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