 	return ceph_fmt_xattr(val, size, "%lld", ci->i_files + ci->i_subdirs);
204 }
205 
206 static ssize_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
207 				       size_t size)
208 {
209 	return ceph_fmt_xattr(val, size, "%lld", ci->i_files);
210 }
211 
212 static ssize_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
213 					 size_t size)
214 {
215 	return ceph_fmt_xattr(val, size, "%lld", ci->i_subdirs);
216 }
217 
218 static ssize_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
219 					  size_t size)
220 {
221 	return ceph_fmt_xattr(val, size, "%lld",
222 				ci->i_rfiles + ci->i_rsubdirs);
223 }
224 
225 static ssize_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
226 					size_t size)
227 {
228 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rfiles);
229 }
230 
231 static ssize_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
232 					  size_t size)
233 {
234 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rsubdirs);
235 }
236 
237 static ssize_t ceph_vxattrcb_dir_rsnaps(struct ceph_inode_info *ci, char *val,
238 					  size_t size)
239 {
240 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rsnaps);
241 }
242 
243 static ssize_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
244 					size_t size)
245 {
246 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rbytes);
247 }
248 
249 static ssize_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
250 					size_t size)
251 {
252 	return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_rctime.tv_sec,
253 				ci->i_rctime.tv_nsec);
254 }
255 
256 /* dir pin */
257 static bool ceph_vxattrcb_dir_pin_exists(struct ceph_inode_info *ci)
258 {
259 	return ci->i_dir_pin != -ENODATA;
260 }
261 
262 static ssize_t ceph_vxattrcb_dir_pin(struct ceph_inode_info *ci, char *val,
263 				     size_t size)
264 {
265 	return ceph_fmt_xattr(val, size, "%d", (int)ci->i_dir_pin);
266 }
267 
268 /* quotas */
269 static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
270 {
271 	bool ret = false;
272 	spin_lock(&ci->i_ceph_lock);
273 	if ((ci->i_max_files || ci->i_max_bytes) &&
274 	    ci->i_vino.snap == CEPH_NOSNAP &&
275 	    ci->i_snap_realm &&
276 	    ci->i_snap_realm->ino == ci->i_vino.ino)
277 		ret = true;
278 	spin_unlock(&ci->i_ceph_lock);
279 	return ret;
280 }
281 
282 static ssize_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
283 				   size_t size)
284 {
285 	return ceph_fmt_xattr(val, size, "max_bytes=%llu max_files=%llu",
286 				ci->i_max_bytes, ci->i_max_files);
287 }
288 
289 static ssize_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
290 					     char *val, size_t size)
291 {
292 	return ceph_fmt_xattr(val, size, "%llu", ci->i_max_bytes);
293 }
294 
295 static ssize_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
296 					     char *val, size_t size)
297 {
298 	return ceph_fmt_xattr(val, size, "%llu", ci->i_max_files);
299 }
300 
301 /* snapshots */
302 static bool ceph_vxattrcb_snap_btime_exists(struct ceph_inode_info *ci)
303 {
304 	return (ci->i_snap_btime.tv_sec != 0 || ci->i_snap_btime.tv_nsec != 0);
305 }
306 
307 static ssize_t ceph_vxattrcb_snap_btime(struct ceph_inode_info *ci, char *val,
308 					size_t size)
309 {
310 	return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_snap_btime.tv_sec,
311 				ci->i_snap_btime.tv_nsec);
312 }
313 
314 static ssize_t ceph_vxattrcb_cluster_fsid(struct ceph_inode_info *ci,
315 					  char *val, size_t size)
316 {
317 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
318 
319 	return ceph_fmt_xattr(val, size, "%pU", &fsc->client->fsid);
320 }
321 
322 static ssize_t ceph_vxattrcb_client_id(struct ceph_inode_info *ci,
323 				       char *val, size_t size)
324 {
325 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
326 
327 	return ceph_fmt_xattr(val, size, "client%lld",
328 			      ceph_client_gid(fsc->client));
329 }
330 
331 static ssize_t ceph_vxattrcb_caps(struct ceph_inode_info *ci, char *val,
332 					size_t size)
333 {
334 	int issued;
335 
336 	spin_lock(&ci->i_ceph_lock);
337 	issued = __ceph_caps_issued(ci, NULL);
338 	spin_unlock(&ci->i_ceph_lock);
339 
340 	return ceph_fmt_xattr(val, size, "%s/0x%x",
341 			      ceph_cap_string(issued), issued);
342 }
343 
344 static ssize_t ceph_vxattrcb_auth_mds(struct ceph_inode_info *ci,
345 				       char *val, size_t size)
346 {
347 	int ret;
348 
349 	spin_lock(&ci->i_ceph_lock);
350 	ret = ceph_fmt_xattr(val, size, "%d",
351 			     ci->i_auth_cap ? ci->i_auth_cap->session->s_mds : -1);
352 	spin_unlock(&ci->i_ceph_lock);
353 	return ret;
354 }
355 
356 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
357 static bool ceph_vxattrcb_fscrypt_auth_exists(struct ceph_inode_info *ci)
358 {
359 	return ci->fscrypt_auth_len;
360 }
361 
362 static ssize_t ceph_vxattrcb_fscrypt_auth(struct ceph_inode_info *ci,
363 					  char *val, size_t size)
364 {
365 	if (size) {
366 		if (size < ci->fscrypt_auth_len)
367 			return -ERANGE;
368 		memcpy(val, ci->fscrypt_auth, ci->fscrypt_auth_len);
369 	}
370 	return ci->fscrypt_auth_len;
371 }
372 #endif /* CONFIG_FS_ENCRYPTION */
373 
374 #define CEPH_XATTR_NAME(_type, _name)	XATTR_CEPH_PREFIX #_type "." #_name
375 #define CEPH_XATTR_NAME2(_type, _name, _name2)	\
376 	XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
377 
378 #define XATTR_NAME_CEPH(_type, _name, _flags)				\
379 	{								\
380 		.name = CEPH_XATTR_NAME(_type, _name),			\
381 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
382 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
383 		.exists_cb = NULL,					\
384 		.flags = (VXATTR_FLAG_READONLY | _flags),		\
385 	}
386 #define XATTR_RSTAT_FIELD(_type, _name)			\
387 	XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
388 #define XATTR_RSTAT_FIELD_UPDATABLE(_type, _name)			\
389 	{								\
390 		.name = CEPH_XATTR_NAME(_type, _name),			\
391 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)),	\
392 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
393 		.exists_cb = NULL,					\
394 		.flags = VXATTR_FLAG_RSTAT,				\
395 	}
396 #define XATTR_LAYOUT_FIELD(_type, _name, _field)			\
397 	{								\
398 		.name = CEPH_XATTR_NAME2(_type, _name, _field),	\
399 		.name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
400 		.getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
401 		.exists_cb = ceph_vxattrcb_layout_exists,	\
402 		.flags = VXATTR_FLAG_HIDDEN,			\
403 	}
404 #define XATTR_QUOTA_FIELD(_type, _name)					\
405 	{								\
406 		.name = CEPH_XATTR_NAME(_type, _name),			\
407 		.name_size = sizeof(CEPH_XATTR_NAME(_type, _name)),	\
408 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
409 		.exists_cb = ceph_vxattrcb_quota_exists,		\
410 		.flags = VXATTR_FLAG_HIDDEN,				\
411 	}
412 
413 static struct ceph_vxattr ceph_dir_vxattrs[] = {
414 	{
415 		.name = "ceph.dir.layout",
416 		.name_size = sizeof("ceph.dir.layout"),
417 		.getxattr_cb = ceph_vxattrcb_layout,
418 		.exists_cb = ceph_vxattrcb_layout_exists,
419 		.flags = VXATTR_FLAG_HIDDEN,
420 	},
421 	XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
422 	XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
423 	XATTR_LAYOUT_FIELD(dir, layout, object_size),
424 	XATTR_LAYOUT_FIELD(dir, layout, pool),
425 	XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
426 	XATTR_NAME_CEPH(dir, entries, VXATTR_FLAG_DIRSTAT),
427 	XATTR_NAME_CEPH(dir, files, VXATTR_FLAG_DIRSTAT),
428 	XATTR_NAME_CEPH(dir, subdirs, VXATTR_FLAG_DIRSTAT),
429 	XATTR_RSTAT_FIELD(dir, rentries),
430 	XATTR_RSTAT_FIELD(dir, rfiles),
431 	XATTR_RSTAT_FIELD(dir, rsubdirs),
432 	XATTR_RSTAT_FIELD(dir, rsnaps),
433 	XATTR_RSTAT_FIELD(dir, rbytes),
434 	XATTR_RSTAT_FIELD_UPDATABLE(dir, rctime),
435 	{
436 		.name = "ceph.dir.pin",
437 		.name_size = sizeof("ceph.dir.pin"),
438 		.getxattr_cb = ceph_vxattrcb_dir_pin,
439 		.exists_cb = ceph_vxattrcb_dir_pin_exists,
440 		.flags = VXATTR_FLAG_HIDDEN,
441 	},
442 	{
443 		.name = "ceph.quota",
444 		.name_size = sizeof("ceph.quota"),
445 		.getxattr_cb = ceph_vxattrcb_quota,
446 		.exists_cb = ceph_vxattrcb_quota_exists,
447 		.flags = VXATTR_FLAG_HIDDEN,
448 	},
449 	XATTR_QUOTA_FIELD(quota, max_bytes),
450 	XATTR_QUOTA_FIELD(quota, max_files),
451 	{
452 		.name = "ceph.snap.btime",
453 		.name_size = sizeof("ceph.snap.btime"),
454 		.getxattr_cb = ceph_vxattrcb_snap_btime,
455 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
456 		.flags = VXATTR_FLAG_READONLY,
457 	},
458 	{
459 		.name = "ceph.caps",
460 		.name_size = sizeof("ceph.caps"),
461 		.getxattr_cb = ceph_vxattrcb_caps,
462 		.exists_cb = NULL,
463 		.flags = VXATTR_FLAG_HIDDEN,
464 	},
465 	{ .name = NULL, 0 }	/* Required table terminator */
466 };
467 
468 /* files */
469 
470 static struct ceph_vxattr ceph_file_vxattrs[] = {
471 	{
472 		.name = "ceph.file.layout",
473 		.name_size = sizeof("ceph.file.layout"),
474 		.getxattr_cb = ceph_vxattrcb_layout,
475 		.exists_cb = ceph_vxattrcb_layout_exists,
476 		.flags = VXATTR_FLAG_HIDDEN,
477 	},
478 	XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
479 	XATTR_LAYOUT_FIELD(file, layout, stripe_count),
480 	XATTR_LAYOUT_FIELD(file, layout, object_size),
481 	XATTR_LAYOUT_FIELD(file, layout, pool),
482 	XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
483 	{
484 		.name = "ceph.snap.btime",
485 		.name_size = sizeof("ceph.snap.btime"),
486 		.getxattr_cb = ceph_vxattrcb_snap_btime,
487 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
488 		.flags = VXATTR_FLAG_READONLY,
489 	},
490 	{
491 		.name = "ceph.caps",
492 		.name_size = sizeof("ceph.caps"),
493 		.getxattr_cb = ceph_vxattrcb_caps,
494 		.exists_cb = NULL,
495 		.flags = VXATTR_FLAG_HIDDEN,
496 	},
497 	{ .name = NULL, 0 }	/* Required table terminator */
498 };
499 
500 static struct ceph_vxattr ceph_common_vxattrs[] = {
501 	{
502 		.name = "ceph.cluster_fsid",
503 		.name_size = sizeof("ceph.cluster_fsid"),
504 		.getxattr_cb = ceph_vxattrcb_cluster_fsid,
505 		.exists_cb = NULL,
506 		.flags = VXATTR_FLAG_READONLY,
507 	},
508 	{
509 		.name = "ceph.client_id",
510 		.name_size = sizeof("ceph.client_id"),
511 		.getxattr_cb = ceph_vxattrcb_client_id,
512 		.exists_cb = NULL,
513 		.flags = VXATTR_FLAG_READONLY,
514 	},
515 	{
516 		.name = "ceph.auth_mds",
517 		.name_size = sizeof("ceph.auth_mds"),
518 		.getxattr_cb = ceph_vxattrcb_auth_mds,
519 		.exists_cb = NULL,
520 		.flags = VXATTR_FLAG_READONLY,
521 	},
522 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
523 	{
524 		.name = "ceph.fscrypt.auth",
525 		.name_size = sizeof("ceph.fscrypt.auth"),
526 		.getxattr_cb = ceph_vxattrcb_fscrypt_auth,
527 		.exists_cb = ceph_vxattrcb_fscrypt_auth_exists,
528 		.flags = VXATTR_FLAG_READONLY,
529 	},
530 #endif /* CONFIG_FS_ENCRYPTION */
531 	{ .name = NULL, 0 }	/* Required table terminator */
532 };
533 
534 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
535 {
536 	if (S_ISDIR(inode->i_mode))
537 		return ceph_dir_vxattrs;
538 	else if (S_ISREG(inode->i_mode))
539 		return ceph_file_vxattrs;
540 	return NULL;
541 }
542 
543 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
544 						const char *name)
545 {
546 	struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
547 
548 	if (vxattr) {
549 		while (vxattr->name) {
550 			if (!strcmp(vxattr->name, name))
551 				return vxattr;
552 			vxattr++;
553 		}
554 	}
555 
556 	vxattr = ceph_common_vxattrs;
557 	while (vxattr->name) {
558 		if (!strcmp(vxattr->name, name))
559 			return vxattr;
560 		vxattr++;
561 	}
562 
563 	return NULL;
564 }
565 
566 #define MAX_XATTR_VAL_PRINT_LEN 256
567 
568 static int __set_xattr(struct ceph_inode_info *ci,
569 			   const char *name, int name_len,
570 			   const char *val, int val_len,
571 			   int flags, int update_xattr,
572 			   struct ceph_inode_xattr **newxattr)
573 {
574 	struct inode *inode = &ci->netfs.inode;
575 	struct ceph_client *cl = ceph_inode_to_client(inode);
576 	struct rb_node **p;
577 	struct rb_node *parent = NULL;
578 	struct ceph_inode_xattr *xattr = NULL;
579 	int c;
580 	int new = 0;
581 
582 	p = &ci->i_xattrs.index.rb_node;
583 	while (*p) {
584 		parent = *p;
585 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
586 		c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
587 		if (c < 0)
588 			p = &(*p)->rb_left;
589 		else if (c > 0)
590 			p = &(*p)->rb_right;
591 		else {
592 			if (name_len == xattr->name_len)
593 				break;
594 			else if (name_len < xattr->name_len)
595 				p = &(*p)->rb_left;
596 			else
597 				p = &(*p)->rb_right;
598 		}
599 		xattr = NULL;
600 	}
601 
602 	if (update_xattr) {
603 		int err = 0;
604 
605 		if (xattr && (flags & XATTR_CREATE))
606 			err = -EEXIST;
607 		else if (!xattr && (flags & XATTR_REPLACE))
608 			err = -ENODATA;
609 		if (err) {
610 			kfree(name);
611 			kfree(val);
612 			kfree(*newxattr);
613 			return err;
614 		}
615 		if (update_xattr < 0) {
616 			if (xattr)
617 				__remove_xattr(ci, xattr);
618 			kfree(name);
619 			kfree(*newxattr);
620 			return 0;
621 		}
622 	}
623 
624 	if (!xattr) {
625 		new = 1;
626 		xattr = *newxattr;
627 		xattr->name = name;
628 		xattr->name_len = name_len;
629 		xattr->should_free_name = update_xattr;
630 
631 		ci->i_xattrs.count++;
632 		doutc(cl, "count=%d\n", ci->i_xattrs.count);
633 	} else {
634 		kfree(*newxattr);
635 		*newxattr = NULL;
636 		if (xattr->should_free_val)
637 			kfree(xattr->val);
638 
639 		if (update_xattr) {
640 			kfree(name);
641 			name = xattr->name;
642 		}
643 		ci->i_xattrs.names_size -= xattr->name_len;
644 		ci->i_xattrs.vals_size -= xattr->val_len;
645 	}
646 	ci->i_xattrs.names_size += name_len;
647 	ci->i_xattrs.vals_size += val_len;
648 	if (val)
649 		xattr->val = val;
650 	else
651 		xattr->val = "";
652 
653 	xattr->val_len = val_len;
654 	xattr->dirty = update_xattr;
655 	xattr->should_free_val = (val && update_xattr);
656 
657 	if (new) {
658 		rb_link_node(&xattr->node, parent, p);
659 		rb_insert_color(&xattr->node, &ci->i_xattrs.index);
660 		doutc(cl, "p=%p\n", p);
661 	}
662 
663 	doutc(cl, "added %p %llx.%llx xattr %p %.*s=%.*s%s\n", inode,
664 	      ceph_vinop(inode), xattr, name_len, name, min(val_len,
665 	      MAX_XATTR_VAL_PRINT_LEN), val,
666 	      val_len > MAX_XATTR_VAL_PRINT_LEN ? "..." : "");
667 
668 	return 0;
669 }
670 
671 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
672 			   const char *name)
673 {
674 	struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
675 	struct rb_node **p;
676 	struct rb_node *parent = NULL;
677 	struct ceph_inode_xattr *xattr = NULL;
678 	int name_len = strlen(name);
679 	int c;
680 
681 	p = &ci->i_xattrs.index.rb_node;
682 	while (*p) {
683 		parent = *p;
684 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
685 		c = strncmp(name, xattr->name, xattr->name_len);
686 		if (c == 0 && name_len > xattr->name_len)
687 			c = 1;
688 		if (c < 0)
689 			p = &(*p)->rb_left;
690 		else if (c > 0)
691 			p = &(*p)->rb_right;
692 		else {
693 			int len = min(xattr->val_len, MAX_XATTR_VAL_PRINT_LEN);
694 
695 			doutc(cl, "%s found %.*s%s\n", name, len, xattr->val,
696 			      xattr->val_len > len ? "..." : "");
697 			return xattr;
698 		}
699 	}
700 
701 	doutc(cl, "%s not found\n", name);
702 
703 	return NULL;
704 }
705 
706 static void __free_xattr(struct ceph_inode_xattr *xattr)
707 {
708 	BUG_ON(!xattr);
709 
710 	if (xattr->should_free_name)
711 		kfree(xattr->name);
712 	if (xattr->should_free_val)
713 		kfree(xattr->val);
714 
715 	kfree(xattr);
716 }
717 
718 static int __remove_xattr(struct ceph_inode_info *ci,
719 			  struct ceph_inode_xattr *xattr)
720 {
721 	if (!xattr)
722 		return -ENODATA;
723 
724 	rb_erase(&xattr->node, &ci->i_xattrs.index);
725 
726 	if (xattr->should_free_name)
727 		kfree(xattr->name);
728 	if (xattr->should_free_val)
729 		kfree(xattr->val);
730 
731 	ci->i_xattrs.names_size -= xattr->name_len;
732 	ci->i_xattrs.vals_size -= xattr->val_len;
733 	ci->i_xattrs.count--;
734 	kfree(xattr);
735 
736 	return 0;
737 }
738 
739 static char *__copy_xattr_names(struct ceph_inode_info *ci,
740 				char *dest)
741 {
742 	struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
743 	struct rb_node *p;
744 	struct ceph_inode_xattr *xattr = NULL;
745 
746 	p = rb_first(&ci->i_xattrs.index);
747 	doutc(cl, "count=%d\n", ci->i_xattrs.count);
748 
749 	while (p) {
750 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
751 		memcpy(dest, xattr->name, xattr->name_len);
752 		dest[xattr->name_len] = '\0';
753 
754 		doutc(cl, "dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
755 		      xattr->name_len, ci->i_xattrs.names_size);
756 
757 		dest += xattr->name_len + 1;
758 		p = rb_next(p);
759 	}
760 
761 	return dest;
762 }
763 
764 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
765 {
766 	struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
767 	struct rb_node *p, *tmp;
768 	struct ceph_inode_xattr *xattr = NULL;
769 
770 	p = rb_first(&ci->i_xattrs.index);
771 
772 	doutc(cl, "p=%p\n", p);
773 
774 	while (p) {
775 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
776 		tmp = p;
777 		p = rb_next(tmp);
778 		doutc(cl, "next p=%p (%.*s)\n", p, xattr->name_len, xattr->name);
779 		rb_erase(tmp, &ci->i_xattrs.index);
780 
781 		__free_xattr(xattr);
782 	}
783 
784 	ci->i_xattrs.names_size = 0;
785 	ci->i_xattrs.vals_size = 0;
786 	ci->i_xattrs.index_version = 0;
787 	ci->i_xattrs.count = 0;
788 	ci->i_xattrs.index = RB_ROOT;
789 }
790 
791 static int __build_xattrs(struct inode *inode)
792 	__releases(ci->i_ceph_lock)
793 	__acquires(ci->i_ceph_lock)
794 {
795 	struct ceph_client *cl = ceph_inode_to_client(inode);
796 	u32 namelen;
797 	u32 numattr = 0;
798 	void *p, *end;
799 	u32 len;
800 	const char *name, *val;
801 	struct ceph_inode_info *ci = ceph_inode(inode);
802 	u64 xattr_version;
803 	struct ceph_inode_xattr **xattrs = NULL;
804 	int err = 0;
805 	int i;
806 
807 	doutc(cl, "len=%d\n",
808 	      ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
809 
810 	if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
811 		return 0; /* already built */
812 
813 	__ceph_destroy_xattrs(ci);
814 
815 start:
816 	/* updated internal xattr rb tree */
817 	if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
818 		p = ci->i_xattrs.blob->vec.iov_base;
819 		end = p + ci->i_xattrs.blob->vec.iov_len;
820 		ceph_decode_32_safe(&p, end, numattr, bad);
821 		xattr_version = ci->i_xattrs.version;
822 		spin_unlock(&ci->i_ceph_lock);
823 
824 		xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
825 				 GFP_NOFS);
826 		err = -ENOMEM;
827 		if (!xattrs)
828 			goto bad_lock;
829 
830 		for (i = 0; i < numattr; i++) {
831 			xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
832 					    GFP_NOFS);
833 			if (!xattrs[i])
834 				goto bad_lock;
835 		}
836 
837 		spin_lock(&ci->i_ceph_lock);
838 		if (ci->i_xattrs.version != xattr_version) {
839 			/* lost a race, retry */
840 			for (i = 0; i < numattr; i++)
841 				kfree(xattrs[i]);
842 			kfree(xattrs);
843 			xattrs = NULL;
844 			goto start;
845 		}
846 		err = -EIO;
847 		while (numattr--) {
848 			ceph_decode_32_safe(&p, end, len, bad);
849 			namelen = len;
850 			name = p;
851 			p += len;
852 			ceph_decode_32_safe(&p, end, len, bad);
853 			val = p;
854 			p += len;
855 
856 			err = __set_xattr(ci, name, namelen, val, len,
857 					  0, 0, &xattrs[numattr]);
858 
859 			if (err < 0)
860 				goto bad;
861 		}
862 		kfree(xattrs);
863 	}
864 	ci->i_xattrs.index_version = ci->i_xattrs.version;
865 	ci->i_xattrs.dirty = false;
866 
867 	return err;
868 bad_lock:
869 	spin_lock(&ci->i_ceph_lock);
870 bad:
871 	if (xattrs) {
872 		for (i = 0; i < numattr; i++)
873 			kfree(xattrs[i]);
874 		kfree(xattrs);
875 	}
876 	ci->i_xattrs.names_size = 0;
877 	return err;
878 }
879 
880 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
881 				    int val_size)
882 {
883 	struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
884 
885 	/*
886 	 * 4 bytes for the length, and additional 4 bytes per each xattr name,
887 	 * 4 bytes per each value
888 	 */
889 	int size = 4 + ci->i_xattrs.count*(4 + 4) +
890 			     ci->i_xattrs.names_size +
891 			     ci->i_xattrs.vals_size;
892 	doutc(cl, "c=%d names.size=%d vals.size=%d\n", ci->i_xattrs.count,
893 	      ci->i_xattrs.names_size, ci->i_xattrs.vals_size);
894 
895 	if (name_size)
896 		size += 4 + 4 + name_size + val_size;
897 
898 	return size;
899 }
900 
901 /*
902  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
903  * and swap into place.  It returns the old i_xattrs.blob (or NULL) so
904  * that it can be freed by the caller as the i_ceph_lock is likely to be
905  * held.
906  */
907 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
908 {
909 	struct inode *inode = &ci->netfs.inode;
910 	struct ceph_client *cl = ceph_inode_to_client(inode);
911 	struct rb_node *p;
912 	struct ceph_inode_xattr *xattr = NULL;
913 	struct ceph_buffer *old_blob = NULL;
914 	void *dest;
915 
916 	doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
917 	if (ci->i_xattrs.dirty) {
918 		int need = __get_required_blob_size(ci, 0, 0);
919 
920 		BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
921 
922 		p = rb_first(&ci->i_xattrs.index);
923 		dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
924 
925 		ceph_encode_32(&dest, ci->i_xattrs.count);
926 		while (p) {
927 			xattr = rb_entry(p, struct ceph_inode_xattr, node);
928 
929 			ceph_encode_32(&dest, xattr->name_len);
930 			memcpy(dest, xattr->name, xattr->name_len);
931 			dest += xattr->name_len;
932 			ceph_encode_32(&dest, xattr->val_len);
933 			memcpy(dest, xattr->val, xattr->val_len);
934 			dest += xattr->val_len;
935 
936 			p = rb_next(p);
937 		}
938 
939 		/* adjust buffer len; it may be larger than we need */
940 		ci->i_xattrs.prealloc_blob->vec.iov_len =
941 			dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
942 
943 		if (ci->i_xattrs.blob)
944 			old_blob = ci->i_xattrs.blob;
945 		ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
946 		ci->i_xattrs.prealloc_blob = NULL;
947 		ci->i_xattrs.dirty = false;
948 		ci->i_xattrs.version++;
949 	}
950 
951 	return old_blob;
952 }
953 
954 static inline int __get_request_mask(struct inode *in) {
955 	struct ceph_mds_request *req = current->journal_info;
956 	int mask = 0;
957 	if (req && req->r_target_inode == in) {
958 		if (req->r_op == CEPH_MDS_OP_LOOKUP ||
959 		    req->r_op == CEPH_MDS_OP_LOOKUPINO ||
960 		    req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
961 		    req->r_op == CEPH_MDS_OP_GETATTR) {
962 			mask = le32_to_cpu(req->r_args.getattr.mask);
963 		} else if (req->r_op == CEPH_MDS_OP_OPEN ||
964 			   req->r_op == CEPH_MDS_OP_CREATE) {
965 			mask = le32_to_cpu(req->r_args.open.mask);
966 		}
967 	}
968 	return mask;
969 }
970 
971 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
972 		      size_t size)
973 {
974 	struct ceph_client *cl = ceph_inode_to_client(inode);
975 	struct ceph_inode_info *ci = ceph_inode(inode);
976 	struct ceph_inode_xattr *xattr;
977 	struct ceph_vxattr *vxattr;
978 	int req_mask;
979 	ssize_t err;
980 
981 	if (strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
982 		goto handle_non_vxattrs;
983 
984 	/* let's see if a virtual xattr was requested */
985 	vxattr = ceph_match_vxattr(inode, name);
986 	if (vxattr) {
987 		int mask = 0;
988 		if (vxattr->flags & VXATTR_FLAG_RSTAT)
989 			mask |= CEPH_STAT_RSTAT;
990 		if (vxattr->flags & VXATTR_FLAG_DIRSTAT)
991 			mask |= CEPH_CAP_FILE_SHARED;
992 		err = ceph_do_getattr(inode, mask, true);
993 		if (err)
994 			return err;
995 		err = -ENODATA;
996 		if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
997 			err = vxattr->getxattr_cb(ci, value, size);
998 			if (size && size < err)
999 				err = -ERANGE;
1000 		}
1001 		return err;
1002 	} else {
1003 		err = ceph_do_getvxattr(inode, name, value, size);
1004 		/* this would happen with a new client and old server combo */
1005 		if (err == -EOPNOTSUPP)
1006 			err = -ENODATA;
1007 		return err;
1008 	}
1009 handle_non_vxattrs:
1010 	req_mask = __get_request_mask(inode);
1011 
1012 	spin_lock(&ci->i_ceph_lock);
1013 	doutc(cl, "%p %llx.%llx name '%s' ver=%lld index_ver=%lld\n", inode,
1014 	      ceph_vinop(inode), name, ci->i_xattrs.version,
1015 	      ci->i_xattrs.index_version);
1016 
1017 	if (ci->i_xattrs.version == 0 ||
1018 	    !((req_mask & CEPH_CAP_XATTR_SHARED) ||
1019 	      __ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1))) {
1020 		spin_unlock(&ci->i_ceph_lock);
1021 
1022 		/* security module gets xattr while filling trace */
1023 		if (current->journal_info) {
1024 			pr_warn_ratelimited_client(cl,
1025 				"sync %p %llx.%llx during filling trace\n",
1026 				inode, ceph_vinop(inode));
1027 			return -EBUSY;
1028 		}
1029 
1030 		/* get xattrs from mds (if we don't already have them) */
1031 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1032 		if (err)
1033 			return err;
1034 		spin_lock(&ci->i_ceph_lock);
1035 	}
1036 
1037 	err = __build_xattrs(inode);
1038 	if (err < 0)
1039 		goto out;
1040 
1041 	err = -ENODATA;  /* == ENOATTR */
1042 	xattr = __get_xattr(ci, name);
1043 	if (!xattr)
1044 		goto out;
1045 
1046 	err = -ERANGE;
1047 	if (size && size < xattr->val_len)
1048 		goto out;
1049 
1050 	err = xattr->val_len;
1051 	if (size == 0)
1052 		goto out;
1053 
1054 	memcpy(value, xattr->val, xattr->val_len);
1055 
1056 	if (current->journal_info &&
1057 	    !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
1058 	    security_ismaclabel(name + XATTR_SECURITY_PREFIX_LEN))
1059 		ci->i_ceph_flags |= CEPH_I_SEC_INITED;
1060 out:
1061 	spin_unlock(&ci->i_ceph_lock);
1062 	return err;
1063 }
1064 
1065 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
1066 {
1067 	struct inode *inode = d_inode(dentry);
1068 	struct ceph_client *cl = ceph_inode_to_client(inode);
1069 	struct ceph_inode_info *ci = ceph_inode(inode);
1070 	bool len_only = (size == 0);
1071 	u32 namelen;
1072 	int err;
1073 
1074 	spin_lock(&ci->i_ceph_lock);
1075 	doutc(cl, "%p %llx.%llx ver=%lld index_ver=%lld\n", inode,
1076 	      ceph_vinop(inode), ci->i_xattrs.version,
1077 	      ci->i_xattrs.index_version);
1078 
1079 	if (ci->i_xattrs.version == 0 ||
1080 	    !__ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1)) {
1081 		spin_unlock(&ci->i_ceph_lock);
1082 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1083 		if (err)
1084 			return err;
1085 		spin_lock(&ci->i_ceph_lock);
1086 	}
1087 
1088 	err = __build_xattrs(inode);
1089 	if (err < 0)
1090 		goto out;
1091 
1092 	/* add 1 byte for each xattr due to the null termination */
1093 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
1094 	if (!len_only) {
1095 		if (namelen > size) {
1096 			err = -ERANGE;
1097 			goto out;
1098 		}
1099 		names = __copy_xattr_names(ci, names);
1100 		size -= namelen;
1101 	}
1102 	err = namelen;
1103 out:
1104 	spin_unlock(&ci->i_ceph_lock);
1105 	return err;
1106 }
1107 
1108 static int ceph_sync_setxattr(struct inode *inode, const char *name,
1109 			      const char *value, size_t size, int flags)
1110 {
1111 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
1112 	struct ceph_client *cl = ceph_inode_to_client(inode);
1113 	struct ceph_inode_info *ci = ceph_inode(inode);
1114 	struct ceph_mds_request *req;
1115 	struct ceph_mds_client *mdsc = fsc->mdsc;
1116 	struct ceph_osd_client *osdc = &fsc->client->osdc;
1117 	struct ceph_pagelist *pagelist = NULL;
1118 	int op = CEPH_MDS_OP_SETXATTR;
1119 	int err;
1120 
1121 	if (size > 0) {
1122 		/* copy value into pagelist */
1123 		pagelist = ceph_pagelist_alloc(GFP_NOFS);
1124 		if (!pagelist)
1125 			return -ENOMEM;
1126 
1127 		err = ceph_pagelist_append(pagelist, value, size);
1128 		if (err)
1129 			goto out;
1130 	} else if (!value) {
1131 		if (flags & CEPH_XATTR_REPLACE)
1132 			op = CEPH_MDS_OP_RMXATTR;
1133 		else
1134 			flags |= CEPH_XATTR_REMOVE;
1135 	}
1136 
1137 	doutc(cl, "name %s value size %zu\n", name, size);
1138 
1139 	/* do request */
1140 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1141 	if (IS_ERR(req)) {
1142 		err = PTR_ERR(req);
1143 		goto out;
1144 	}
1145 
1146 	req->r_path2 = kstrdup(name, GFP_NOFS);
1147 	if (!req->r_path2) {
1148 		ceph_mdsc_put_request(req);
1149 		err = -ENOMEM;
1150 		goto out;
1151 	}
1152 
1153 	if (op == CEPH_MDS_OP_SETXATTR) {
1154 		req->r_args.setxattr.flags = cpu_to_le32(flags);
1155 		req->r_args.setxattr.osdmap_epoch =
1156 			cpu_to_le32(osdc->osdmap->epoch);
1157 		req->r_pagelist = pagelist;
1158 		pagelist = NULL;
1159 	}
1160 
1161 	req->r_inode = inode;
1162 	ihold(inode);
1163 	req->r_num_caps = 1;
1164 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1165 
1166 	doutc(cl, "xattr.ver (before): %lld\n", ci->i_xattrs.version);
1167 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1168 	ceph_mdsc_put_request(req);
1169 	doutc(cl, "xattr.ver (after): %lld\n", ci->i_xattrs.version);
1170 
1171 out:
1172 	if (pagelist)
1173 		ceph_pagelist_release(pagelist);
1174 	return err;
1175 }
1176 
1177 int __ceph_setxattr(struct inode *inode, const char *name,
1178 			const void *value, size_t size, int flags)
1179 {
1180 	struct ceph_client *cl = ceph_inode_to_client(inode);
1181 	struct ceph_vxattr *vxattr;
1182 	struct ceph_inode_info *ci = ceph_inode(inode);
1183 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
1184 	struct ceph_cap_flush *prealloc_cf = NULL;
1185 	struct ceph_buffer *old_blob = NULL;
1186 	int issued;
1187 	int err;
1188 	int dirty = 0;
1189 	int name_len = strlen(name);
1190 	int val_len = size;
1191 	char *newname = NULL;
1192 	char *newval = NULL;
1193 	struct ceph_inode_xattr *xattr = NULL;
1194 	int required_blob_size;
1195 	bool check_realm = false;
1196 	bool lock_snap_rwsem = false;
1197 
1198 	if (ceph_snap(inode) != CEPH_NOSNAP)
1199 		return -EROFS;
1200 
1201 	vxattr = ceph_match_vxattr(inode, name);
1202 	if (vxattr) {
1203 		if (vxattr->flags & VXATTR_FLAG_READONLY)
1204 			return -EOPNOTSUPP;
1205 		if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1206 			check_realm = true;
1207 	}
1208 
1209 	/* pass any unhandled ceph.* xattrs through to the MDS */
1210 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1211 		goto do_sync_unlocked;
1212 
1213 	/* preallocate memory for xattr name, value, index node */
1214 	err = -ENOMEM;
1215 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
1216 	if (!newname)
1217 		goto out;
1218 
1219 	if (val_len) {
1220 		newval = kmemdup(value, val_len, GFP_NOFS);
1221 		if (!newval)
1222 			goto out;
1223 	}
1224 
1225 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1226 	if (!xattr)
1227 		goto out;
1228 
1229 	prealloc_cf = ceph_alloc_cap_flush();
1230 	if (!prealloc_cf)
1231 		goto out;
1232 
1233 	spin_lock(&ci->i_ceph_lock);
1234 retry:
1235 	issued = __ceph_caps_issued(ci, NULL);
1236 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1237 	if ((ci->i_xattrs.version == 0) || !(issued & CEPH_CAP_XATTR_EXCL) ||
1238 	    (required_blob_size > mdsc->mdsmap->m_max_xattr_size)) {
1239 		doutc(cl, "sync version: %llu size: %d max: %llu\n",
1240 		      ci->i_xattrs.version, required_blob_size,
1241 		      mdsc->mdsmap->m_max_xattr_size);
1242 		goto do_sync;
1243 	}
1244 
1245 	if (!lock_snap_rwsem && !ci->i_head_snapc) {
1246 		lock_snap_rwsem = true;
1247 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
1248 			spin_unlock(&ci->i_ceph_lock);
1249 			down_read(&mdsc->snap_rwsem);
1250 			spin_lock(&ci->i_ceph_lock);
1251 			goto retry;
1252 		}
1253 	}
1254 
1255 	doutc(cl, "%p %llx.%llx name '%s' issued %s\n", inode,
1256 	      ceph_vinop(inode), name, ceph_cap_string(issued));
1257 	__build_xattrs(inode);
1258 
1259 	if (!ci->i_xattrs.prealloc_blob ||
1260 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1261 		struct ceph_buffer *blob;
1262 
1263 		spin_unlock(&ci->i_ceph_lock);
1264 		ceph_buffer_put(old_blob); /* Shouldn't be required */
1265 		doutc(cl, " pre-allocating new blob size=%d\n",
1266 		      required_blob_size);
1267 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1268 		if (!blob)
1269 			goto do_sync_unlocked;
1270 		spin_lock(&ci->i_ceph_lock);
1271 		/* prealloc_blob can't be released while holding i_ceph_lock */
1272 		if (ci->i_xattrs.prealloc_blob)
1273 			old_blob = ci->i_xattrs.prealloc_blob;
1274 		ci->i_xattrs.prealloc_blob = blob;
1275 		goto retry;
1276 	}
1277 
1278 	err = __set_xattr(ci, newname, name_len, newval, val_len,
1279 			  flags, value ? 1 : -1, &xattr);
1280 
1281 	if (!err) {
1282 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1283 					       &prealloc_cf);
1284 		ci->i_xattrs.dirty = true;
1285 		inode_set_ctime_current(inode);
1286 	}
1287 
1288 	spin_unlock(&ci->i_ceph_lock);
1289 	ceph_buffer_put(old_blob);
1290 	if (lock_snap_rwsem)
1291 		up_read(&mdsc->snap_rwsem);
1292 	if (dirty)
1293 		__mark_inode_dirty(inode, dirty);
1294 	ceph_free_cap_flush(prealloc_cf);
1295 	return err;
1296 
1297 do_sync:
1298 	spin_unlock(&ci->i_ceph_lock);
1299 do_sync_unlocked:
1300 	if (lock_snap_rwsem)
1301 		up_read(&mdsc->snap_rwsem);
1302 
1303 	/* security module set xattr while filling trace */
1304 	if (current->journal_info) {
1305 		pr_warn_ratelimited_client(cl,
1306 				"sync %p %llx.%llx during filling trace\n",
1307 				inode, ceph_vinop(inode));
1308 		err = -EBUSY;
1309 	} else {
1310 		err = ceph_sync_setxattr(inode, name, value, size, flags);
1311 		if (err >= 0 && check_realm) {
1312 			/* check if snaprealm was created for quota inode */
1313 			spin_lock(&ci->i_ceph_lock);
1314 			if ((ci->i_max_files || ci->i_max_bytes) &&
1315 			    !(ci->i_snap_realm &&
1316 			      ci->i_snap_realm->ino == ci->i_vino.ino))
1317 				err = -EOPNOTSUPP;
1318 			spin_unlock(&ci->i_ceph_lock);
1319 		}
1320 	}
1321 out:
1322 	ceph_free_cap_flush(prealloc_cf);
1323 	kfree(newname);
1324 	kfree(newval);
1325 	kfree(xattr);
1326 	return err;
1327 }
1328 
1329 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1330 				  struct dentry *dentry, struct inode *inode,
1331 				  const char *name, void *value, size_t size)
1332 {
1333 	if (!ceph_is_valid_xattr(name))
1334 		return -EOPNOTSUPP;
1335 	return __ceph_getxattr(inode, name, value, size);
1336 }
1337 
1338 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1339 				  struct mnt_idmap *idmap,
1340 				  struct dentry *unused, struct inode *inode,
1341 				  const char *name, const void *value,
1342 				  size_t size, int flags)
1343 {
1344 	if (!ceph_is_valid_xattr(name))
1345 		return -EOPNOTSUPP;
1346 	return __ceph_setxattr(inode, name, value, size, flags);
1347 }
1348 
1349 static const struct xattr_handler ceph_other_xattr_handler = {
1350 	.prefix = "",  /* match any name => handlers called with full name */
1351 	.get = ceph_get_xattr_handler,
1352 	.set = ceph_set_xattr_handler,
1353 };
1354 
1355 #ifdef CONFIG_SECURITY
1356 bool ceph_security_xattr_wanted(struct inode *in)
1357 {
1358 	return in->i_security != NULL;
1359 }
1360 
1361 bool ceph_security_xattr_deadlock(struct inode *in)
1362 {
1363 	struct ceph_inode_info *ci;
1364 	bool ret;
1365 	if (!in->i_security)
1366 		return false;
1367 	ci = ceph_inode(in);
1368 	spin_lock(&ci->i_ceph_lock);
1369 	ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1370 	      !(ci->i_xattrs.version > 0 &&
1371 		__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1372 	spin_unlock(&ci->i_ceph_lock);
1373 	return ret;
1374 }
1375 
1376 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1377 int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1378 			   struct ceph_acl_sec_ctx *as_ctx)
1379 {
1380 	struct ceph_pagelist *pagelist = as_ctx->pagelist;
1381 	const char *name;
1382 	size_t name_len;
1383 	int err;
1384 
1385 	err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1386 					    &name, &as_ctx->sec_ctx,
1387 					    &as_ctx->sec_ctxlen);
1388 	if (err < 0) {
1389 		WARN_ON_ONCE(err != -EOPNOTSUPP);
1390 		err = 0; /* do nothing */
1391 		goto out;
1392 	}
1393 
1394 	err = -ENOMEM;
1395 	if (!pagelist) {
1396 		pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1397 		if (!pagelist)
1398 			goto out;
1399 		err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1400 		if (err)
1401 			goto out;
1402 		ceph_pagelist_encode_32(pagelist, 1);
1403 	}
1404 
1405 	/*
1406 	 * FIXME: Make security_dentry_init_security() generic. Currently
1407 	 * It only supports single security module and only selinux has
1408 	 * dentry_init_security hook.
1409 	 */
1410 	name_len = strlen(name);
1411 	err = ceph_pagelist_reserve(pagelist,
1412 				    4 * 2 + name_len + as_ctx->sec_ctxlen);
1413 	if (err)
1414 		goto out;
1415 
1416 	if (as_ctx->pagelist) {
1417 		/* update count of KV pairs */
1418 		BUG_ON(pagelist->length <= sizeof(__le32));
1419 		if (list_is_singular(&pagelist->head)) {
1420 			le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1421 		} else {
1422 			struct page *page = list_first_entry(&pagelist->head,
1423 							     struct page, lru);
1424 			void *addr = kmap_atomic(page);
1425 			le32_add_cpu((__le32*)addr, 1);
1426 			kunmap_atomic(addr);
1427 		}
1428 	} else {
1429 		as_ctx->pagelist = pagelist;
1430 	}
1431 
1432 	ceph_pagelist_encode_32(pagelist, name_len);
1433 	ceph_pagelist_append(pagelist, name, name_len);
1434 
1435 	ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1436 	ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1437 
1438 	err = 0;
1439 out:
1440 	if (pagelist && !as_ctx->pagelist)
1441 		ceph_pagelist_release(pagelist);
1442 	return err;
1443 }
1444 #endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1445 #endif /* CONFIG_SECURITY */
1446 
1447 void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1448 {
1449 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1450 	posix_acl_release(as_ctx->acl);
1451 	posix_acl_release(as_ctx->default_acl);
1452 #endif
1453 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1454 	security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1455 #endif
1456 #ifdef CONFIG_FS_ENCRYPTION
1457 	kfree(as_ctx->fscrypt_auth);
1458 #endif
1459 	if (as_ctx->pagelist)
1460 		ceph_pagelist_release(as_ctx->pagelist);
1461 }
1462 
1463 /*
1464  * List of handlers for synthetic system.* attributes. Other
1465  * attributes are handled directly.
1466  */
1467 const struct xattr_handler * const ceph_xattr_handlers[] = {
1468 	&ceph_other_xattr_handler,
1469 	NULL,
1470 };
1471