xattr.c

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