xref: /openbmc/linux/fs/ceph/xattr.c (revision 1b36955c)
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_client(ci->netfs.inode.i_sb);
61 	struct ceph_osd_client *osdc = &fsc->client->osdc;
62 	struct ceph_string *pool_ns;
63 	s64 pool = ci->i_layout.pool_id;
64 	const char *pool_name;
65 	const char *ns_field = " pool_namespace=";
66 	char buf[128];
67 	size_t len, total_len = 0;
68 	ssize_t ret;
69 
70 	pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
71 
72 	dout("ceph_vxattrcb_layout %p\n", &ci->netfs.inode);
73 	down_read(&osdc->lock);
74 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
75 	if (pool_name) {
76 		len = snprintf(buf, sizeof(buf),
77 		"stripe_unit=%u stripe_count=%u object_size=%u pool=",
78 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
79 	        ci->i_layout.object_size);
80 		total_len = len + strlen(pool_name);
81 	} else {
82 		len = snprintf(buf, sizeof(buf),
83 		"stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
84 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
85 		ci->i_layout.object_size, pool);
86 		total_len = len;
87 	}
88 
89 	if (pool_ns)
90 		total_len += strlen(ns_field) + pool_ns->len;
91 
92 	ret = total_len;
93 	if (size >= total_len) {
94 		memcpy(val, buf, len);
95 		ret = len;
96 		if (pool_name) {
97 			len = strlen(pool_name);
98 			memcpy(val + ret, pool_name, len);
99 			ret += len;
100 		}
101 		if (pool_ns) {
102 			len = strlen(ns_field);
103 			memcpy(val + ret, ns_field, len);
104 			ret += len;
105 			memcpy(val + ret, pool_ns->str, pool_ns->len);
106 			ret += pool_ns->len;
107 		}
108 	}
109 	up_read(&osdc->lock);
110 	ceph_put_string(pool_ns);
111 	return ret;
112 }
113 
114 /*
115  * The convention with strings in xattrs is that they should not be NULL
116  * terminated, since we're returning the length with them. snprintf always
117  * NULL terminates however, so call it on a temporary buffer and then memcpy
118  * the result into place.
119  */
120 static __printf(3, 4)
121 int ceph_fmt_xattr(char *val, size_t size, const char *fmt, ...)
122 {
123 	int ret;
124 	va_list args;
125 	char buf[96]; /* NB: reevaluate size if new vxattrs are added */
126 
127 	va_start(args, fmt);
128 	ret = vsnprintf(buf, size ? sizeof(buf) : 0, fmt, args);
129 	va_end(args);
130 
131 	/* Sanity check */
132 	if (size && ret + 1 > sizeof(buf)) {
133 		WARN_ONCE(true, "Returned length too big (%d)", ret);
134 		return -E2BIG;
135 	}
136 
137 	if (ret <= size)
138 		memcpy(val, buf, ret);
139 	return ret;
140 }
141 
142 static ssize_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
143 						char *val, size_t size)
144 {
145 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_unit);
146 }
147 
148 static ssize_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
149 						 char *val, size_t size)
150 {
151 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_count);
152 }
153 
154 static ssize_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
155 						char *val, size_t size)
156 {
157 	return ceph_fmt_xattr(val, size, "%u", ci->i_layout.object_size);
158 }
159 
160 static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
161 					 char *val, size_t size)
162 {
163 	ssize_t ret;
164 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->netfs.inode.i_sb);
165 	struct ceph_osd_client *osdc = &fsc->client->osdc;
166 	s64 pool = ci->i_layout.pool_id;
167 	const char *pool_name;
168 
169 	down_read(&osdc->lock);
170 	pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
171 	if (pool_name) {
172 		ret = strlen(pool_name);
173 		if (ret <= size)
174 			memcpy(val, pool_name, ret);
175 	} else {
176 		ret = ceph_fmt_xattr(val, size, "%lld", pool);
177 	}
178 	up_read(&osdc->lock);
179 	return ret;
180 }
181 
182 static ssize_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
183 						   char *val, size_t size)
184 {
185 	ssize_t ret = 0;
186 	struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
187 
188 	if (ns) {
189 		ret = ns->len;
190 		if (ret <= size)
191 			memcpy(val, ns->str, ret);
192 		ceph_put_string(ns);
193 	}
194 	return ret;
195 }
196 
197 /* directories */
198 
199 static ssize_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
200 					 size_t size)
201 {
202 	return ceph_fmt_xattr(val, size, "%lld", ci->i_files + ci->i_subdirs);
203 }
204 
205 static ssize_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
206 				       size_t size)
207 {
208 	return ceph_fmt_xattr(val, size, "%lld", ci->i_files);
209 }
210 
211 static ssize_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
212 					 size_t size)
213 {
214 	return ceph_fmt_xattr(val, size, "%lld", ci->i_subdirs);
215 }
216 
217 static ssize_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
218 					  size_t size)
219 {
220 	return ceph_fmt_xattr(val, size, "%lld",
221 				ci->i_rfiles + ci->i_rsubdirs);
222 }
223 
224 static ssize_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
225 					size_t size)
226 {
227 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rfiles);
228 }
229 
230 static ssize_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
231 					  size_t size)
232 {
233 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rsubdirs);
234 }
235 
236 static ssize_t ceph_vxattrcb_dir_rsnaps(struct ceph_inode_info *ci, char *val,
237 					  size_t size)
238 {
239 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rsnaps);
240 }
241 
242 static ssize_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
243 					size_t size)
244 {
245 	return ceph_fmt_xattr(val, size, "%lld", ci->i_rbytes);
246 }
247 
248 static ssize_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
249 					size_t size)
250 {
251 	return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_rctime.tv_sec,
252 				ci->i_rctime.tv_nsec);
253 }
254 
255 /* dir pin */
256 static bool ceph_vxattrcb_dir_pin_exists(struct ceph_inode_info *ci)
257 {
258 	return ci->i_dir_pin != -ENODATA;
259 }
260 
261 static ssize_t ceph_vxattrcb_dir_pin(struct ceph_inode_info *ci, char *val,
262 				     size_t size)
263 {
264 	return ceph_fmt_xattr(val, size, "%d", (int)ci->i_dir_pin);
265 }
266 
267 /* quotas */
268 static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
269 {
270 	bool ret = false;
271 	spin_lock(&ci->i_ceph_lock);
272 	if ((ci->i_max_files || ci->i_max_bytes) &&
273 	    ci->i_vino.snap == CEPH_NOSNAP &&
274 	    ci->i_snap_realm &&
275 	    ci->i_snap_realm->ino == ci->i_vino.ino)
276 		ret = true;
277 	spin_unlock(&ci->i_ceph_lock);
278 	return ret;
279 }
280 
281 static ssize_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
282 				   size_t size)
283 {
284 	return ceph_fmt_xattr(val, size, "max_bytes=%llu max_files=%llu",
285 				ci->i_max_bytes, ci->i_max_files);
286 }
287 
288 static ssize_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
289 					     char *val, size_t size)
290 {
291 	return ceph_fmt_xattr(val, size, "%llu", ci->i_max_bytes);
292 }
293 
294 static ssize_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
295 					     char *val, size_t size)
296 {
297 	return ceph_fmt_xattr(val, size, "%llu", ci->i_max_files);
298 }
299 
300 /* snapshots */
301 static bool ceph_vxattrcb_snap_btime_exists(struct ceph_inode_info *ci)
302 {
303 	return (ci->i_snap_btime.tv_sec != 0 || ci->i_snap_btime.tv_nsec != 0);
304 }
305 
306 static ssize_t ceph_vxattrcb_snap_btime(struct ceph_inode_info *ci, char *val,
307 					size_t size)
308 {
309 	return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_snap_btime.tv_sec,
310 				ci->i_snap_btime.tv_nsec);
311 }
312 
313 static ssize_t ceph_vxattrcb_cluster_fsid(struct ceph_inode_info *ci,
314 					  char *val, size_t size)
315 {
316 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->netfs.inode.i_sb);
317 
318 	return ceph_fmt_xattr(val, size, "%pU", &fsc->client->fsid);
319 }
320 
321 static ssize_t ceph_vxattrcb_client_id(struct ceph_inode_info *ci,
322 				       char *val, size_t size)
323 {
324 	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->netfs.inode.i_sb);
325 
326 	return ceph_fmt_xattr(val, size, "client%lld",
327 			      ceph_client_gid(fsc->client));
328 }
329 
330 static ssize_t ceph_vxattrcb_caps(struct ceph_inode_info *ci, char *val,
331 					size_t size)
332 {
333 	int issued;
334 
335 	spin_lock(&ci->i_ceph_lock);
336 	issued = __ceph_caps_issued(ci, NULL);
337 	spin_unlock(&ci->i_ceph_lock);
338 
339 	return ceph_fmt_xattr(val, size, "%s/0x%x",
340 			      ceph_cap_string(issued), issued);
341 }
342 
343 static ssize_t ceph_vxattrcb_auth_mds(struct ceph_inode_info *ci,
344 				       char *val, size_t size)
345 {
346 	int ret;
347 
348 	spin_lock(&ci->i_ceph_lock);
349 	ret = ceph_fmt_xattr(val, size, "%d",
350 			     ci->i_auth_cap ? ci->i_auth_cap->session->s_mds : -1);
351 	spin_unlock(&ci->i_ceph_lock);
352 	return ret;
353 }
354 
355 #define CEPH_XATTR_NAME(_type, _name)	XATTR_CEPH_PREFIX #_type "." #_name
356 #define CEPH_XATTR_NAME2(_type, _name, _name2)	\
357 	XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
358 
359 #define XATTR_NAME_CEPH(_type, _name, _flags)				\
360 	{								\
361 		.name = CEPH_XATTR_NAME(_type, _name),			\
362 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
363 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
364 		.exists_cb = NULL,					\
365 		.flags = (VXATTR_FLAG_READONLY | _flags),		\
366 	}
367 #define XATTR_RSTAT_FIELD(_type, _name)			\
368 	XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
369 #define XATTR_RSTAT_FIELD_UPDATABLE(_type, _name)			\
370 	{								\
371 		.name = CEPH_XATTR_NAME(_type, _name),			\
372 		.name_size = sizeof (CEPH_XATTR_NAME(_type, _name)),	\
373 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
374 		.exists_cb = NULL,					\
375 		.flags = VXATTR_FLAG_RSTAT,				\
376 	}
377 #define XATTR_LAYOUT_FIELD(_type, _name, _field)			\
378 	{								\
379 		.name = CEPH_XATTR_NAME2(_type, _name, _field),	\
380 		.name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
381 		.getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
382 		.exists_cb = ceph_vxattrcb_layout_exists,	\
383 		.flags = VXATTR_FLAG_HIDDEN,			\
384 	}
385 #define XATTR_QUOTA_FIELD(_type, _name)					\
386 	{								\
387 		.name = CEPH_XATTR_NAME(_type, _name),			\
388 		.name_size = sizeof(CEPH_XATTR_NAME(_type, _name)),	\
389 		.getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name,	\
390 		.exists_cb = ceph_vxattrcb_quota_exists,		\
391 		.flags = VXATTR_FLAG_HIDDEN,				\
392 	}
393 
394 static struct ceph_vxattr ceph_dir_vxattrs[] = {
395 	{
396 		.name = "ceph.dir.layout",
397 		.name_size = sizeof("ceph.dir.layout"),
398 		.getxattr_cb = ceph_vxattrcb_layout,
399 		.exists_cb = ceph_vxattrcb_layout_exists,
400 		.flags = VXATTR_FLAG_HIDDEN,
401 	},
402 	XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
403 	XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
404 	XATTR_LAYOUT_FIELD(dir, layout, object_size),
405 	XATTR_LAYOUT_FIELD(dir, layout, pool),
406 	XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
407 	XATTR_NAME_CEPH(dir, entries, VXATTR_FLAG_DIRSTAT),
408 	XATTR_NAME_CEPH(dir, files, VXATTR_FLAG_DIRSTAT),
409 	XATTR_NAME_CEPH(dir, subdirs, VXATTR_FLAG_DIRSTAT),
410 	XATTR_RSTAT_FIELD(dir, rentries),
411 	XATTR_RSTAT_FIELD(dir, rfiles),
412 	XATTR_RSTAT_FIELD(dir, rsubdirs),
413 	XATTR_RSTAT_FIELD(dir, rsnaps),
414 	XATTR_RSTAT_FIELD(dir, rbytes),
415 	XATTR_RSTAT_FIELD_UPDATABLE(dir, rctime),
416 	{
417 		.name = "ceph.dir.pin",
418 		.name_size = sizeof("ceph.dir.pin"),
419 		.getxattr_cb = ceph_vxattrcb_dir_pin,
420 		.exists_cb = ceph_vxattrcb_dir_pin_exists,
421 		.flags = VXATTR_FLAG_HIDDEN,
422 	},
423 	{
424 		.name = "ceph.quota",
425 		.name_size = sizeof("ceph.quota"),
426 		.getxattr_cb = ceph_vxattrcb_quota,
427 		.exists_cb = ceph_vxattrcb_quota_exists,
428 		.flags = VXATTR_FLAG_HIDDEN,
429 	},
430 	XATTR_QUOTA_FIELD(quota, max_bytes),
431 	XATTR_QUOTA_FIELD(quota, max_files),
432 	{
433 		.name = "ceph.snap.btime",
434 		.name_size = sizeof("ceph.snap.btime"),
435 		.getxattr_cb = ceph_vxattrcb_snap_btime,
436 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
437 		.flags = VXATTR_FLAG_READONLY,
438 	},
439 	{
440 		.name = "ceph.caps",
441 		.name_size = sizeof("ceph.caps"),
442 		.getxattr_cb = ceph_vxattrcb_caps,
443 		.exists_cb = NULL,
444 		.flags = VXATTR_FLAG_HIDDEN,
445 	},
446 	{ .name = NULL, 0 }	/* Required table terminator */
447 };
448 
449 /* files */
450 
451 static struct ceph_vxattr ceph_file_vxattrs[] = {
452 	{
453 		.name = "ceph.file.layout",
454 		.name_size = sizeof("ceph.file.layout"),
455 		.getxattr_cb = ceph_vxattrcb_layout,
456 		.exists_cb = ceph_vxattrcb_layout_exists,
457 		.flags = VXATTR_FLAG_HIDDEN,
458 	},
459 	XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
460 	XATTR_LAYOUT_FIELD(file, layout, stripe_count),
461 	XATTR_LAYOUT_FIELD(file, layout, object_size),
462 	XATTR_LAYOUT_FIELD(file, layout, pool),
463 	XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
464 	{
465 		.name = "ceph.snap.btime",
466 		.name_size = sizeof("ceph.snap.btime"),
467 		.getxattr_cb = ceph_vxattrcb_snap_btime,
468 		.exists_cb = ceph_vxattrcb_snap_btime_exists,
469 		.flags = VXATTR_FLAG_READONLY,
470 	},
471 	{
472 		.name = "ceph.caps",
473 		.name_size = sizeof("ceph.caps"),
474 		.getxattr_cb = ceph_vxattrcb_caps,
475 		.exists_cb = NULL,
476 		.flags = VXATTR_FLAG_HIDDEN,
477 	},
478 	{ .name = NULL, 0 }	/* Required table terminator */
479 };
480 
481 static struct ceph_vxattr ceph_common_vxattrs[] = {
482 	{
483 		.name = "ceph.cluster_fsid",
484 		.name_size = sizeof("ceph.cluster_fsid"),
485 		.getxattr_cb = ceph_vxattrcb_cluster_fsid,
486 		.exists_cb = NULL,
487 		.flags = VXATTR_FLAG_READONLY,
488 	},
489 	{
490 		.name = "ceph.client_id",
491 		.name_size = sizeof("ceph.client_id"),
492 		.getxattr_cb = ceph_vxattrcb_client_id,
493 		.exists_cb = NULL,
494 		.flags = VXATTR_FLAG_READONLY,
495 	},
496 	{
497 		.name = "ceph.auth_mds",
498 		.name_size = sizeof("ceph.auth_mds"),
499 		.getxattr_cb = ceph_vxattrcb_auth_mds,
500 		.exists_cb = NULL,
501 		.flags = VXATTR_FLAG_READONLY,
502 	},
503 	{ .name = NULL, 0 }	/* Required table terminator */
504 };
505 
506 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
507 {
508 	if (S_ISDIR(inode->i_mode))
509 		return ceph_dir_vxattrs;
510 	else if (S_ISREG(inode->i_mode))
511 		return ceph_file_vxattrs;
512 	return NULL;
513 }
514 
515 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
516 						const char *name)
517 {
518 	struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
519 
520 	if (vxattr) {
521 		while (vxattr->name) {
522 			if (!strcmp(vxattr->name, name))
523 				return vxattr;
524 			vxattr++;
525 		}
526 	}
527 
528 	vxattr = ceph_common_vxattrs;
529 	while (vxattr->name) {
530 		if (!strcmp(vxattr->name, name))
531 			return vxattr;
532 		vxattr++;
533 	}
534 
535 	return NULL;
536 }
537 
538 #define MAX_XATTR_VAL_PRINT_LEN 256
539 
540 static int __set_xattr(struct ceph_inode_info *ci,
541 			   const char *name, int name_len,
542 			   const char *val, int val_len,
543 			   int flags, int update_xattr,
544 			   struct ceph_inode_xattr **newxattr)
545 {
546 	struct rb_node **p;
547 	struct rb_node *parent = NULL;
548 	struct ceph_inode_xattr *xattr = NULL;
549 	int c;
550 	int new = 0;
551 
552 	p = &ci->i_xattrs.index.rb_node;
553 	while (*p) {
554 		parent = *p;
555 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
556 		c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
557 		if (c < 0)
558 			p = &(*p)->rb_left;
559 		else if (c > 0)
560 			p = &(*p)->rb_right;
561 		else {
562 			if (name_len == xattr->name_len)
563 				break;
564 			else if (name_len < xattr->name_len)
565 				p = &(*p)->rb_left;
566 			else
567 				p = &(*p)->rb_right;
568 		}
569 		xattr = NULL;
570 	}
571 
572 	if (update_xattr) {
573 		int err = 0;
574 
575 		if (xattr && (flags & XATTR_CREATE))
576 			err = -EEXIST;
577 		else if (!xattr && (flags & XATTR_REPLACE))
578 			err = -ENODATA;
579 		if (err) {
580 			kfree(name);
581 			kfree(val);
582 			kfree(*newxattr);
583 			return err;
584 		}
585 		if (update_xattr < 0) {
586 			if (xattr)
587 				__remove_xattr(ci, xattr);
588 			kfree(name);
589 			kfree(*newxattr);
590 			return 0;
591 		}
592 	}
593 
594 	if (!xattr) {
595 		new = 1;
596 		xattr = *newxattr;
597 		xattr->name = name;
598 		xattr->name_len = name_len;
599 		xattr->should_free_name = update_xattr;
600 
601 		ci->i_xattrs.count++;
602 		dout("%s count=%d\n", __func__, ci->i_xattrs.count);
603 	} else {
604 		kfree(*newxattr);
605 		*newxattr = NULL;
606 		if (xattr->should_free_val)
607 			kfree(xattr->val);
608 
609 		if (update_xattr) {
610 			kfree(name);
611 			name = xattr->name;
612 		}
613 		ci->i_xattrs.names_size -= xattr->name_len;
614 		ci->i_xattrs.vals_size -= xattr->val_len;
615 	}
616 	ci->i_xattrs.names_size += name_len;
617 	ci->i_xattrs.vals_size += val_len;
618 	if (val)
619 		xattr->val = val;
620 	else
621 		xattr->val = "";
622 
623 	xattr->val_len = val_len;
624 	xattr->dirty = update_xattr;
625 	xattr->should_free_val = (val && update_xattr);
626 
627 	if (new) {
628 		rb_link_node(&xattr->node, parent, p);
629 		rb_insert_color(&xattr->node, &ci->i_xattrs.index);
630 		dout("%s p=%p\n", __func__, p);
631 	}
632 
633 	dout("%s added %llx.%llx xattr %p %.*s=%.*s%s\n", __func__,
634 	     ceph_vinop(&ci->netfs.inode), xattr, name_len, name,
635 	     min(val_len, MAX_XATTR_VAL_PRINT_LEN), val,
636 	     val_len > MAX_XATTR_VAL_PRINT_LEN ? "..." : "");
637 
638 	return 0;
639 }
640 
641 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
642 			   const char *name)
643 {
644 	struct rb_node **p;
645 	struct rb_node *parent = NULL;
646 	struct ceph_inode_xattr *xattr = NULL;
647 	int name_len = strlen(name);
648 	int c;
649 
650 	p = &ci->i_xattrs.index.rb_node;
651 	while (*p) {
652 		parent = *p;
653 		xattr = rb_entry(parent, struct ceph_inode_xattr, node);
654 		c = strncmp(name, xattr->name, xattr->name_len);
655 		if (c == 0 && name_len > xattr->name_len)
656 			c = 1;
657 		if (c < 0)
658 			p = &(*p)->rb_left;
659 		else if (c > 0)
660 			p = &(*p)->rb_right;
661 		else {
662 			int len = min(xattr->val_len, MAX_XATTR_VAL_PRINT_LEN);
663 
664 			dout("%s %s: found %.*s%s\n", __func__, name, len,
665 			     xattr->val, xattr->val_len > len ? "..." : "");
666 			return xattr;
667 		}
668 	}
669 
670 	dout("%s %s: not found\n", __func__, name);
671 
672 	return NULL;
673 }
674 
675 static void __free_xattr(struct ceph_inode_xattr *xattr)
676 {
677 	BUG_ON(!xattr);
678 
679 	if (xattr->should_free_name)
680 		kfree(xattr->name);
681 	if (xattr->should_free_val)
682 		kfree(xattr->val);
683 
684 	kfree(xattr);
685 }
686 
687 static int __remove_xattr(struct ceph_inode_info *ci,
688 			  struct ceph_inode_xattr *xattr)
689 {
690 	if (!xattr)
691 		return -ENODATA;
692 
693 	rb_erase(&xattr->node, &ci->i_xattrs.index);
694 
695 	if (xattr->should_free_name)
696 		kfree(xattr->name);
697 	if (xattr->should_free_val)
698 		kfree(xattr->val);
699 
700 	ci->i_xattrs.names_size -= xattr->name_len;
701 	ci->i_xattrs.vals_size -= xattr->val_len;
702 	ci->i_xattrs.count--;
703 	kfree(xattr);
704 
705 	return 0;
706 }
707 
708 static char *__copy_xattr_names(struct ceph_inode_info *ci,
709 				char *dest)
710 {
711 	struct rb_node *p;
712 	struct ceph_inode_xattr *xattr = NULL;
713 
714 	p = rb_first(&ci->i_xattrs.index);
715 	dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
716 
717 	while (p) {
718 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
719 		memcpy(dest, xattr->name, xattr->name_len);
720 		dest[xattr->name_len] = '\0';
721 
722 		dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
723 		     xattr->name_len, ci->i_xattrs.names_size);
724 
725 		dest += xattr->name_len + 1;
726 		p = rb_next(p);
727 	}
728 
729 	return dest;
730 }
731 
732 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
733 {
734 	struct rb_node *p, *tmp;
735 	struct ceph_inode_xattr *xattr = NULL;
736 
737 	p = rb_first(&ci->i_xattrs.index);
738 
739 	dout("__ceph_destroy_xattrs p=%p\n", p);
740 
741 	while (p) {
742 		xattr = rb_entry(p, struct ceph_inode_xattr, node);
743 		tmp = p;
744 		p = rb_next(tmp);
745 		dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
746 		     xattr->name_len, xattr->name);
747 		rb_erase(tmp, &ci->i_xattrs.index);
748 
749 		__free_xattr(xattr);
750 	}
751 
752 	ci->i_xattrs.names_size = 0;
753 	ci->i_xattrs.vals_size = 0;
754 	ci->i_xattrs.index_version = 0;
755 	ci->i_xattrs.count = 0;
756 	ci->i_xattrs.index = RB_ROOT;
757 }
758 
759 static int __build_xattrs(struct inode *inode)
760 	__releases(ci->i_ceph_lock)
761 	__acquires(ci->i_ceph_lock)
762 {
763 	u32 namelen;
764 	u32 numattr = 0;
765 	void *p, *end;
766 	u32 len;
767 	const char *name, *val;
768 	struct ceph_inode_info *ci = ceph_inode(inode);
769 	u64 xattr_version;
770 	struct ceph_inode_xattr **xattrs = NULL;
771 	int err = 0;
772 	int i;
773 
774 	dout("__build_xattrs() len=%d\n",
775 	     ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
776 
777 	if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
778 		return 0; /* already built */
779 
780 	__ceph_destroy_xattrs(ci);
781 
782 start:
783 	/* updated internal xattr rb tree */
784 	if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
785 		p = ci->i_xattrs.blob->vec.iov_base;
786 		end = p + ci->i_xattrs.blob->vec.iov_len;
787 		ceph_decode_32_safe(&p, end, numattr, bad);
788 		xattr_version = ci->i_xattrs.version;
789 		spin_unlock(&ci->i_ceph_lock);
790 
791 		xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
792 				 GFP_NOFS);
793 		err = -ENOMEM;
794 		if (!xattrs)
795 			goto bad_lock;
796 
797 		for (i = 0; i < numattr; i++) {
798 			xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
799 					    GFP_NOFS);
800 			if (!xattrs[i])
801 				goto bad_lock;
802 		}
803 
804 		spin_lock(&ci->i_ceph_lock);
805 		if (ci->i_xattrs.version != xattr_version) {
806 			/* lost a race, retry */
807 			for (i = 0; i < numattr; i++)
808 				kfree(xattrs[i]);
809 			kfree(xattrs);
810 			xattrs = NULL;
811 			goto start;
812 		}
813 		err = -EIO;
814 		while (numattr--) {
815 			ceph_decode_32_safe(&p, end, len, bad);
816 			namelen = len;
817 			name = p;
818 			p += len;
819 			ceph_decode_32_safe(&p, end, len, bad);
820 			val = p;
821 			p += len;
822 
823 			err = __set_xattr(ci, name, namelen, val, len,
824 					  0, 0, &xattrs[numattr]);
825 
826 			if (err < 0)
827 				goto bad;
828 		}
829 		kfree(xattrs);
830 	}
831 	ci->i_xattrs.index_version = ci->i_xattrs.version;
832 	ci->i_xattrs.dirty = false;
833 
834 	return err;
835 bad_lock:
836 	spin_lock(&ci->i_ceph_lock);
837 bad:
838 	if (xattrs) {
839 		for (i = 0; i < numattr; i++)
840 			kfree(xattrs[i]);
841 		kfree(xattrs);
842 	}
843 	ci->i_xattrs.names_size = 0;
844 	return err;
845 }
846 
847 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
848 				    int val_size)
849 {
850 	/*
851 	 * 4 bytes for the length, and additional 4 bytes per each xattr name,
852 	 * 4 bytes per each value
853 	 */
854 	int size = 4 + ci->i_xattrs.count*(4 + 4) +
855 			     ci->i_xattrs.names_size +
856 			     ci->i_xattrs.vals_size;
857 	dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
858 	     ci->i_xattrs.count, ci->i_xattrs.names_size,
859 	     ci->i_xattrs.vals_size);
860 
861 	if (name_size)
862 		size += 4 + 4 + name_size + val_size;
863 
864 	return size;
865 }
866 
867 /*
868  * If there are dirty xattrs, reencode xattrs into the prealloc_blob
869  * and swap into place.  It returns the old i_xattrs.blob (or NULL) so
870  * that it can be freed by the caller as the i_ceph_lock is likely to be
871  * held.
872  */
873 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
874 {
875 	struct rb_node *p;
876 	struct ceph_inode_xattr *xattr = NULL;
877 	struct ceph_buffer *old_blob = NULL;
878 	void *dest;
879 
880 	dout("__build_xattrs_blob %p\n", &ci->netfs.inode);
881 	if (ci->i_xattrs.dirty) {
882 		int need = __get_required_blob_size(ci, 0, 0);
883 
884 		BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
885 
886 		p = rb_first(&ci->i_xattrs.index);
887 		dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
888 
889 		ceph_encode_32(&dest, ci->i_xattrs.count);
890 		while (p) {
891 			xattr = rb_entry(p, struct ceph_inode_xattr, node);
892 
893 			ceph_encode_32(&dest, xattr->name_len);
894 			memcpy(dest, xattr->name, xattr->name_len);
895 			dest += xattr->name_len;
896 			ceph_encode_32(&dest, xattr->val_len);
897 			memcpy(dest, xattr->val, xattr->val_len);
898 			dest += xattr->val_len;
899 
900 			p = rb_next(p);
901 		}
902 
903 		/* adjust buffer len; it may be larger than we need */
904 		ci->i_xattrs.prealloc_blob->vec.iov_len =
905 			dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
906 
907 		if (ci->i_xattrs.blob)
908 			old_blob = ci->i_xattrs.blob;
909 		ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
910 		ci->i_xattrs.prealloc_blob = NULL;
911 		ci->i_xattrs.dirty = false;
912 		ci->i_xattrs.version++;
913 	}
914 
915 	return old_blob;
916 }
917 
918 static inline int __get_request_mask(struct inode *in) {
919 	struct ceph_mds_request *req = current->journal_info;
920 	int mask = 0;
921 	if (req && req->r_target_inode == in) {
922 		if (req->r_op == CEPH_MDS_OP_LOOKUP ||
923 		    req->r_op == CEPH_MDS_OP_LOOKUPINO ||
924 		    req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
925 		    req->r_op == CEPH_MDS_OP_GETATTR) {
926 			mask = le32_to_cpu(req->r_args.getattr.mask);
927 		} else if (req->r_op == CEPH_MDS_OP_OPEN ||
928 			   req->r_op == CEPH_MDS_OP_CREATE) {
929 			mask = le32_to_cpu(req->r_args.open.mask);
930 		}
931 	}
932 	return mask;
933 }
934 
935 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
936 		      size_t size)
937 {
938 	struct ceph_inode_info *ci = ceph_inode(inode);
939 	struct ceph_inode_xattr *xattr;
940 	struct ceph_vxattr *vxattr;
941 	int req_mask;
942 	ssize_t err;
943 
944 	if (strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
945 		goto handle_non_vxattrs;
946 
947 	/* let's see if a virtual xattr was requested */
948 	vxattr = ceph_match_vxattr(inode, name);
949 	if (vxattr) {
950 		int mask = 0;
951 		if (vxattr->flags & VXATTR_FLAG_RSTAT)
952 			mask |= CEPH_STAT_RSTAT;
953 		if (vxattr->flags & VXATTR_FLAG_DIRSTAT)
954 			mask |= CEPH_CAP_FILE_SHARED;
955 		err = ceph_do_getattr(inode, mask, true);
956 		if (err)
957 			return err;
958 		err = -ENODATA;
959 		if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
960 			err = vxattr->getxattr_cb(ci, value, size);
961 			if (size && size < err)
962 				err = -ERANGE;
963 		}
964 		return err;
965 	} else {
966 		err = ceph_do_getvxattr(inode, name, value, size);
967 		/* this would happen with a new client and old server combo */
968 		if (err == -EOPNOTSUPP)
969 			err = -ENODATA;
970 		return err;
971 	}
972 handle_non_vxattrs:
973 	req_mask = __get_request_mask(inode);
974 
975 	spin_lock(&ci->i_ceph_lock);
976 	dout("getxattr %p name '%s' ver=%lld index_ver=%lld\n", inode, name,
977 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
978 
979 	if (ci->i_xattrs.version == 0 ||
980 	    !((req_mask & CEPH_CAP_XATTR_SHARED) ||
981 	      __ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1))) {
982 		spin_unlock(&ci->i_ceph_lock);
983 
984 		/* security module gets xattr while filling trace */
985 		if (current->journal_info) {
986 			pr_warn_ratelimited("sync getxattr %p "
987 					    "during filling trace\n", inode);
988 			return -EBUSY;
989 		}
990 
991 		/* get xattrs from mds (if we don't already have them) */
992 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
993 		if (err)
994 			return err;
995 		spin_lock(&ci->i_ceph_lock);
996 	}
997 
998 	err = __build_xattrs(inode);
999 	if (err < 0)
1000 		goto out;
1001 
1002 	err = -ENODATA;  /* == ENOATTR */
1003 	xattr = __get_xattr(ci, name);
1004 	if (!xattr)
1005 		goto out;
1006 
1007 	err = -ERANGE;
1008 	if (size && size < xattr->val_len)
1009 		goto out;
1010 
1011 	err = xattr->val_len;
1012 	if (size == 0)
1013 		goto out;
1014 
1015 	memcpy(value, xattr->val, xattr->val_len);
1016 
1017 	if (current->journal_info &&
1018 	    !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
1019 	    security_ismaclabel(name + XATTR_SECURITY_PREFIX_LEN))
1020 		ci->i_ceph_flags |= CEPH_I_SEC_INITED;
1021 out:
1022 	spin_unlock(&ci->i_ceph_lock);
1023 	return err;
1024 }
1025 
1026 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
1027 {
1028 	struct inode *inode = d_inode(dentry);
1029 	struct ceph_inode_info *ci = ceph_inode(inode);
1030 	bool len_only = (size == 0);
1031 	u32 namelen;
1032 	int err;
1033 
1034 	spin_lock(&ci->i_ceph_lock);
1035 	dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
1036 	     ci->i_xattrs.version, ci->i_xattrs.index_version);
1037 
1038 	if (ci->i_xattrs.version == 0 ||
1039 	    !__ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1)) {
1040 		spin_unlock(&ci->i_ceph_lock);
1041 		err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1042 		if (err)
1043 			return err;
1044 		spin_lock(&ci->i_ceph_lock);
1045 	}
1046 
1047 	err = __build_xattrs(inode);
1048 	if (err < 0)
1049 		goto out;
1050 
1051 	/* add 1 byte for each xattr due to the null termination */
1052 	namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
1053 	if (!len_only) {
1054 		if (namelen > size) {
1055 			err = -ERANGE;
1056 			goto out;
1057 		}
1058 		names = __copy_xattr_names(ci, names);
1059 		size -= namelen;
1060 	}
1061 	err = namelen;
1062 out:
1063 	spin_unlock(&ci->i_ceph_lock);
1064 	return err;
1065 }
1066 
1067 static int ceph_sync_setxattr(struct inode *inode, const char *name,
1068 			      const char *value, size_t size, int flags)
1069 {
1070 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1071 	struct ceph_inode_info *ci = ceph_inode(inode);
1072 	struct ceph_mds_request *req;
1073 	struct ceph_mds_client *mdsc = fsc->mdsc;
1074 	struct ceph_osd_client *osdc = &fsc->client->osdc;
1075 	struct ceph_pagelist *pagelist = NULL;
1076 	int op = CEPH_MDS_OP_SETXATTR;
1077 	int err;
1078 
1079 	if (size > 0) {
1080 		/* copy value into pagelist */
1081 		pagelist = ceph_pagelist_alloc(GFP_NOFS);
1082 		if (!pagelist)
1083 			return -ENOMEM;
1084 
1085 		err = ceph_pagelist_append(pagelist, value, size);
1086 		if (err)
1087 			goto out;
1088 	} else if (!value) {
1089 		if (flags & CEPH_XATTR_REPLACE)
1090 			op = CEPH_MDS_OP_RMXATTR;
1091 		else
1092 			flags |= CEPH_XATTR_REMOVE;
1093 	}
1094 
1095 	dout("setxattr value size: %zu\n", size);
1096 
1097 	/* do request */
1098 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1099 	if (IS_ERR(req)) {
1100 		err = PTR_ERR(req);
1101 		goto out;
1102 	}
1103 
1104 	req->r_path2 = kstrdup(name, GFP_NOFS);
1105 	if (!req->r_path2) {
1106 		ceph_mdsc_put_request(req);
1107 		err = -ENOMEM;
1108 		goto out;
1109 	}
1110 
1111 	if (op == CEPH_MDS_OP_SETXATTR) {
1112 		req->r_args.setxattr.flags = cpu_to_le32(flags);
1113 		req->r_args.setxattr.osdmap_epoch =
1114 			cpu_to_le32(osdc->osdmap->epoch);
1115 		req->r_pagelist = pagelist;
1116 		pagelist = NULL;
1117 	}
1118 
1119 	req->r_inode = inode;
1120 	ihold(inode);
1121 	req->r_num_caps = 1;
1122 	req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1123 
1124 	dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
1125 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1126 	ceph_mdsc_put_request(req);
1127 	dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
1128 
1129 out:
1130 	if (pagelist)
1131 		ceph_pagelist_release(pagelist);
1132 	return err;
1133 }
1134 
1135 int __ceph_setxattr(struct inode *inode, const char *name,
1136 			const void *value, size_t size, int flags)
1137 {
1138 	struct ceph_vxattr *vxattr;
1139 	struct ceph_inode_info *ci = ceph_inode(inode);
1140 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1141 	struct ceph_cap_flush *prealloc_cf = NULL;
1142 	struct ceph_buffer *old_blob = NULL;
1143 	int issued;
1144 	int err;
1145 	int dirty = 0;
1146 	int name_len = strlen(name);
1147 	int val_len = size;
1148 	char *newname = NULL;
1149 	char *newval = NULL;
1150 	struct ceph_inode_xattr *xattr = NULL;
1151 	int required_blob_size;
1152 	bool check_realm = false;
1153 	bool lock_snap_rwsem = false;
1154 
1155 	if (ceph_snap(inode) != CEPH_NOSNAP)
1156 		return -EROFS;
1157 
1158 	vxattr = ceph_match_vxattr(inode, name);
1159 	if (vxattr) {
1160 		if (vxattr->flags & VXATTR_FLAG_READONLY)
1161 			return -EOPNOTSUPP;
1162 		if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1163 			check_realm = true;
1164 	}
1165 
1166 	/* pass any unhandled ceph.* xattrs through to the MDS */
1167 	if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1168 		goto do_sync_unlocked;
1169 
1170 	/* preallocate memory for xattr name, value, index node */
1171 	err = -ENOMEM;
1172 	newname = kmemdup(name, name_len + 1, GFP_NOFS);
1173 	if (!newname)
1174 		goto out;
1175 
1176 	if (val_len) {
1177 		newval = kmemdup(value, val_len, GFP_NOFS);
1178 		if (!newval)
1179 			goto out;
1180 	}
1181 
1182 	xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1183 	if (!xattr)
1184 		goto out;
1185 
1186 	prealloc_cf = ceph_alloc_cap_flush();
1187 	if (!prealloc_cf)
1188 		goto out;
1189 
1190 	spin_lock(&ci->i_ceph_lock);
1191 retry:
1192 	issued = __ceph_caps_issued(ci, NULL);
1193 	required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1194 	if ((ci->i_xattrs.version == 0) || !(issued & CEPH_CAP_XATTR_EXCL) ||
1195 	    (required_blob_size > mdsc->mdsmap->m_max_xattr_size)) {
1196 		dout("%s do sync setxattr: version: %llu size: %d max: %llu\n",
1197 		     __func__, ci->i_xattrs.version, required_blob_size,
1198 		     mdsc->mdsmap->m_max_xattr_size);
1199 		goto do_sync;
1200 	}
1201 
1202 	if (!lock_snap_rwsem && !ci->i_head_snapc) {
1203 		lock_snap_rwsem = true;
1204 		if (!down_read_trylock(&mdsc->snap_rwsem)) {
1205 			spin_unlock(&ci->i_ceph_lock);
1206 			down_read(&mdsc->snap_rwsem);
1207 			spin_lock(&ci->i_ceph_lock);
1208 			goto retry;
1209 		}
1210 	}
1211 
1212 	dout("setxattr %p name '%s' issued %s\n", inode, name,
1213 	     ceph_cap_string(issued));
1214 	__build_xattrs(inode);
1215 
1216 	if (!ci->i_xattrs.prealloc_blob ||
1217 	    required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1218 		struct ceph_buffer *blob;
1219 
1220 		spin_unlock(&ci->i_ceph_lock);
1221 		ceph_buffer_put(old_blob); /* Shouldn't be required */
1222 		dout(" pre-allocating new blob size=%d\n", required_blob_size);
1223 		blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1224 		if (!blob)
1225 			goto do_sync_unlocked;
1226 		spin_lock(&ci->i_ceph_lock);
1227 		/* prealloc_blob can't be released while holding i_ceph_lock */
1228 		if (ci->i_xattrs.prealloc_blob)
1229 			old_blob = ci->i_xattrs.prealloc_blob;
1230 		ci->i_xattrs.prealloc_blob = blob;
1231 		goto retry;
1232 	}
1233 
1234 	err = __set_xattr(ci, newname, name_len, newval, val_len,
1235 			  flags, value ? 1 : -1, &xattr);
1236 
1237 	if (!err) {
1238 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1239 					       &prealloc_cf);
1240 		ci->i_xattrs.dirty = true;
1241 		inode_set_ctime_current(inode);
1242 	}
1243 
1244 	spin_unlock(&ci->i_ceph_lock);
1245 	ceph_buffer_put(old_blob);
1246 	if (lock_snap_rwsem)
1247 		up_read(&mdsc->snap_rwsem);
1248 	if (dirty)
1249 		__mark_inode_dirty(inode, dirty);
1250 	ceph_free_cap_flush(prealloc_cf);
1251 	return err;
1252 
1253 do_sync:
1254 	spin_unlock(&ci->i_ceph_lock);
1255 do_sync_unlocked:
1256 	if (lock_snap_rwsem)
1257 		up_read(&mdsc->snap_rwsem);
1258 
1259 	/* security module set xattr while filling trace */
1260 	if (current->journal_info) {
1261 		pr_warn_ratelimited("sync setxattr %p "
1262 				    "during filling trace\n", inode);
1263 		err = -EBUSY;
1264 	} else {
1265 		err = ceph_sync_setxattr(inode, name, value, size, flags);
1266 		if (err >= 0 && check_realm) {
1267 			/* check if snaprealm was created for quota inode */
1268 			spin_lock(&ci->i_ceph_lock);
1269 			if ((ci->i_max_files || ci->i_max_bytes) &&
1270 			    !(ci->i_snap_realm &&
1271 			      ci->i_snap_realm->ino == ci->i_vino.ino))
1272 				err = -EOPNOTSUPP;
1273 			spin_unlock(&ci->i_ceph_lock);
1274 		}
1275 	}
1276 out:
1277 	ceph_free_cap_flush(prealloc_cf);
1278 	kfree(newname);
1279 	kfree(newval);
1280 	kfree(xattr);
1281 	return err;
1282 }
1283 
1284 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1285 				  struct dentry *dentry, struct inode *inode,
1286 				  const char *name, void *value, size_t size)
1287 {
1288 	if (!ceph_is_valid_xattr(name))
1289 		return -EOPNOTSUPP;
1290 	return __ceph_getxattr(inode, name, value, size);
1291 }
1292 
1293 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1294 				  struct mnt_idmap *idmap,
1295 				  struct dentry *unused, struct inode *inode,
1296 				  const char *name, const void *value,
1297 				  size_t size, int flags)
1298 {
1299 	if (!ceph_is_valid_xattr(name))
1300 		return -EOPNOTSUPP;
1301 	return __ceph_setxattr(inode, name, value, size, flags);
1302 }
1303 
1304 static const struct xattr_handler ceph_other_xattr_handler = {
1305 	.prefix = "",  /* match any name => handlers called with full name */
1306 	.get = ceph_get_xattr_handler,
1307 	.set = ceph_set_xattr_handler,
1308 };
1309 
1310 #ifdef CONFIG_SECURITY
1311 bool ceph_security_xattr_wanted(struct inode *in)
1312 {
1313 	return in->i_security != NULL;
1314 }
1315 
1316 bool ceph_security_xattr_deadlock(struct inode *in)
1317 {
1318 	struct ceph_inode_info *ci;
1319 	bool ret;
1320 	if (!in->i_security)
1321 		return false;
1322 	ci = ceph_inode(in);
1323 	spin_lock(&ci->i_ceph_lock);
1324 	ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1325 	      !(ci->i_xattrs.version > 0 &&
1326 		__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1327 	spin_unlock(&ci->i_ceph_lock);
1328 	return ret;
1329 }
1330 
1331 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1332 int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1333 			   struct ceph_acl_sec_ctx *as_ctx)
1334 {
1335 	struct ceph_pagelist *pagelist = as_ctx->pagelist;
1336 	const char *name;
1337 	size_t name_len;
1338 	int err;
1339 
1340 	err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1341 					    &name, &as_ctx->sec_ctx,
1342 					    &as_ctx->sec_ctxlen);
1343 	if (err < 0) {
1344 		WARN_ON_ONCE(err != -EOPNOTSUPP);
1345 		err = 0; /* do nothing */
1346 		goto out;
1347 	}
1348 
1349 	err = -ENOMEM;
1350 	if (!pagelist) {
1351 		pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1352 		if (!pagelist)
1353 			goto out;
1354 		err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1355 		if (err)
1356 			goto out;
1357 		ceph_pagelist_encode_32(pagelist, 1);
1358 	}
1359 
1360 	/*
1361 	 * FIXME: Make security_dentry_init_security() generic. Currently
1362 	 * It only supports single security module and only selinux has
1363 	 * dentry_init_security hook.
1364 	 */
1365 	name_len = strlen(name);
1366 	err = ceph_pagelist_reserve(pagelist,
1367 				    4 * 2 + name_len + as_ctx->sec_ctxlen);
1368 	if (err)
1369 		goto out;
1370 
1371 	if (as_ctx->pagelist) {
1372 		/* update count of KV pairs */
1373 		BUG_ON(pagelist->length <= sizeof(__le32));
1374 		if (list_is_singular(&pagelist->head)) {
1375 			le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1376 		} else {
1377 			struct page *page = list_first_entry(&pagelist->head,
1378 							     struct page, lru);
1379 			void *addr = kmap_atomic(page);
1380 			le32_add_cpu((__le32*)addr, 1);
1381 			kunmap_atomic(addr);
1382 		}
1383 	} else {
1384 		as_ctx->pagelist = pagelist;
1385 	}
1386 
1387 	ceph_pagelist_encode_32(pagelist, name_len);
1388 	ceph_pagelist_append(pagelist, name, name_len);
1389 
1390 	ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1391 	ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1392 
1393 	err = 0;
1394 out:
1395 	if (pagelist && !as_ctx->pagelist)
1396 		ceph_pagelist_release(pagelist);
1397 	return err;
1398 }
1399 #endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1400 #endif /* CONFIG_SECURITY */
1401 
1402 void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1403 {
1404 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1405 	posix_acl_release(as_ctx->acl);
1406 	posix_acl_release(as_ctx->default_acl);
1407 #endif
1408 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1409 	security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1410 #endif
1411 	if (as_ctx->pagelist)
1412 		ceph_pagelist_release(as_ctx->pagelist);
1413 }
1414 
1415 /*
1416  * List of handlers for synthetic system.* attributes. Other
1417  * attributes are handled directly.
1418  */
1419 const struct xattr_handler *ceph_xattr_handlers[] = {
1420 	&ceph_other_xattr_handler,
1421 	NULL,
1422 };
1423