Loading...
1#include <linux/ceph/ceph_debug.h>
2
3#include "super.h"
4#include "mds_client.h"
5
6#include <linux/ceph/decode.h>
7
8#include <linux/xattr.h>
9#include <linux/posix_acl_xattr.h>
10#include <linux/slab.h>
11
12#define XATTR_CEPH_PREFIX "ceph."
13#define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
14
15static int __remove_xattr(struct ceph_inode_info *ci,
16 struct ceph_inode_xattr *xattr);
17
18/*
19 * List of handlers for synthetic system.* attributes. Other
20 * attributes are handled directly.
21 */
22const struct xattr_handler *ceph_xattr_handlers[] = {
23#ifdef CONFIG_CEPH_FS_POSIX_ACL
24 &posix_acl_access_xattr_handler,
25 &posix_acl_default_xattr_handler,
26#endif
27 NULL,
28};
29
30static bool ceph_is_valid_xattr(const char *name)
31{
32 return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
33 !strncmp(name, XATTR_SECURITY_PREFIX,
34 XATTR_SECURITY_PREFIX_LEN) ||
35 !strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) ||
36 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
37 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
38}
39
40/*
41 * These define virtual xattrs exposing the recursive directory
42 * statistics and layout metadata.
43 */
44struct ceph_vxattr {
45 char *name;
46 size_t name_size; /* strlen(name) + 1 (for '\0') */
47 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
48 size_t size);
49 bool readonly, hidden;
50 bool (*exists_cb)(struct ceph_inode_info *ci);
51};
52
53/* layouts */
54
55static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
56{
57 size_t s;
58 char *p = (char *)&ci->i_layout;
59
60 for (s = 0; s < sizeof(ci->i_layout); s++, p++)
61 if (*p)
62 return true;
63 return false;
64}
65
66static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
67 size_t size)
68{
69 int ret;
70 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
71 struct ceph_osd_client *osdc = &fsc->client->osdc;
72 s64 pool = ceph_file_layout_pg_pool(ci->i_layout);
73 const char *pool_name;
74 char buf[128];
75
76 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
77 down_read(&osdc->map_sem);
78 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
79 if (pool_name) {
80 size_t len = strlen(pool_name);
81 ret = snprintf(buf, sizeof(buf),
82 "stripe_unit=%lld stripe_count=%lld object_size=%lld pool=",
83 (unsigned long long)ceph_file_layout_su(ci->i_layout),
84 (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
85 (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
86 if (!size) {
87 ret += len;
88 } else if (ret + len > size) {
89 ret = -ERANGE;
90 } else {
91 memcpy(val, buf, ret);
92 memcpy(val + ret, pool_name, len);
93 ret += len;
94 }
95 } else {
96 ret = snprintf(buf, sizeof(buf),
97 "stripe_unit=%lld stripe_count=%lld object_size=%lld pool=%lld",
98 (unsigned long long)ceph_file_layout_su(ci->i_layout),
99 (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout),
100 (unsigned long long)ceph_file_layout_object_size(ci->i_layout),
101 (unsigned long long)pool);
102 if (size) {
103 if (ret <= size)
104 memcpy(val, buf, ret);
105 else
106 ret = -ERANGE;
107 }
108 }
109 up_read(&osdc->map_sem);
110 return ret;
111}
112
113static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
114 char *val, size_t size)
115{
116 return snprintf(val, size, "%lld",
117 (unsigned long long)ceph_file_layout_su(ci->i_layout));
118}
119
120static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
121 char *val, size_t size)
122{
123 return snprintf(val, size, "%lld",
124 (unsigned long long)ceph_file_layout_stripe_count(ci->i_layout));
125}
126
127static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
128 char *val, size_t size)
129{
130 return snprintf(val, size, "%lld",
131 (unsigned long long)ceph_file_layout_object_size(ci->i_layout));
132}
133
134static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
135 char *val, size_t size)
136{
137 int ret;
138 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
139 struct ceph_osd_client *osdc = &fsc->client->osdc;
140 s64 pool = ceph_file_layout_pg_pool(ci->i_layout);
141 const char *pool_name;
142
143 down_read(&osdc->map_sem);
144 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
145 if (pool_name)
146 ret = snprintf(val, size, "%s", pool_name);
147 else
148 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
149 up_read(&osdc->map_sem);
150 return ret;
151}
152
153/* directories */
154
155static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
156 size_t size)
157{
158 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
159}
160
161static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
162 size_t size)
163{
164 return snprintf(val, size, "%lld", ci->i_files);
165}
166
167static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
168 size_t size)
169{
170 return snprintf(val, size, "%lld", ci->i_subdirs);
171}
172
173static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
174 size_t size)
175{
176 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
177}
178
179static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
180 size_t size)
181{
182 return snprintf(val, size, "%lld", ci->i_rfiles);
183}
184
185static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
186 size_t size)
187{
188 return snprintf(val, size, "%lld", ci->i_rsubdirs);
189}
190
191static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
192 size_t size)
193{
194 return snprintf(val, size, "%lld", ci->i_rbytes);
195}
196
197static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
198 size_t size)
199{
200 return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
201 (long)ci->i_rctime.tv_nsec);
202}
203
204
205#define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
206#define CEPH_XATTR_NAME2(_type, _name, _name2) \
207 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
208
209#define XATTR_NAME_CEPH(_type, _name) \
210 { \
211 .name = CEPH_XATTR_NAME(_type, _name), \
212 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
213 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
214 .readonly = true, \
215 .hidden = false, \
216 .exists_cb = NULL, \
217 }
218#define XATTR_LAYOUT_FIELD(_type, _name, _field) \
219 { \
220 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
221 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
222 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
223 .readonly = false, \
224 .hidden = true, \
225 .exists_cb = ceph_vxattrcb_layout_exists, \
226 }
227
228static struct ceph_vxattr ceph_dir_vxattrs[] = {
229 {
230 .name = "ceph.dir.layout",
231 .name_size = sizeof("ceph.dir.layout"),
232 .getxattr_cb = ceph_vxattrcb_layout,
233 .readonly = false,
234 .hidden = true,
235 .exists_cb = ceph_vxattrcb_layout_exists,
236 },
237 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
238 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
239 XATTR_LAYOUT_FIELD(dir, layout, object_size),
240 XATTR_LAYOUT_FIELD(dir, layout, pool),
241 XATTR_NAME_CEPH(dir, entries),
242 XATTR_NAME_CEPH(dir, files),
243 XATTR_NAME_CEPH(dir, subdirs),
244 XATTR_NAME_CEPH(dir, rentries),
245 XATTR_NAME_CEPH(dir, rfiles),
246 XATTR_NAME_CEPH(dir, rsubdirs),
247 XATTR_NAME_CEPH(dir, rbytes),
248 XATTR_NAME_CEPH(dir, rctime),
249 { .name = NULL, 0 } /* Required table terminator */
250};
251static size_t ceph_dir_vxattrs_name_size; /* total size of all names */
252
253/* files */
254
255static struct ceph_vxattr ceph_file_vxattrs[] = {
256 {
257 .name = "ceph.file.layout",
258 .name_size = sizeof("ceph.file.layout"),
259 .getxattr_cb = ceph_vxattrcb_layout,
260 .readonly = false,
261 .hidden = true,
262 .exists_cb = ceph_vxattrcb_layout_exists,
263 },
264 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
265 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
266 XATTR_LAYOUT_FIELD(file, layout, object_size),
267 XATTR_LAYOUT_FIELD(file, layout, pool),
268 { .name = NULL, 0 } /* Required table terminator */
269};
270static size_t ceph_file_vxattrs_name_size; /* total size of all names */
271
272static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
273{
274 if (S_ISDIR(inode->i_mode))
275 return ceph_dir_vxattrs;
276 else if (S_ISREG(inode->i_mode))
277 return ceph_file_vxattrs;
278 return NULL;
279}
280
281static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
282{
283 if (vxattrs == ceph_dir_vxattrs)
284 return ceph_dir_vxattrs_name_size;
285 if (vxattrs == ceph_file_vxattrs)
286 return ceph_file_vxattrs_name_size;
287 BUG();
288
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 */
296static 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
310void __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
316void ceph_xattr_exit(void)
317{
318 ceph_dir_vxattrs_name_size = 0;
319 ceph_file_vxattrs_name_size = 0;
320}
321
322static 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
338static 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 if (xattr && (flags & XATTR_CREATE))
373 err = -EEXIST;
374 else if (!xattr && (flags & XATTR_REPLACE))
375 err = -ENODATA;
376 if (err) {
377 kfree(name);
378 kfree(val);
379 return err;
380 }
381 if (update_xattr < 0) {
382 if (xattr)
383 __remove_xattr(ci, xattr);
384 kfree(name);
385 return 0;
386 }
387 }
388
389 if (!xattr) {
390 new = 1;
391 xattr = *newxattr;
392 xattr->name = name;
393 xattr->name_len = name_len;
394 xattr->should_free_name = update_xattr;
395
396 ci->i_xattrs.count++;
397 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
398 } else {
399 kfree(*newxattr);
400 *newxattr = NULL;
401 if (xattr->should_free_val)
402 kfree((void *)xattr->val);
403
404 if (update_xattr) {
405 kfree((void *)name);
406 name = xattr->name;
407 }
408 ci->i_xattrs.names_size -= xattr->name_len;
409 ci->i_xattrs.vals_size -= xattr->val_len;
410 }
411 ci->i_xattrs.names_size += name_len;
412 ci->i_xattrs.vals_size += val_len;
413 if (val)
414 xattr->val = val;
415 else
416 xattr->val = "";
417
418 xattr->val_len = val_len;
419 xattr->dirty = update_xattr;
420 xattr->should_free_val = (val && update_xattr);
421
422 if (new) {
423 rb_link_node(&xattr->node, parent, p);
424 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
425 dout("__set_xattr_val p=%p\n", p);
426 }
427
428 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
429 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
430
431 return 0;
432}
433
434static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
435 const char *name)
436{
437 struct rb_node **p;
438 struct rb_node *parent = NULL;
439 struct ceph_inode_xattr *xattr = NULL;
440 int name_len = strlen(name);
441 int c;
442
443 p = &ci->i_xattrs.index.rb_node;
444 while (*p) {
445 parent = *p;
446 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
447 c = strncmp(name, xattr->name, xattr->name_len);
448 if (c == 0 && name_len > xattr->name_len)
449 c = 1;
450 if (c < 0)
451 p = &(*p)->rb_left;
452 else if (c > 0)
453 p = &(*p)->rb_right;
454 else {
455 dout("__get_xattr %s: found %.*s\n", name,
456 xattr->val_len, xattr->val);
457 return xattr;
458 }
459 }
460
461 dout("__get_xattr %s: not found\n", name);
462
463 return NULL;
464}
465
466static void __free_xattr(struct ceph_inode_xattr *xattr)
467{
468 BUG_ON(!xattr);
469
470 if (xattr->should_free_name)
471 kfree((void *)xattr->name);
472 if (xattr->should_free_val)
473 kfree((void *)xattr->val);
474
475 kfree(xattr);
476}
477
478static int __remove_xattr(struct ceph_inode_info *ci,
479 struct ceph_inode_xattr *xattr)
480{
481 if (!xattr)
482 return -ENODATA;
483
484 rb_erase(&xattr->node, &ci->i_xattrs.index);
485
486 if (xattr->should_free_name)
487 kfree((void *)xattr->name);
488 if (xattr->should_free_val)
489 kfree((void *)xattr->val);
490
491 ci->i_xattrs.names_size -= xattr->name_len;
492 ci->i_xattrs.vals_size -= xattr->val_len;
493 ci->i_xattrs.count--;
494 kfree(xattr);
495
496 return 0;
497}
498
499static int __remove_xattr_by_name(struct ceph_inode_info *ci,
500 const char *name)
501{
502 struct rb_node **p;
503 struct ceph_inode_xattr *xattr;
504 int err;
505
506 p = &ci->i_xattrs.index.rb_node;
507 xattr = __get_xattr(ci, name);
508 err = __remove_xattr(ci, xattr);
509 return err;
510}
511
512static char *__copy_xattr_names(struct ceph_inode_info *ci,
513 char *dest)
514{
515 struct rb_node *p;
516 struct ceph_inode_xattr *xattr = NULL;
517
518 p = rb_first(&ci->i_xattrs.index);
519 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
520
521 while (p) {
522 xattr = rb_entry(p, struct ceph_inode_xattr, node);
523 memcpy(dest, xattr->name, xattr->name_len);
524 dest[xattr->name_len] = '\0';
525
526 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
527 xattr->name_len, ci->i_xattrs.names_size);
528
529 dest += xattr->name_len + 1;
530 p = rb_next(p);
531 }
532
533 return dest;
534}
535
536void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
537{
538 struct rb_node *p, *tmp;
539 struct ceph_inode_xattr *xattr = NULL;
540
541 p = rb_first(&ci->i_xattrs.index);
542
543 dout("__ceph_destroy_xattrs p=%p\n", p);
544
545 while (p) {
546 xattr = rb_entry(p, struct ceph_inode_xattr, node);
547 tmp = p;
548 p = rb_next(tmp);
549 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
550 xattr->name_len, xattr->name);
551 rb_erase(tmp, &ci->i_xattrs.index);
552
553 __free_xattr(xattr);
554 }
555
556 ci->i_xattrs.names_size = 0;
557 ci->i_xattrs.vals_size = 0;
558 ci->i_xattrs.index_version = 0;
559 ci->i_xattrs.count = 0;
560 ci->i_xattrs.index = RB_ROOT;
561}
562
563static int __build_xattrs(struct inode *inode)
564 __releases(ci->i_ceph_lock)
565 __acquires(ci->i_ceph_lock)
566{
567 u32 namelen;
568 u32 numattr = 0;
569 void *p, *end;
570 u32 len;
571 const char *name, *val;
572 struct ceph_inode_info *ci = ceph_inode(inode);
573 int xattr_version;
574 struct ceph_inode_xattr **xattrs = NULL;
575 int err = 0;
576 int i;
577
578 dout("__build_xattrs() len=%d\n",
579 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
580
581 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
582 return 0; /* already built */
583
584 __ceph_destroy_xattrs(ci);
585
586start:
587 /* updated internal xattr rb tree */
588 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
589 p = ci->i_xattrs.blob->vec.iov_base;
590 end = p + ci->i_xattrs.blob->vec.iov_len;
591 ceph_decode_32_safe(&p, end, numattr, bad);
592 xattr_version = ci->i_xattrs.version;
593 spin_unlock(&ci->i_ceph_lock);
594
595 xattrs = kcalloc(numattr, sizeof(struct ceph_xattr *),
596 GFP_NOFS);
597 err = -ENOMEM;
598 if (!xattrs)
599 goto bad_lock;
600 memset(xattrs, 0, numattr*sizeof(struct ceph_xattr *));
601 for (i = 0; i < numattr; i++) {
602 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
603 GFP_NOFS);
604 if (!xattrs[i])
605 goto bad_lock;
606 }
607
608 spin_lock(&ci->i_ceph_lock);
609 if (ci->i_xattrs.version != xattr_version) {
610 /* lost a race, retry */
611 for (i = 0; i < numattr; i++)
612 kfree(xattrs[i]);
613 kfree(xattrs);
614 xattrs = NULL;
615 goto start;
616 }
617 err = -EIO;
618 while (numattr--) {
619 ceph_decode_32_safe(&p, end, len, bad);
620 namelen = len;
621 name = p;
622 p += len;
623 ceph_decode_32_safe(&p, end, len, bad);
624 val = p;
625 p += len;
626
627 err = __set_xattr(ci, name, namelen, val, len,
628 0, 0, &xattrs[numattr]);
629
630 if (err < 0)
631 goto bad;
632 }
633 kfree(xattrs);
634 }
635 ci->i_xattrs.index_version = ci->i_xattrs.version;
636 ci->i_xattrs.dirty = false;
637
638 return err;
639bad_lock:
640 spin_lock(&ci->i_ceph_lock);
641bad:
642 if (xattrs) {
643 for (i = 0; i < numattr; i++)
644 kfree(xattrs[i]);
645 kfree(xattrs);
646 }
647 ci->i_xattrs.names_size = 0;
648 return err;
649}
650
651static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
652 int val_size)
653{
654 /*
655 * 4 bytes for the length, and additional 4 bytes per each xattr name,
656 * 4 bytes per each value
657 */
658 int size = 4 + ci->i_xattrs.count*(4 + 4) +
659 ci->i_xattrs.names_size +
660 ci->i_xattrs.vals_size;
661 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
662 ci->i_xattrs.count, ci->i_xattrs.names_size,
663 ci->i_xattrs.vals_size);
664
665 if (name_size)
666 size += 4 + 4 + name_size + val_size;
667
668 return size;
669}
670
671/*
672 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
673 * and swap into place.
674 */
675void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
676{
677 struct rb_node *p;
678 struct ceph_inode_xattr *xattr = NULL;
679 void *dest;
680
681 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
682 if (ci->i_xattrs.dirty) {
683 int need = __get_required_blob_size(ci, 0, 0);
684
685 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
686
687 p = rb_first(&ci->i_xattrs.index);
688 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
689
690 ceph_encode_32(&dest, ci->i_xattrs.count);
691 while (p) {
692 xattr = rb_entry(p, struct ceph_inode_xattr, node);
693
694 ceph_encode_32(&dest, xattr->name_len);
695 memcpy(dest, xattr->name, xattr->name_len);
696 dest += xattr->name_len;
697 ceph_encode_32(&dest, xattr->val_len);
698 memcpy(dest, xattr->val, xattr->val_len);
699 dest += xattr->val_len;
700
701 p = rb_next(p);
702 }
703
704 /* adjust buffer len; it may be larger than we need */
705 ci->i_xattrs.prealloc_blob->vec.iov_len =
706 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
707
708 if (ci->i_xattrs.blob)
709 ceph_buffer_put(ci->i_xattrs.blob);
710 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
711 ci->i_xattrs.prealloc_blob = NULL;
712 ci->i_xattrs.dirty = false;
713 ci->i_xattrs.version++;
714 }
715}
716
717ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
718 size_t size)
719{
720 struct ceph_inode_info *ci = ceph_inode(inode);
721 int err;
722 struct ceph_inode_xattr *xattr;
723 struct ceph_vxattr *vxattr = NULL;
724
725 if (!ceph_is_valid_xattr(name))
726 return -ENODATA;
727
728 /* let's see if a virtual xattr was requested */
729 vxattr = ceph_match_vxattr(inode, name);
730 if (vxattr && !(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
731 err = vxattr->getxattr_cb(ci, value, size);
732 return err;
733 }
734
735 spin_lock(&ci->i_ceph_lock);
736 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
737 ci->i_xattrs.version, ci->i_xattrs.index_version);
738
739 if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
740 (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
741 goto get_xattr;
742 } else {
743 spin_unlock(&ci->i_ceph_lock);
744 /* get xattrs from mds (if we don't already have them) */
745 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
746 if (err)
747 return err;
748 }
749
750 spin_lock(&ci->i_ceph_lock);
751
752 err = __build_xattrs(inode);
753 if (err < 0)
754 goto out;
755
756get_xattr:
757 err = -ENODATA; /* == ENOATTR */
758 xattr = __get_xattr(ci, name);
759 if (!xattr)
760 goto out;
761
762 err = -ERANGE;
763 if (size && size < xattr->val_len)
764 goto out;
765
766 err = xattr->val_len;
767 if (size == 0)
768 goto out;
769
770 memcpy(value, xattr->val, xattr->val_len);
771
772out:
773 spin_unlock(&ci->i_ceph_lock);
774 return err;
775}
776
777ssize_t ceph_getxattr(struct dentry *dentry, const char *name, void *value,
778 size_t size)
779{
780 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
781 return generic_getxattr(dentry, name, value, size);
782
783 return __ceph_getxattr(dentry->d_inode, name, value, size);
784}
785
786ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
787{
788 struct inode *inode = dentry->d_inode;
789 struct ceph_inode_info *ci = ceph_inode(inode);
790 struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
791 u32 vir_namelen = 0;
792 u32 namelen;
793 int err;
794 u32 len;
795 int i;
796
797 spin_lock(&ci->i_ceph_lock);
798 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
799 ci->i_xattrs.version, ci->i_xattrs.index_version);
800
801 if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1) &&
802 (ci->i_xattrs.index_version >= ci->i_xattrs.version)) {
803 goto list_xattr;
804 } else {
805 spin_unlock(&ci->i_ceph_lock);
806 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR);
807 if (err)
808 return err;
809 }
810
811 spin_lock(&ci->i_ceph_lock);
812
813 err = __build_xattrs(inode);
814 if (err < 0)
815 goto out;
816
817list_xattr:
818 /*
819 * Start with virtual dir xattr names (if any) (including
820 * terminating '\0' characters for each).
821 */
822 vir_namelen = ceph_vxattrs_name_size(vxattrs);
823
824 /* adding 1 byte per each variable due to the null termination */
825 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
826 err = -ERANGE;
827 if (size && vir_namelen + namelen > size)
828 goto out;
829
830 err = namelen + vir_namelen;
831 if (size == 0)
832 goto out;
833
834 names = __copy_xattr_names(ci, names);
835
836 /* virtual xattr names, too */
837 err = namelen;
838 if (vxattrs) {
839 for (i = 0; vxattrs[i].name; i++) {
840 if (!vxattrs[i].hidden &&
841 !(vxattrs[i].exists_cb &&
842 !vxattrs[i].exists_cb(ci))) {
843 len = sprintf(names, "%s", vxattrs[i].name);
844 names += len + 1;
845 err += len + 1;
846 }
847 }
848 }
849
850out:
851 spin_unlock(&ci->i_ceph_lock);
852 return err;
853}
854
855static int ceph_sync_setxattr(struct dentry *dentry, const char *name,
856 const char *value, size_t size, int flags)
857{
858 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
859 struct inode *inode = dentry->d_inode;
860 struct ceph_inode_info *ci = ceph_inode(inode);
861 struct ceph_mds_request *req;
862 struct ceph_mds_client *mdsc = fsc->mdsc;
863 int err;
864 int i, nr_pages;
865 struct page **pages = NULL;
866 void *kaddr;
867
868 /* copy value into some pages */
869 nr_pages = calc_pages_for(0, size);
870 if (nr_pages) {
871 pages = kmalloc(sizeof(pages[0])*nr_pages, GFP_NOFS);
872 if (!pages)
873 return -ENOMEM;
874 err = -ENOMEM;
875 for (i = 0; i < nr_pages; i++) {
876 pages[i] = __page_cache_alloc(GFP_NOFS);
877 if (!pages[i]) {
878 nr_pages = i;
879 goto out;
880 }
881 kaddr = kmap(pages[i]);
882 memcpy(kaddr, value + i*PAGE_CACHE_SIZE,
883 min(PAGE_CACHE_SIZE, size-i*PAGE_CACHE_SIZE));
884 }
885 }
886
887 dout("setxattr value=%.*s\n", (int)size, value);
888
889 if (!value)
890 flags |= CEPH_XATTR_REMOVE;
891
892 /* do request */
893 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETXATTR,
894 USE_AUTH_MDS);
895 if (IS_ERR(req)) {
896 err = PTR_ERR(req);
897 goto out;
898 }
899 req->r_inode = inode;
900 ihold(inode);
901 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
902 req->r_num_caps = 1;
903 req->r_args.setxattr.flags = cpu_to_le32(flags);
904 req->r_path2 = kstrdup(name, GFP_NOFS);
905
906 req->r_pages = pages;
907 req->r_num_pages = nr_pages;
908 req->r_data_len = size;
909
910 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
911 err = ceph_mdsc_do_request(mdsc, NULL, req);
912 ceph_mdsc_put_request(req);
913 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
914
915out:
916 if (pages) {
917 for (i = 0; i < nr_pages; i++)
918 __free_page(pages[i]);
919 kfree(pages);
920 }
921 return err;
922}
923
924int __ceph_setxattr(struct dentry *dentry, const char *name,
925 const void *value, size_t size, int flags)
926{
927 struct inode *inode = dentry->d_inode;
928 struct ceph_vxattr *vxattr;
929 struct ceph_inode_info *ci = ceph_inode(inode);
930 int issued;
931 int err;
932 int dirty = 0;
933 int name_len = strlen(name);
934 int val_len = size;
935 char *newname = NULL;
936 char *newval = NULL;
937 struct ceph_inode_xattr *xattr = NULL;
938 int required_blob_size;
939
940 if (!ceph_is_valid_xattr(name))
941 return -EOPNOTSUPP;
942
943 vxattr = ceph_match_vxattr(inode, name);
944 if (vxattr && vxattr->readonly)
945 return -EOPNOTSUPP;
946
947 /* pass any unhandled ceph.* xattrs through to the MDS */
948 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
949 goto do_sync_unlocked;
950
951 /* preallocate memory for xattr name, value, index node */
952 err = -ENOMEM;
953 newname = kmemdup(name, name_len + 1, GFP_NOFS);
954 if (!newname)
955 goto out;
956
957 if (val_len) {
958 newval = kmemdup(value, val_len, GFP_NOFS);
959 if (!newval)
960 goto out;
961 }
962
963 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
964 if (!xattr)
965 goto out;
966
967 spin_lock(&ci->i_ceph_lock);
968retry:
969 issued = __ceph_caps_issued(ci, NULL);
970 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
971 if (!(issued & CEPH_CAP_XATTR_EXCL))
972 goto do_sync;
973 __build_xattrs(inode);
974
975 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
976
977 if (!ci->i_xattrs.prealloc_blob ||
978 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
979 struct ceph_buffer *blob;
980
981 spin_unlock(&ci->i_ceph_lock);
982 dout(" preaallocating new blob size=%d\n", required_blob_size);
983 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
984 if (!blob)
985 goto out;
986 spin_lock(&ci->i_ceph_lock);
987 if (ci->i_xattrs.prealloc_blob)
988 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
989 ci->i_xattrs.prealloc_blob = blob;
990 goto retry;
991 }
992
993 err = __set_xattr(ci, newname, name_len, newval, val_len,
994 flags, value ? 1 : -1, &xattr);
995
996 if (!err) {
997 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
998 ci->i_xattrs.dirty = true;
999 inode->i_ctime = CURRENT_TIME;
1000 }
1001
1002 spin_unlock(&ci->i_ceph_lock);
1003 if (dirty)
1004 __mark_inode_dirty(inode, dirty);
1005 return err;
1006
1007do_sync:
1008 spin_unlock(&ci->i_ceph_lock);
1009do_sync_unlocked:
1010 err = ceph_sync_setxattr(dentry, name, value, size, flags);
1011out:
1012 kfree(newname);
1013 kfree(newval);
1014 kfree(xattr);
1015 return err;
1016}
1017
1018int ceph_setxattr(struct dentry *dentry, const char *name,
1019 const void *value, size_t size, int flags)
1020{
1021 if (ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
1022 return -EROFS;
1023
1024 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1025 return generic_setxattr(dentry, name, value, size, flags);
1026
1027 return __ceph_setxattr(dentry, name, value, size, flags);
1028}
1029
1030static int ceph_send_removexattr(struct dentry *dentry, const char *name)
1031{
1032 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
1033 struct ceph_mds_client *mdsc = fsc->mdsc;
1034 struct inode *inode = dentry->d_inode;
1035 struct ceph_mds_request *req;
1036 int err;
1037
1038 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RMXATTR,
1039 USE_AUTH_MDS);
1040 if (IS_ERR(req))
1041 return PTR_ERR(req);
1042 req->r_inode = inode;
1043 ihold(inode);
1044 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1045 req->r_num_caps = 1;
1046 req->r_path2 = kstrdup(name, GFP_NOFS);
1047
1048 err = ceph_mdsc_do_request(mdsc, NULL, req);
1049 ceph_mdsc_put_request(req);
1050 return err;
1051}
1052
1053int __ceph_removexattr(struct dentry *dentry, const char *name)
1054{
1055 struct inode *inode = dentry->d_inode;
1056 struct ceph_vxattr *vxattr;
1057 struct ceph_inode_info *ci = ceph_inode(inode);
1058 int issued;
1059 int err;
1060 int required_blob_size;
1061 int dirty;
1062
1063 if (!ceph_is_valid_xattr(name))
1064 return -EOPNOTSUPP;
1065
1066 vxattr = ceph_match_vxattr(inode, name);
1067 if (vxattr && vxattr->readonly)
1068 return -EOPNOTSUPP;
1069
1070 /* pass any unhandled ceph.* xattrs through to the MDS */
1071 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1072 goto do_sync_unlocked;
1073
1074 err = -ENOMEM;
1075 spin_lock(&ci->i_ceph_lock);
1076retry:
1077 issued = __ceph_caps_issued(ci, NULL);
1078 dout("removexattr %p issued %s\n", inode, ceph_cap_string(issued));
1079
1080 if (!(issued & CEPH_CAP_XATTR_EXCL))
1081 goto do_sync;
1082 __build_xattrs(inode);
1083
1084 required_blob_size = __get_required_blob_size(ci, 0, 0);
1085
1086 if (!ci->i_xattrs.prealloc_blob ||
1087 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1088 struct ceph_buffer *blob;
1089
1090 spin_unlock(&ci->i_ceph_lock);
1091 dout(" preaallocating new blob size=%d\n", required_blob_size);
1092 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1093 if (!blob)
1094 goto out;
1095 spin_lock(&ci->i_ceph_lock);
1096 if (ci->i_xattrs.prealloc_blob)
1097 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1098 ci->i_xattrs.prealloc_blob = blob;
1099 goto retry;
1100 }
1101
1102 err = __remove_xattr_by_name(ceph_inode(inode), name);
1103
1104 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL);
1105 ci->i_xattrs.dirty = true;
1106 inode->i_ctime = CURRENT_TIME;
1107 spin_unlock(&ci->i_ceph_lock);
1108 if (dirty)
1109 __mark_inode_dirty(inode, dirty);
1110 return err;
1111do_sync:
1112 spin_unlock(&ci->i_ceph_lock);
1113do_sync_unlocked:
1114 err = ceph_send_removexattr(dentry, name);
1115out:
1116 return err;
1117}
1118
1119int ceph_removexattr(struct dentry *dentry, const char *name)
1120{
1121 if (ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
1122 return -EROFS;
1123
1124 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1125 return generic_removexattr(dentry, name);
1126
1127 return __ceph_removexattr(dentry, name);
1128}
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
46/* layouts */
47
48static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
49{
50 struct ceph_file_layout *fl = &ci->i_layout;
51 return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
52 fl->object_size > 0 || fl->pool_id >= 0 ||
53 rcu_dereference_raw(fl->pool_ns) != NULL);
54}
55
56static ssize_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
57 size_t size)
58{
59 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
60 struct ceph_osd_client *osdc = &fsc->client->osdc;
61 struct ceph_string *pool_ns;
62 s64 pool = ci->i_layout.pool_id;
63 const char *pool_name;
64 const char *ns_field = " pool_namespace=";
65 char buf[128];
66 size_t len, total_len = 0;
67 ssize_t ret;
68
69 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
70
71 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
72 down_read(&osdc->lock);
73 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
74 if (pool_name) {
75 len = snprintf(buf, sizeof(buf),
76 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
77 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
78 ci->i_layout.object_size);
79 total_len = len + strlen(pool_name);
80 } else {
81 len = snprintf(buf, sizeof(buf),
82 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
83 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
84 ci->i_layout.object_size, pool);
85 total_len = len;
86 }
87
88 if (pool_ns)
89 total_len += strlen(ns_field) + pool_ns->len;
90
91 ret = total_len;
92 if (size >= total_len) {
93 memcpy(val, buf, len);
94 ret = len;
95 if (pool_name) {
96 len = strlen(pool_name);
97 memcpy(val + ret, pool_name, len);
98 ret += len;
99 }
100 if (pool_ns) {
101 len = strlen(ns_field);
102 memcpy(val + ret, ns_field, len);
103 ret += len;
104 memcpy(val + ret, pool_ns->str, pool_ns->len);
105 ret += pool_ns->len;
106 }
107 }
108 up_read(&osdc->lock);
109 ceph_put_string(pool_ns);
110 return ret;
111}
112
113/*
114 * The convention with strings in xattrs is that they should not be NULL
115 * terminated, since we're returning the length with them. snprintf always
116 * NULL terminates however, so call it on a temporary buffer and then memcpy
117 * the result into place.
118 */
119static int ceph_fmt_xattr(char *val, size_t size, const char *fmt, ...)
120{
121 int ret;
122 va_list args;
123 char buf[96]; /* NB: reevaluate size if new vxattrs are added */
124
125 va_start(args, fmt);
126 ret = vsnprintf(buf, size ? sizeof(buf) : 0, fmt, args);
127 va_end(args);
128
129 /* Sanity check */
130 if (size && ret + 1 > sizeof(buf)) {
131 WARN_ONCE(true, "Returned length too big (%d)", ret);
132 return -E2BIG;
133 }
134
135 if (ret <= size)
136 memcpy(val, buf, ret);
137 return ret;
138}
139
140static ssize_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
141 char *val, size_t size)
142{
143 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_unit);
144}
145
146static ssize_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
147 char *val, size_t size)
148{
149 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_count);
150}
151
152static ssize_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
153 char *val, size_t size)
154{
155 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.object_size);
156}
157
158static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
159 char *val, size_t size)
160{
161 ssize_t ret;
162 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
163 struct ceph_osd_client *osdc = &fsc->client->osdc;
164 s64 pool = ci->i_layout.pool_id;
165 const char *pool_name;
166
167 down_read(&osdc->lock);
168 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
169 if (pool_name) {
170 ret = strlen(pool_name);
171 if (ret <= size)
172 memcpy(val, pool_name, ret);
173 } else {
174 ret = ceph_fmt_xattr(val, size, "%lld", pool);
175 }
176 up_read(&osdc->lock);
177 return ret;
178}
179
180static ssize_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
181 char *val, size_t size)
182{
183 ssize_t ret = 0;
184 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
185
186 if (ns) {
187 ret = ns->len;
188 if (ret <= size)
189 memcpy(val, ns->str, ret);
190 ceph_put_string(ns);
191 }
192 return ret;
193}
194
195/* directories */
196
197static ssize_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
198 size_t size)
199{
200 return ceph_fmt_xattr(val, size, "%lld", ci->i_files + ci->i_subdirs);
201}
202
203static ssize_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
204 size_t size)
205{
206 return ceph_fmt_xattr(val, size, "%lld", ci->i_files);
207}
208
209static ssize_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
210 size_t size)
211{
212 return ceph_fmt_xattr(val, size, "%lld", ci->i_subdirs);
213}
214
215static ssize_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
216 size_t size)
217{
218 return ceph_fmt_xattr(val, size, "%lld",
219 ci->i_rfiles + ci->i_rsubdirs);
220}
221
222static ssize_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
223 size_t size)
224{
225 return ceph_fmt_xattr(val, size, "%lld", ci->i_rfiles);
226}
227
228static ssize_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
229 size_t size)
230{
231 return ceph_fmt_xattr(val, size, "%lld", ci->i_rsubdirs);
232}
233
234static ssize_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
235 size_t size)
236{
237 return ceph_fmt_xattr(val, size, "%lld", ci->i_rbytes);
238}
239
240static ssize_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
241 size_t size)
242{
243 return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_rctime.tv_sec,
244 ci->i_rctime.tv_nsec);
245}
246
247/* dir pin */
248static bool ceph_vxattrcb_dir_pin_exists(struct ceph_inode_info *ci)
249{
250 return ci->i_dir_pin != -ENODATA;
251}
252
253static ssize_t ceph_vxattrcb_dir_pin(struct ceph_inode_info *ci, char *val,
254 size_t size)
255{
256 return ceph_fmt_xattr(val, size, "%d", (int)ci->i_dir_pin);
257}
258
259/* quotas */
260static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
261{
262 bool ret = false;
263 spin_lock(&ci->i_ceph_lock);
264 if ((ci->i_max_files || ci->i_max_bytes) &&
265 ci->i_vino.snap == CEPH_NOSNAP &&
266 ci->i_snap_realm &&
267 ci->i_snap_realm->ino == ci->i_vino.ino)
268 ret = true;
269 spin_unlock(&ci->i_ceph_lock);
270 return ret;
271}
272
273static ssize_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
274 size_t size)
275{
276 return ceph_fmt_xattr(val, size, "max_bytes=%llu max_files=%llu",
277 ci->i_max_bytes, ci->i_max_files);
278}
279
280static ssize_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
281 char *val, size_t size)
282{
283 return ceph_fmt_xattr(val, size, "%llu", ci->i_max_bytes);
284}
285
286static ssize_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
287 char *val, size_t size)
288{
289 return ceph_fmt_xattr(val, size, "%llu", ci->i_max_files);
290}
291
292/* snapshots */
293static bool ceph_vxattrcb_snap_btime_exists(struct ceph_inode_info *ci)
294{
295 return (ci->i_snap_btime.tv_sec != 0 || ci->i_snap_btime.tv_nsec != 0);
296}
297
298static ssize_t ceph_vxattrcb_snap_btime(struct ceph_inode_info *ci, char *val,
299 size_t size)
300{
301 return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_snap_btime.tv_sec,
302 ci->i_snap_btime.tv_nsec);
303}
304
305#define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
306#define CEPH_XATTR_NAME2(_type, _name, _name2) \
307 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
308
309#define XATTR_NAME_CEPH(_type, _name, _flags) \
310 { \
311 .name = CEPH_XATTR_NAME(_type, _name), \
312 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
313 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
314 .exists_cb = NULL, \
315 .flags = (VXATTR_FLAG_READONLY | _flags), \
316 }
317#define XATTR_RSTAT_FIELD(_type, _name) \
318 XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
319#define XATTR_LAYOUT_FIELD(_type, _name, _field) \
320 { \
321 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
322 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
323 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
324 .exists_cb = ceph_vxattrcb_layout_exists, \
325 .flags = VXATTR_FLAG_HIDDEN, \
326 }
327#define XATTR_QUOTA_FIELD(_type, _name) \
328 { \
329 .name = CEPH_XATTR_NAME(_type, _name), \
330 .name_size = sizeof(CEPH_XATTR_NAME(_type, _name)), \
331 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
332 .exists_cb = ceph_vxattrcb_quota_exists, \
333 .flags = VXATTR_FLAG_HIDDEN, \
334 }
335
336static struct ceph_vxattr ceph_dir_vxattrs[] = {
337 {
338 .name = "ceph.dir.layout",
339 .name_size = sizeof("ceph.dir.layout"),
340 .getxattr_cb = ceph_vxattrcb_layout,
341 .exists_cb = ceph_vxattrcb_layout_exists,
342 .flags = VXATTR_FLAG_HIDDEN,
343 },
344 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
345 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
346 XATTR_LAYOUT_FIELD(dir, layout, object_size),
347 XATTR_LAYOUT_FIELD(dir, layout, pool),
348 XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
349 XATTR_NAME_CEPH(dir, entries, 0),
350 XATTR_NAME_CEPH(dir, files, 0),
351 XATTR_NAME_CEPH(dir, subdirs, 0),
352 XATTR_RSTAT_FIELD(dir, rentries),
353 XATTR_RSTAT_FIELD(dir, rfiles),
354 XATTR_RSTAT_FIELD(dir, rsubdirs),
355 XATTR_RSTAT_FIELD(dir, rbytes),
356 XATTR_RSTAT_FIELD(dir, rctime),
357 {
358 .name = "ceph.dir.pin",
359 .name_size = sizeof("ceph.dir.pin"),
360 .getxattr_cb = ceph_vxattrcb_dir_pin,
361 .exists_cb = ceph_vxattrcb_dir_pin_exists,
362 .flags = VXATTR_FLAG_HIDDEN,
363 },
364 {
365 .name = "ceph.quota",
366 .name_size = sizeof("ceph.quota"),
367 .getxattr_cb = ceph_vxattrcb_quota,
368 .exists_cb = ceph_vxattrcb_quota_exists,
369 .flags = VXATTR_FLAG_HIDDEN,
370 },
371 XATTR_QUOTA_FIELD(quota, max_bytes),
372 XATTR_QUOTA_FIELD(quota, max_files),
373 {
374 .name = "ceph.snap.btime",
375 .name_size = sizeof("ceph.snap.btime"),
376 .getxattr_cb = ceph_vxattrcb_snap_btime,
377 .exists_cb = ceph_vxattrcb_snap_btime_exists,
378 .flags = VXATTR_FLAG_READONLY,
379 },
380 { .name = NULL, 0 } /* Required table terminator */
381};
382
383/* files */
384
385static struct ceph_vxattr ceph_file_vxattrs[] = {
386 {
387 .name = "ceph.file.layout",
388 .name_size = sizeof("ceph.file.layout"),
389 .getxattr_cb = ceph_vxattrcb_layout,
390 .exists_cb = ceph_vxattrcb_layout_exists,
391 .flags = VXATTR_FLAG_HIDDEN,
392 },
393 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
394 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
395 XATTR_LAYOUT_FIELD(file, layout, object_size),
396 XATTR_LAYOUT_FIELD(file, layout, pool),
397 XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
398 {
399 .name = "ceph.snap.btime",
400 .name_size = sizeof("ceph.snap.btime"),
401 .getxattr_cb = ceph_vxattrcb_snap_btime,
402 .exists_cb = ceph_vxattrcb_snap_btime_exists,
403 .flags = VXATTR_FLAG_READONLY,
404 },
405 { .name = NULL, 0 } /* Required table terminator */
406};
407
408static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
409{
410 if (S_ISDIR(inode->i_mode))
411 return ceph_dir_vxattrs;
412 else if (S_ISREG(inode->i_mode))
413 return ceph_file_vxattrs;
414 return NULL;
415}
416
417static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
418 const char *name)
419{
420 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
421
422 if (vxattr) {
423 while (vxattr->name) {
424 if (!strcmp(vxattr->name, name))
425 return vxattr;
426 vxattr++;
427 }
428 }
429
430 return NULL;
431}
432
433static int __set_xattr(struct ceph_inode_info *ci,
434 const char *name, int name_len,
435 const char *val, int val_len,
436 int flags, int update_xattr,
437 struct ceph_inode_xattr **newxattr)
438{
439 struct rb_node **p;
440 struct rb_node *parent = NULL;
441 struct ceph_inode_xattr *xattr = NULL;
442 int c;
443 int new = 0;
444
445 p = &ci->i_xattrs.index.rb_node;
446 while (*p) {
447 parent = *p;
448 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
449 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
450 if (c < 0)
451 p = &(*p)->rb_left;
452 else if (c > 0)
453 p = &(*p)->rb_right;
454 else {
455 if (name_len == xattr->name_len)
456 break;
457 else if (name_len < xattr->name_len)
458 p = &(*p)->rb_left;
459 else
460 p = &(*p)->rb_right;
461 }
462 xattr = NULL;
463 }
464
465 if (update_xattr) {
466 int err = 0;
467
468 if (xattr && (flags & XATTR_CREATE))
469 err = -EEXIST;
470 else if (!xattr && (flags & XATTR_REPLACE))
471 err = -ENODATA;
472 if (err) {
473 kfree(name);
474 kfree(val);
475 kfree(*newxattr);
476 return err;
477 }
478 if (update_xattr < 0) {
479 if (xattr)
480 __remove_xattr(ci, xattr);
481 kfree(name);
482 kfree(*newxattr);
483 return 0;
484 }
485 }
486
487 if (!xattr) {
488 new = 1;
489 xattr = *newxattr;
490 xattr->name = name;
491 xattr->name_len = name_len;
492 xattr->should_free_name = update_xattr;
493
494 ci->i_xattrs.count++;
495 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
496 } else {
497 kfree(*newxattr);
498 *newxattr = NULL;
499 if (xattr->should_free_val)
500 kfree((void *)xattr->val);
501
502 if (update_xattr) {
503 kfree((void *)name);
504 name = xattr->name;
505 }
506 ci->i_xattrs.names_size -= xattr->name_len;
507 ci->i_xattrs.vals_size -= xattr->val_len;
508 }
509 ci->i_xattrs.names_size += name_len;
510 ci->i_xattrs.vals_size += val_len;
511 if (val)
512 xattr->val = val;
513 else
514 xattr->val = "";
515
516 xattr->val_len = val_len;
517 xattr->dirty = update_xattr;
518 xattr->should_free_val = (val && update_xattr);
519
520 if (new) {
521 rb_link_node(&xattr->node, parent, p);
522 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
523 dout("__set_xattr_val p=%p\n", p);
524 }
525
526 dout("__set_xattr_val added %llx.%llx xattr %p %.*s=%.*s\n",
527 ceph_vinop(&ci->vfs_inode), xattr, name_len, name, val_len, val);
528
529 return 0;
530}
531
532static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
533 const char *name)
534{
535 struct rb_node **p;
536 struct rb_node *parent = NULL;
537 struct ceph_inode_xattr *xattr = NULL;
538 int name_len = strlen(name);
539 int c;
540
541 p = &ci->i_xattrs.index.rb_node;
542 while (*p) {
543 parent = *p;
544 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
545 c = strncmp(name, xattr->name, xattr->name_len);
546 if (c == 0 && name_len > xattr->name_len)
547 c = 1;
548 if (c < 0)
549 p = &(*p)->rb_left;
550 else if (c > 0)
551 p = &(*p)->rb_right;
552 else {
553 dout("__get_xattr %s: found %.*s\n", name,
554 xattr->val_len, xattr->val);
555 return xattr;
556 }
557 }
558
559 dout("__get_xattr %s: not found\n", name);
560
561 return NULL;
562}
563
564static void __free_xattr(struct ceph_inode_xattr *xattr)
565{
566 BUG_ON(!xattr);
567
568 if (xattr->should_free_name)
569 kfree((void *)xattr->name);
570 if (xattr->should_free_val)
571 kfree((void *)xattr->val);
572
573 kfree(xattr);
574}
575
576static int __remove_xattr(struct ceph_inode_info *ci,
577 struct ceph_inode_xattr *xattr)
578{
579 if (!xattr)
580 return -ENODATA;
581
582 rb_erase(&xattr->node, &ci->i_xattrs.index);
583
584 if (xattr->should_free_name)
585 kfree((void *)xattr->name);
586 if (xattr->should_free_val)
587 kfree((void *)xattr->val);
588
589 ci->i_xattrs.names_size -= xattr->name_len;
590 ci->i_xattrs.vals_size -= xattr->val_len;
591 ci->i_xattrs.count--;
592 kfree(xattr);
593
594 return 0;
595}
596
597static char *__copy_xattr_names(struct ceph_inode_info *ci,
598 char *dest)
599{
600 struct rb_node *p;
601 struct ceph_inode_xattr *xattr = NULL;
602
603 p = rb_first(&ci->i_xattrs.index);
604 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
605
606 while (p) {
607 xattr = rb_entry(p, struct ceph_inode_xattr, node);
608 memcpy(dest, xattr->name, xattr->name_len);
609 dest[xattr->name_len] = '\0';
610
611 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
612 xattr->name_len, ci->i_xattrs.names_size);
613
614 dest += xattr->name_len + 1;
615 p = rb_next(p);
616 }
617
618 return dest;
619}
620
621void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
622{
623 struct rb_node *p, *tmp;
624 struct ceph_inode_xattr *xattr = NULL;
625
626 p = rb_first(&ci->i_xattrs.index);
627
628 dout("__ceph_destroy_xattrs p=%p\n", p);
629
630 while (p) {
631 xattr = rb_entry(p, struct ceph_inode_xattr, node);
632 tmp = p;
633 p = rb_next(tmp);
634 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
635 xattr->name_len, xattr->name);
636 rb_erase(tmp, &ci->i_xattrs.index);
637
638 __free_xattr(xattr);
639 }
640
641 ci->i_xattrs.names_size = 0;
642 ci->i_xattrs.vals_size = 0;
643 ci->i_xattrs.index_version = 0;
644 ci->i_xattrs.count = 0;
645 ci->i_xattrs.index = RB_ROOT;
646}
647
648static int __build_xattrs(struct inode *inode)
649 __releases(ci->i_ceph_lock)
650 __acquires(ci->i_ceph_lock)
651{
652 u32 namelen;
653 u32 numattr = 0;
654 void *p, *end;
655 u32 len;
656 const char *name, *val;
657 struct ceph_inode_info *ci = ceph_inode(inode);
658 int xattr_version;
659 struct ceph_inode_xattr **xattrs = NULL;
660 int err = 0;
661 int i;
662
663 dout("__build_xattrs() len=%d\n",
664 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
665
666 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
667 return 0; /* already built */
668
669 __ceph_destroy_xattrs(ci);
670
671start:
672 /* updated internal xattr rb tree */
673 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
674 p = ci->i_xattrs.blob->vec.iov_base;
675 end = p + ci->i_xattrs.blob->vec.iov_len;
676 ceph_decode_32_safe(&p, end, numattr, bad);
677 xattr_version = ci->i_xattrs.version;
678 spin_unlock(&ci->i_ceph_lock);
679
680 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
681 GFP_NOFS);
682 err = -ENOMEM;
683 if (!xattrs)
684 goto bad_lock;
685
686 for (i = 0; i < numattr; i++) {
687 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
688 GFP_NOFS);
689 if (!xattrs[i])
690 goto bad_lock;
691 }
692
693 spin_lock(&ci->i_ceph_lock);
694 if (ci->i_xattrs.version != xattr_version) {
695 /* lost a race, retry */
696 for (i = 0; i < numattr; i++)
697 kfree(xattrs[i]);
698 kfree(xattrs);
699 xattrs = NULL;
700 goto start;
701 }
702 err = -EIO;
703 while (numattr--) {
704 ceph_decode_32_safe(&p, end, len, bad);
705 namelen = len;
706 name = p;
707 p += len;
708 ceph_decode_32_safe(&p, end, len, bad);
709 val = p;
710 p += len;
711
712 err = __set_xattr(ci, name, namelen, val, len,
713 0, 0, &xattrs[numattr]);
714
715 if (err < 0)
716 goto bad;
717 }
718 kfree(xattrs);
719 }
720 ci->i_xattrs.index_version = ci->i_xattrs.version;
721 ci->i_xattrs.dirty = false;
722
723 return err;
724bad_lock:
725 spin_lock(&ci->i_ceph_lock);
726bad:
727 if (xattrs) {
728 for (i = 0; i < numattr; i++)
729 kfree(xattrs[i]);
730 kfree(xattrs);
731 }
732 ci->i_xattrs.names_size = 0;
733 return err;
734}
735
736static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
737 int val_size)
738{
739 /*
740 * 4 bytes for the length, and additional 4 bytes per each xattr name,
741 * 4 bytes per each value
742 */
743 int size = 4 + ci->i_xattrs.count*(4 + 4) +
744 ci->i_xattrs.names_size +
745 ci->i_xattrs.vals_size;
746 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
747 ci->i_xattrs.count, ci->i_xattrs.names_size,
748 ci->i_xattrs.vals_size);
749
750 if (name_size)
751 size += 4 + 4 + name_size + val_size;
752
753 return size;
754}
755
756/*
757 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
758 * and swap into place. It returns the old i_xattrs.blob (or NULL) so
759 * that it can be freed by the caller as the i_ceph_lock is likely to be
760 * held.
761 */
762struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
763{
764 struct rb_node *p;
765 struct ceph_inode_xattr *xattr = NULL;
766 struct ceph_buffer *old_blob = NULL;
767 void *dest;
768
769 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
770 if (ci->i_xattrs.dirty) {
771 int need = __get_required_blob_size(ci, 0, 0);
772
773 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
774
775 p = rb_first(&ci->i_xattrs.index);
776 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
777
778 ceph_encode_32(&dest, ci->i_xattrs.count);
779 while (p) {
780 xattr = rb_entry(p, struct ceph_inode_xattr, node);
781
782 ceph_encode_32(&dest, xattr->name_len);
783 memcpy(dest, xattr->name, xattr->name_len);
784 dest += xattr->name_len;
785 ceph_encode_32(&dest, xattr->val_len);
786 memcpy(dest, xattr->val, xattr->val_len);
787 dest += xattr->val_len;
788
789 p = rb_next(p);
790 }
791
792 /* adjust buffer len; it may be larger than we need */
793 ci->i_xattrs.prealloc_blob->vec.iov_len =
794 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
795
796 if (ci->i_xattrs.blob)
797 old_blob = ci->i_xattrs.blob;
798 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
799 ci->i_xattrs.prealloc_blob = NULL;
800 ci->i_xattrs.dirty = false;
801 ci->i_xattrs.version++;
802 }
803
804 return old_blob;
805}
806
807static inline int __get_request_mask(struct inode *in) {
808 struct ceph_mds_request *req = current->journal_info;
809 int mask = 0;
810 if (req && req->r_target_inode == in) {
811 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
812 req->r_op == CEPH_MDS_OP_LOOKUPINO ||
813 req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
814 req->r_op == CEPH_MDS_OP_GETATTR) {
815 mask = le32_to_cpu(req->r_args.getattr.mask);
816 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
817 req->r_op == CEPH_MDS_OP_CREATE) {
818 mask = le32_to_cpu(req->r_args.open.mask);
819 }
820 }
821 return mask;
822}
823
824ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
825 size_t size)
826{
827 struct ceph_inode_info *ci = ceph_inode(inode);
828 struct ceph_inode_xattr *xattr;
829 struct ceph_vxattr *vxattr = NULL;
830 int req_mask;
831 ssize_t err;
832
833 /* let's see if a virtual xattr was requested */
834 vxattr = ceph_match_vxattr(inode, name);
835 if (vxattr) {
836 int mask = 0;
837 if (vxattr->flags & VXATTR_FLAG_RSTAT)
838 mask |= CEPH_STAT_RSTAT;
839 err = ceph_do_getattr(inode, mask, true);
840 if (err)
841 return err;
842 err = -ENODATA;
843 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
844 err = vxattr->getxattr_cb(ci, value, size);
845 if (size && size < err)
846 err = -ERANGE;
847 }
848 return err;
849 }
850
851 req_mask = __get_request_mask(inode);
852
853 spin_lock(&ci->i_ceph_lock);
854 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
855 ci->i_xattrs.version, ci->i_xattrs.index_version);
856
857 if (ci->i_xattrs.version == 0 ||
858 !((req_mask & CEPH_CAP_XATTR_SHARED) ||
859 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
860 spin_unlock(&ci->i_ceph_lock);
861
862 /* security module gets xattr while filling trace */
863 if (current->journal_info) {
864 pr_warn_ratelimited("sync getxattr %p "
865 "during filling trace\n", inode);
866 return -EBUSY;
867 }
868
869 /* get xattrs from mds (if we don't already have them) */
870 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
871 if (err)
872 return err;
873 spin_lock(&ci->i_ceph_lock);
874 }
875
876 err = __build_xattrs(inode);
877 if (err < 0)
878 goto out;
879
880 err = -ENODATA; /* == ENOATTR */
881 xattr = __get_xattr(ci, name);
882 if (!xattr)
883 goto out;
884
885 err = -ERANGE;
886 if (size && size < xattr->val_len)
887 goto out;
888
889 err = xattr->val_len;
890 if (size == 0)
891 goto out;
892
893 memcpy(value, xattr->val, xattr->val_len);
894
895 if (current->journal_info &&
896 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
897 security_ismaclabel(name + XATTR_SECURITY_PREFIX_LEN))
898 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
899out:
900 spin_unlock(&ci->i_ceph_lock);
901 return err;
902}
903
904ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
905{
906 struct inode *inode = d_inode(dentry);
907 struct ceph_inode_info *ci = ceph_inode(inode);
908 bool len_only = (size == 0);
909 u32 namelen;
910 int err;
911
912 spin_lock(&ci->i_ceph_lock);
913 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
914 ci->i_xattrs.version, ci->i_xattrs.index_version);
915
916 if (ci->i_xattrs.version == 0 ||
917 !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
918 spin_unlock(&ci->i_ceph_lock);
919 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
920 if (err)
921 return err;
922 spin_lock(&ci->i_ceph_lock);
923 }
924
925 err = __build_xattrs(inode);
926 if (err < 0)
927 goto out;
928
929 /* add 1 byte for each xattr due to the null termination */
930 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
931 if (!len_only) {
932 if (namelen > size) {
933 err = -ERANGE;
934 goto out;
935 }
936 names = __copy_xattr_names(ci, names);
937 size -= namelen;
938 }
939 err = namelen;
940out:
941 spin_unlock(&ci->i_ceph_lock);
942 return err;
943}
944
945static int ceph_sync_setxattr(struct inode *inode, const char *name,
946 const char *value, size_t size, int flags)
947{
948 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
949 struct ceph_inode_info *ci = ceph_inode(inode);
950 struct ceph_mds_request *req;
951 struct ceph_mds_client *mdsc = fsc->mdsc;
952 struct ceph_pagelist *pagelist = NULL;
953 int op = CEPH_MDS_OP_SETXATTR;
954 int err;
955
956 if (size > 0) {
957 /* copy value into pagelist */
958 pagelist = ceph_pagelist_alloc(GFP_NOFS);
959 if (!pagelist)
960 return -ENOMEM;
961
962 err = ceph_pagelist_append(pagelist, value, size);
963 if (err)
964 goto out;
965 } else if (!value) {
966 if (flags & CEPH_XATTR_REPLACE)
967 op = CEPH_MDS_OP_RMXATTR;
968 else
969 flags |= CEPH_XATTR_REMOVE;
970 }
971
972 dout("setxattr value=%.*s\n", (int)size, value);
973
974 /* do request */
975 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
976 if (IS_ERR(req)) {
977 err = PTR_ERR(req);
978 goto out;
979 }
980
981 req->r_path2 = kstrdup(name, GFP_NOFS);
982 if (!req->r_path2) {
983 ceph_mdsc_put_request(req);
984 err = -ENOMEM;
985 goto out;
986 }
987
988 if (op == CEPH_MDS_OP_SETXATTR) {
989 req->r_args.setxattr.flags = cpu_to_le32(flags);
990 req->r_pagelist = pagelist;
991 pagelist = NULL;
992 }
993
994 req->r_inode = inode;
995 ihold(inode);
996 req->r_num_caps = 1;
997 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
998
999 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
1000 err = ceph_mdsc_do_request(mdsc, NULL, req);
1001 ceph_mdsc_put_request(req);
1002 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
1003
1004out:
1005 if (pagelist)
1006 ceph_pagelist_release(pagelist);
1007 return err;
1008}
1009
1010int __ceph_setxattr(struct inode *inode, const char *name,
1011 const void *value, size_t size, int flags)
1012{
1013 struct ceph_vxattr *vxattr;
1014 struct ceph_inode_info *ci = ceph_inode(inode);
1015 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1016 struct ceph_cap_flush *prealloc_cf = NULL;
1017 struct ceph_buffer *old_blob = NULL;
1018 int issued;
1019 int err;
1020 int dirty = 0;
1021 int name_len = strlen(name);
1022 int val_len = size;
1023 char *newname = NULL;
1024 char *newval = NULL;
1025 struct ceph_inode_xattr *xattr = NULL;
1026 int required_blob_size;
1027 bool check_realm = false;
1028 bool lock_snap_rwsem = false;
1029
1030 if (ceph_snap(inode) != CEPH_NOSNAP)
1031 return -EROFS;
1032
1033 vxattr = ceph_match_vxattr(inode, name);
1034 if (vxattr) {
1035 if (vxattr->flags & VXATTR_FLAG_READONLY)
1036 return -EOPNOTSUPP;
1037 if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1038 check_realm = true;
1039 }
1040
1041 /* pass any unhandled ceph.* xattrs through to the MDS */
1042 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1043 goto do_sync_unlocked;
1044
1045 /* preallocate memory for xattr name, value, index node */
1046 err = -ENOMEM;
1047 newname = kmemdup(name, name_len + 1, GFP_NOFS);
1048 if (!newname)
1049 goto out;
1050
1051 if (val_len) {
1052 newval = kmemdup(value, val_len, GFP_NOFS);
1053 if (!newval)
1054 goto out;
1055 }
1056
1057 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1058 if (!xattr)
1059 goto out;
1060
1061 prealloc_cf = ceph_alloc_cap_flush();
1062 if (!prealloc_cf)
1063 goto out;
1064
1065 spin_lock(&ci->i_ceph_lock);
1066retry:
1067 issued = __ceph_caps_issued(ci, NULL);
1068 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1069 goto do_sync;
1070
1071 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1072 lock_snap_rwsem = true;
1073 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1074 spin_unlock(&ci->i_ceph_lock);
1075 down_read(&mdsc->snap_rwsem);
1076 spin_lock(&ci->i_ceph_lock);
1077 goto retry;
1078 }
1079 }
1080
1081 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1082 __build_xattrs(inode);
1083
1084 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1085
1086 if (!ci->i_xattrs.prealloc_blob ||
1087 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1088 struct ceph_buffer *blob;
1089
1090 spin_unlock(&ci->i_ceph_lock);
1091 ceph_buffer_put(old_blob); /* Shouldn't be required */
1092 dout(" pre-allocating new blob size=%d\n", required_blob_size);
1093 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1094 if (!blob)
1095 goto do_sync_unlocked;
1096 spin_lock(&ci->i_ceph_lock);
1097 /* prealloc_blob can't be released while holding i_ceph_lock */
1098 if (ci->i_xattrs.prealloc_blob)
1099 old_blob = ci->i_xattrs.prealloc_blob;
1100 ci->i_xattrs.prealloc_blob = blob;
1101 goto retry;
1102 }
1103
1104 err = __set_xattr(ci, newname, name_len, newval, val_len,
1105 flags, value ? 1 : -1, &xattr);
1106
1107 if (!err) {
1108 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1109 &prealloc_cf);
1110 ci->i_xattrs.dirty = true;
1111 inode->i_ctime = current_time(inode);
1112 }
1113
1114 spin_unlock(&ci->i_ceph_lock);
1115 ceph_buffer_put(old_blob);
1116 if (lock_snap_rwsem)
1117 up_read(&mdsc->snap_rwsem);
1118 if (dirty)
1119 __mark_inode_dirty(inode, dirty);
1120 ceph_free_cap_flush(prealloc_cf);
1121 return err;
1122
1123do_sync:
1124 spin_unlock(&ci->i_ceph_lock);
1125do_sync_unlocked:
1126 if (lock_snap_rwsem)
1127 up_read(&mdsc->snap_rwsem);
1128
1129 /* security module set xattr while filling trace */
1130 if (current->journal_info) {
1131 pr_warn_ratelimited("sync setxattr %p "
1132 "during filling trace\n", inode);
1133 err = -EBUSY;
1134 } else {
1135 err = ceph_sync_setxattr(inode, name, value, size, flags);
1136 if (err >= 0 && check_realm) {
1137 /* check if snaprealm was created for quota inode */
1138 spin_lock(&ci->i_ceph_lock);
1139 if ((ci->i_max_files || ci->i_max_bytes) &&
1140 !(ci->i_snap_realm &&
1141 ci->i_snap_realm->ino == ci->i_vino.ino))
1142 err = -EOPNOTSUPP;
1143 spin_unlock(&ci->i_ceph_lock);
1144 }
1145 }
1146out:
1147 ceph_free_cap_flush(prealloc_cf);
1148 kfree(newname);
1149 kfree(newval);
1150 kfree(xattr);
1151 return err;
1152}
1153
1154static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1155 struct dentry *dentry, struct inode *inode,
1156 const char *name, void *value, size_t size)
1157{
1158 if (!ceph_is_valid_xattr(name))
1159 return -EOPNOTSUPP;
1160 return __ceph_getxattr(inode, name, value, size);
1161}
1162
1163static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1164 struct dentry *unused, struct inode *inode,
1165 const char *name, const void *value,
1166 size_t size, int flags)
1167{
1168 if (!ceph_is_valid_xattr(name))
1169 return -EOPNOTSUPP;
1170 return __ceph_setxattr(inode, name, value, size, flags);
1171}
1172
1173static const struct xattr_handler ceph_other_xattr_handler = {
1174 .prefix = "", /* match any name => handlers called with full name */
1175 .get = ceph_get_xattr_handler,
1176 .set = ceph_set_xattr_handler,
1177};
1178
1179#ifdef CONFIG_SECURITY
1180bool ceph_security_xattr_wanted(struct inode *in)
1181{
1182 return in->i_security != NULL;
1183}
1184
1185bool ceph_security_xattr_deadlock(struct inode *in)
1186{
1187 struct ceph_inode_info *ci;
1188 bool ret;
1189 if (!in->i_security)
1190 return false;
1191 ci = ceph_inode(in);
1192 spin_lock(&ci->i_ceph_lock);
1193 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1194 !(ci->i_xattrs.version > 0 &&
1195 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1196 spin_unlock(&ci->i_ceph_lock);
1197 return ret;
1198}
1199
1200#ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1201int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1202 struct ceph_acl_sec_ctx *as_ctx)
1203{
1204 struct ceph_pagelist *pagelist = as_ctx->pagelist;
1205 const char *name;
1206 size_t name_len;
1207 int err;
1208
1209 err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1210 &as_ctx->sec_ctx,
1211 &as_ctx->sec_ctxlen);
1212 if (err < 0) {
1213 WARN_ON_ONCE(err != -EOPNOTSUPP);
1214 err = 0; /* do nothing */
1215 goto out;
1216 }
1217
1218 err = -ENOMEM;
1219 if (!pagelist) {
1220 pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1221 if (!pagelist)
1222 goto out;
1223 err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1224 if (err)
1225 goto out;
1226 ceph_pagelist_encode_32(pagelist, 1);
1227 }
1228
1229 /*
1230 * FIXME: Make security_dentry_init_security() generic. Currently
1231 * It only supports single security module and only selinux has
1232 * dentry_init_security hook.
1233 */
1234 name = XATTR_NAME_SELINUX;
1235 name_len = strlen(name);
1236 err = ceph_pagelist_reserve(pagelist,
1237 4 * 2 + name_len + as_ctx->sec_ctxlen);
1238 if (err)
1239 goto out;
1240
1241 if (as_ctx->pagelist) {
1242 /* update count of KV pairs */
1243 BUG_ON(pagelist->length <= sizeof(__le32));
1244 if (list_is_singular(&pagelist->head)) {
1245 le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1246 } else {
1247 struct page *page = list_first_entry(&pagelist->head,
1248 struct page, lru);
1249 void *addr = kmap_atomic(page);
1250 le32_add_cpu((__le32*)addr, 1);
1251 kunmap_atomic(addr);
1252 }
1253 } else {
1254 as_ctx->pagelist = pagelist;
1255 }
1256
1257 ceph_pagelist_encode_32(pagelist, name_len);
1258 ceph_pagelist_append(pagelist, name, name_len);
1259
1260 ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1261 ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1262
1263 err = 0;
1264out:
1265 if (pagelist && !as_ctx->pagelist)
1266 ceph_pagelist_release(pagelist);
1267 return err;
1268}
1269#endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1270#endif /* CONFIG_SECURITY */
1271
1272void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1273{
1274#ifdef CONFIG_CEPH_FS_POSIX_ACL
1275 posix_acl_release(as_ctx->acl);
1276 posix_acl_release(as_ctx->default_acl);
1277#endif
1278#ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1279 security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1280#endif
1281 if (as_ctx->pagelist)
1282 ceph_pagelist_release(as_ctx->pagelist);
1283}
1284
1285/*
1286 * List of handlers for synthetic system.* attributes. Other
1287 * attributes are handled directly.
1288 */
1289const struct xattr_handler *ceph_xattr_handlers[] = {
1290#ifdef CONFIG_CEPH_FS_POSIX_ACL
1291 &posix_acl_access_xattr_handler,
1292 &posix_acl_default_xattr_handler,
1293#endif
1294 &ceph_other_xattr_handler,
1295 NULL,
1296};